Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1161269AbbBDP1l (ORCPT ); Wed, 4 Feb 2015 10:27:41 -0500 Received: from mail-bn1on0114.outbound.protection.outlook.com ([157.56.110.114]:46688 "EHLO na01-bn1-obe.outbound.protection.outlook.com" rhost-flags-OK-OK-OK-FAIL) by vger.kernel.org with ESMTP id S965563AbbBDP1c (ORCPT ); Wed, 4 Feb 2015 10:27:32 -0500 From: Emil Medve To: , , CC: Geoff Thorpe Subject: [RFC 02/10] fsl_qman: Add drivers for the Freescale DPAA QMan Date: Wed, 4 Feb 2015 08:48:34 -0600 Message-ID: <1423061322-16059-3-git-send-email-Emilian.Medve@Freescale.com> X-Mailer: git-send-email 2.2.2 In-Reply-To: <1423061322-16059-1-git-send-email-Emilian.Medve@Freescale.com> References: <1423061322-16059-1-git-send-email-Emilian.Medve@Freescale.com> X-EOPAttributedMessage: 0 Authentication-Results: spf=fail (sender IP is 192.88.168.50) smtp.mailfrom=Emilian.Medve@Freescale.com; freescale.mail.onmicrosoft.com; dkim=none (message not signed) header.d=none; X-Forefront-Antispam-Report: CIP:192.88.168.50;CTRY:US;IPV:NLI;EFV:NLI;SFV:NSPM;SFS:(10019020)(6009001)(339900001)(52314003)(51234002)(36756003)(2950100001)(104016003)(62966003)(50226001)(92566002)(229853001)(47776003)(77156002)(105606002)(77096005)(106466001)(87936001)(85426001)(46102003)(19580395003)(19580405001)(6806004)(50466002)(76176999)(575784001)(50986999)(2201001)(86362001)(48376002)(2101003)(559001)(579004);DIR:OUT;SFP:1102;SCL:1;SRVR:BN3PR0301MB1187;H:tx30smr01.am.freescale.net;FPR:;SPF:Fail;MLV:sfv;LANG:en; MIME-Version: 1.0 Content-Type: text/plain X-Microsoft-Antispam: UriScan:; X-Microsoft-Antispam: BCL:0;PCL:0;RULEID:;SRVR:BN3PR0301MB1187; X-Exchange-Antispam-Report-Test: UriScan:; X-Exchange-Antispam-Report-CFA-Test: BCL:0;PCL:0;RULEID:(601004);SRVR:BN3PR0301MB1187; X-Forefront-PRVS: 04772EA191 X-Exchange-Antispam-Report-CFA-Test: BCL:0;PCL:0;RULEID:;SRVR:BN3PR0301MB1187; X-OriginatorOrg: freescale.com X-MS-Exchange-CrossTenant-OriginalArrivalTime: 04 Feb 2015 14:53:05.6817 (UTC) X-MS-Exchange-CrossTenant-Id: 710a03f5-10f6-4d38-9ff4-a80b81da590d X-MS-Exchange-CrossTenant-OriginalAttributedTenantConnectingIp: TenantId=710a03f5-10f6-4d38-9ff4-a80b81da590d;Ip=[192.88.168.50] X-MS-Exchange-CrossTenant-FromEntityHeader: HybridOnPrem X-MS-Exchange-Transport-CrossTenantHeadersStamped: BN3PR0301MB1187 Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 270611 Lines: 8254 From: Geoff Thorpe Change-Id: I075944acf740dbaae861104c17a9ff7247dec1be Signed-off-by: Geoff Thorpe --- drivers/staging/fsl_qbman/Kconfig | 76 +- drivers/staging/fsl_qbman/Makefile | 4 + drivers/staging/fsl_qbman/bman_high.c | 5 +- drivers/staging/fsl_qbman/dpa_alloc.c | 169 ++ drivers/staging/fsl_qbman/dpa_sys.h | 60 +- drivers/staging/fsl_qbman/qbman_driver.c | 44 + drivers/staging/fsl_qbman/qman_config.c | 991 ++++++++++++ drivers/staging/fsl_qbman/qman_driver.c | 503 ++++++ drivers/staging/fsl_qbman/qman_high.c | 2566 ++++++++++++++++++++++++++++++ drivers/staging/fsl_qbman/qman_low.h | 1302 +++++++++++++++ drivers/staging/fsl_qbman/qman_private.h | 275 ++++ drivers/staging/fsl_qbman/qman_utility.c | 129 ++ include/linux/fsl_bman.h | 4 +- include/linux/fsl_qman.h | 1949 +++++++++++++++++++++++ 14 files changed, 8071 insertions(+), 6 deletions(-) create mode 100644 drivers/staging/fsl_qbman/qman_config.c create mode 100644 drivers/staging/fsl_qbman/qman_driver.c create mode 100644 drivers/staging/fsl_qbman/qman_high.c create mode 100644 drivers/staging/fsl_qbman/qman_low.h create mode 100644 drivers/staging/fsl_qbman/qman_private.h create mode 100644 drivers/staging/fsl_qbman/qman_utility.c create mode 100644 include/linux/fsl_qman.h diff --git a/drivers/staging/fsl_qbman/Kconfig b/drivers/staging/fsl_qbman/Kconfig index 9bf4e67..db90fe5 100644 --- a/drivers/staging/fsl_qbman/Kconfig +++ b/drivers/staging/fsl_qbman/Kconfig @@ -1,9 +1,10 @@ config FSL_DPA - bool "Freescale Datapath Buffer management" + bool "Freescale Datapath Queue and Buffer management" depends on HAS_FSL_QBMAN default y + select FSL_QMAN_FQ_LOOKUP if PPC64 -menu "Freescale Datapath BMan options" +menu "Freescale Datapath QMan/BMan options" depends on FSL_DPA config FSL_DPA_CHECKING @@ -49,4 +50,75 @@ config FSL_BMAN_CONFIG endif # FSL_BMAN +config FSL_QMAN + bool "Freescale Queue Manager (QMan) support" + default y + +if FSL_QMAN + +config FSL_QMAN_POLL_LIMIT + int + default 32 + +config FSL_QMAN_CONFIG + bool "QMan device management" + default y + ---help--- + If this linux image is running natively, you need this option. If this + linux image is running as a guest OS under the hypervisor, only one + guest OS ("the control plane") needs this option. + +# H/w settings that can be hard-coded for now. +config FSL_QMAN_FQD_SZ + int "size of Frame Queue Descriptor region" + default 10 + ---help--- + This is the size of the FQD region defined as: PAGE_SIZE * (2^value) + ex: 10 => PAGE_SIZE * (2^10) + Note: Default device-trees now require minimum Kconfig setting of 10. + +# Corenet initiator settings. Stash request queues are 4-deep to match cores' +# ability to snart. Stash priority is 3, other priorities are 2. +config FSL_QMAN_CI_SCHED_CFG_SRCCIV + int + depends on FSL_QMAN_CONFIG + default 4 +config FSL_QMAN_CI_SCHED_CFG_SRQ_W + int + depends on FSL_QMAN_CONFIG + default 3 +config FSL_QMAN_CI_SCHED_CFG_RW_W + int + depends on FSL_QMAN_CONFIG + default 2 +config FSL_QMAN_CI_SCHED_CFG_BMAN_W + int + depends on FSL_QMAN_CONFIG + default 2 + +# portal interrupt settings +config FSL_QMAN_PIRQ_DQRR_ITHRESH + int + default 12 +config FSL_QMAN_PIRQ_MR_ITHRESH + int + default 4 +config FSL_QMAN_PIRQ_IPERIOD + int + default 100 + +# 64 bit kernel support +config FSL_QMAN_FQ_LOOKUP + bool + default n + +config FSL_QMAN_INIT_TIMEOUT + int "timeout for qman init stage, in seconds" + default 10 + ---help--- + The timeout setting to quit the initialization loop for non-control + partition in case the control partition fails to boot-up. + +endif # FSL_QMAN + endmenu diff --git a/drivers/staging/fsl_qbman/Makefile b/drivers/staging/fsl_qbman/Makefile index d6e3605..399d87e 100644 --- a/drivers/staging/fsl_qbman/Makefile +++ b/drivers/staging/fsl_qbman/Makefile @@ -5,3 +5,7 @@ obj-$(CONFIG_HAS_FSL_QBMAN) += qbman_driver.o # Bman obj-$(CONFIG_FSL_BMAN) += bman_high.o obj-$(CONFIG_FSL_BMAN_CONFIG) += bman_config.o bman_driver.o + +# Qman +obj-$(CONFIG_FSL_QMAN) += qman_high.o qman_utility.o +obj-$(CONFIG_FSL_QMAN_CONFIG) += qman_config.o qman_driver.o diff --git a/drivers/staging/fsl_qbman/bman_high.c b/drivers/staging/fsl_qbman/bman_high.c index 20765a4..3caad72 100644 --- a/drivers/staging/fsl_qbman/bman_high.c +++ b/drivers/staging/fsl_qbman/bman_high.c @@ -65,6 +65,9 @@ struct bman_portal { u8 alloced; }; +/* For an explanation of the locking, redirection, or affine-portal logic, + * please consult the Qman driver for details. This is the same, only simpler + * (no fiddly Qman-specific bits.) */ #ifdef CONFIG_FSL_DPA_PORTAL_SHARE #define PORTAL_IRQ_LOCK(p, irqflags) \ do { \ @@ -170,7 +173,7 @@ static void depletion_unlink(struct bman_pool *pool) PORTAL_IRQ_UNLOCK(pool->portal, irqflags); } -/* In the case that the application's core loop calls +/* In the case that the application's core loop calls qman_poll() and * bman_poll(), we ought to balance how often we incur the overheads of the * slow-path poll. We'll use two decrementer sources. The idle decrementer * constant is used when the last slow-poll detected no work to do, and the busy diff --git a/drivers/staging/fsl_qbman/dpa_alloc.c b/drivers/staging/fsl_qbman/dpa_alloc.c index 8289670..ab36ed3 100644 --- a/drivers/staging/fsl_qbman/dpa_alloc.c +++ b/drivers/staging/fsl_qbman/dpa_alloc.c @@ -30,6 +30,7 @@ */ #include "dpa_sys.h" +#include #include /* This interface is needed in a few places and though it's not specific to @@ -402,3 +403,171 @@ int bman_reserve_bpid_range(u32 bpid, u32 count) } EXPORT_SYMBOL(bman_reserve_bpid_range); #endif /* CONFIG_FSL_BMAN */ + +#ifdef CONFIG_FSL_QMAN +static DECLARE_DPA_ALLOC(fqalloc); /* FQID allocator */ +static DECLARE_DPA_ALLOC(qpalloc); /* pool-channel allocator */ +static DECLARE_DPA_ALLOC(cgralloc); /* CGR ID allocator */ + +/* FQID allocator front-end */ + +int qman_alloc_fqid_range(u32 *result, u32 count, u32 align, int partial) +{ + return dpa_alloc_new(&fqalloc, result, count, align, partial); +} +EXPORT_SYMBOL(qman_alloc_fqid_range); + +static int fq_cleanup(u32 fqid) +{ + return qman_shutdown_fq(fqid) == 0; +} +void qman_release_fqid_range(u32 fqid, u32 count) +{ + u32 total_invalid = release_id_range(&fqalloc, fqid, count, fq_cleanup); + if (total_invalid) + pr_err("FQID range [%d..%d] (%d) had %d leaks\n", + fqid, fqid + count - 1, count, total_invalid); +} +EXPORT_SYMBOL(qman_release_fqid_range); + +int qman_reserve_fqid_range(u32 fqid, u32 count) +{ + return dpa_alloc_reserve(&fqalloc, fqid, count); +} +EXPORT_SYMBOL(qman_reserve_fqid_range); + +void qman_seed_fqid_range(u32 fqid, u32 count) +{ + dpa_alloc_seed(&fqalloc, fqid, count); +} +EXPORT_SYMBOL(qman_seed_fqid_range); + +/* Pool-channel allocator front-end */ + +int qman_alloc_pool_range(u32 *result, u32 count, u32 align, int partial) +{ + return dpa_alloc_new(&qpalloc, result, count, align, partial); +} +EXPORT_SYMBOL(qman_alloc_pool_range); + +static int qpool_cleanup(u32 qp) +{ + /* We query all FQDs starting from + * FQID 1 until we get an "invalid FQID" error, looking for non-OOS FQDs + * whose destination channel is the pool-channel being released. + * When a non-OOS FQD is found we attempt to clean it up */ + struct qman_fq fq = { + .fqid = 1 + }; + int err; + do { + struct qm_mcr_queryfq_np np; + err = qman_query_fq_np(&fq, &np); + if (err) + /* FQID range exceeded, found no problems */ + return 1; + if ((np.state & QM_MCR_NP_STATE_MASK) != QM_MCR_NP_STATE_OOS) { + struct qm_fqd fqd; + err = qman_query_fq(&fq, &fqd); + BUG_ON(err); + if (fqd.dest.channel == qp) { + /* The channel is the FQ's target, clean it */ + if (qman_shutdown_fq(fq.fqid) != 0) + /* Couldn't shut down the FQ + so the pool must be leaked */ + return 0; + } + } + /* Move to the next FQID */ + fq.fqid++; + } while (1); +} +void qman_release_pool_range(u32 qp, u32 count) +{ + u32 total_invalid = release_id_range(&qpalloc, qp, + count, qpool_cleanup); + if (total_invalid) { + /* Pool channels are almost always used individually */ + if (count == 1) + pr_err("Pool channel 0x%x had %d leaks\n", + qp, total_invalid); + else + pr_err("Pool channels [%d..%d] (%d) had %d leaks\n", + qp, qp + count - 1, count, total_invalid); + } +} +EXPORT_SYMBOL(qman_release_pool_range); + + +void qman_seed_pool_range(u32 poolid, u32 count) +{ + dpa_alloc_seed(&qpalloc, poolid, count); + +} +EXPORT_SYMBOL(qman_seed_pool_range); + +int qman_reserve_pool_range(u32 poolid, u32 count) +{ + return dpa_alloc_reserve(&qpalloc, poolid, count); +} +EXPORT_SYMBOL(qman_reserve_pool_range); + + +/* CGR ID allocator front-end */ + +int qman_alloc_cgrid_range(u32 *result, u32 count, u32 align, int partial) +{ + return dpa_alloc_new(&cgralloc, result, count, align, partial); +} +EXPORT_SYMBOL(qman_alloc_cgrid_range); + +static int cqr_cleanup(u32 cgrid) +{ + /* We query all FQDs starting from + * FQID 1 until we get an "invalid FQID" error, looking for non-OOS FQDs + * whose CGR is the CGR being released. + */ + struct qman_fq fq = { + .fqid = 1 + }; + int err; + do { + struct qm_mcr_queryfq_np np; + err = qman_query_fq_np(&fq, &np); + if (err) + /* FQID range exceeded, found no problems */ + return 1; + if ((np.state & QM_MCR_NP_STATE_MASK) != QM_MCR_NP_STATE_OOS) { + struct qm_fqd fqd; + err = qman_query_fq(&fq, &fqd); + BUG_ON(err); + if ((fqd.fq_ctrl & QM_FQCTRL_CGE) && + (fqd.cgid == cgrid)) { + pr_err("CRGID 0x%x is being used by FQID 0x%x," + " CGR will be leaked\n", + cgrid, fq.fqid); + return 1; + } + } + /* Move to the next FQID */ + fq.fqid++; + } while (1); +} + +void qman_release_cgrid_range(u32 cgrid, u32 count) +{ + u32 total_invalid = release_id_range(&cgralloc, cgrid, + count, cqr_cleanup); + if (total_invalid) + pr_err("CGRID range [%d..%d] (%d) had %d leaks\n", + cgrid, cgrid + count - 1, count, total_invalid); +} +EXPORT_SYMBOL(qman_release_cgrid_range); + +void qman_seed_cgrid_range(u32 cgrid, u32 count) +{ + dpa_alloc_seed(&cgralloc, cgrid, count); + +} +EXPORT_SYMBOL(qman_seed_cgrid_range); +#endif /* CONFIG_FSL_QMAN */ diff --git a/drivers/staging/fsl_qbman/dpa_sys.h b/drivers/staging/fsl_qbman/dpa_sys.h index 19adf70..86b280c 100644 --- a/drivers/staging/fsl_qbman/dpa_sys.h +++ b/drivers/staging/fsl_qbman/dpa_sys.h @@ -179,10 +179,68 @@ static inline void copy_bytes(void *dest, const void *src, size_t sz) #endif /************/ +/* RB-trees */ +/************/ + +/* We encapsulate RB-trees so that its easier to use non-linux forms in + * non-linux systems. This also encapsulates the extra plumbing that linux code + * usually provides when using RB-trees. This encapsulation assumes that the + * data type held by the tree is u32. */ + +struct dpa_rbtree { + struct rb_root root; +}; +#define DPA_RBTREE { .root = RB_ROOT } + +static inline void dpa_rbtree_init(struct dpa_rbtree *tree) +{ + tree->root = RB_ROOT; +} + +#define IMPLEMENT_DPA_RBTREE(name, type, node_field, val_field) \ +static inline int name##_push(struct dpa_rbtree *tree, type *obj) \ +{ \ + struct rb_node *parent = NULL, **p = &tree->root.rb_node; \ + while (*p) { \ + u32 item; \ + parent = *p; \ + item = rb_entry(parent, type, node_field)->val_field; \ + if (obj->val_field < item) \ + p = &parent->rb_left; \ + else if (obj->val_field > item) \ + p = &parent->rb_right; \ + else \ + return -EBUSY; \ + } \ + rb_link_node(&obj->node_field, parent, p); \ + rb_insert_color(&obj->node_field, &tree->root); \ + return 0; \ +} \ +static inline void name##_del(struct dpa_rbtree *tree, type *obj) \ +{ \ + rb_erase(&obj->node_field, &tree->root); \ +} \ +static inline type *name##_find(struct dpa_rbtree *tree, u32 val) \ +{ \ + type *ret; \ + struct rb_node *p = tree->root.rb_node; \ + while (p) { \ + ret = rb_entry(p, type, node_field); \ + if (val < ret->val_field) \ + p = p->rb_left; \ + else if (val > ret->val_field) \ + p = p->rb_right; \ + else \ + return ret; \ + } \ + return NULL; \ +} + +/************/ /* Bootargs */ /************/ -/* Bman has "bportals=", they use the same syntax +/* Qman has "qportals=" and Bman has "bportals=", they use the same syntax * though; a comma-separated list of items, each item being a cpu index and/or a * range of cpu indices, and each item optionally be prefixed by "s" to indicate * that the portal associated with that cpu should be shared. See bman_driver.c diff --git a/drivers/staging/fsl_qbman/qbman_driver.c b/drivers/staging/fsl_qbman/qbman_driver.c index 265c6f0..ef8eca3 100644 --- a/drivers/staging/fsl_qbman/qbman_driver.c +++ b/drivers/staging/fsl_qbman/qbman_driver.c @@ -30,11 +30,55 @@ */ #include +#include "qman_private.h" #include "bman_private.h" static __init int qbman_init(void) { + struct device_node *dn; + u32 is_portal_available; + bman_init(); + qman_init(); + + is_portal_available = 0; + for_each_compatible_node(dn, NULL, "fsl,qman-portal") { + if (!of_device_is_available(dn)) + continue; + else + is_portal_available = 1; + } + + if (!qman_have_ccsr() && is_portal_available) { + struct qman_fq fq = { + .fqid = 1 + }; + struct qm_mcr_queryfq_np np; + int err, retry = CONFIG_FSL_QMAN_INIT_TIMEOUT; + struct timespec nowts, diffts, startts = current_kernel_time(); + /* Loop while querying given fqid succeeds or time out */ + while (1) { + err = qman_query_fq_np(&fq, &np); + if (!err) { + /* success, control-plane has configured QMan */ + break; + } else if (err != -ERANGE) { + pr_err("QMan: I/O error, continuing anyway\n"); + break; + } + nowts = current_kernel_time(); + diffts = timespec_sub(nowts, startts); + if (diffts.tv_sec > 0) { + if (!retry--) { + pr_err("QMan: time out, control-plane" + " dead?\n"); + break; + } + pr_warn("QMan: polling for the control-plane" + " (%d)\n", retry); + } + } + } return 0; } diff --git a/drivers/staging/fsl_qbman/qman_config.c b/drivers/staging/fsl_qbman/qman_config.c new file mode 100644 index 0000000..03ee68a --- /dev/null +++ b/drivers/staging/fsl_qbman/qman_config.c @@ -0,0 +1,991 @@ +/* Copyright 2008-2012 Freescale Semiconductor, Inc. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions are met: + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * * Neither the name of Freescale Semiconductor nor the + * names of its contributors may be used to endorse or promote products + * derived from this software without specific prior written permission. + * + * + * ALTERNATIVELY, this software may be distributed under the terms of the + * GNU General Public License ("GPL") as published by the Free Software + * Foundation, either version 2 of that License or (at your option) any + * later version. + * + * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED + * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE + * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY + * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; + * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND + * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#include +#include "qman_private.h" + +/* Last updated for v00.800 of the BG */ + +/* Register offsets */ +#define REG_QCSP_LIO_CFG(n) (0x0000 + ((n) * 0x10)) +#define REG_QCSP_IO_CFG(n) (0x0004 + ((n) * 0x10)) +#define REG_QCSP_DD_CFG(n) (0x000c + ((n) * 0x10)) +#define REG_DD_CFG 0x0200 +#define REG_DCP_CFG(n) (0x0300 + ((n) * 0x10)) +#define REG_DCP_DD_CFG(n) (0x0304 + ((n) * 0x10)) +#define REG_DCP_DLM_AVG(n) (0x030c + ((n) * 0x10)) +#define REG_PFDR_FPC 0x0400 +#define REG_PFDR_FP_HEAD 0x0404 +#define REG_PFDR_FP_TAIL 0x0408 +#define REG_PFDR_FP_LWIT 0x0410 +#define REG_PFDR_CFG 0x0414 +#define REG_SFDR_CFG 0x0500 +#define REG_SFDR_IN_USE 0x0504 +#define REG_WQ_CS_CFG(n) (0x0600 + ((n) * 0x04)) +#define REG_WQ_DEF_ENC_WQID 0x0630 +#define REG_WQ_SC_DD_CFG(n) (0x640 + ((n) * 0x04)) +#define REG_WQ_PC_DD_CFG(n) (0x680 + ((n) * 0x04)) +#define REG_WQ_DC0_DD_CFG(n) (0x6c0 + ((n) * 0x04)) +#define REG_WQ_DC1_DD_CFG(n) (0x700 + ((n) * 0x04)) +#define REG_WQ_DCn_DD_CFG(n) (0x6c0 + ((n) * 0x40)) /* n=2,3 */ +#define REG_CM_CFG 0x0800 +#define REG_ECSR 0x0a00 +#define REG_ECIR 0x0a04 +#define REG_EADR 0x0a08 +#define REG_ECIR2 0x0a0c +#define REG_EDATA(n) (0x0a10 + ((n) * 0x04)) +#define REG_SBEC(n) (0x0a80 + ((n) * 0x04)) +#define REG_MCR 0x0b00 +#define REG_MCP(n) (0x0b04 + ((n) * 0x04)) +#define REG_MISC_CFG 0x0be0 +#define REG_HID_CFG 0x0bf0 +#define REG_IDLE_STAT 0x0bf4 +#define REG_IP_REV_1 0x0bf8 +#define REG_IP_REV_2 0x0bfc +#define REG_FQD_BARE 0x0c00 +#define REG_PFDR_BARE 0x0c20 +#define REG_offset_BAR 0x0004 /* relative to REG_[FQD|PFDR]_BARE */ +#define REG_offset_AR 0x0010 /* relative to REG_[FQD|PFDR]_BARE */ +#define REG_QCSP_BARE 0x0c80 +#define REG_QCSP_BAR 0x0c84 +#define REG_CI_SCHED_CFG 0x0d00 +#define REG_SRCIDR 0x0d04 +#define REG_LIODNR 0x0d08 +#define REG_CI_RLM_AVG 0x0d14 +#define REG_ERR_ISR 0x0e00 /* + "enum qm_isr_reg" */ +#define REG_REV3_QCSP_LIO_CFG(n) (0x1000 + ((n) * 0x10)) +#define REG_REV3_QCSP_IO_CFG(n) (0x1004 + ((n) * 0x10)) +#define REG_REV3_QCSP_DD_CFG(n) (0x100c + ((n) * 0x10)) + +/* Assists for QMAN_MCR */ +#define MCR_INIT_PFDR 0x01000000 +#define MCR_get_rslt(v) (u8)((v) >> 24) +#define MCR_rslt_idle(r) (!rslt || (rslt >= 0xf0)) +#define MCR_rslt_ok(r) (rslt == 0xf0) +#define MCR_rslt_eaccess(r) (rslt == 0xf8) +#define MCR_rslt_inval(r) (rslt == 0xff) + +struct qman; + +/* Follows WQ_CS_CFG0-5 */ +enum qm_wq_class { + qm_wq_portal = 0, + qm_wq_pool = 1, + qm_wq_fman0 = 2, + qm_wq_fman1 = 3, + qm_wq_caam = 4, + qm_wq_pme = 5, + qm_wq_first = qm_wq_portal, + qm_wq_last = qm_wq_pme +}; + +/* Follows FQD_[BARE|BAR|AR] and PFDR_[BARE|BAR|AR] */ +enum qm_memory { + qm_memory_fqd, + qm_memory_pfdr +}; + +/* Used by all error interrupt registers except 'inhibit' */ +#define QM_EIRQ_CIDE 0x20000000 /* Corenet Initiator Data Error */ +#define QM_EIRQ_CTDE 0x10000000 /* Corenet Target Data Error */ +#define QM_EIRQ_CITT 0x08000000 /* Corenet Invalid Target Transaction */ +#define QM_EIRQ_PLWI 0x04000000 /* PFDR Low Watermark */ +#define QM_EIRQ_MBEI 0x02000000 /* Multi-bit ECC Error */ +#define QM_EIRQ_SBEI 0x01000000 /* Single-bit ECC Error */ +#define QM_EIRQ_PEBI 0x00800000 /* PFDR Enqueues Blocked Interrupt */ +#define QM_EIRQ_IFSI 0x00020000 /* Invalid FQ Flow Control State */ +#define QM_EIRQ_ICVI 0x00010000 /* Invalid Command Verb */ +#define QM_EIRQ_IDDI 0x00000800 /* Invalid Dequeue (Direct-connect) */ +#define QM_EIRQ_IDFI 0x00000400 /* Invalid Dequeue FQ */ +#define QM_EIRQ_IDSI 0x00000200 /* Invalid Dequeue Source */ +#define QM_EIRQ_IDQI 0x00000100 /* Invalid Dequeue Queue */ +#define QM_EIRQ_IECE 0x00000010 /* Invalid Enqueue Configuration */ +#define QM_EIRQ_IEOI 0x00000008 /* Invalid Enqueue Overflow */ +#define QM_EIRQ_IESI 0x00000004 /* Invalid Enqueue State */ +#define QM_EIRQ_IECI 0x00000002 /* Invalid Enqueue Channel */ +#define QM_EIRQ_IEQI 0x00000001 /* Invalid Enqueue Queue */ + +/* QMAN_ECIR valid error bit */ +#define PORTAL_ECSR_ERR (QM_EIRQ_IEQI | QM_EIRQ_IESI | QM_EIRQ_IEOI | \ + QM_EIRQ_IDQI | QM_EIRQ_IDSI | QM_EIRQ_IDFI | \ + QM_EIRQ_IDDI | QM_EIRQ_ICVI | QM_EIRQ_IFSI) +#define FQID_ECSR_ERR (QM_EIRQ_IEQI | QM_EIRQ_IECI | QM_EIRQ_IESI | \ + QM_EIRQ_IEOI | QM_EIRQ_IDQI | QM_EIRQ_IDFI | \ + QM_EIRQ_IFSI) + +union qman_ecir { + u32 ecir_raw; + struct { + u32 __reserved:2; + u32 portal_type:1; + u32 portal_num:5; + u32 fqid:24; + } __packed info; +}; + +union qman_ecir2 { + u32 ecir2_raw; + struct { + u32 portal_type:1; + u32 __reserved:21; + u32 portal_num:10; + } __packed info; +}; + +union qman_eadr { + u32 eadr_raw; + struct { + u32 __reserved1:4; + u32 memid:4; + u32 __reserved2:12; + u32 eadr:12; + } __packed info; + struct { + u32 __reserved1:3; + u32 memid:5; + u32 __reserved:8; + u32 eadr:16; + } __packed info_rev3; +}; + +struct qman_hwerr_txt { + u32 mask; + const char *txt; +}; + +#define QMAN_HWE_TXT(a, b) { .mask = QM_EIRQ_##a, .txt = b } + +static const struct qman_hwerr_txt qman_hwerr_txts[] = { + QMAN_HWE_TXT(CIDE, "Corenet Initiator Data Error"), + QMAN_HWE_TXT(CTDE, "Corenet Target Data Error"), + QMAN_HWE_TXT(CITT, "Corenet Invalid Target Transaction"), + QMAN_HWE_TXT(PLWI, "PFDR Low Watermark"), + QMAN_HWE_TXT(MBEI, "Multi-bit ECC Error"), + QMAN_HWE_TXT(SBEI, "Single-bit ECC Error"), + QMAN_HWE_TXT(PEBI, "PFDR Enqueues Blocked Interrupt"), + QMAN_HWE_TXT(ICVI, "Invalid Command Verb"), + QMAN_HWE_TXT(IFSI, "Invalid Flow Control State"), + QMAN_HWE_TXT(IDDI, "Invalid Dequeue (Direct-connect)"), + QMAN_HWE_TXT(IDFI, "Invalid Dequeue FQ"), + QMAN_HWE_TXT(IDSI, "Invalid Dequeue Source"), + QMAN_HWE_TXT(IDQI, "Invalid Dequeue Queue"), + QMAN_HWE_TXT(IECE, "Invalid Enqueue Configuration"), + QMAN_HWE_TXT(IEOI, "Invalid Enqueue Overflow"), + QMAN_HWE_TXT(IESI, "Invalid Enqueue State"), + QMAN_HWE_TXT(IECI, "Invalid Enqueue Channel"), + QMAN_HWE_TXT(IEQI, "Invalid Enqueue Queue") +}; +#define QMAN_HWE_COUNT (sizeof(qman_hwerr_txts)/sizeof(struct qman_hwerr_txt)) + +struct qman_error_info_mdata { + u16 addr_mask; + u16 bits; + const char *txt; +}; + +#define QMAN_ERR_MDATA(a, b, c) { .addr_mask = a, .bits = b, .txt = c} +static const struct qman_error_info_mdata error_mdata[] = { + QMAN_ERR_MDATA(0x01FF, 24, "FQD cache tag memory 0"), + QMAN_ERR_MDATA(0x01FF, 24, "FQD cache tag memory 1"), + QMAN_ERR_MDATA(0x01FF, 24, "FQD cache tag memory 2"), + QMAN_ERR_MDATA(0x01FF, 24, "FQD cache tag memory 3"), + QMAN_ERR_MDATA(0x0FFF, 512, "FQD cache memory"), + QMAN_ERR_MDATA(0x07FF, 128, "SFDR memory"), + QMAN_ERR_MDATA(0x01FF, 72, "WQ context memory"), + QMAN_ERR_MDATA(0x00FF, 240, "CGR memory"), + QMAN_ERR_MDATA(0x00FF, 302, "Internal Order Restoration List memory"), + QMAN_ERR_MDATA(0x01FF, 256, "SW portal ring memory"), +}; +#define QMAN_ERR_MDATA_COUNT \ + (sizeof(error_mdata)/sizeof(struct qman_error_info_mdata)) + +/* Add this in Kconfig */ +#define QMAN_ERRS_TO_UNENABLE (QM_EIRQ_PLWI | QM_EIRQ_PEBI) + +/** + * qm_err_isr__ - Manipulate global interrupt registers + * @v: for accessors that write values, this is the 32-bit value + * + * Manipulates QMAN_ERR_ISR, QMAN_ERR_IER, QMAN_ERR_ISDR, QMAN_ERR_IIR. All + * manipulations except qm_err_isr_[un]inhibit() use 32-bit masks composed of + * the QM_EIRQ_*** definitions. Note that "qm_err_isr_enable_write" means + * "write the enable register" rather than "enable the write register"! + */ +#define qm_err_isr_status_read(qm) \ + __qm_err_isr_read(qm, qm_isr_status) +#define qm_err_isr_status_clear(qm, m) \ + __qm_err_isr_write(qm, qm_isr_status, m) +#define qm_err_isr_enable_read(qm) \ + __qm_err_isr_read(qm, qm_isr_enable) +#define qm_err_isr_enable_write(qm, v) \ + __qm_err_isr_write(qm, qm_isr_enable, v) +#define qm_err_isr_disable_read(qm) \ + __qm_err_isr_read(qm, qm_isr_disable) +#define qm_err_isr_disable_write(qm, v) \ + __qm_err_isr_write(qm, qm_isr_disable, v) +#define qm_err_isr_inhibit(qm) \ + __qm_err_isr_write(qm, qm_isr_inhibit, 1) +#define qm_err_isr_uninhibit(qm) \ + __qm_err_isr_write(qm, qm_isr_inhibit, 0) + +/* + * TODO: unimplemented registers + * + * Keeping a list here of Qman registers I have not yet covered; + * QCSP_DD_IHRSR, QCSP_DD_IHRFR, QCSP_DD_HASR, + * DCP_DD_IHRSR, DCP_DD_IHRFR, DCP_DD_HASR, CM_CFG, + * QMAN_EECC, QMAN_SBET, QMAN_EINJ, QMAN_SBEC0-12 + */ + +/* Encapsulate "struct qman *" as a cast of the register space address. */ + +static struct qman *qm_create(void *regs) +{ + return (struct qman *)regs; +} + +static inline u32 __qm_in(struct qman *qm, u32 offset) +{ + return in_be32((void *)qm + offset); +} +static inline void __qm_out(struct qman *qm, u32 offset, u32 val) +{ + out_be32((void *)qm + offset, val); +} +#define qm_in(reg) __qm_in(qm, REG_##reg) +#define qm_out(reg, val) __qm_out(qm, REG_##reg, val) + +static u32 __qm_err_isr_read(struct qman *qm, enum qm_isr_reg n) +{ + return __qm_in(qm, REG_ERR_ISR + (n << 2)); +} + +static void __qm_err_isr_write(struct qman *qm, enum qm_isr_reg n, u32 val) +{ + __qm_out(qm, REG_ERR_ISR + (n << 2), val); +} + +static void qm_set_dc(struct qman *qm, enum qm_dc_portal portal, + int ed, u8 sernd) +{ + DPA_ASSERT(!ed || (portal == qm_dc_portal_fman0) || + (portal == qm_dc_portal_fman1)); + if ((qman_ip_rev & 0xFF00) >= QMAN_REV30) + qm_out(DCP_CFG(portal), (ed ? 0x1000 : 0) | (sernd & 0x3ff)); + else + qm_out(DCP_CFG(portal), (ed ? 0x100 : 0) | (sernd & 0x1f)); +} + +static void qm_set_wq_scheduling(struct qman *qm, enum qm_wq_class wq_class, + u8 cs_elev, u8 csw2, u8 csw3, u8 csw4, u8 csw5, + u8 csw6, u8 csw7) +{ + qm_out(WQ_CS_CFG(wq_class), ((cs_elev & 0xff) << 24) | + ((csw2 & 0x7) << 20) | ((csw3 & 0x7) << 16) | + ((csw4 & 0x7) << 12) | ((csw5 & 0x7) << 8) | + ((csw6 & 0x7) << 4) | (csw7 & 0x7)); +} + +static void qm_set_hid(struct qman *qm) +{ + qm_out(HID_CFG, 0); +} + +static void qm_set_corenet_initiator(struct qman *qm) +{ + qm_out(CI_SCHED_CFG, + 0x80000000 | /* write srcciv enable */ + (CONFIG_FSL_QMAN_CI_SCHED_CFG_SRCCIV << 24) | + (CONFIG_FSL_QMAN_CI_SCHED_CFG_SRQ_W << 8) | + (CONFIG_FSL_QMAN_CI_SCHED_CFG_RW_W << 4) | + CONFIG_FSL_QMAN_CI_SCHED_CFG_BMAN_W); +} + +static void qm_get_version(struct qman *qm, u16 *id, u8 *major, u8 *minor) +{ + u32 v = qm_in(IP_REV_1); + *id = (v >> 16); + *major = (v >> 8) & 0xff; + *minor = v & 0xff; +} + +static void qm_set_memory(struct qman *qm, enum qm_memory memory, u64 ba, + int enable, int prio, int stash, u32 size) +{ + u32 offset = (memory == qm_memory_fqd) ? REG_FQD_BARE : REG_PFDR_BARE; + u32 exp = ilog2(size); + /* choke if size isn't within range */ + DPA_ASSERT((size >= 4096) && (size <= 1073741824) && + is_power_of_2(size)); + /* choke if 'ba' has lower-alignment than 'size' */ + DPA_ASSERT(!(ba & (size - 1))); + __qm_out(qm, offset, upper_32_bits(ba)); + __qm_out(qm, offset + REG_offset_BAR, lower_32_bits(ba)); + __qm_out(qm, offset + REG_offset_AR, + (enable ? 0x80000000 : 0) | + (prio ? 0x40000000 : 0) | + (stash ? 0x20000000 : 0) | + (exp - 1)); +} + +static void qm_set_pfdr_threshold(struct qman *qm, u32 th, u8 k) +{ + qm_out(PFDR_FP_LWIT, th & 0xffffff); + qm_out(PFDR_CFG, k); +} + +static void qm_set_sfdr_threshold(struct qman *qm, u16 th) +{ + qm_out(SFDR_CFG, th & 0x3ff); +} + +static int qm_init_pfdr(struct qman *qm, u32 pfdr_start, u32 num) +{ + u8 rslt = MCR_get_rslt(qm_in(MCR)); + + DPA_ASSERT(pfdr_start && !(pfdr_start & 7) && !(num & 7) && num); + /* Make sure the command interface is 'idle' */ + if (!MCR_rslt_idle(rslt)) + panic("QMAN_MCR isn't idle"); + + /* Write the MCR command params then the verb */ + qm_out(MCP(0), pfdr_start); + /* TODO: remove this - it's a workaround for a model bug that is + * corrected in more recent versions. We use the workaround until + * everyone has upgraded. */ + qm_out(MCP(1), (pfdr_start + num - 16)); + lwsync(); + qm_out(MCR, MCR_INIT_PFDR); + /* Poll for the result */ + do { + rslt = MCR_get_rslt(qm_in(MCR)); + } while (!MCR_rslt_idle(rslt)); + if (MCR_rslt_ok(rslt)) + return 0; + if (MCR_rslt_eaccess(rslt)) + return -EACCES; + if (MCR_rslt_inval(rslt)) + return -EINVAL; + pr_crit("Unexpected result from MCR_INIT_PFDR: %02x\n", rslt); + return -ENOSYS; +} + +/*****************/ +/* Config driver */ +/*****************/ + +/* We support only one of these */ +static struct qman *qm; +static struct device_node *qm_node; + +/* And this state belongs to 'qm'. It is set during fsl_qman_init(), but used + * during qman_init_ccsr(). */ +static dma_addr_t fqd_a, pfdr_a; +static size_t fqd_sz, pfdr_sz; + +static int qman_fqd(struct reserved_mem *rmem) +{ + fqd_a = rmem->base; + fqd_sz = rmem->size; + + WARN_ON(!(fqd_a && fqd_sz)); + + return 0; +} +RESERVEDMEM_OF_DECLARE(qman_fqd, "fsl,qman-fqd", qman_fqd); + +static int qman_pfdr(struct reserved_mem *rmem) +{ + pfdr_a = rmem->base; + pfdr_sz = rmem->size; + + WARN_ON(!(pfdr_a && pfdr_sz)); + + return 0; +} +RESERVEDMEM_OF_DECLARE(qman_fbpr, "fsl,qman-pfdr", qman_pfdr); + +static __init int parse_mem_property(struct device_node *node, + dma_addr_t *addr, size_t *sz, int zero) +{ + if (zero) { + /* map as cacheable, non-guarded */ + void __iomem *tmpp = ioremap_prot(*addr, *sz, 0); + memset_io(tmpp, 0, *sz); + flush_dcache_range((unsigned long)tmpp, + (unsigned long)tmpp + *sz); + iounmap(tmpp); + } + + return 0; +} + +/* TODO: + * - there is obviously no handling of errors, + * - the calls to qm_set_memory() hard-code the priority and CPC-stashing for + * both memory resources to zero. + */ +static int __init fsl_qman_init(struct device_node *node) +{ + struct resource res; + u32 __iomem *regs; + int ret; + u16 id; + u8 major, minor; + ret = of_address_to_resource(node, 0, &res); + if (ret) { + pr_err("Can't get %s property '%s'\n", node->full_name, "reg"); + return ret; + } + + ret = parse_mem_property(node, &fqd_a, &fqd_sz, 1); + BUG_ON(ret); + ret = parse_mem_property(node, &pfdr_a, &pfdr_sz, 0); + BUG_ON(ret); + + /* Global configuration */ + regs = ioremap(res.start, res.end - res.start + 1); + qm = qm_create(regs); + qm_node = node; + qm_get_version(qm, &id, &major, &minor); + pr_info("Qman ver:%04x,%02x,%02x\n", id, major, minor); + if (!qman_ip_rev) { + if ((major == 1) && (minor == 0)) { + pr_err("QMAN rev1.0 on P4080 rev1 is not supported!\n"); + iounmap(regs); + return -ENODEV; + } else if ((major == 1) && (minor == 1)) + qman_ip_rev = QMAN_REV11; + else if ((major == 1) && (minor == 2)) + qman_ip_rev = QMAN_REV12; + else if ((major == 2) && (minor == 0)) + qman_ip_rev = QMAN_REV20; + else if ((major == 3) && (minor == 0)) + qman_ip_rev = QMAN_REV30; + else if ((major == 3) && (minor == 1)) + qman_ip_rev = QMAN_REV31; + else { + pr_warn("unknown Qman version, default to rev1.1\n"); + qman_ip_rev = QMAN_REV11; + } + } + + return 0; +} + +int qman_have_ccsr(void) +{ + return qm ? 1 : 0; +} + +__init void qman_init_early(void) +{ + struct device_node *dn; + int ret; + + for_each_compatible_node(dn, NULL, "fsl,qman") { + if (qm) + pr_err("%s: only one 'fsl,qman' allowed\n", + dn->full_name); + else { + if (!of_device_is_available(dn)) + continue; + + ret = fsl_qman_init(dn); + BUG_ON(ret); + } + } +} + +static void log_edata_bits(u32 bit_count) +{ + u32 i, j, mask = 0xffffffff; + + pr_warn("Qman ErrInt, EDATA:\n"); + i = bit_count/32; + if (bit_count%32) { + i++; + mask = ~(mask << bit_count%32); + } + j = 16-i; + pr_warn(" 0x%08x\n", qm_in(EDATA(j)) & mask); + j++; + for (; j < 16; j++) + pr_warn(" 0x%08x\n", qm_in(EDATA(j))); +} + +static void log_additional_error_info(u32 isr_val, u32 ecsr_val) +{ + union qman_ecir ecir_val; + union qman_eadr eadr_val; + + ecir_val.ecir_raw = qm_in(ECIR); + /* Is portal info valid */ + if ((qman_ip_rev & 0xFF00) >= QMAN_REV30) { + union qman_ecir2 ecir2_val; + ecir2_val.ecir2_raw = qm_in(ECIR2); + if (ecsr_val & PORTAL_ECSR_ERR) { + pr_warn("Qman ErrInt: %s id %d\n", + (ecir2_val.info.portal_type) ? + "DCP" : "SWP", ecir2_val.info.portal_num); + } + if (ecsr_val & (FQID_ECSR_ERR | QM_EIRQ_IECE)) { + pr_warn("Qman ErrInt: ecir.fqid 0x%x\n", + ecir_val.info.fqid); + } + if (ecsr_val & (QM_EIRQ_SBEI|QM_EIRQ_MBEI)) { + eadr_val.eadr_raw = qm_in(EADR); + pr_warn("Qman ErrInt: EADR Memory: %s, 0x%x\n", + error_mdata[eadr_val.info_rev3.memid].txt, + error_mdata[eadr_val.info_rev3.memid].addr_mask + & eadr_val.info_rev3.eadr); + log_edata_bits( + error_mdata[eadr_val.info_rev3.memid].bits); + } + } else { + if (ecsr_val & PORTAL_ECSR_ERR) { + pr_warn("Qman ErrInt: %s id %d\n", + (ecir_val.info.portal_type) ? + "DCP" : "SWP", ecir_val.info.portal_num); + } + if (ecsr_val & FQID_ECSR_ERR) { + pr_warn("Qman ErrInt: ecir.fqid 0x%x\n", + ecir_val.info.fqid); + } + if (ecsr_val & (QM_EIRQ_SBEI|QM_EIRQ_MBEI)) { + eadr_val.eadr_raw = qm_in(EADR); + pr_warn("Qman ErrInt: EADR Memory: %s, 0x%x\n", + error_mdata[eadr_val.info.memid].txt, + error_mdata[eadr_val.info.memid].addr_mask + & eadr_val.info.eadr); + log_edata_bits(error_mdata[eadr_val.info.memid].bits); + } + } +} + +/* Qman interrupt handler */ +static irqreturn_t qman_isr(int irq, void *ptr) +{ + u32 isr_val, ier_val, ecsr_val, isr_mask, i; + + ier_val = qm_err_isr_enable_read(qm); + isr_val = qm_err_isr_status_read(qm); + ecsr_val = qm_in(ECSR); + isr_mask = isr_val & ier_val; + + if (!isr_mask) + return IRQ_NONE; + for (i = 0; i < QMAN_HWE_COUNT; i++) { + if (qman_hwerr_txts[i].mask & isr_mask) { + pr_warn("Qman ErrInt: %s\n", qman_hwerr_txts[i].txt); + if (qman_hwerr_txts[i].mask & ecsr_val) { + log_additional_error_info(isr_mask, ecsr_val); + /* Re-arm error capture registers */ + qm_out(ECSR, ecsr_val); + } + if (qman_hwerr_txts[i].mask & QMAN_ERRS_TO_UNENABLE) { + pr_devel("Qman un-enabling error 0x%x\n", + qman_hwerr_txts[i].mask); + ier_val &= ~qman_hwerr_txts[i].mask; + qm_err_isr_enable_write(qm, ier_val); + } + } + } + qm_err_isr_status_clear(qm, isr_val); + return IRQ_HANDLED; +} + +static int __bind_irq(void) +{ + int ret, err_irq; + + err_irq = of_irq_to_resource(qm_node, 0, NULL); + if (err_irq == NO_IRQ) { + pr_info("Can't get %s property '%s'\n", qm_node->full_name, + "interrupts"); + return -ENODEV; + } + ret = request_irq(err_irq, qman_isr, IRQF_SHARED, "qman-err", qm_node); + if (ret) { + pr_err("request_irq() failed %d for '%s'\n", ret, + qm_node->full_name); + return -ENODEV; + } + /* Write-to-clear any stale bits, (eg. starvation being asserted prior + * to resource allocation during driver init). */ + qm_err_isr_status_clear(qm, 0xffffffff); + /* Enable Error Interrupts */ + qm_err_isr_enable_write(qm, 0xffffffff); + return 0; +} + +int qman_init_ccsr(struct device_node *node) +{ + int ret; + if (!qman_have_ccsr()) + return 0; + if (node != qm_node) + return -EINVAL; + /* FQD memory */ + qm_set_memory(qm, qm_memory_fqd, fqd_a, 1, 0, 0, fqd_sz); + /* PFDR memory */ + qm_set_memory(qm, qm_memory_pfdr, pfdr_a, 1, 0, 0, pfdr_sz); + qm_init_pfdr(qm, 8, pfdr_sz / 64 - 8); + /* thresholds */ + qm_set_pfdr_threshold(qm, 512, 64); + qm_set_sfdr_threshold(qm, 128); + /* clear stale PEBI bit from interrupt status register */ + qm_err_isr_status_clear(qm, QM_EIRQ_PEBI); + /* corenet initiator settings */ + qm_set_corenet_initiator(qm); + /* HID settings */ + qm_set_hid(qm); + /* Set scheduling weights to defaults */ + for (ret = qm_wq_first; ret <= qm_wq_last; ret++) + qm_set_wq_scheduling(qm, ret, 0, 0, 0, 0, 0, 0, 0); + /* We are not prepared to accept ERNs for hardware enqueues */ + qm_set_dc(qm, qm_dc_portal_fman0, 1, 0); + qm_set_dc(qm, qm_dc_portal_fman1, 1, 0); + /* Initialise Error Interrupt Handler */ + ret = __bind_irq(); + if (ret) + return ret; + return 0; +} + +#define LIO_CFG_LIODN_MASK 0x0fff0000 +void qman_liodn_fixup(u16 channel) +{ + static int done; + static u32 liodn_offset; + u32 before, after; + int idx = channel - QM_CHANNEL_SWPORTAL0; + + if (!qman_have_ccsr()) + return; + if ((qman_ip_rev & 0xFF00) >= QMAN_REV30) + before = qm_in(REV3_QCSP_LIO_CFG(idx)); + else + before = qm_in(QCSP_LIO_CFG(idx)); + if (!done) { + liodn_offset = before & LIO_CFG_LIODN_MASK; + done = 1; + return; + } + after = (before & (~LIO_CFG_LIODN_MASK)) | liodn_offset; + if ((qman_ip_rev & 0xFF00) >= QMAN_REV30) + qm_out(REV3_QCSP_LIO_CFG(idx), after); + else + qm_out(QCSP_LIO_CFG(idx), after); +} + +#define IO_CFG_SDEST_MASK 0x00ff0000 +int qman_set_sdest(u16 channel, unsigned int cpu_idx) +{ + int idx = channel - QM_CHANNEL_SWPORTAL0; + u32 before, after; + + if (!qman_have_ccsr()) + return -ENODEV; + + if ((qman_ip_rev & 0xFF00) >= QMAN_REV30) { + before = qm_in(REV3_QCSP_IO_CFG(idx)); + /* Each pair of vcpu share the same SRQ(SDEST) */ + cpu_idx /= 2; + after = (before & (~IO_CFG_SDEST_MASK)) | (cpu_idx << 16); + qm_out(REV3_QCSP_IO_CFG(idx), after); + } else { + before = qm_in(QCSP_IO_CFG(idx)); + after = (before & (~IO_CFG_SDEST_MASK)) | (cpu_idx << 16); + qm_out(QCSP_IO_CFG(idx), after); + } + return 0; +} + +#define MISC_CFG_WPM_MASK 0x00000002 +int qm_set_wpm(int wpm) +{ + u32 before; + u32 after; + + if (!qman_have_ccsr()) + return -ENODEV; + + before = qm_in(MISC_CFG); + after = (before & (~MISC_CFG_WPM_MASK)) | (wpm << 1); + qm_out(MISC_CFG, after); + return 0; +} + +int qm_get_wpm(int *wpm) +{ + u32 before; + + if (!qman_have_ccsr()) + return -ENODEV; + + before = qm_in(MISC_CFG); + *wpm = (before & MISC_CFG_WPM_MASK) >> 1; + return 0; +} + +#ifdef CONFIG_SYSFS + +#define DRV_NAME "fsl-qman" + +static ssize_t show_pfdr_fpc(struct device *dev, + struct device_attribute *dev_attr, char *buf) +{ + return snprintf(buf, PAGE_SIZE, "%u\n", qm_in(PFDR_FPC)); +}; + +static ssize_t show_dlm_avg(struct device *dev, + struct device_attribute *dev_attr, char *buf) +{ + u32 data; + int i; + + if (!sscanf(dev_attr->attr.name, "dcp%d_dlm_avg", &i)) + return -EINVAL; + data = qm_in(DCP_DLM_AVG(i)); + return snprintf(buf, PAGE_SIZE, "%d.%08d\n", data>>8, + (data & 0x000000ff)*390625); +}; + +static ssize_t set_dlm_avg(struct device *dev, + struct device_attribute *dev_attr, const char *buf, size_t count) +{ + unsigned long val; + int i; + + if (!sscanf(dev_attr->attr.name, "dcp%d_dlm_avg", &i)) + return -EINVAL; + if (kstrtoul(buf, 0, &val)) { + dev_dbg(dev, "invalid input %s\n", buf); + return -EINVAL; + } + qm_out(DCP_DLM_AVG(i), val); + return count; +}; + +static ssize_t show_pfdr_cfg(struct device *dev, + struct device_attribute *dev_attr, char *buf) +{ + return snprintf(buf, PAGE_SIZE, "%u\n", qm_in(PFDR_CFG)); +}; + +static ssize_t set_pfdr_cfg(struct device *dev, + struct device_attribute *dev_attr, const char *buf, size_t count) +{ + unsigned long val; + + if (kstrtoul(buf, 0, &val)) { + dev_dbg(dev, "invalid input %s\n", buf); + return -EINVAL; + } + qm_out(PFDR_CFG, val); + return count; +}; + +static ssize_t show_sfdr_in_use(struct device *dev, + struct device_attribute *dev_attr, char *buf) +{ + return snprintf(buf, PAGE_SIZE, "%u\n", qm_in(SFDR_IN_USE)); +}; + +static ssize_t show_idle_stat(struct device *dev, + struct device_attribute *dev_attr, char *buf) +{ + return snprintf(buf, PAGE_SIZE, "%u\n", qm_in(IDLE_STAT)); +}; + +static ssize_t show_ci_rlm_avg(struct device *dev, + struct device_attribute *dev_attr, char *buf) +{ + u32 data = qm_in(CI_RLM_AVG); + return snprintf(buf, PAGE_SIZE, "%d.%08d\n", data>>8, + (data & 0x000000ff)*390625); +}; + +static ssize_t set_ci_rlm_avg(struct device *dev, + struct device_attribute *dev_attr, const char *buf, size_t count) +{ + unsigned long val; + + if (kstrtoul(buf, 0, &val)) { + dev_dbg(dev, "invalid input %s\n", buf); + return -EINVAL; + } + qm_out(CI_RLM_AVG, val); + return count; +}; + +static ssize_t show_err_isr(struct device *dev, + struct device_attribute *dev_attr, char *buf) +{ + return snprintf(buf, PAGE_SIZE, "0x%08x\n", qm_in(ERR_ISR)); +}; + + +static ssize_t show_sbec(struct device *dev, + struct device_attribute *dev_attr, char *buf) +{ + int i; + + if (!sscanf(dev_attr->attr.name, "sbec_%d", &i)) + return -EINVAL; + return snprintf(buf, PAGE_SIZE, "%u\n", qm_in(SBEC(i))); +}; + +static DEVICE_ATTR(pfdr_fpc, S_IRUSR, show_pfdr_fpc, NULL); +static DEVICE_ATTR(pfdr_cfg, S_IRUSR, show_pfdr_cfg, set_pfdr_cfg); +static DEVICE_ATTR(idle_stat, S_IRUSR, show_idle_stat, NULL); +static DEVICE_ATTR(ci_rlm_avg, (S_IRUSR|S_IWUSR), + show_ci_rlm_avg, set_ci_rlm_avg); +static DEVICE_ATTR(err_isr, S_IRUSR, show_err_isr, NULL); +static DEVICE_ATTR(sfdr_in_use, S_IRUSR, show_sfdr_in_use, NULL); + +static DEVICE_ATTR(dcp0_dlm_avg, (S_IRUSR|S_IWUSR), show_dlm_avg, set_dlm_avg); +static DEVICE_ATTR(dcp1_dlm_avg, (S_IRUSR|S_IWUSR), show_dlm_avg, set_dlm_avg); +static DEVICE_ATTR(dcp2_dlm_avg, (S_IRUSR|S_IWUSR), show_dlm_avg, set_dlm_avg); +static DEVICE_ATTR(dcp3_dlm_avg, (S_IRUSR|S_IWUSR), show_dlm_avg, set_dlm_avg); + +static DEVICE_ATTR(sbec_0, S_IRUSR, show_sbec, NULL); +static DEVICE_ATTR(sbec_1, S_IRUSR, show_sbec, NULL); +static DEVICE_ATTR(sbec_2, S_IRUSR, show_sbec, NULL); +static DEVICE_ATTR(sbec_3, S_IRUSR, show_sbec, NULL); +static DEVICE_ATTR(sbec_4, S_IRUSR, show_sbec, NULL); +static DEVICE_ATTR(sbec_5, S_IRUSR, show_sbec, NULL); +static DEVICE_ATTR(sbec_6, S_IRUSR, show_sbec, NULL); +static DEVICE_ATTR(sbec_7, S_IRUSR, show_sbec, NULL); +static DEVICE_ATTR(sbec_8, S_IRUSR, show_sbec, NULL); +static DEVICE_ATTR(sbec_9, S_IRUSR, show_sbec, NULL); +static DEVICE_ATTR(sbec_10, S_IRUSR, show_sbec, NULL); +static DEVICE_ATTR(sbec_11, S_IRUSR, show_sbec, NULL); +static DEVICE_ATTR(sbec_12, S_IRUSR, show_sbec, NULL); +static DEVICE_ATTR(sbec_13, S_IRUSR, show_sbec, NULL); +static DEVICE_ATTR(sbec_14, S_IRUSR, show_sbec, NULL); + +static struct attribute *qman_dev_attributes[] = { + &dev_attr_pfdr_fpc.attr, + &dev_attr_pfdr_cfg.attr, + &dev_attr_idle_stat.attr, + &dev_attr_ci_rlm_avg.attr, + &dev_attr_err_isr.attr, + &dev_attr_dcp0_dlm_avg.attr, + &dev_attr_dcp1_dlm_avg.attr, + &dev_attr_dcp2_dlm_avg.attr, + &dev_attr_dcp3_dlm_avg.attr, + /* sfdr_in_use will be added if necessary */ + NULL +}; + +static struct attribute *qman_dev_ecr_attributes[] = { + &dev_attr_sbec_0.attr, + &dev_attr_sbec_1.attr, + &dev_attr_sbec_2.attr, + &dev_attr_sbec_3.attr, + &dev_attr_sbec_4.attr, + &dev_attr_sbec_5.attr, + &dev_attr_sbec_6.attr, + &dev_attr_sbec_7.attr, + &dev_attr_sbec_8.attr, + &dev_attr_sbec_9.attr, + &dev_attr_sbec_10.attr, + &dev_attr_sbec_11.attr, + &dev_attr_sbec_12.attr, + &dev_attr_sbec_13.attr, + &dev_attr_sbec_14.attr, + NULL +}; + +/* root level */ +static const struct attribute_group qman_dev_attr_grp = { + .name = NULL, + .attrs = qman_dev_attributes +}; +static const struct attribute_group qman_dev_ecr_grp = { + .name = "error_capture", + .attrs = qman_dev_ecr_attributes +}; + +static int of_fsl_qman_remove(struct platform_device *ofdev) +{ + sysfs_remove_group(&ofdev->dev.kobj, &qman_dev_attr_grp); + return 0; +}; + +static int of_fsl_qman_probe(struct platform_device *ofdev) +{ + int ret; + struct device *dev = &ofdev->dev; + + ret = sysfs_create_group(&dev->kobj, &qman_dev_attr_grp); + if (ret) + goto done; + ret = sysfs_add_file_to_group(&dev->kobj, + &dev_attr_sfdr_in_use.attr, qman_dev_attr_grp.name); + if (ret) + goto del_group_0; + ret = sysfs_create_group(&dev->kobj, &qman_dev_ecr_grp); + if (ret) + goto del_group_0; + + goto done; + +del_group_0: + sysfs_remove_group(&dev->kobj, &qman_dev_attr_grp); +done: + if (ret) + dev_err(dev, "Cannot create dev attributes ret=%d\n", ret); + return ret; +}; + +static const struct of_device_id of_fsl_qman_ids[] = { + { + .compatible = "fsl,qman", + }, + {} +}; +MODULE_DEVICE_TABLE(of, of_fsl_qman_ids); + +static struct platform_driver of_fsl_qman_driver = { + .driver = { + .name = DRV_NAME, + .of_match_table = of_fsl_qman_ids, + }, + .probe = of_fsl_qman_probe, + .remove = of_fsl_qman_remove, +}; + +module_platform_driver(of_fsl_qman_driver); + +#endif /* CONFIG_SYSFS */ diff --git a/drivers/staging/fsl_qbman/qman_driver.c b/drivers/staging/fsl_qbman/qman_driver.c new file mode 100644 index 0000000..5ca6221 --- /dev/null +++ b/drivers/staging/fsl_qbman/qman_driver.c @@ -0,0 +1,503 @@ +/* Copyright 2008-2012 Freescale Semiconductor, Inc. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions are met: + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * * Neither the name of Freescale Semiconductor nor the + * names of its contributors may be used to endorse or promote products + * derived from this software without specific prior written permission. + * + * + * ALTERNATIVELY, this software may be distributed under the terms of the + * GNU General Public License ("GPL") as published by the Free Software + * Foundation, either version 2 of that License or (at your option) any + * later version. + * + * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED + * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE + * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY + * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; + * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND + * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#include "qman_private.h" + +/* Global variable containing revision id (even on non-control plane systems + * where CCSR isn't available) */ +u16 qman_ip_rev; +EXPORT_SYMBOL(qman_ip_rev); +u16 qm_channel_pool1 = QMAN_CHANNEL_POOL1; +EXPORT_SYMBOL(qm_channel_pool1); +u16 qm_channel_caam = QMAN_CHANNEL_CAAM; +EXPORT_SYMBOL(qm_channel_caam); +u16 qm_channel_pme = QMAN_CHANNEL_PME; +EXPORT_SYMBOL(qm_channel_pme); +u16 qm_channel_dce = QMAN_CHANNEL_DCE; +EXPORT_SYMBOL(qm_channel_dce); + +/* size of the fqd region in bytes */ +#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP +static u32 fqd_size = (PAGE_SIZE << CONFIG_FSL_QMAN_FQD_SZ); +#endif + +/* For these variables, and the portal-initialisation logic, the + * comments in bman_driver.c apply here so won't be repeated. */ +static struct qman_portal *shared_portals[NR_CPUS]; +static int num_shared_portals; +static int shared_portals_idx; +static LIST_HEAD(unused_pcfgs); + +/* A SDQCR mask comprising all the available/visible pool channels */ +static u32 pools_sdqcr; + +#define STR_ERR_NOPROP "No '%s' property in node %s\n" +#define STR_ERR_CELL "'%s' is not a %d-cell range in node %s\n" + +static void qman_get_ip_revision(struct device_node *dn) +{ + u16 ip_rev = 0; + for_each_compatible_node(dn, NULL, "fsl,qman-portal") { + if (!of_device_is_available(dn)) + continue; + if (of_device_is_compatible(dn, "fsl,qman-portal-1.0") || + of_device_is_compatible(dn, "fsl,qman-portal-1.0.0")) { + pr_err("QMAN rev1.0 on P4080 rev1 is not supported!\n"); + BUG_ON(1); + } else if (of_device_is_compatible(dn, "fsl,qman-portal-1.1") || + of_device_is_compatible(dn, "fsl,qman-portal-1.1.0")) { + ip_rev = QMAN_REV11; + } else if (of_device_is_compatible(dn, "fsl,qman-portal-1.2") || + of_device_is_compatible(dn, "fsl,qman-portal-1.2.0")) { + ip_rev = QMAN_REV12; + } else if (of_device_is_compatible(dn, "fsl,qman-portal-2.0") || + of_device_is_compatible(dn, "fsl,qman-portal-2.0.0")) { + ip_rev = QMAN_REV20; + } else if (of_device_is_compatible(dn, + "fsl,qman-portal-3.0.0")) { + ip_rev = QMAN_REV30; + } else if (of_device_is_compatible(dn, + "fsl,qman-portal-3.0.1")) { + ip_rev = QMAN_REV30; + } else if (of_device_is_compatible(dn, + "fsl,qman-portal-3.1.0")) { + ip_rev = QMAN_REV31; + } else if (of_device_is_compatible(dn, + "fsl,qman-portal-3.1.1")) { + ip_rev = QMAN_REV31; + } else if (of_device_is_compatible(dn, + "fsl,qman-portal-3.1.2")) { + ip_rev = QMAN_REV31; + } else if (of_device_is_compatible(dn, + "fsl,qman-portal-3.1.3")) { + ip_rev = QMAN_REV31; + } else { + pr_warn("unknown QMan version in portal node," + "default to rev1.1\n"); + ip_rev = QMAN_REV11; + } + + if (!qman_ip_rev) { + if (ip_rev) { + qman_ip_rev = ip_rev; + } else { + pr_warn("unknown Qman version," + " default to rev1.1\n"); + qman_ip_rev = QMAN_REV11; + } + } else if (ip_rev && (qman_ip_rev != ip_rev)) + pr_warn("Revision=0x%04x, but portal '%s' has" + " 0x%04x\n", + qman_ip_rev, dn->full_name, ip_rev); + if (qman_ip_rev == ip_rev) + break; + } +} + +/* Parse a portal node, perform generic mapping duties and return the config. It + * is not known at this stage for what purpose (or even if) the portal will be + * used. */ +static struct qm_portal_config * __init parse_pcfg(struct device_node *node) +{ + struct qm_portal_config *pcfg; + const u32 *channel; + int irq, ret; + + pcfg = kmalloc(sizeof(*pcfg), GFP_KERNEL); + if (!pcfg) { + pr_err("can't allocate portal config"); + return NULL; + } + + /* + * This is a *horrible hack*, but the IOMMU/PAMU driver needs a + * 'struct device' in order to get the PAMU stashing setup and the QMan + * portal [driver] won't function at all without ring stashing + * + * Making the QMan portal driver nice and proper is part of the + * upstreaming effort + */ + pcfg->dev.bus = &platform_bus_type; + pcfg->dev.of_node = node; +#ifdef CONFIG_IOMMU_API + pcfg->dev.archdata.iommu_domain = NULL; +#endif + + ret = of_address_to_resource(node, DPA_PORTAL_CE, + &pcfg->addr_phys[DPA_PORTAL_CE]); + if (ret) { + pr_err("Can't get %s property '%s'\n", node->full_name, + "reg::CE"); + goto err; + } + ret = of_address_to_resource(node, DPA_PORTAL_CI, + &pcfg->addr_phys[DPA_PORTAL_CI]); + if (ret) { + pr_err("Can't get %s property '%s'\n", node->full_name, + "reg::CI"); + goto err; + } + + channel = of_get_property(node, "fsl,qman-channel-id", &ret); + if (!channel || (ret != 4)) { + pr_err("Can't get %s property '%s'\n", node->full_name, + "fsl,qman-channel-id"); + goto err; + } + pcfg->public_cfg.channel = *channel; + pcfg->public_cfg.cpu = -1; + irq = irq_of_parse_and_map(node, 0); + if (irq == NO_IRQ) { + pr_err("Can't get %s property '%s'\n", node->full_name, + "interrupts"); + goto err; + } + pcfg->public_cfg.irq = irq; +#ifdef CONFIG_FSL_QMAN_CONFIG + /* We need the same LIODN offset for all portals */ + qman_liodn_fixup(pcfg->public_cfg.channel); +#endif + + pcfg->addr_virt[DPA_PORTAL_CE] = ioremap_prot( + pcfg->addr_phys[DPA_PORTAL_CE].start, + resource_size(&pcfg->addr_phys[DPA_PORTAL_CE]), + 0); + pcfg->addr_virt[DPA_PORTAL_CI] = ioremap_prot( + pcfg->addr_phys[DPA_PORTAL_CI].start, + resource_size(&pcfg->addr_phys[DPA_PORTAL_CI]), + _PAGE_GUARDED | _PAGE_NO_CACHE); + + return pcfg; +err: + kfree(pcfg); + return NULL; +} + +static struct qm_portal_config *get_pcfg(struct list_head *list) +{ + struct qm_portal_config *pcfg; + if (list_empty(list)) + return NULL; + pcfg = list_entry(list->prev, struct qm_portal_config, list); + list_del(&pcfg->list); + return pcfg; +} + +static void portal_set_cpu(struct qm_portal_config *pcfg, int cpu) +{ +#ifdef CONFIG_FSL_PAMU + int ret; + int window_count = 1; + struct iommu_domain_geometry geom_attr; + struct pamu_stash_attribute stash_attr; + + pcfg->iommu_domain = iommu_domain_alloc(&platform_bus_type); + if (!pcfg->iommu_domain) { + pr_err(KBUILD_MODNAME ":%s(): iommu_domain_alloc() failed", + __func__); + goto _no_iommu; + } + geom_attr.aperture_start = 0; + geom_attr.aperture_end = + ((dma_addr_t)1 << min(8 * sizeof(dma_addr_t), (size_t)36)) - 1; + geom_attr.force_aperture = true; + ret = iommu_domain_set_attr(pcfg->iommu_domain, DOMAIN_ATTR_GEOMETRY, + &geom_attr); + if (ret < 0) { + pr_err(KBUILD_MODNAME ":%s(): iommu_domain_set_attr() = %d", + __func__, ret); + goto _iommu_domain_free; + } + ret = iommu_domain_set_attr(pcfg->iommu_domain, DOMAIN_ATTR_WINDOWS, + &window_count); + if (ret < 0) { + pr_err(KBUILD_MODNAME ":%s(): iommu_domain_set_attr() = %d", + __func__, ret); + goto _iommu_domain_free; + } + stash_attr.cpu = cpu; + stash_attr.cache = PAMU_ATTR_CACHE_L1; + ret = iommu_domain_set_attr(pcfg->iommu_domain, + DOMAIN_ATTR_FSL_PAMU_STASH, + &stash_attr); + if (ret < 0) { + pr_err(KBUILD_MODNAME ":%s(): iommu_domain_set_attr() = %d", + __func__, ret); + goto _iommu_domain_free; + } + ret = iommu_domain_window_enable(pcfg->iommu_domain, 0, 0, 1ULL << 36, + IOMMU_READ | IOMMU_WRITE); + if (ret < 0) { + pr_err(KBUILD_MODNAME ":%s(): iommu_domain_window_enable() = %d", + __func__, ret); + goto _iommu_domain_free; + } + ret = iommu_attach_device(pcfg->iommu_domain, &pcfg->dev); + if (ret < 0) { + pr_err(KBUILD_MODNAME ":%s(): iommu_device_attach() = %d", + __func__, ret); + goto _iommu_domain_free; + } + ret = iommu_domain_set_attr(pcfg->iommu_domain, + DOMAIN_ATTR_FSL_PAMU_ENABLE, + &window_count); + if (ret < 0) { + pr_err(KBUILD_MODNAME ":%s(): iommu_domain_set_attr() = %d", + __func__, ret); + goto _iommu_detach_device; + } + +_no_iommu: +#endif +#ifdef CONFIG_FSL_QMAN_CONFIG + if (qman_set_sdest(pcfg->public_cfg.channel, cpu)) +#endif + pr_warn("Failed to set QMan portal's stash request queue\n"); + + return; + +#ifdef CONFIG_FSL_PAMU +_iommu_detach_device: + iommu_detach_device(pcfg->iommu_domain, NULL); +_iommu_domain_free: + iommu_domain_free(pcfg->iommu_domain); + pcfg->iommu_domain = NULL; +#endif +} + +static struct qman_portal *init_pcfg(struct qm_portal_config *pcfg) +{ + struct qman_portal *p; + + pcfg->iommu_domain = NULL; + portal_set_cpu(pcfg, pcfg->public_cfg.cpu); + p = qman_create_affine_portal(pcfg, NULL); + if (p) { + u32 irq_sources = 0; + /* Determine what should be interrupt-vs-poll driven */ +#ifdef CONFIG_FSL_DPA_PIRQ_SLOW + irq_sources |= QM_PIRQ_EQCI | QM_PIRQ_EQRI | QM_PIRQ_MRI | + QM_PIRQ_CSCI; +#endif +#ifdef CONFIG_FSL_DPA_PIRQ_FAST + irq_sources |= QM_PIRQ_DQRI; +#endif + qman_p_irqsource_add(p, irq_sources); + pr_info("Qman portal %sinitialised, cpu %d\n", + pcfg->public_cfg.is_shared ? "(shared) " : "", + pcfg->public_cfg.cpu); + } else + pr_crit("Qman portal failure on cpu %d\n", + pcfg->public_cfg.cpu); + return p; +} + +static void init_slave(int cpu) +{ + struct qman_portal *p; + struct cpumask oldmask = *tsk_cpus_allowed(current); + set_cpus_allowed_ptr(current, get_cpu_mask(cpu)); + p = qman_create_affine_slave(shared_portals[shared_portals_idx++], cpu); + if (!p) + pr_err("Qman slave portal failure on cpu %d\n", cpu); + else + pr_info("Qman portal %sinitialised, cpu %d\n", "(slave) ", cpu); + set_cpus_allowed_ptr(current, &oldmask); + if (shared_portals_idx >= num_shared_portals) + shared_portals_idx = 0; +} + +static struct cpumask want_unshared __initdata; +static struct cpumask want_shared __initdata; + +static int __init parse_qportals(char *str) +{ + return parse_portals_bootarg(str, &want_shared, &want_unshared, + "qportals"); +} +__setup("qportals=", parse_qportals); + +static void qman_portal_update_sdest(const struct qm_portal_config *pcfg, + unsigned int cpu) +{ + struct pamu_stash_attribute stash_attr; + int ret; + + if (pcfg->iommu_domain) { + stash_attr.cpu = cpu; + stash_attr.cache = PAMU_ATTR_CACHE_L1; + ret = iommu_domain_set_attr(pcfg->iommu_domain, + DOMAIN_ATTR_FSL_PAMU_STASH, &stash_attr); + if (ret < 0) { + pr_err("Failed to update pamu stash setting\n"); + return; + } + } +#ifdef CONFIG_FSL_QMAN_CONFIG + if (qman_set_sdest(pcfg->public_cfg.channel, cpu)) +#endif + pr_warn("Failed to update portal's stash request queue\n"); +} + +static void qman_offline_cpu(unsigned int cpu) +{ + struct qman_portal *p; + const struct qm_portal_config *pcfg; + p = (struct qman_portal *)affine_portals[cpu]; + if (p) { + pcfg = qman_get_qm_portal_config(p); + if (pcfg) { + irq_set_affinity(pcfg->public_cfg.irq, cpumask_of(0)); + qman_portal_update_sdest(pcfg, 0); + } + } +} + +__init int qman_init(void) +{ + struct cpumask slave_cpus; + struct cpumask unshared_cpus = *cpu_none_mask; + struct cpumask shared_cpus = *cpu_none_mask; + LIST_HEAD(unshared_pcfgs); + LIST_HEAD(shared_pcfgs); + struct device_node *dn; + struct qm_portal_config *pcfg; + struct qman_portal *p; + int cpu; +#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP + int ret; +#endif + struct cpumask offline_cpus; + + /* Initialise the Qman (CCSR) device */ + for_each_compatible_node(dn, NULL, "fsl,qman") { + if (!qman_init_ccsr(dn)) + pr_info("Qman err interrupt handler present\n"); + else + pr_err("Qman CCSR setup failed\n"); + } +#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP + /* Setup lookup table for FQ demux */ + ret = qman_setup_fq_lookup_table(fqd_size/64); + if (ret) + return ret; +#endif + + /* Get qman ip revision */ + qman_get_ip_revision(dn); + if ((qman_ip_rev & 0xff00) >= QMAN_REV30) { + qm_channel_pool1 = QMAN_CHANNEL_POOL1_REV3; + qm_channel_caam = QMAN_CHANNEL_CAAM_REV3; + qm_channel_pme = QMAN_CHANNEL_PME_REV3; + } + + memset(affine_portals, 0, sizeof(void *) * num_possible_cpus()); + /* Initialise portals. See bman_driver.c for comments */ + for_each_compatible_node(dn, NULL, "fsl,qman-portal") { + if (!of_device_is_available(dn)) + continue; + pcfg = parse_pcfg(dn); + if (pcfg) { + pcfg->public_cfg.pools = pools_sdqcr; + list_add_tail(&pcfg->list, &unused_pcfgs); + } + } + for_each_possible_cpu(cpu) { + if (cpumask_test_cpu(cpu, &want_shared)) { + pcfg = get_pcfg(&unused_pcfgs); + if (!pcfg) + break; + pcfg->public_cfg.cpu = cpu; + list_add_tail(&pcfg->list, &shared_pcfgs); + cpumask_set_cpu(cpu, &shared_cpus); + } + if (cpumask_test_cpu(cpu, &want_unshared)) { + if (cpumask_test_cpu(cpu, &shared_cpus)) + continue; + pcfg = get_pcfg(&unused_pcfgs); + if (!pcfg) + break; + pcfg->public_cfg.cpu = cpu; + list_add_tail(&pcfg->list, &unshared_pcfgs); + cpumask_set_cpu(cpu, &unshared_cpus); + } + } + if (list_empty(&shared_pcfgs) && list_empty(&unshared_pcfgs)) { + for_each_possible_cpu(cpu) { + pcfg = get_pcfg(&unused_pcfgs); + if (!pcfg) + break; + pcfg->public_cfg.cpu = cpu; + list_add_tail(&pcfg->list, &unshared_pcfgs); + cpumask_set_cpu(cpu, &unshared_cpus); + } + } + cpumask_andnot(&slave_cpus, cpu_possible_mask, &shared_cpus); + cpumask_andnot(&slave_cpus, &slave_cpus, &unshared_cpus); + if (cpumask_empty(&slave_cpus)) { + if (!list_empty(&shared_pcfgs)) { + cpumask_or(&unshared_cpus, &unshared_cpus, + &shared_cpus); + cpumask_clear(&shared_cpus); + list_splice_tail(&shared_pcfgs, &unshared_pcfgs); + INIT_LIST_HEAD(&shared_pcfgs); + } + } else { + if (list_empty(&shared_pcfgs)) { + pcfg = get_pcfg(&unshared_pcfgs); + if (!pcfg) { + pr_crit("No QMan portals available!\n"); + return 0; + } + cpumask_clear_cpu(pcfg->public_cfg.cpu, &unshared_cpus); + cpumask_set_cpu(pcfg->public_cfg.cpu, &shared_cpus); + list_add_tail(&pcfg->list, &shared_pcfgs); + } + } + list_for_each_entry(pcfg, &unshared_pcfgs, list) { + pcfg->public_cfg.is_shared = 0; + p = init_pcfg(pcfg); + } + list_for_each_entry(pcfg, &shared_pcfgs, list) { + pcfg->public_cfg.is_shared = 1; + p = init_pcfg(pcfg); + if (p) + shared_portals[num_shared_portals++] = p; + } + if (!cpumask_empty(&slave_cpus)) + for_each_cpu(cpu, &slave_cpus) + init_slave(cpu); + pr_info("Qman portals initialised\n"); + cpumask_andnot(&offline_cpus, cpu_possible_mask, cpu_online_mask); + for_each_cpu(cpu, &offline_cpus) + qman_offline_cpu(cpu); + return 0; +} diff --git a/drivers/staging/fsl_qbman/qman_high.c b/drivers/staging/fsl_qbman/qman_high.c new file mode 100644 index 0000000..39840be --- /dev/null +++ b/drivers/staging/fsl_qbman/qman_high.c @@ -0,0 +1,2566 @@ +/* Copyright 2008-2012 Freescale Semiconductor, Inc. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions are met: + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * * Neither the name of Freescale Semiconductor nor the + * names of its contributors may be used to endorse or promote products + * derived from this software without specific prior written permission. + * + * + * ALTERNATIVELY, this software may be distributed under the terms of the + * GNU General Public License ("GPL") as published by the Free Software + * Foundation, either version 2 of that License or (at your option) any + * later version. + * + * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED + * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE + * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY + * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; + * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND + * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#include "qman_low.h" + +/* Compilation constants */ +#define DQRR_MAXFILL 15 +#define EQCR_ITHRESH 4 /* if EQCR congests, interrupt threshold */ +#define IRQNAME "QMan portal %d" +#define MAX_IRQNAME 16 /* big enough for "QMan portal %d" */ + +/* Divide 'n' by 'd', rounding down if 'r' is negative, rounding up if it's + * positive, and rounding to the closest value if it's zero. NB, this macro + * implicitly upgrades parameters to unsigned 64-bit, so feed it with types + * that are compatible with this. NB, these arguments should not be expressions + * unless it is safe for them to be evaluated multiple times. Eg. do not pass + * in "some_value++" as a parameter to the macro! */ +#define ROUNDING(n, d, r) \ + (((r) < 0) ? div64_u64((n), (d)) : \ + (((r) > 0) ? div64_u64(((n) + (d) - 1), (d)) : \ + div64_u64(((n) + ((d) / 2)), (d)))) + +/* Lock/unlock frame queues, subject to the "LOCKED" flag. This is about + * inter-processor locking only. Note, FQLOCK() is always called either under a + * local_irq_save() or from interrupt context - hence there's no need for irq + * protection (and indeed, attempting to nest irq-protection doesn't work, as + * the "irq en/disable" machinery isn't recursive...). */ +#define FQLOCK(fq) \ + do { \ + struct qman_fq *__fq478 = (fq); \ + if (fq_isset(__fq478, QMAN_FQ_FLAG_LOCKED)) \ + spin_lock(&__fq478->fqlock); \ + } while (0) +#define FQUNLOCK(fq) \ + do { \ + struct qman_fq *__fq478 = (fq); \ + if (fq_isset(__fq478, QMAN_FQ_FLAG_LOCKED)) \ + spin_unlock(&__fq478->fqlock); \ + } while (0) + +static inline void fq_set(struct qman_fq *fq, u32 mask) +{ + set_bits(mask, &fq->flags); +} +static inline void fq_clear(struct qman_fq *fq, u32 mask) +{ + clear_bits(mask, &fq->flags); +} +static inline int fq_isset(struct qman_fq *fq, u32 mask) +{ + return fq->flags & mask; +} +static inline int fq_isclear(struct qman_fq *fq, u32 mask) +{ + return !(fq->flags & mask); +} + +struct qman_portal { + struct qm_portal p; + unsigned long bits; /* PORTAL_BITS_*** - dynamic, strictly internal */ + unsigned long irq_sources; + u32 use_eqcr_ci_stashing; + u32 slowpoll; /* only used when interrupts are off */ + struct qman_fq *vdqcr_owned; /* only 1 volatile dequeue at a time */ +#ifdef CONFIG_FSL_DPA_CAN_WAIT_SYNC + struct qman_fq *eqci_owned; /* only 1 enqueue WAIT_SYNC at a time */ +#endif +#ifdef CONFIG_FSL_DPA_PORTAL_SHARE + raw_spinlock_t sharing_lock; /* only used if is_shared */ + int is_shared; + struct qman_portal *sharing_redirect; +#endif + u32 sdqcr; + int dqrr_disable_ref; + /* A portal-specific handler for DCP ERNs. If this is NULL, the global + * handler is called instead. */ + qman_cb_dc_ern cb_dc_ern; + /* When the cpu-affine portal is activated, this is non-NULL */ + const struct qm_portal_config *config; + /* This is needed for providing a non-NULL device to dma_map_***() */ + struct platform_device *pdev; + struct dpa_rbtree retire_table; + char irqname[MAX_IRQNAME]; + /* 2-element array. cgrs[0] is mask, cgrs[1] is snapshot. */ + struct qman_cgrs *cgrs; + /* linked-list of CSCN handlers. */ + struct list_head cgr_cbs; + /* list lock */ + spinlock_t cgr_lock; + /* track if memory was allocated by the driver */ + u8 alloced; +}; + +#ifdef CONFIG_FSL_DPA_PORTAL_SHARE +#define PORTAL_IRQ_LOCK(p, irqflags) \ + do { \ + if ((p)->is_shared) \ + raw_spin_lock_irqsave(&(p)->sharing_lock, irqflags); \ + else \ + local_irq_save(irqflags); \ + } while (0) +#define PORTAL_IRQ_UNLOCK(p, irqflags) \ + do { \ + if ((p)->is_shared) \ + raw_spin_unlock_irqrestore(&(p)->sharing_lock, \ + irqflags); \ + else \ + local_irq_restore(irqflags); \ + } while (0) +#else +#define PORTAL_IRQ_LOCK(p, irqflags) local_irq_save(irqflags) +#define PORTAL_IRQ_UNLOCK(p, irqflags) local_irq_restore(irqflags) +#endif + +/* Global handler for DCP ERNs. Used when the portal receiving the message does + * not have a portal-specific handler. */ +static qman_cb_dc_ern cb_dc_ern; + +static cpumask_t affine_mask; +static DEFINE_SPINLOCK(affine_mask_lock); +static u16 affine_channels[NR_CPUS]; +static DEFINE_PER_CPU(struct qman_portal, qman_affine_portal); +void *affine_portals[NR_CPUS]; + +/* "raw" gets the cpu-local struct whether it's a redirect or not. */ +static inline struct qman_portal *get_raw_affine_portal(void) +{ + return &get_cpu_var(qman_affine_portal); +} +/* For ops that can redirect, this obtains the portal to use */ +#ifdef CONFIG_FSL_DPA_PORTAL_SHARE +static inline struct qman_portal *get_affine_portal(void) +{ + struct qman_portal *p = get_raw_affine_portal(); + if (p->sharing_redirect) + return p->sharing_redirect; + return p; +} +#else +#define get_affine_portal() get_raw_affine_portal() +#endif +/* For every "get", there must be a "put" */ +static inline void put_affine_portal(void) +{ + put_cpu_var(qman_affine_portal); +} +/* Exception: poll functions assume the caller is cpu-affine and in no risk of + * re-entrance, which are the two reasons we usually use the get/put_cpu_var() + * semantic - ie. to disable pre-emption. Some use-cases expect the execution + * context to remain as non-atomic during poll-triggered callbacks as it was + * when the poll API was first called (eg. NAPI), so we go out of our way in + * this case to not disable pre-emption. */ +static inline struct qman_portal *get_poll_portal(void) +{ + return &__get_cpu_var(qman_affine_portal); +} +#define put_poll_portal() + +/* This gives a FQID->FQ lookup to cover the fact that we can't directly demux + * retirement notifications (the fact they are sometimes h/w-consumed means that + * contextB isn't always a s/w demux - and as we can't know which case it is + * when looking at the notification, we have to use the slow lookup for all of + * them). NB, it's possible to have multiple FQ objects refer to the same FQID + * (though at most one of them should be the consumer), so this table isn't for + * all FQs - FQs are added when retirement commands are issued, and removed when + * they complete, which also massively reduces the size of this table. */ +IMPLEMENT_DPA_RBTREE(fqtree, struct qman_fq, node, fqid); + +/* This is what everything can wait on, even if it migrates to a different cpu + * to the one whose affine portal it is waiting on. */ +static DECLARE_WAIT_QUEUE_HEAD(affine_queue); + +static inline int table_push_fq(struct qman_portal *p, struct qman_fq *fq) +{ + int ret = fqtree_push(&p->retire_table, fq); + if (ret) + pr_err("ERROR: double FQ-retirement %d\n", fq->fqid); + return ret; +} + +static inline void table_del_fq(struct qman_portal *p, struct qman_fq *fq) +{ + fqtree_del(&p->retire_table, fq); +} + +static inline struct qman_fq *table_find_fq(struct qman_portal *p, u32 fqid) +{ + return fqtree_find(&p->retire_table, fqid); +} + +#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP +static void **qman_fq_lookup_table; +static size_t qman_fq_lookup_table_size; + +int qman_setup_fq_lookup_table(size_t num_entries) +{ + num_entries++; + /* Allocate 1 more entry since the first entry is not used */ + qman_fq_lookup_table = vzalloc((num_entries * sizeof(void *))); + if (!qman_fq_lookup_table) { + pr_err("QMan: Could not allocate fq lookup table\n"); + return -ENOMEM; + } + qman_fq_lookup_table_size = num_entries; + pr_info("QMan: Allocated lookup table at %p, entry count %lu\n", + qman_fq_lookup_table, + (unsigned long)qman_fq_lookup_table_size); + return 0; +} + +/* global structure that maintains fq object mapping */ +static DEFINE_SPINLOCK(fq_hash_table_lock); + +static int find_empty_fq_table_entry(u32 *entry, struct qman_fq *fq) +{ + u32 i; + + spin_lock(&fq_hash_table_lock); + /* Can't use index zero because this has special meaning + * in context_b field. */ + for (i = 1; i < qman_fq_lookup_table_size; i++) { + if (qman_fq_lookup_table[i] == NULL) { + *entry = i; + qman_fq_lookup_table[i] = fq; + spin_unlock(&fq_hash_table_lock); + return 0; + } + } + spin_unlock(&fq_hash_table_lock); + return -ENOMEM; +} + +static void clear_fq_table_entry(u32 entry) +{ + spin_lock(&fq_hash_table_lock); + BUG_ON(entry >= qman_fq_lookup_table_size); + qman_fq_lookup_table[entry] = NULL; + spin_unlock(&fq_hash_table_lock); +} + +static inline struct qman_fq *get_fq_table_entry(u32 entry) +{ + BUG_ON(entry >= qman_fq_lookup_table_size); + return qman_fq_lookup_table[entry]; +} +#endif + +/* In the case that slow- and fast-path handling are both done by qman_poll() + * (ie. because there is no interrupt handling), we ought to balance how often + * we do the fast-path poll versus the slow-path poll. We'll use two decrementer + * sources, so we call the fast poll 'n' times before calling the slow poll + * once. The idle decrementer constant is used when the last slow-poll detected + * no work to do, and the busy decrementer constant when the last slow-poll had + * work to do. */ +#define SLOW_POLL_IDLE 1000 +#define SLOW_POLL_BUSY 10 +static u32 __poll_portal_slow(struct qman_portal *p, u32 is); +static inline unsigned int __poll_portal_fast(struct qman_portal *p, + unsigned int poll_limit); + +/* Portal interrupt handler */ +static irqreturn_t portal_isr(__always_unused int irq, void *ptr) +{ + struct qman_portal *p = ptr; + /* + * The CSCI source is cleared inside __poll_portal_slow(), because + * it could race against a Query Congestion State command also given + * as part of the handling of this interrupt source. We mustn't + * clear it a second time in this top-level function. + */ + u32 clear = QM_DQAVAIL_MASK | (p->irq_sources & ~QM_PIRQ_CSCI); + u32 is = qm_isr_status_read(&p->p) & p->irq_sources; + /* DQRR-handling if it's interrupt-driven */ + if (is & QM_PIRQ_DQRI) + __poll_portal_fast(p, CONFIG_FSL_QMAN_POLL_LIMIT); + /* Handling of anything else that's interrupt-driven */ + clear |= __poll_portal_slow(p, is); + qm_isr_status_clear(&p->p, clear); + return IRQ_HANDLED; +} + +/* This inner version is used privately by qman_create_affine_portal(), as well + * as by the exported qman_stop_dequeues(). */ +static inline void qman_stop_dequeues_ex(struct qman_portal *p) +{ + unsigned long irqflags __maybe_unused; + PORTAL_IRQ_LOCK(p, irqflags); + if (!(p->dqrr_disable_ref++)) + qm_dqrr_set_maxfill(&p->p, 0); + PORTAL_IRQ_UNLOCK(p, irqflags); +} + +static int drain_mr_fqrni(struct qm_portal *p) +{ + const struct qm_mr_entry *msg; +loop: + msg = qm_mr_current(p); + if (!msg) { + /* if MR was full and h/w had other FQRNI entries to produce, we + * need to allow it time to produce those entries once the + * existing entries are consumed. A worst-case situation + * (fully-loaded system) means h/w sequencers may have to do 3-4 + * other things before servicing the portal's MR pump, each of + * which (if slow) may take ~50 qman cycles (which is ~200 + * processor cycles). So rounding up and then multiplying this + * worst-case estimate by a factor of 10, just to be + * ultra-paranoid, goes as high as 10,000 cycles. NB, we consume + * one entry at a time, so h/w has an opportunity to produce new + * entries well before the ring has been fully consumed, so + * we're being *really* paranoid here. */ + u64 now, then = mfatb(); + do { + now = mfatb(); + } while ((then + 10000) > now); + msg = qm_mr_current(p); + if (!msg) + return 0; + } + if ((msg->verb & QM_MR_VERB_TYPE_MASK) != QM_MR_VERB_FQRNI) { + /* We aren't draining anything but FQRNIs */ + pr_err("QMan found verb 0x%x in MR\n", msg->verb); + return -1; + } + qm_mr_next(p); + qm_mr_cci_consume(p, 1); + goto loop; +} + +struct qman_portal *qman_create_portal( + struct qman_portal *portal, + const struct qm_portal_config *config, + const struct qman_cgrs *cgrs) +{ + struct qm_portal *__p; + char buf[16]; + int ret; + u32 isdr; + + if (!portal) { + portal = kmalloc(sizeof(*portal), GFP_KERNEL); + if (!portal) + return portal; + portal->alloced = 1; + } else + portal->alloced = 0; + + __p = &portal->p; + + portal->use_eqcr_ci_stashing = ((qman_ip_rev >= QMAN_REV30) ? + 1 : 0); + + /* prep the low-level portal struct with the mapped addresses from the + * config, everything that follows depends on it and "config" is more + * for (de)reference... */ + __p->addr.addr_ce = config->addr_virt[DPA_PORTAL_CE]; + __p->addr.addr_ci = config->addr_virt[DPA_PORTAL_CI]; + /* + * If CI-stashing is used, the current defaults use a threshold of 3, + * and stash with high-than-DQRR priority. + */ + if (qm_eqcr_init(__p, qm_eqcr_pvb, + portal->use_eqcr_ci_stashing ? 3 : 0, 1)) { + pr_err("Qman EQCR initialisation failed\n"); + goto fail_eqcr; + } + if (qm_dqrr_init(__p, config, qm_dqrr_dpush, qm_dqrr_pvb, + qm_dqrr_cdc, DQRR_MAXFILL)) { + pr_err("Qman DQRR initialisation failed\n"); + goto fail_dqrr; + } + if (qm_mr_init(__p, qm_mr_pvb, qm_mr_cci)) { + pr_err("Qman MR initialisation failed\n"); + goto fail_mr; + } + if (qm_mc_init(__p)) { + pr_err("Qman MC initialisation failed\n"); + goto fail_mc; + } + if (qm_isr_init(__p)) { + pr_err("Qman ISR initialisation failed\n"); + goto fail_isr; + } + /* static interrupt-gating controls */ + qm_dqrr_set_ithresh(__p, CONFIG_FSL_QMAN_PIRQ_DQRR_ITHRESH); + qm_mr_set_ithresh(__p, CONFIG_FSL_QMAN_PIRQ_MR_ITHRESH); + qm_isr_set_iperiod(__p, CONFIG_FSL_QMAN_PIRQ_IPERIOD); + portal->cgrs = kmalloc(2 * sizeof(*cgrs), GFP_KERNEL); + if (!portal->cgrs) + goto fail_cgrs; + /* initial snapshot is no-depletion */ + qman_cgrs_init(&portal->cgrs[1]); + if (cgrs) + portal->cgrs[0] = *cgrs; + else + /* if the given mask is NULL, assume all CGRs can be seen */ + qman_cgrs_fill(&portal->cgrs[0]); + INIT_LIST_HEAD(&portal->cgr_cbs); + spin_lock_init(&portal->cgr_lock); + portal->bits = 0; + portal->slowpoll = 0; +#ifdef CONFIG_FSL_DPA_CAN_WAIT_SYNC + portal->eqci_owned = NULL; +#endif +#ifdef CONFIG_FSL_DPA_PORTAL_SHARE + raw_spin_lock_init(&portal->sharing_lock); + portal->is_shared = config->public_cfg.is_shared; + portal->sharing_redirect = NULL; +#endif + portal->sdqcr = QM_SDQCR_SOURCE_CHANNELS | QM_SDQCR_COUNT_UPTO3 | + QM_SDQCR_DEDICATED_PRECEDENCE | QM_SDQCR_TYPE_PRIO_QOS | + QM_SDQCR_TOKEN_SET(0xab) | QM_SDQCR_CHANNELS_DEDICATED; + portal->dqrr_disable_ref = 0; + portal->cb_dc_ern = NULL; + sprintf(buf, "qportal-%d", config->public_cfg.channel); + portal->pdev = platform_device_alloc(buf, -1); + if (!portal->pdev) + goto fail_devalloc; + if (dma_set_mask(&portal->pdev->dev, DMA_BIT_MASK(40))) + goto fail_devadd; + ret = platform_device_add(portal->pdev); + if (ret) + goto fail_devadd; + dpa_rbtree_init(&portal->retire_table); + isdr = 0xffffffff; + qm_isr_disable_write(__p, isdr); + portal->irq_sources = 0; + qm_isr_enable_write(__p, portal->irq_sources); + qm_isr_status_clear(__p, 0xffffffff); + snprintf(portal->irqname, MAX_IRQNAME, IRQNAME, config->public_cfg.cpu); + if (request_irq(config->public_cfg.irq, portal_isr, 0, portal->irqname, + portal)) { + pr_err("request_irq() failed\n"); + goto fail_irq; + } + if ((config->public_cfg.cpu != -1) && + irq_can_set_affinity(config->public_cfg.irq) && + irq_set_affinity(config->public_cfg.irq, + cpumask_of(config->public_cfg.cpu))) { + pr_err("irq_set_affinity() failed\n"); + goto fail_affinity; + } + + /* Need EQCR to be empty before continuing */ + isdr ^= QM_PIRQ_EQCI; + qm_isr_disable_write(__p, isdr); + ret = qm_eqcr_get_fill(__p); + if (ret) { + pr_err("Qman EQCR unclean\n"); + goto fail_eqcr_empty; + } + isdr ^= (QM_PIRQ_DQRI | QM_PIRQ_MRI); + qm_isr_disable_write(__p, isdr); + if (qm_dqrr_current(__p) != NULL) { + pr_err("Qman DQRR unclean\n"); + qm_dqrr_cdc_consume_n(__p, 0xffff); + } + if (qm_mr_current(__p) != NULL) { + /* special handling, drain just in case it's a few FQRNIs */ + if (drain_mr_fqrni(__p)) { + const struct qm_mr_entry *e = qm_mr_current(__p); + pr_err("Qman MR unclean, MR VERB 0x%x, " + "rc 0x%x\n, addr 0x%x", + e->verb, e->ern.rc, e->ern.fd.addr_lo); + goto fail_dqrr_mr_empty; + } + } + /* Success */ + portal->config = config; + qm_isr_disable_write(__p, 0); + qm_isr_uninhibit(__p); + /* Write a sane SDQCR */ + qm_dqrr_sdqcr_set(__p, portal->sdqcr); + return portal; +fail_dqrr_mr_empty: +fail_eqcr_empty: +fail_affinity: + free_irq(config->public_cfg.irq, portal); +fail_irq: + platform_device_del(portal->pdev); +fail_devadd: + platform_device_put(portal->pdev); +fail_devalloc: + kfree(portal->cgrs); +fail_cgrs: + qm_isr_finish(__p); +fail_isr: + qm_mc_finish(__p); +fail_mc: + qm_mr_finish(__p); +fail_mr: + qm_dqrr_finish(__p); +fail_dqrr: + qm_eqcr_finish(__p); +fail_eqcr: + return NULL; +} + +struct qman_portal *qman_create_affine_portal( + const struct qm_portal_config *config, + const struct qman_cgrs *cgrs) +{ + struct qman_portal *res; + struct qman_portal *portal; + + portal = &per_cpu(qman_affine_portal, config->public_cfg.cpu); + res = qman_create_portal(portal, config, cgrs); + if (res) { + spin_lock(&affine_mask_lock); + cpumask_set_cpu(config->public_cfg.cpu, &affine_mask); + affine_channels[config->public_cfg.cpu] = + config->public_cfg.channel; + affine_portals[config->public_cfg.cpu] = portal; + spin_unlock(&affine_mask_lock); + } + return res; +} + +/* These checks are BUG_ON()s because the driver is already supposed to avoid + * these cases. */ +struct qman_portal *qman_create_affine_slave(struct qman_portal *redirect, + int cpu) +{ +#ifdef CONFIG_FSL_DPA_PORTAL_SHARE + struct qman_portal *p; + p = &per_cpu(qman_affine_portal, cpu); + /* Check that we don't already have our own portal */ + BUG_ON(p->config); + /* Check that we aren't already slaving to another portal */ + BUG_ON(p->is_shared); + /* Check that 'redirect' is prepared to have us */ + BUG_ON(!redirect->config->public_cfg.is_shared); + /* These are the only elements to initialise when redirecting */ + p->irq_sources = 0; + p->sharing_redirect = redirect; + affine_portals[cpu] = p; + return p; +#else + BUG(); + return NULL; +#endif +} + +void qman_destroy_portal(struct qman_portal *qm) +{ + const struct qm_portal_config *pcfg; + + /* Stop dequeues on the portal */ + qm_dqrr_sdqcr_set(&qm->p, 0); + + /* NB we do this to "quiesce" EQCR. If we add enqueue-completions or + * something related to QM_PIRQ_EQCI, this may need fixing. + * Also, due to the prefetching model used for CI updates in the enqueue + * path, this update will only invalidate the CI cacheline *after* + * working on it, so we need to call this twice to ensure a full update + * irrespective of where the enqueue processing was at when the teardown + * began. */ + qm_eqcr_cce_update(&qm->p); + qm_eqcr_cce_update(&qm->p); + pcfg = qm->config; + + free_irq(pcfg->public_cfg.irq, qm); + + kfree(qm->cgrs); + qm_isr_finish(&qm->p); + qm_mc_finish(&qm->p); + qm_mr_finish(&qm->p); + qm_dqrr_finish(&qm->p); + qm_eqcr_finish(&qm->p); + + platform_device_del(qm->pdev); + platform_device_put(qm->pdev); + + qm->config = NULL; + if (qm->alloced) + kfree(qm); +} + +const struct qm_portal_config *qman_destroy_affine_portal(void) +{ + /* We don't want to redirect if we're a slave, use "raw" */ + struct qman_portal *qm = get_raw_affine_portal(); + const struct qm_portal_config *pcfg; + int cpu; +#ifdef CONFIG_FSL_DPA_PORTAL_SHARE + if (qm->sharing_redirect) { + qm->sharing_redirect = NULL; + put_affine_portal(); + return NULL; + } + qm->is_shared = 0; +#endif + pcfg = qm->config; + cpu = pcfg->public_cfg.cpu; + + qman_destroy_portal(qm); + + spin_lock(&affine_mask_lock); + cpumask_clear_cpu(cpu, &affine_mask); + spin_unlock(&affine_mask_lock); + put_affine_portal(); + return pcfg; +} + +const struct qman_portal_config *qman_p_get_portal_config(struct qman_portal *p) +{ + return &p->config->public_cfg; +} +EXPORT_SYMBOL(qman_p_get_portal_config); + +const struct qman_portal_config *qman_get_portal_config(void) +{ + struct qman_portal *p = get_affine_portal(); + const struct qman_portal_config *ret = qman_p_get_portal_config(p); + put_affine_portal(); + return ret; +} +EXPORT_SYMBOL(qman_get_portal_config); + +/* Inline helper to reduce nesting in __poll_portal_slow() */ +static inline void fq_state_change(struct qman_portal *p, struct qman_fq *fq, + const struct qm_mr_entry *msg, u8 verb) +{ + FQLOCK(fq); + switch (verb) { + case QM_MR_VERB_FQRL: + DPA_ASSERT(fq_isset(fq, QMAN_FQ_STATE_ORL)); + fq_clear(fq, QMAN_FQ_STATE_ORL); + table_del_fq(p, fq); + break; + case QM_MR_VERB_FQRN: + DPA_ASSERT((fq->state == qman_fq_state_parked) || + (fq->state == qman_fq_state_sched)); + DPA_ASSERT(fq_isset(fq, QMAN_FQ_STATE_CHANGING)); + fq_clear(fq, QMAN_FQ_STATE_CHANGING); + if (msg->fq.fqs & QM_MR_FQS_NOTEMPTY) + fq_set(fq, QMAN_FQ_STATE_NE); + if (msg->fq.fqs & QM_MR_FQS_ORLPRESENT) + fq_set(fq, QMAN_FQ_STATE_ORL); + else + table_del_fq(p, fq); + fq->state = qman_fq_state_retired; + break; + case QM_MR_VERB_FQPN: + DPA_ASSERT(fq->state == qman_fq_state_sched); + DPA_ASSERT(fq_isclear(fq, QMAN_FQ_STATE_CHANGING)); + fq->state = qman_fq_state_parked; + } + FQUNLOCK(fq); +} + +static u32 __poll_portal_slow(struct qman_portal *p, u32 is) +{ + const struct qm_mr_entry *msg; + + if (is & QM_PIRQ_CSCI) { + struct qman_cgrs rr, c; + struct qm_mc_result *mcr; + struct qman_cgr *cgr; + unsigned long irqflags __maybe_unused; + + spin_lock_irqsave(&p->cgr_lock, irqflags); + /* + * The CSCI bit must be cleared _before_ issuing the + * Query Congestion State command, to ensure that a long + * CGR State Change callback cannot miss an intervening + * state change. + */ + qm_isr_status_clear(&p->p, QM_PIRQ_CSCI); + qm_mc_start(&p->p); + qm_mc_commit(&p->p, QM_MCC_VERB_QUERYCONGESTION); + while (!(mcr = qm_mc_result(&p->p))) + cpu_relax(); + /* mask out the ones I'm not interested in */ + qman_cgrs_and(&rr, (const struct qman_cgrs *) + &mcr->querycongestion.state, &p->cgrs[0]); + /* check previous snapshot for delta, enter/exit congestion */ + qman_cgrs_xor(&c, &rr, &p->cgrs[1]); + /* update snapshot */ + qman_cgrs_cp(&p->cgrs[1], &rr); + /* Invoke callback */ + list_for_each_entry(cgr, &p->cgr_cbs, node) + if (cgr->cb && qman_cgrs_get(&c, cgr->cgrid)) + cgr->cb(p, cgr, qman_cgrs_get(&rr, cgr->cgrid)); + spin_unlock_irqrestore(&p->cgr_lock, irqflags); + } + +#ifdef CONFIG_FSL_DPA_CAN_WAIT_SYNC + if (is & QM_PIRQ_EQCI) { + unsigned long irqflags; + PORTAL_IRQ_LOCK(p, irqflags); + p->eqci_owned = NULL; + PORTAL_IRQ_UNLOCK(p, irqflags); + wake_up(&affine_queue); + } +#endif + + if (is & QM_PIRQ_EQRI) { + unsigned long irqflags __maybe_unused; + PORTAL_IRQ_LOCK(p, irqflags); + qm_eqcr_cce_update(&p->p); + qm_eqcr_set_ithresh(&p->p, 0); + PORTAL_IRQ_UNLOCK(p, irqflags); + wake_up(&affine_queue); + } + + if (is & QM_PIRQ_MRI) { + struct qman_fq *fq; + u8 verb, num = 0; +mr_loop: + qm_mr_pvb_update(&p->p); + msg = qm_mr_current(&p->p); + if (!msg) + goto mr_done; + verb = msg->verb & QM_MR_VERB_TYPE_MASK; + /* The message is a software ERN iff the 0x20 bit is set */ + if (verb & 0x20) { + switch (verb) { + case QM_MR_VERB_FQRNI: + /* nada, we drop FQRNIs on the floor */ + break; + case QM_MR_VERB_FQRN: + case QM_MR_VERB_FQRL: + /* Lookup in the retirement table */ + fq = table_find_fq(p, msg->fq.fqid); + BUG_ON(!fq); + fq_state_change(p, fq, msg, verb); + if (fq->cb.fqs) + fq->cb.fqs(p, fq, msg); + break; + case QM_MR_VERB_FQPN: + /* Parked */ +#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP + fq = get_fq_table_entry(msg->fq.contextB); +#else + fq = (void *)(uintptr_t)msg->fq.contextB; +#endif + fq_state_change(p, fq, msg, verb); + if (fq->cb.fqs) + fq->cb.fqs(p, fq, msg); + break; + case QM_MR_VERB_DC_ERN: + /* DCP ERN */ + if (p->cb_dc_ern) + p->cb_dc_ern(p, msg); + else if (cb_dc_ern) + cb_dc_ern(p, msg); + else { + static int warn_once; + if (!warn_once) { + pr_crit("Leaking DCP ERNs!\n"); + warn_once = 1; + } + } + break; + default: + pr_crit("Invalid MR verb 0x%02x\n", verb); + } + } else { + /* Its a software ERN */ +#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP + fq = get_fq_table_entry(msg->ern.tag); +#else + fq = (void *)(uintptr_t)msg->ern.tag; +#endif + fq->cb.ern(p, fq, msg); + } + num++; + qm_mr_next(&p->p); + goto mr_loop; +mr_done: + qm_mr_cci_consume(&p->p, num); + } + /* + * QM_PIRQ_CSCI has already been cleared, as part of its specific + * processing. If that interrupt source has meanwhile been re-asserted, + * we mustn't clear it here (or in the top-level interrupt handler). + */ + return is & (QM_PIRQ_EQCI | QM_PIRQ_EQRI | QM_PIRQ_MRI); +} + +/* remove some slowish-path stuff from the "fast path" and make sure it isn't + * inlined. */ +static noinline void clear_vdqcr(struct qman_portal *p, struct qman_fq *fq) +{ + p->vdqcr_owned = NULL; + FQLOCK(fq); + fq_clear(fq, QMAN_FQ_STATE_VDQCR); + FQUNLOCK(fq); + wake_up(&affine_queue); +} + +/* Look: no locks, no irq_save()s, no preempt_disable()s! :-) The only states + * that would conflict with other things if they ran at the same time on the + * same cpu are; + * + * (i) setting/clearing vdqcr_owned, and + * (ii) clearing the NE (Not Empty) flag. + * + * Both are safe. Because; + * + * (i) this clearing can only occur after qman_volatile_dequeue() has set the + * vdqcr_owned field (which it does before setting VDQCR), and + * qman_volatile_dequeue() blocks interrupts and preemption while this is + * done so that we can't interfere. + * (ii) the NE flag is only cleared after qman_retire_fq() has set it, and as + * with (i) that API prevents us from interfering until it's safe. + * + * The good thing is that qman_volatile_dequeue() and qman_retire_fq() run far + * less frequently (ie. per-FQ) than __poll_portal_fast() does, so the nett + * advantage comes from this function not having to "lock" anything at all. + * + * Note also that the callbacks are invoked at points which are safe against the + * above potential conflicts, but that this function itself is not re-entrant + * (this is because the function tracks one end of each FIFO in the portal and + * we do *not* want to lock that). So the consequence is that it is safe for + * user callbacks to call into any Qman API *except* qman_poll() (as that's the + * sole API that could be invoking the callback through this function). + */ +static inline unsigned int __poll_portal_fast(struct qman_portal *p, + unsigned int poll_limit) +{ + const struct qm_dqrr_entry *dq; + struct qman_fq *fq; + enum qman_cb_dqrr_result res; + unsigned int limit = 0; + +loop: + qm_dqrr_pvb_update(&p->p); + dq = qm_dqrr_current(&p->p); + if (!dq) + goto done; + if (dq->stat & QM_DQRR_STAT_UNSCHEDULED) { + /* VDQCR: don't trust contextB as the FQ may have been + * configured for h/w consumption and we're draining it + * post-retirement. */ + fq = p->vdqcr_owned; + /* We only set QMAN_FQ_STATE_NE when retiring, so we only need + * to check for clearing it when doing volatile dequeues. It's + * one less thing to check in the critical path (SDQCR). */ + if (dq->stat & QM_DQRR_STAT_FQ_EMPTY) + fq_clear(fq, QMAN_FQ_STATE_NE); + /* this is duplicated from the SDQCR code, but we have stuff to + * do before *and* after this callback, and we don't want + * multiple if()s in the critical path (SDQCR). */ + res = fq->cb.dqrr(p, fq, dq); + if (res == qman_cb_dqrr_stop) + goto done; + /* Check for VDQCR completion */ + if (dq->stat & QM_DQRR_STAT_DQCR_EXPIRED) + clear_vdqcr(p, fq); + } else { + /* SDQCR: contextB points to the FQ */ +#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP + fq = get_fq_table_entry(dq->contextB); +#else + fq = (void *)(uintptr_t)dq->contextB; +#endif + /* Now let the callback do its stuff */ + res = fq->cb.dqrr(p, fq, dq); + /* The callback can request that we exit without consuming this + * entry nor advancing; */ + if (res == qman_cb_dqrr_stop) + goto done; + } + /* Interpret 'dq' from a driver perspective. */ + /* Parking isn't possible unless HELDACTIVE was set. NB, + * FORCEELIGIBLE implies HELDACTIVE, so we only need to + * check for HELDACTIVE to cover both. */ + DPA_ASSERT((dq->stat & QM_DQRR_STAT_FQ_HELDACTIVE) || + (res != qman_cb_dqrr_park)); + /* Defer just means "skip it, I'll consume it myself later on" */ + if (res != qman_cb_dqrr_defer) + qm_dqrr_cdc_consume_1ptr(&p->p, dq, (res == qman_cb_dqrr_park)); + /* Move forward */ + qm_dqrr_next(&p->p); + /* Entry processed and consumed, increment our counter. The callback can + * request that we exit after consuming the entry, and we also exit if + * we reach our processing limit, so loop back only if neither of these + * conditions is met. */ + if ((++limit < poll_limit) && (res != qman_cb_dqrr_consume_stop)) + goto loop; +done: + return limit; +} + +u32 qman_irqsource_get(void) +{ + /* "irqsource" and "poll" APIs mustn't redirect when sharing, they + * should shut the user out if they are not the primary CPU hosting the + * portal. That's why we use the "raw" interface. */ + struct qman_portal *p = get_raw_affine_portal(); + u32 ret = p->irq_sources & QM_PIRQ_VISIBLE; + put_affine_portal(); + return ret; +} +EXPORT_SYMBOL(qman_irqsource_get); + +int qman_p_irqsource_add(struct qman_portal *p, u32 bits __maybe_unused) +{ + __maybe_unused unsigned long irqflags; +#ifdef CONFIG_FSL_DPA_PORTAL_SHARE + if (p->sharing_redirect) + return -EINVAL; + else +#endif + { + PORTAL_IRQ_LOCK(p, irqflags); + set_bits(bits & QM_PIRQ_VISIBLE, &p->irq_sources); + qm_isr_enable_write(&p->p, p->irq_sources); + PORTAL_IRQ_UNLOCK(p, irqflags); + } + return 0; +} +EXPORT_SYMBOL(qman_p_irqsource_add); + +int qman_irqsource_add(u32 bits __maybe_unused) +{ + struct qman_portal *p = get_raw_affine_portal(); + int ret; + ret = qman_p_irqsource_add(p, bits); + put_affine_portal(); + return ret; +} +EXPORT_SYMBOL(qman_irqsource_add); + +int qman_p_irqsource_remove(struct qman_portal *p, u32 bits) +{ + __maybe_unused unsigned long irqflags; + u32 ier; +#ifdef CONFIG_FSL_DPA_PORTAL_SHARE + if (p->sharing_redirect) { + put_affine_portal(); + return -EINVAL; + } +#endif + /* Our interrupt handler only processes+clears status register bits that + * are in p->irq_sources. As we're trimming that mask, if one of them + * were to assert in the status register just before we remove it from + * the enable register, there would be an interrupt-storm when we + * release the IRQ lock. So we wait for the enable register update to + * take effect in h/w (by reading it back) and then clear all other bits + * in the status register. Ie. we clear them from ISR once it's certain + * IER won't allow them to reassert. */ + PORTAL_IRQ_LOCK(p, irqflags); + bits &= QM_PIRQ_VISIBLE; + clear_bits(bits, &p->irq_sources); + qm_isr_enable_write(&p->p, p->irq_sources); + ier = qm_isr_enable_read(&p->p); + /* Using "~ier" (rather than "bits" or "~p->irq_sources") creates a + * data-dependency, ie. to protect against re-ordering. */ + qm_isr_status_clear(&p->p, ~ier); + PORTAL_IRQ_UNLOCK(p, irqflags); + return 0; +} +EXPORT_SYMBOL(qman_p_irqsource_remove); + +int qman_irqsource_remove(u32 bits) +{ + struct qman_portal *p = get_raw_affine_portal(); + int ret; + ret = qman_p_irqsource_remove(p, bits); + put_affine_portal(); + return ret; +} +EXPORT_SYMBOL(qman_irqsource_remove); + +const cpumask_t *qman_affine_cpus(void) +{ + return &affine_mask; +} +EXPORT_SYMBOL(qman_affine_cpus); + +u16 qman_affine_channel(int cpu) +{ + if (cpu < 0) { + struct qman_portal *portal = get_raw_affine_portal(); +#ifdef CONFIG_FSL_DPA_PORTAL_SHARE + BUG_ON(portal->sharing_redirect); +#endif + cpu = portal->config->public_cfg.cpu; + put_affine_portal(); + } + BUG_ON(!cpumask_test_cpu(cpu, &affine_mask)); + return affine_channels[cpu]; +} +EXPORT_SYMBOL(qman_affine_channel); + +void *qman_get_affine_portal(int cpu) +{ + return affine_portals[cpu]; +} +EXPORT_SYMBOL(qman_get_affine_portal); + +int qman_p_poll_dqrr(struct qman_portal *p, unsigned int limit) +{ + int ret; + +#ifdef CONFIG_FSL_DPA_PORTAL_SHARE + if (unlikely(p->sharing_redirect)) + ret = -EINVAL; + else +#endif + { + BUG_ON(p->irq_sources & QM_PIRQ_DQRI); + ret = __poll_portal_fast(p, limit); + } + return ret; +} +EXPORT_SYMBOL(qman_p_poll_dqrr); + +int qman_poll_dqrr(unsigned int limit) +{ + struct qman_portal *p = get_poll_portal(); + int ret; + ret = qman_p_poll_dqrr(p, limit); + put_poll_portal(); + return ret; +} +EXPORT_SYMBOL(qman_poll_dqrr); + +u32 qman_p_poll_slow(struct qman_portal *p) +{ + u32 ret; +#ifdef CONFIG_FSL_DPA_PORTAL_SHARE + if (unlikely(p->sharing_redirect)) + ret = (u32)-1; + else +#endif + { + u32 is = qm_isr_status_read(&p->p) & ~p->irq_sources; + ret = __poll_portal_slow(p, is); + qm_isr_status_clear(&p->p, ret); + } + return ret; +} +EXPORT_SYMBOL(qman_p_poll_slow); + +u32 qman_poll_slow(void) +{ + struct qman_portal *p = get_poll_portal(); + u32 ret; + ret = qman_p_poll_slow(p); + put_poll_portal(); + return ret; +} +EXPORT_SYMBOL(qman_poll_slow); + +/* Legacy wrapper */ +void qman_p_poll(struct qman_portal *p) +{ +#ifdef CONFIG_FSL_DPA_PORTAL_SHARE + if (unlikely(p->sharing_redirect)) + return; +#endif + if ((~p->irq_sources) & QM_PIRQ_SLOW) { + if (!(p->slowpoll--)) { + u32 is = qm_isr_status_read(&p->p) & ~p->irq_sources; + u32 active = __poll_portal_slow(p, is); + if (active) { + qm_isr_status_clear(&p->p, active); + p->slowpoll = SLOW_POLL_BUSY; + } else + p->slowpoll = SLOW_POLL_IDLE; + } + } + if ((~p->irq_sources) & QM_PIRQ_DQRI) + __poll_portal_fast(p, CONFIG_FSL_QMAN_POLL_LIMIT); +} +EXPORT_SYMBOL(qman_p_poll); + +void qman_poll(void) +{ + struct qman_portal *p = get_poll_portal(); + qman_p_poll(p); + put_poll_portal(); +} +EXPORT_SYMBOL(qman_poll); + +void qman_p_stop_dequeues(struct qman_portal *p) +{ + qman_stop_dequeues_ex(p); +} +EXPORT_SYMBOL(qman_p_stop_dequeues); + +void qman_stop_dequeues(void) +{ + struct qman_portal *p = get_affine_portal(); + qman_p_stop_dequeues(p); + put_affine_portal(); +} +EXPORT_SYMBOL(qman_stop_dequeues); + +void qman_p_start_dequeues(struct qman_portal *p) +{ + unsigned long irqflags __maybe_unused; + PORTAL_IRQ_LOCK(p, irqflags); + DPA_ASSERT(p->dqrr_disable_ref > 0); + if (!(--p->dqrr_disable_ref)) + qm_dqrr_set_maxfill(&p->p, DQRR_MAXFILL); + PORTAL_IRQ_UNLOCK(p, irqflags); +} +EXPORT_SYMBOL(qman_p_start_dequeues); + +void qman_start_dequeues(void) +{ + struct qman_portal *p = get_affine_portal(); + qman_p_start_dequeues(p); + put_affine_portal(); +} +EXPORT_SYMBOL(qman_start_dequeues); + +void qman_p_static_dequeue_add(struct qman_portal *p, u32 pools) +{ + unsigned long irqflags __maybe_unused; + PORTAL_IRQ_LOCK(p, irqflags); + pools &= p->config->public_cfg.pools; + p->sdqcr |= pools; + qm_dqrr_sdqcr_set(&p->p, p->sdqcr); + PORTAL_IRQ_UNLOCK(p, irqflags); +} +EXPORT_SYMBOL(qman_p_static_dequeue_add); + +void qman_static_dequeue_add(u32 pools) +{ + struct qman_portal *p = get_affine_portal(); + qman_p_static_dequeue_add(p, pools); + put_affine_portal(); +} +EXPORT_SYMBOL(qman_static_dequeue_add); + +void qman_p_static_dequeue_del(struct qman_portal *p, u32 pools) +{ + unsigned long irqflags __maybe_unused; + PORTAL_IRQ_LOCK(p, irqflags); + pools &= p->config->public_cfg.pools; + p->sdqcr &= ~pools; + qm_dqrr_sdqcr_set(&p->p, p->sdqcr); + PORTAL_IRQ_UNLOCK(p, irqflags); +} +EXPORT_SYMBOL(qman_p_static_dequeue_del); + +void qman_static_dequeue_del(u32 pools) +{ + struct qman_portal *p = get_affine_portal(); + qman_p_static_dequeue_del(p, pools); + put_affine_portal(); +} +EXPORT_SYMBOL(qman_static_dequeue_del); + +u32 qman_p_static_dequeue_get(struct qman_portal *p) +{ + return p->sdqcr; +} +EXPORT_SYMBOL(qman_p_static_dequeue_get); + +u32 qman_static_dequeue_get(void) +{ + struct qman_portal *p = get_affine_portal(); + u32 ret = qman_p_static_dequeue_get(p); + put_affine_portal(); + return ret; +} +EXPORT_SYMBOL(qman_static_dequeue_get); + +void qman_p_dca(struct qman_portal *p, struct qm_dqrr_entry *dq, + int park_request) +{ + qm_dqrr_cdc_consume_1ptr(&p->p, dq, park_request); +} +EXPORT_SYMBOL(qman_p_dca); + +void qman_dca(struct qm_dqrr_entry *dq, int park_request) +{ + struct qman_portal *p = get_affine_portal(); + qman_p_dca(p, dq, park_request); + put_affine_portal(); +} +EXPORT_SYMBOL(qman_dca); + +/*******************/ +/* Frame queue API */ +/*******************/ + +static const char *mcr_result_str(u8 result) +{ + switch (result) { + case QM_MCR_RESULT_NULL: + return "QM_MCR_RESULT_NULL"; + case QM_MCR_RESULT_OK: + return "QM_MCR_RESULT_OK"; + case QM_MCR_RESULT_ERR_FQID: + return "QM_MCR_RESULT_ERR_FQID"; + case QM_MCR_RESULT_ERR_FQSTATE: + return "QM_MCR_RESULT_ERR_FQSTATE"; + case QM_MCR_RESULT_ERR_NOTEMPTY: + return "QM_MCR_RESULT_ERR_NOTEMPTY"; + case QM_MCR_RESULT_PENDING: + return "QM_MCR_RESULT_PENDING"; + case QM_MCR_RESULT_ERR_BADCOMMAND: + return "QM_MCR_RESULT_ERR_BADCOMMAND"; + } + return ""; +} + +int qman_create_fq(u32 fqid, u32 flags, struct qman_fq *fq) +{ + struct qm_fqd fqd; + struct qm_mcr_queryfq_np np; + struct qm_mc_command *mcc; + struct qm_mc_result *mcr; + struct qman_portal *p; + unsigned long irqflags __maybe_unused; + + if (flags & QMAN_FQ_FLAG_DYNAMIC_FQID) { + int ret = qman_alloc_fqid(&fqid); + if (ret) + return ret; + } + spin_lock_init(&fq->fqlock); + fq->fqid = fqid; + fq->flags = flags; + fq->state = qman_fq_state_oos; + fq->cgr_groupid = 0; +#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP + if (unlikely(find_empty_fq_table_entry(&fq->key, fq))) + return -ENOMEM; +#endif + if (!(flags & QMAN_FQ_FLAG_AS_IS) || (flags & QMAN_FQ_FLAG_NO_MODIFY)) + return 0; + /* Everything else is AS_IS support */ + p = get_affine_portal(); + PORTAL_IRQ_LOCK(p, irqflags); + mcc = qm_mc_start(&p->p); + mcc->queryfq.fqid = fqid; + qm_mc_commit(&p->p, QM_MCC_VERB_QUERYFQ); + while (!(mcr = qm_mc_result(&p->p))) + cpu_relax(); + DPA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == QM_MCC_VERB_QUERYFQ); + if (mcr->result != QM_MCR_RESULT_OK) { + pr_err("QUERYFQ failed: %s\n", mcr_result_str(mcr->result)); + goto err; + } + fqd = mcr->queryfq.fqd; + mcc = qm_mc_start(&p->p); + mcc->queryfq_np.fqid = fqid; + qm_mc_commit(&p->p, QM_MCC_VERB_QUERYFQ_NP); + while (!(mcr = qm_mc_result(&p->p))) + cpu_relax(); + DPA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == QM_MCC_VERB_QUERYFQ_NP); + if (mcr->result != QM_MCR_RESULT_OK) { + pr_err("QUERYFQ_NP failed: %s\n", mcr_result_str(mcr->result)); + goto err; + } + np = mcr->queryfq_np; + /* Phew, have queryfq and queryfq_np results, stitch together + * the FQ object from those. */ + fq->cgr_groupid = fqd.cgid; + switch (np.state & QM_MCR_NP_STATE_MASK) { + case QM_MCR_NP_STATE_OOS: + break; + case QM_MCR_NP_STATE_RETIRED: + fq->state = qman_fq_state_retired; + if (np.frm_cnt) + fq_set(fq, QMAN_FQ_STATE_NE); + break; + case QM_MCR_NP_STATE_TEN_SCHED: + case QM_MCR_NP_STATE_TRU_SCHED: + case QM_MCR_NP_STATE_ACTIVE: + fq->state = qman_fq_state_sched; + if (np.state & QM_MCR_NP_STATE_R) + fq_set(fq, QMAN_FQ_STATE_CHANGING); + break; + case QM_MCR_NP_STATE_PARKED: + fq->state = qman_fq_state_parked; + break; + default: + DPA_ASSERT(NULL == "invalid FQ state"); + } + if (fqd.fq_ctrl & QM_FQCTRL_CGE) + fq->state |= QMAN_FQ_STATE_CGR_EN; + PORTAL_IRQ_UNLOCK(p, irqflags); + put_affine_portal(); + return 0; +err: + PORTAL_IRQ_UNLOCK(p, irqflags); + put_affine_portal(); + if (flags & QMAN_FQ_FLAG_DYNAMIC_FQID) + qman_release_fqid(fqid); + return -EIO; +} +EXPORT_SYMBOL(qman_create_fq); + +void qman_destroy_fq(struct qman_fq *fq, u32 flags __maybe_unused) +{ + + /* We don't need to lock the FQ as it is a pre-condition that the FQ be + * quiesced. Instead, run some checks. */ + switch (fq->state) { + case qman_fq_state_parked: + DPA_ASSERT(flags & QMAN_FQ_DESTROY_PARKED); + case qman_fq_state_oos: + if (fq_isset(fq, QMAN_FQ_FLAG_DYNAMIC_FQID)) + qman_release_fqid(fq->fqid); +#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP + clear_fq_table_entry(fq->key); +#endif + return; + default: + break; + } + DPA_ASSERT(NULL == "qman_free_fq() on unquiesced FQ!"); +} +EXPORT_SYMBOL(qman_destroy_fq); + +u32 qman_fq_fqid(struct qman_fq *fq) +{ + return fq->fqid; +} +EXPORT_SYMBOL(qman_fq_fqid); + +void qman_fq_state(struct qman_fq *fq, enum qman_fq_state *state, u32 *flags) +{ + if (state) + *state = fq->state; + if (flags) + *flags = fq->flags; +} +EXPORT_SYMBOL(qman_fq_state); + +int qman_init_fq(struct qman_fq *fq, u32 flags, struct qm_mcc_initfq *opts) +{ + struct qm_mc_command *mcc; + struct qm_mc_result *mcr; + struct qman_portal *p; + unsigned long irqflags __maybe_unused; + u8 res, myverb = (flags & QMAN_INITFQ_FLAG_SCHED) ? + QM_MCC_VERB_INITFQ_SCHED : QM_MCC_VERB_INITFQ_PARKED; + + if ((fq->state != qman_fq_state_oos) && + (fq->state != qman_fq_state_parked)) + return -EINVAL; +#ifdef CONFIG_FSL_DPA_CHECKING + if (unlikely(fq_isset(fq, QMAN_FQ_FLAG_NO_MODIFY))) + return -EINVAL; +#endif + if (opts && (opts->we_mask & QM_INITFQ_WE_OAC)) { + /* And can't be set at the same time as TDTHRESH */ + if (opts->we_mask & QM_INITFQ_WE_TDTHRESH) + return -EINVAL; + } + /* Issue an INITFQ_[PARKED|SCHED] management command */ + p = get_affine_portal(); + PORTAL_IRQ_LOCK(p, irqflags); + FQLOCK(fq); + if (unlikely((fq_isset(fq, QMAN_FQ_STATE_CHANGING)) || + ((fq->state != qman_fq_state_oos) && + (fq->state != qman_fq_state_parked)))) { + FQUNLOCK(fq); + PORTAL_IRQ_UNLOCK(p, irqflags); + put_affine_portal(); + return -EBUSY; + } + mcc = qm_mc_start(&p->p); + if (opts) + mcc->initfq = *opts; + mcc->initfq.fqid = fq->fqid; + mcc->initfq.count = 0; + /* If the FQ does *not* have the TO_DCPORTAL flag, contextB is set as a + * demux pointer. Otherwise, the caller-provided value is allowed to + * stand, don't overwrite it. */ + if (fq_isclear(fq, QMAN_FQ_FLAG_TO_DCPORTAL)) { + dma_addr_t phys_fq; + mcc->initfq.we_mask |= QM_INITFQ_WE_CONTEXTB; +#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP + mcc->initfq.fqd.context_b = fq->key; +#else + mcc->initfq.fqd.context_b = (u32)(uintptr_t)fq; +#endif + /* and the physical address - NB, if the user wasn't trying to + * set CONTEXTA, clear the stashing settings. */ + if (!(mcc->initfq.we_mask & QM_INITFQ_WE_CONTEXTA)) { + mcc->initfq.we_mask |= QM_INITFQ_WE_CONTEXTA; + memset(&mcc->initfq.fqd.context_a, 0, + sizeof(mcc->initfq.fqd.context_a)); + } else { + phys_fq = dma_map_single(&p->pdev->dev, fq, sizeof(*fq), + DMA_TO_DEVICE); + qm_fqd_stashing_set64(&mcc->initfq.fqd, phys_fq); + } + } + if (flags & QMAN_INITFQ_FLAG_LOCAL) { + mcc->initfq.fqd.dest.channel = p->config->public_cfg.channel; + if (!(mcc->initfq.we_mask & QM_INITFQ_WE_DESTWQ)) { + mcc->initfq.we_mask |= QM_INITFQ_WE_DESTWQ; + mcc->initfq.fqd.dest.wq = 4; + } + } + qm_mc_commit(&p->p, myverb); + while (!(mcr = qm_mc_result(&p->p))) + cpu_relax(); + DPA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == myverb); + res = mcr->result; + if (res != QM_MCR_RESULT_OK) { + FQUNLOCK(fq); + PORTAL_IRQ_UNLOCK(p, irqflags); + put_affine_portal(); + return -EIO; + } + if (opts) { + if (opts->we_mask & QM_INITFQ_WE_FQCTRL) { + if (opts->fqd.fq_ctrl & QM_FQCTRL_CGE) + fq_set(fq, QMAN_FQ_STATE_CGR_EN); + else + fq_clear(fq, QMAN_FQ_STATE_CGR_EN); + } + if (opts->we_mask & QM_INITFQ_WE_CGID) + fq->cgr_groupid = opts->fqd.cgid; + } + fq->state = (flags & QMAN_INITFQ_FLAG_SCHED) ? + qman_fq_state_sched : qman_fq_state_parked; + FQUNLOCK(fq); + PORTAL_IRQ_UNLOCK(p, irqflags); + put_affine_portal(); + return 0; +} +EXPORT_SYMBOL(qman_init_fq); + +int qman_schedule_fq(struct qman_fq *fq) +{ + struct qm_mc_command *mcc; + struct qm_mc_result *mcr; + struct qman_portal *p; + unsigned long irqflags __maybe_unused; + int ret = 0; + u8 res; + + if (fq->state != qman_fq_state_parked) + return -EINVAL; +#ifdef CONFIG_FSL_DPA_CHECKING + if (unlikely(fq_isset(fq, QMAN_FQ_FLAG_NO_MODIFY))) + return -EINVAL; +#endif + /* Issue a ALTERFQ_SCHED management command */ + p = get_affine_portal(); + PORTAL_IRQ_LOCK(p, irqflags); + FQLOCK(fq); + if (unlikely((fq_isset(fq, QMAN_FQ_STATE_CHANGING)) || + (fq->state != qman_fq_state_parked))) { + ret = -EBUSY; + goto out; + } + mcc = qm_mc_start(&p->p); + mcc->alterfq.fqid = fq->fqid; + qm_mc_commit(&p->p, QM_MCC_VERB_ALTER_SCHED); + while (!(mcr = qm_mc_result(&p->p))) + cpu_relax(); + DPA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == QM_MCR_VERB_ALTER_SCHED); + res = mcr->result; + if (res != QM_MCR_RESULT_OK) { + ret = -EIO; + goto out; + } + fq->state = qman_fq_state_sched; +out: + FQUNLOCK(fq); + PORTAL_IRQ_UNLOCK(p, irqflags); + put_affine_portal(); + return ret; +} +EXPORT_SYMBOL(qman_schedule_fq); + +int qman_retire_fq(struct qman_fq *fq, u32 *flags) +{ + struct qm_mc_command *mcc; + struct qm_mc_result *mcr; + struct qman_portal *p; + unsigned long irqflags __maybe_unused; + int rval; + u8 res; + + if ((fq->state != qman_fq_state_parked) && + (fq->state != qman_fq_state_sched)) + return -EINVAL; +#ifdef CONFIG_FSL_DPA_CHECKING + if (unlikely(fq_isset(fq, QMAN_FQ_FLAG_NO_MODIFY))) + return -EINVAL; +#endif + p = get_affine_portal(); + PORTAL_IRQ_LOCK(p, irqflags); + FQLOCK(fq); + if (unlikely((fq_isset(fq, QMAN_FQ_STATE_CHANGING)) || + (fq->state == qman_fq_state_retired) || + (fq->state == qman_fq_state_oos))) { + rval = -EBUSY; + goto out; + } + rval = table_push_fq(p, fq); + if (rval) + goto out; + mcc = qm_mc_start(&p->p); + mcc->alterfq.fqid = fq->fqid; + qm_mc_commit(&p->p, QM_MCC_VERB_ALTER_RETIRE); + while (!(mcr = qm_mc_result(&p->p))) + cpu_relax(); + DPA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == QM_MCR_VERB_ALTER_RETIRE); + res = mcr->result; + /* "Elegant" would be to treat OK/PENDING the same way; set CHANGING, + * and defer the flags until FQRNI or FQRN (respectively) show up. But + * "Friendly" is to process OK immediately, and not set CHANGING. We do + * friendly, otherwise the caller doesn't necessarily have a fully + * "retired" FQ on return even if the retirement was immediate. However + * this does mean some code duplication between here and + * fq_state_change(). */ + if (likely(res == QM_MCR_RESULT_OK)) { + rval = 0; + /* Process 'fq' right away, we'll ignore FQRNI */ + if (mcr->alterfq.fqs & QM_MCR_FQS_NOTEMPTY) + fq_set(fq, QMAN_FQ_STATE_NE); + if (mcr->alterfq.fqs & QM_MCR_FQS_ORLPRESENT) + fq_set(fq, QMAN_FQ_STATE_ORL); + else + table_del_fq(p, fq); + if (flags) + *flags = fq->flags; + fq->state = qman_fq_state_retired; + if (fq->cb.fqs) { + /* Another issue with supporting "immediate" retirement + * is that we're forced to drop FQRNIs, because by the + * time they're seen it may already be "too late" (the + * fq may have been OOS'd and free()'d already). But if + * the upper layer wants a callback whether it's + * immediate or not, we have to fake a "MR" entry to + * look like an FQRNI... */ + struct qm_mr_entry msg; + msg.verb = QM_MR_VERB_FQRNI; + msg.fq.fqs = mcr->alterfq.fqs; + msg.fq.fqid = fq->fqid; +#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP + msg.fq.contextB = fq->key; +#else + msg.fq.contextB = (u32)(uintptr_t)fq; +#endif + fq->cb.fqs(p, fq, &msg); + } + } else if (res == QM_MCR_RESULT_PENDING) { + rval = 1; + fq_set(fq, QMAN_FQ_STATE_CHANGING); + } else { + rval = -EIO; + table_del_fq(p, fq); + } +out: + FQUNLOCK(fq); + PORTAL_IRQ_UNLOCK(p, irqflags); + put_affine_portal(); + return rval; +} +EXPORT_SYMBOL(qman_retire_fq); + +int qman_oos_fq(struct qman_fq *fq) +{ + struct qm_mc_command *mcc; + struct qm_mc_result *mcr; + struct qman_portal *p; + unsigned long irqflags __maybe_unused; + int ret = 0; + u8 res; + + if (fq->state != qman_fq_state_retired) + return -EINVAL; +#ifdef CONFIG_FSL_DPA_CHECKING + if (unlikely(fq_isset(fq, QMAN_FQ_FLAG_NO_MODIFY))) + return -EINVAL; +#endif + p = get_affine_portal(); + PORTAL_IRQ_LOCK(p, irqflags); + FQLOCK(fq); + if (unlikely((fq_isset(fq, QMAN_FQ_STATE_BLOCKOOS)) || + (fq->state != qman_fq_state_retired))) { + ret = -EBUSY; + goto out; + } + mcc = qm_mc_start(&p->p); + mcc->alterfq.fqid = fq->fqid; + qm_mc_commit(&p->p, QM_MCC_VERB_ALTER_OOS); + while (!(mcr = qm_mc_result(&p->p))) + cpu_relax(); + DPA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == QM_MCR_VERB_ALTER_OOS); + res = mcr->result; + if (res != QM_MCR_RESULT_OK) { + ret = -EIO; + goto out; + } + fq->state = qman_fq_state_oos; +out: + FQUNLOCK(fq); + PORTAL_IRQ_UNLOCK(p, irqflags); + put_affine_portal(); + return ret; +} +EXPORT_SYMBOL(qman_oos_fq); + +int qman_fq_flow_control(struct qman_fq *fq, int xon) +{ + struct qm_mc_command *mcc; + struct qm_mc_result *mcr; + struct qman_portal *p; + unsigned long irqflags __maybe_unused; + int ret = 0; + u8 res; + u8 myverb; + + if ((fq->state == qman_fq_state_oos) || + (fq->state == qman_fq_state_retired) || + (fq->state == qman_fq_state_parked)) + return -EINVAL; + +#ifdef CONFIG_FSL_DPA_CHECKING + if (unlikely(fq_isset(fq, QMAN_FQ_FLAG_NO_MODIFY))) + return -EINVAL; +#endif + /* Issue a ALTER_FQXON or ALTER_FQXOFF management command */ + p = get_affine_portal(); + PORTAL_IRQ_LOCK(p, irqflags); + FQLOCK(fq); + if (unlikely((fq_isset(fq, QMAN_FQ_STATE_CHANGING)) || + (fq->state == qman_fq_state_parked) || + (fq->state == qman_fq_state_oos) || + (fq->state == qman_fq_state_retired))) { + ret = -EBUSY; + goto out; + } + mcc = qm_mc_start(&p->p); + mcc->alterfq.fqid = fq->fqid; + mcc->alterfq.count = 0; + myverb = xon ? QM_MCC_VERB_ALTER_FQXON : QM_MCC_VERB_ALTER_FQXOFF; + + qm_mc_commit(&p->p, myverb); + while (!(mcr = qm_mc_result(&p->p))) + cpu_relax(); + DPA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == myverb); + + res = mcr->result; + if (res != QM_MCR_RESULT_OK) { + ret = -EIO; + goto out; + } +out: + FQUNLOCK(fq); + PORTAL_IRQ_UNLOCK(p, irqflags); + put_affine_portal(); + return ret; +} +EXPORT_SYMBOL(qman_fq_flow_control); + +int qman_query_fq(struct qman_fq *fq, struct qm_fqd *fqd) +{ + struct qm_mc_command *mcc; + struct qm_mc_result *mcr; + struct qman_portal *p = get_affine_portal(); + unsigned long irqflags __maybe_unused; + u8 res; + + PORTAL_IRQ_LOCK(p, irqflags); + mcc = qm_mc_start(&p->p); + mcc->queryfq.fqid = fq->fqid; + qm_mc_commit(&p->p, QM_MCC_VERB_QUERYFQ); + while (!(mcr = qm_mc_result(&p->p))) + cpu_relax(); + DPA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == QM_MCR_VERB_QUERYFQ); + res = mcr->result; + if (res == QM_MCR_RESULT_OK) + *fqd = mcr->queryfq.fqd; + PORTAL_IRQ_UNLOCK(p, irqflags); + put_affine_portal(); + if (res != QM_MCR_RESULT_OK) + return -EIO; + return 0; +} +EXPORT_SYMBOL(qman_query_fq); + +int qman_query_fq_np(struct qman_fq *fq, struct qm_mcr_queryfq_np *np) +{ + struct qm_mc_command *mcc; + struct qm_mc_result *mcr; + struct qman_portal *p = get_affine_portal(); + unsigned long irqflags __maybe_unused; + u8 res; + + PORTAL_IRQ_LOCK(p, irqflags); + mcc = qm_mc_start(&p->p); + mcc->queryfq.fqid = fq->fqid; + qm_mc_commit(&p->p, QM_MCC_VERB_QUERYFQ_NP); + while (!(mcr = qm_mc_result(&p->p))) + cpu_relax(); + DPA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == QM_MCR_VERB_QUERYFQ_NP); + res = mcr->result; + if (res == QM_MCR_RESULT_OK) + *np = mcr->queryfq_np; + PORTAL_IRQ_UNLOCK(p, irqflags); + put_affine_portal(); + if (res == QM_MCR_RESULT_ERR_FQID) + return -ERANGE; + else if (res != QM_MCR_RESULT_OK) + return -EIO; + return 0; +} +EXPORT_SYMBOL(qman_query_fq_np); + +int qman_query_wq(u8 query_dedicated, struct qm_mcr_querywq *wq) +{ + struct qm_mc_command *mcc; + struct qm_mc_result *mcr; + struct qman_portal *p = get_affine_portal(); + unsigned long irqflags __maybe_unused; + u8 res, myverb; + + PORTAL_IRQ_LOCK(p, irqflags); + myverb = (query_dedicated) ? QM_MCR_VERB_QUERYWQ_DEDICATED : + QM_MCR_VERB_QUERYWQ; + mcc = qm_mc_start(&p->p); + mcc->querywq.channel.id = wq->channel.id; + qm_mc_commit(&p->p, myverb); + while (!(mcr = qm_mc_result(&p->p))) + cpu_relax(); + DPA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == myverb); + res = mcr->result; + if (res == QM_MCR_RESULT_OK) + *wq = mcr->querywq; + PORTAL_IRQ_UNLOCK(p, irqflags); + put_affine_portal(); + if (res != QM_MCR_RESULT_OK) { + pr_err("QUERYWQ failed: %s\n", mcr_result_str(res)); + return -EIO; + } + return 0; +} +EXPORT_SYMBOL(qman_query_wq); + +int qman_query_cgr(struct qman_cgr *cgr, struct qm_mcr_querycgr *cgrd) +{ + struct qm_mc_command *mcc; + struct qm_mc_result *mcr; + struct qman_portal *p = get_affine_portal(); + unsigned long irqflags __maybe_unused; + u8 res; + + PORTAL_IRQ_LOCK(p, irqflags); + mcc = qm_mc_start(&p->p); + mcc->querycgr.cgid = cgr->cgrid; + qm_mc_commit(&p->p, QM_MCC_VERB_QUERYCGR); + while (!(mcr = qm_mc_result(&p->p))) + cpu_relax(); + DPA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == QM_MCC_VERB_QUERYCGR); + res = mcr->result; + if (res == QM_MCR_RESULT_OK) + *cgrd = mcr->querycgr; + PORTAL_IRQ_UNLOCK(p, irqflags); + put_affine_portal(); + if (res != QM_MCR_RESULT_OK) { + pr_err("QUERY_CGR failed: %s\n", mcr_result_str(res)); + return -EIO; + } + return 0; +} +EXPORT_SYMBOL(qman_query_cgr); + +/* internal function used as a wait_event() expression */ +static int set_p_vdqcr(struct qman_portal *p, struct qman_fq *fq, u32 vdqcr) +{ + unsigned long irqflags __maybe_unused; + int ret = -EBUSY; + PORTAL_IRQ_LOCK(p, irqflags); + if (!p->vdqcr_owned) { + FQLOCK(fq); + if (fq_isset(fq, QMAN_FQ_STATE_VDQCR)) + goto escape; + fq_set(fq, QMAN_FQ_STATE_VDQCR); + FQUNLOCK(fq); + p->vdqcr_owned = fq; + ret = 0; + } +escape: + PORTAL_IRQ_UNLOCK(p, irqflags); + if (!ret) + qm_dqrr_vdqcr_set(&p->p, vdqcr); + return ret; +} + +static int set_vdqcr(struct qman_portal **p, struct qman_fq *fq, u32 vdqcr) +{ + int ret; + *p = get_affine_portal(); + ret = set_p_vdqcr(*p, fq, vdqcr); + put_affine_portal(); + return ret; +} + +#ifdef CONFIG_FSL_DPA_CAN_WAIT +static int wait_p_vdqcr_start(struct qman_portal *p, struct qman_fq *fq, + u32 vdqcr, u32 flags) +{ + int ret = 0; + if (flags & QMAN_VOLATILE_FLAG_WAIT_INT) + ret = wait_event_interruptible(affine_queue, + !(ret = set_p_vdqcr(p, fq, vdqcr))); + else + wait_event(affine_queue, !(ret = set_p_vdqcr(p, fq, vdqcr))); + return ret; +} + +static int wait_vdqcr_start(struct qman_portal **p, struct qman_fq *fq, + u32 vdqcr, u32 flags) +{ + int ret = 0; + if (flags & QMAN_VOLATILE_FLAG_WAIT_INT) + ret = wait_event_interruptible(affine_queue, + !(ret = set_vdqcr(p, fq, vdqcr))); + else + wait_event(affine_queue, !(ret = set_vdqcr(p, fq, vdqcr))); + return ret; +} +#endif + +int qman_p_volatile_dequeue(struct qman_portal *p, struct qman_fq *fq, + u32 flags __maybe_unused, u32 vdqcr) +{ + int ret; + + if ((fq->state != qman_fq_state_parked) && + (fq->state != qman_fq_state_retired)) + return -EINVAL; + if (vdqcr & QM_VDQCR_FQID_MASK) + return -EINVAL; + if (fq_isset(fq, QMAN_FQ_STATE_VDQCR)) + return -EBUSY; + vdqcr = (vdqcr & ~QM_VDQCR_FQID_MASK) | fq->fqid; +#ifdef CONFIG_FSL_DPA_CAN_WAIT + if (flags & QMAN_VOLATILE_FLAG_WAIT) + ret = wait_p_vdqcr_start(p, fq, vdqcr, flags); + else +#endif + ret = set_p_vdqcr(p, fq, vdqcr); + if (ret) + return ret; + /* VDQCR is set */ +#ifdef CONFIG_FSL_DPA_CAN_WAIT + if (flags & QMAN_VOLATILE_FLAG_FINISH) { + if (flags & QMAN_VOLATILE_FLAG_WAIT_INT) + /* NB: don't propagate any error - the caller wouldn't + * know whether the VDQCR was issued or not. A signal + * could arrive after returning anyway, so the caller + * can check signal_pending() if that's an issue. */ + wait_event_interruptible(affine_queue, + !fq_isset(fq, QMAN_FQ_STATE_VDQCR)); + else + wait_event(affine_queue, + !fq_isset(fq, QMAN_FQ_STATE_VDQCR)); + } +#endif + return 0; +} +EXPORT_SYMBOL(qman_p_volatile_dequeue); + +int qman_volatile_dequeue(struct qman_fq *fq, u32 flags __maybe_unused, + u32 vdqcr) +{ + struct qman_portal *p; + int ret; + + if ((fq->state != qman_fq_state_parked) && + (fq->state != qman_fq_state_retired)) + return -EINVAL; + if (vdqcr & QM_VDQCR_FQID_MASK) + return -EINVAL; + if (fq_isset(fq, QMAN_FQ_STATE_VDQCR)) + return -EBUSY; + vdqcr = (vdqcr & ~QM_VDQCR_FQID_MASK) | fq->fqid; +#ifdef CONFIG_FSL_DPA_CAN_WAIT + if (flags & QMAN_VOLATILE_FLAG_WAIT) + ret = wait_vdqcr_start(&p, fq, vdqcr, flags); + else +#endif + ret = set_vdqcr(&p, fq, vdqcr); + if (ret) + return ret; + /* VDQCR is set */ +#ifdef CONFIG_FSL_DPA_CAN_WAIT + if (flags & QMAN_VOLATILE_FLAG_FINISH) { + if (flags & QMAN_VOLATILE_FLAG_WAIT_INT) + /* NB: don't propagate any error - the caller wouldn't + * know whether the VDQCR was issued or not. A signal + * could arrive after returning anyway, so the caller + * can check signal_pending() if that's an issue. */ + wait_event_interruptible(affine_queue, + !fq_isset(fq, QMAN_FQ_STATE_VDQCR)); + else + wait_event(affine_queue, + !fq_isset(fq, QMAN_FQ_STATE_VDQCR)); + } +#endif + return 0; +} +EXPORT_SYMBOL(qman_volatile_dequeue); + +static noinline void update_eqcr_ci(struct qman_portal *p, u8 avail) +{ + if (avail) + qm_eqcr_cce_prefetch(&p->p); + else + qm_eqcr_cce_update(&p->p); +} + +int qman_eqcr_is_empty(void) +{ + unsigned long irqflags __maybe_unused; + struct qman_portal *p = get_affine_portal(); + u8 avail; + + PORTAL_IRQ_LOCK(p, irqflags); + update_eqcr_ci(p, 0); + avail = qm_eqcr_get_fill(&p->p); + PORTAL_IRQ_UNLOCK(p, irqflags); + put_affine_portal(); + return avail == 0; +} +EXPORT_SYMBOL(qman_eqcr_is_empty); + +void qman_set_dc_ern(qman_cb_dc_ern handler, int affine) +{ + if (affine) { + unsigned long irqflags __maybe_unused; + struct qman_portal *p = get_affine_portal(); + PORTAL_IRQ_LOCK(p, irqflags); + p->cb_dc_ern = handler; + PORTAL_IRQ_UNLOCK(p, irqflags); + put_affine_portal(); + } else + cb_dc_ern = handler; +} +EXPORT_SYMBOL(qman_set_dc_ern); + +static inline struct qm_eqcr_entry *try_p_eq_start(struct qman_portal *p, + unsigned long *irqflags __maybe_unused, + struct qman_fq *fq, + const struct qm_fd *fd, + u32 flags) +{ + struct qm_eqcr_entry *eq; + u8 avail; + PORTAL_IRQ_LOCK(p, (*irqflags)); +#ifdef CONFIG_FSL_DPA_CAN_WAIT_SYNC + if (unlikely((flags & QMAN_ENQUEUE_FLAG_WAIT) && + (flags & QMAN_ENQUEUE_FLAG_WAIT_SYNC))) { + if (p->eqci_owned) { + PORTAL_IRQ_UNLOCK(p, (*irqflags)); + return NULL; + } + p->eqci_owned = fq; + } +#endif + if (p->use_eqcr_ci_stashing) { + /* + * The stashing case is easy, only update if we need to in + * order to try and liberate ring entries. + */ + eq = qm_eqcr_start_stash(&p->p); + } else { + /* + * The non-stashing case is harder, need to prefetch ahead of + * time. + */ + avail = qm_eqcr_get_avail(&p->p); + if (avail < 2) + update_eqcr_ci(p, avail); + eq = qm_eqcr_start_no_stash(&p->p); + } + + if (unlikely(!eq)) { +#ifdef CONFIG_FSL_DPA_CAN_WAIT_SYNC + if (unlikely((flags & QMAN_ENQUEUE_FLAG_WAIT) && + (flags & QMAN_ENQUEUE_FLAG_WAIT_SYNC))) + p->eqci_owned = NULL; +#endif + PORTAL_IRQ_UNLOCK(p, (*irqflags)); + return NULL; + } + if (flags & QMAN_ENQUEUE_FLAG_DCA) + eq->dca = QM_EQCR_DCA_ENABLE | + ((flags & QMAN_ENQUEUE_FLAG_DCA_PARK) ? + QM_EQCR_DCA_PARK : 0) | + ((flags >> 8) & QM_EQCR_DCA_IDXMASK); + eq->fqid = fq->fqid; +#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP + eq->tag = fq->key; +#else + eq->tag = (u32)(uintptr_t)fq; +#endif + eq->fd = *fd; + return eq; +} + +static inline struct qm_eqcr_entry *try_eq_start(struct qman_portal **p, + unsigned long *irqflags __maybe_unused, + struct qman_fq *fq, + const struct qm_fd *fd, + u32 flags) +{ + struct qm_eqcr_entry *eq; + *p = get_affine_portal(); + eq = try_p_eq_start(*p, irqflags, fq, fd, flags); + if (!eq) + put_affine_portal(); + return eq; +} + +#ifdef CONFIG_FSL_DPA_CAN_WAIT +static noinline struct qm_eqcr_entry *__wait_eq_start(struct qman_portal **p, + unsigned long *irqflags __maybe_unused, + struct qman_fq *fq, + const struct qm_fd *fd, + u32 flags) +{ + struct qm_eqcr_entry *eq = try_eq_start(p, irqflags, fq, fd, flags); + if (!eq) + qm_eqcr_set_ithresh(&(*p)->p, EQCR_ITHRESH); + return eq; +} +static noinline struct qm_eqcr_entry *wait_eq_start(struct qman_portal **p, + unsigned long *irqflags __maybe_unused, + struct qman_fq *fq, + const struct qm_fd *fd, + u32 flags) +{ + struct qm_eqcr_entry *eq; + if (flags & QMAN_ENQUEUE_FLAG_WAIT_INT) + wait_event_interruptible(affine_queue, + (eq = __wait_eq_start(p, irqflags, fq, fd, flags))); + else + wait_event(affine_queue, + (eq = __wait_eq_start(p, irqflags, fq, fd, flags))); + return eq; +} +static noinline struct qm_eqcr_entry *__wait_p_eq_start(struct qman_portal *p, + unsigned long *irqflags __maybe_unused, + struct qman_fq *fq, + const struct qm_fd *fd, + u32 flags) +{ + struct qm_eqcr_entry *eq = try_p_eq_start(p, irqflags, fq, fd, flags); + if (!eq) + qm_eqcr_set_ithresh(&p->p, EQCR_ITHRESH); + return eq; +} +static noinline struct qm_eqcr_entry *wait_p_eq_start(struct qman_portal *p, + unsigned long *irqflags __maybe_unused, + struct qman_fq *fq, + const struct qm_fd *fd, + u32 flags) +{ + struct qm_eqcr_entry *eq; + if (flags & QMAN_ENQUEUE_FLAG_WAIT_INT) + wait_event_interruptible(affine_queue, + (eq = __wait_p_eq_start(p, irqflags, fq, fd, flags))); + else + wait_event(affine_queue, + (eq = __wait_p_eq_start(p, irqflags, fq, fd, flags))); + return eq; +} +#endif + +int qman_p_enqueue(struct qman_portal *p, struct qman_fq *fq, + const struct qm_fd *fd, u32 flags) +{ + struct qm_eqcr_entry *eq; + unsigned long irqflags __maybe_unused; + +#ifdef CONFIG_FSL_DPA_CAN_WAIT + if (flags & QMAN_ENQUEUE_FLAG_WAIT) + eq = wait_p_eq_start(p, &irqflags, fq, fd, flags); + else +#endif + eq = try_p_eq_start(p, &irqflags, fq, fd, flags); + if (!eq) + return -EBUSY; + /* Note: QM_EQCR_VERB_INTERRUPT == QMAN_ENQUEUE_FLAG_WAIT_SYNC */ + qm_eqcr_pvb_commit(&p->p, QM_EQCR_VERB_CMD_ENQUEUE | + (flags & (QM_EQCR_VERB_COLOUR_MASK | QM_EQCR_VERB_INTERRUPT))); + /* Factor the below out, it's used from qman_enqueue_orp() too */ + PORTAL_IRQ_UNLOCK(p, irqflags); +#ifdef CONFIG_FSL_DPA_CAN_WAIT_SYNC + if (unlikely((flags & QMAN_ENQUEUE_FLAG_WAIT) && + (flags & QMAN_ENQUEUE_FLAG_WAIT_SYNC))) { + if (flags & QMAN_ENQUEUE_FLAG_WAIT_INT) + wait_event_interruptible(affine_queue, + (p->eqci_owned != fq)); + else + wait_event(affine_queue, (p->eqci_owned != fq)); + } +#endif + return 0; +} +EXPORT_SYMBOL(qman_p_enqueue); + +int qman_enqueue(struct qman_fq *fq, const struct qm_fd *fd, u32 flags) +{ + struct qman_portal *p; + struct qm_eqcr_entry *eq; + unsigned long irqflags __maybe_unused; + +#ifdef CONFIG_FSL_DPA_CAN_WAIT + if (flags & QMAN_ENQUEUE_FLAG_WAIT) + eq = wait_eq_start(&p, &irqflags, fq, fd, flags); + else +#endif + eq = try_eq_start(&p, &irqflags, fq, fd, flags); + if (!eq) + return -EBUSY; + /* Note: QM_EQCR_VERB_INTERRUPT == QMAN_ENQUEUE_FLAG_WAIT_SYNC */ + qm_eqcr_pvb_commit(&p->p, QM_EQCR_VERB_CMD_ENQUEUE | + (flags & (QM_EQCR_VERB_COLOUR_MASK | QM_EQCR_VERB_INTERRUPT))); + /* Factor the below out, it's used from qman_enqueue_orp() too */ + PORTAL_IRQ_UNLOCK(p, irqflags); + put_affine_portal(); +#ifdef CONFIG_FSL_DPA_CAN_WAIT_SYNC + if (unlikely((flags & QMAN_ENQUEUE_FLAG_WAIT) && + (flags & QMAN_ENQUEUE_FLAG_WAIT_SYNC))) { + if (flags & QMAN_ENQUEUE_FLAG_WAIT_INT) + wait_event_interruptible(affine_queue, + (p->eqci_owned != fq)); + else + wait_event(affine_queue, (p->eqci_owned != fq)); + } +#endif + return 0; +} +EXPORT_SYMBOL(qman_enqueue); + +int qman_p_enqueue_orp(struct qman_portal *p, struct qman_fq *fq, + const struct qm_fd *fd, u32 flags, + struct qman_fq *orp, u16 orp_seqnum) +{ + struct qm_eqcr_entry *eq; + unsigned long irqflags __maybe_unused; + +#ifdef CONFIG_FSL_DPA_CAN_WAIT + if (flags & QMAN_ENQUEUE_FLAG_WAIT) + eq = wait_p_eq_start(p, &irqflags, fq, fd, flags); + else +#endif + eq = try_p_eq_start(p, &irqflags, fq, fd, flags); + if (!eq) + return -EBUSY; + /* Process ORP-specifics here */ + if (flags & QMAN_ENQUEUE_FLAG_NLIS) + orp_seqnum |= QM_EQCR_SEQNUM_NLIS; + else { + orp_seqnum &= ~QM_EQCR_SEQNUM_NLIS; + if (flags & QMAN_ENQUEUE_FLAG_NESN) + orp_seqnum |= QM_EQCR_SEQNUM_NESN; + else + /* No need to check 4 QMAN_ENQUEUE_FLAG_HOLE */ + orp_seqnum &= ~QM_EQCR_SEQNUM_NESN; + } + eq->seqnum = orp_seqnum; + eq->orp = orp->fqid; + /* Note: QM_EQCR_VERB_INTERRUPT == QMAN_ENQUEUE_FLAG_WAIT_SYNC */ + qm_eqcr_pvb_commit(&p->p, QM_EQCR_VERB_ORP | + ((flags & (QMAN_ENQUEUE_FLAG_HOLE | QMAN_ENQUEUE_FLAG_NESN)) ? + 0 : QM_EQCR_VERB_CMD_ENQUEUE) | + (flags & (QM_EQCR_VERB_COLOUR_MASK | QM_EQCR_VERB_INTERRUPT))); + PORTAL_IRQ_UNLOCK(p, irqflags); +#ifdef CONFIG_FSL_DPA_CAN_WAIT_SYNC + if (unlikely((flags & QMAN_ENQUEUE_FLAG_WAIT) && + (flags & QMAN_ENQUEUE_FLAG_WAIT_SYNC))) { + if (flags & QMAN_ENQUEUE_FLAG_WAIT_INT) + wait_event_interruptible(affine_queue, + (p->eqci_owned != fq)); + else + wait_event(affine_queue, (p->eqci_owned != fq)); + } +#endif + return 0; +} +EXPORT_SYMBOL(qman_p_enqueue_orp); + +int qman_enqueue_orp(struct qman_fq *fq, const struct qm_fd *fd, u32 flags, + struct qman_fq *orp, u16 orp_seqnum) +{ + struct qman_portal *p; + struct qm_eqcr_entry *eq; + unsigned long irqflags __maybe_unused; + +#ifdef CONFIG_FSL_DPA_CAN_WAIT + if (flags & QMAN_ENQUEUE_FLAG_WAIT) + eq = wait_eq_start(&p, &irqflags, fq, fd, flags); + else +#endif + eq = try_eq_start(&p, &irqflags, fq, fd, flags); + if (!eq) + return -EBUSY; + /* Process ORP-specifics here */ + if (flags & QMAN_ENQUEUE_FLAG_NLIS) + orp_seqnum |= QM_EQCR_SEQNUM_NLIS; + else { + orp_seqnum &= ~QM_EQCR_SEQNUM_NLIS; + if (flags & QMAN_ENQUEUE_FLAG_NESN) + orp_seqnum |= QM_EQCR_SEQNUM_NESN; + else + /* No need to check 4 QMAN_ENQUEUE_FLAG_HOLE */ + orp_seqnum &= ~QM_EQCR_SEQNUM_NESN; + } + eq->seqnum = orp_seqnum; + eq->orp = orp->fqid; + /* Note: QM_EQCR_VERB_INTERRUPT == QMAN_ENQUEUE_FLAG_WAIT_SYNC */ + qm_eqcr_pvb_commit(&p->p, QM_EQCR_VERB_ORP | + ((flags & (QMAN_ENQUEUE_FLAG_HOLE | QMAN_ENQUEUE_FLAG_NESN)) ? + 0 : QM_EQCR_VERB_CMD_ENQUEUE) | + (flags & (QM_EQCR_VERB_COLOUR_MASK | QM_EQCR_VERB_INTERRUPT))); + PORTAL_IRQ_UNLOCK(p, irqflags); + put_affine_portal(); +#ifdef CONFIG_FSL_DPA_CAN_WAIT_SYNC + if (unlikely((flags & QMAN_ENQUEUE_FLAG_WAIT) && + (flags & QMAN_ENQUEUE_FLAG_WAIT_SYNC))) { + if (flags & QMAN_ENQUEUE_FLAG_WAIT_INT) + wait_event_interruptible(affine_queue, + (p->eqci_owned != fq)); + else + wait_event(affine_queue, (p->eqci_owned != fq)); + } +#endif + return 0; +} +EXPORT_SYMBOL(qman_enqueue_orp); + +int qman_p_enqueue_precommit(struct qman_portal *p, struct qman_fq *fq, + const struct qm_fd *fd, u32 flags, + qman_cb_precommit cb, void *cb_arg) +{ + struct qm_eqcr_entry *eq; + unsigned long irqflags __maybe_unused; + +#ifdef CONFIG_FSL_DPA_CAN_WAIT + if (flags & QMAN_ENQUEUE_FLAG_WAIT) + eq = wait_p_eq_start(p, &irqflags, fq, fd, flags); + else +#endif + eq = try_p_eq_start(p, &irqflags, fq, fd, flags); + if (!eq) + return -EBUSY; + /* invoke user supplied callback function before writing commit verb */ + if (cb(cb_arg)) { + PORTAL_IRQ_UNLOCK(p, irqflags); + return -EINVAL; + } + /* Note: QM_EQCR_VERB_INTERRUPT == QMAN_ENQUEUE_FLAG_WAIT_SYNC */ + qm_eqcr_pvb_commit(&p->p, QM_EQCR_VERB_CMD_ENQUEUE | + (flags & (QM_EQCR_VERB_COLOUR_MASK | QM_EQCR_VERB_INTERRUPT))); + /* Factor the below out, it's used from qman_enqueue_orp() too */ + PORTAL_IRQ_UNLOCK(p, irqflags); +#ifdef CONFIG_FSL_DPA_CAN_WAIT_SYNC + if (unlikely((flags & QMAN_ENQUEUE_FLAG_WAIT) && + (flags & QMAN_ENQUEUE_FLAG_WAIT_SYNC))) { + if (flags & QMAN_ENQUEUE_FLAG_WAIT_INT) + wait_event_interruptible(affine_queue, + (p->eqci_owned != fq)); + else + wait_event(affine_queue, (p->eqci_owned != fq)); + } +#endif + return 0; +} +EXPORT_SYMBOL(qman_p_enqueue_precommit); + +int qman_enqueue_precommit(struct qman_fq *fq, const struct qm_fd *fd, + u32 flags, qman_cb_precommit cb, void *cb_arg) +{ + struct qman_portal *p; + struct qm_eqcr_entry *eq; + unsigned long irqflags __maybe_unused; + +#ifdef CONFIG_FSL_DPA_CAN_WAIT + if (flags & QMAN_ENQUEUE_FLAG_WAIT) + eq = wait_eq_start(&p, &irqflags, fq, fd, flags); + else +#endif + eq = try_eq_start(&p, &irqflags, fq, fd, flags); + if (!eq) + return -EBUSY; + /* invoke user supplied callback function before writing commit verb */ + if (cb(cb_arg)) { + PORTAL_IRQ_UNLOCK(p, irqflags); + put_affine_portal(); + return -EINVAL; + } + /* Note: QM_EQCR_VERB_INTERRUPT == QMAN_ENQUEUE_FLAG_WAIT_SYNC */ + qm_eqcr_pvb_commit(&p->p, QM_EQCR_VERB_CMD_ENQUEUE | + (flags & (QM_EQCR_VERB_COLOUR_MASK | QM_EQCR_VERB_INTERRUPT))); + /* Factor the below out, it's used from qman_enqueue_orp() too */ + PORTAL_IRQ_UNLOCK(p, irqflags); + put_affine_portal(); +#ifdef CONFIG_FSL_DPA_CAN_WAIT_SYNC + if (unlikely((flags & QMAN_ENQUEUE_FLAG_WAIT) && + (flags & QMAN_ENQUEUE_FLAG_WAIT_SYNC))) { + if (flags & QMAN_ENQUEUE_FLAG_WAIT_INT) + wait_event_interruptible(affine_queue, + (p->eqci_owned != fq)); + else + wait_event(affine_queue, (p->eqci_owned != fq)); + } +#endif + return 0; +} +EXPORT_SYMBOL(qman_enqueue_precommit); + +int qman_modify_cgr(struct qman_cgr *cgr, u32 flags, + struct qm_mcc_initcgr *opts) +{ + struct qm_mc_command *mcc; + struct qm_mc_result *mcr; + struct qman_portal *p = get_affine_portal(); + unsigned long irqflags __maybe_unused; + u8 res; + u8 verb = QM_MCC_VERB_MODIFYCGR; + + PORTAL_IRQ_LOCK(p, irqflags); + mcc = qm_mc_start(&p->p); + if (opts) + mcc->initcgr = *opts; + mcc->initcgr.cgid = cgr->cgrid; + if (flags & QMAN_CGR_FLAG_USE_INIT) + verb = QM_MCC_VERB_INITCGR; + qm_mc_commit(&p->p, verb); + while (!(mcr = qm_mc_result(&p->p))) + cpu_relax(); + DPA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == verb); + res = mcr->result; + PORTAL_IRQ_UNLOCK(p, irqflags); + put_affine_portal(); + return (res == QM_MCR_RESULT_OK) ? 0 : -EIO; +} +EXPORT_SYMBOL(qman_modify_cgr); + +#define TARG_MASK(n) (0x80000000 >> (n->config->public_cfg.channel - \ + QM_CHANNEL_SWPORTAL0)) +#define PORTAL_IDX(n) (n->config->public_cfg.channel - QM_CHANNEL_SWPORTAL0) + +int qman_create_cgr(struct qman_cgr *cgr, u32 flags, + struct qm_mcc_initcgr *opts) +{ + unsigned long irqflags __maybe_unused; + struct qm_mcr_querycgr cgr_state; + struct qm_mcc_initcgr local_opts; + int ret; + struct qman_portal *p; + + /* We have to check that the provided CGRID is within the limits of the + * data-structures, for obvious reasons. However we'll let h/w take + * care of determining whether it's within the limits of what exists on + * the SoC. */ + if (cgr->cgrid >= __CGR_NUM) + return -EINVAL; + + p = get_affine_portal(); + + memset(&local_opts, 0, sizeof(struct qm_mcc_initcgr)); + cgr->chan = p->config->public_cfg.channel; + spin_lock_irqsave(&p->cgr_lock, irqflags); + + /* if no opts specified, just add it to the list */ + if (!opts) + goto add_list; + + ret = qman_query_cgr(cgr, &cgr_state); + if (ret) + goto release_lock; + if (opts) + local_opts = *opts; + if ((qman_ip_rev & 0xFF00) >= QMAN_REV30) + local_opts.cgr.cscn_targ_upd_ctrl = + QM_CGR_TARG_UDP_CTRL_WRITE_BIT | PORTAL_IDX(p); + else + /* Overwrite TARG */ + local_opts.cgr.cscn_targ = cgr_state.cgr.cscn_targ | + TARG_MASK(p); + local_opts.we_mask |= QM_CGR_WE_CSCN_TARG; + + /* send init if flags indicate so */ + if (opts && (flags & QMAN_CGR_FLAG_USE_INIT)) + ret = qman_modify_cgr(cgr, QMAN_CGR_FLAG_USE_INIT, &local_opts); + else + ret = qman_modify_cgr(cgr, 0, &local_opts); + if (ret) + goto release_lock; +add_list: + list_add(&cgr->node, &p->cgr_cbs); + + /* Determine if newly added object requires its callback to be called */ + ret = qman_query_cgr(cgr, &cgr_state); + if (ret) { + /* we can't go back, so proceed and return success, but screen + * and wail to the log file */ + pr_crit("CGR HW state partially modified\n"); + ret = 0; + goto release_lock; + } + if (cgr->cb && cgr_state.cgr.cscn_en && qman_cgrs_get(&p->cgrs[1], + cgr->cgrid)) + cgr->cb(p, cgr, 1); +release_lock: + spin_unlock_irqrestore(&p->cgr_lock, irqflags); + put_affine_portal(); + return ret; +} +EXPORT_SYMBOL(qman_create_cgr); + +int qman_create_cgr_to_dcp(struct qman_cgr *cgr, u32 flags, u16 dcp_portal, + struct qm_mcc_initcgr *opts) +{ + unsigned long irqflags __maybe_unused; + struct qm_mcc_initcgr local_opts; + int ret; + + if ((qman_ip_rev & 0xFF00) < QMAN_REV30) { + pr_warn("This QMan version doesn't support to send CSCN to" + " DCP portal\n"); + return -EINVAL; + } + /* We have to check that the provided CGRID is within the limits of the + * data-structures, for obvious reasons. However we'll let h/w take + * care of determining whether it's within the limits of what exists on + * the SoC. + */ + if (cgr->cgrid >= __CGR_NUM) + return -EINVAL; + + memset(&local_opts, 0, sizeof(struct qm_mcc_initcgr)); + if (opts) + local_opts = *opts; + + local_opts.cgr.cscn_targ_upd_ctrl = QM_CGR_TARG_UDP_CTRL_WRITE_BIT | + QM_CGR_TARG_UDP_CTRL_DCP | dcp_portal; + local_opts.we_mask |= QM_CGR_WE_CSCN_TARG; + + /* send init if flags indicate so */ + if (opts && (flags & QMAN_CGR_FLAG_USE_INIT)) + ret = qman_modify_cgr(cgr, QMAN_CGR_FLAG_USE_INIT, + &local_opts); + else + ret = qman_modify_cgr(cgr, 0, &local_opts); + + return ret; +} +EXPORT_SYMBOL(qman_create_cgr_to_dcp); + +int qman_delete_cgr(struct qman_cgr *cgr) +{ + unsigned long irqflags __maybe_unused; + struct qm_mcr_querycgr cgr_state; + struct qm_mcc_initcgr local_opts; + int ret = 0; + struct qman_cgr *i; + struct qman_portal *p = get_affine_portal(); + + if (cgr->chan != p->config->public_cfg.channel) { + pr_crit("Attempting to delete cgr from different portal " + "than it was create: create 0x%x, delete 0x%x\n", + cgr->chan, p->config->public_cfg.channel); + ret = -EINVAL; + goto put_portal; + } + memset(&local_opts, 0, sizeof(struct qm_mcc_initcgr)); + spin_lock_irqsave(&p->cgr_lock, irqflags); + list_del(&cgr->node); + /* + * If there are no other CGR objects for this CGRID in the list, update + * CSCN_TARG accordingly + */ + list_for_each_entry(i, &p->cgr_cbs, node) + if ((i->cgrid == cgr->cgrid) && i->cb) + goto release_lock; + ret = qman_query_cgr(cgr, &cgr_state); + if (ret) { + /* add back to the list */ + list_add(&cgr->node, &p->cgr_cbs); + goto release_lock; + } + /* Overwrite TARG */ + local_opts.we_mask = QM_CGR_WE_CSCN_TARG; + if ((qman_ip_rev & 0xFF00) >= QMAN_REV30) + local_opts.cgr.cscn_targ_upd_ctrl = PORTAL_IDX(p); + else + local_opts.cgr.cscn_targ = cgr_state.cgr.cscn_targ & + ~(TARG_MASK(p)); + ret = qman_modify_cgr(cgr, 0, &local_opts); + if (ret) + /* add back to the list */ + list_add(&cgr->node, &p->cgr_cbs); +release_lock: + spin_unlock_irqrestore(&p->cgr_lock, irqflags); +put_portal: + put_affine_portal(); + return ret; +} +EXPORT_SYMBOL(qman_delete_cgr); + +int qman_set_wpm(int wpm_enable) +{ + return qm_set_wpm(wpm_enable); +} +EXPORT_SYMBOL(qman_set_wpm); + +int qman_get_wpm(int *wpm_enable) +{ + return qm_get_wpm(wpm_enable); +} +EXPORT_SYMBOL(qman_get_wpm); + +int qman_shutdown_fq(u32 fqid) +{ + struct qman_portal *p; + unsigned long irqflags __maybe_unused; + int ret; + struct qm_portal *low_p; + p = get_affine_portal(); + PORTAL_IRQ_LOCK(p, irqflags); + low_p = &p->p; + ret = qm_shutdown_fq(&low_p, 1, fqid); + PORTAL_IRQ_UNLOCK(p, irqflags); + put_affine_portal(); + return ret; +} + +const struct qm_portal_config *qman_get_qm_portal_config( + struct qman_portal *portal) +{ + return portal->sharing_redirect ? NULL : portal->config; +} diff --git a/drivers/staging/fsl_qbman/qman_low.h b/drivers/staging/fsl_qbman/qman_low.h new file mode 100644 index 0000000..a400a2a --- /dev/null +++ b/drivers/staging/fsl_qbman/qman_low.h @@ -0,0 +1,1302 @@ +/* Copyright 2008-2011 Freescale Semiconductor, Inc. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions are met: + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * * Neither the name of Freescale Semiconductor nor the + * names of its contributors may be used to endorse or promote products + * derived from this software without specific prior written permission. + * + * + * ALTERNATIVELY, this software may be distributed under the terms of the + * GNU General Public License ("GPL") as published by the Free Software + * Foundation, either version 2 of that License or (at your option) any + * later version. + * + * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED + * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE + * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY + * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; + * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND + * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#include "qman_private.h" + +/***************************/ +/* Portal register assists */ +/***************************/ + +/* Cache-inhibited register offsets */ +#define QM_REG_EQCR_PI_CINH 0x0000 +#define QM_REG_EQCR_CI_CINH 0x0004 +#define QM_REG_EQCR_ITR 0x0008 +#define QM_REG_DQRR_PI_CINH 0x0040 +#define QM_REG_DQRR_CI_CINH 0x0044 +#define QM_REG_DQRR_ITR 0x0048 +#define QM_REG_DQRR_DCAP 0x0050 +#define QM_REG_DQRR_SDQCR 0x0054 +#define QM_REG_DQRR_VDQCR 0x0058 +#define QM_REG_DQRR_PDQCR 0x005c +#define QM_REG_MR_PI_CINH 0x0080 +#define QM_REG_MR_CI_CINH 0x0084 +#define QM_REG_MR_ITR 0x0088 +#define QM_REG_CFG 0x0100 +#define QM_REG_ISR 0x0e00 +#define QM_REG_IIR 0x0e0c +#define QM_REG_ITPR 0x0e14 + +/* Cache-enabled register offsets */ +#define QM_CL_EQCR 0x0000 +#define QM_CL_DQRR 0x1000 +#define QM_CL_MR 0x2000 +#define QM_CL_EQCR_PI_CENA 0x3000 +#define QM_CL_EQCR_CI_CENA 0x3100 +#define QM_CL_DQRR_PI_CENA 0x3200 +#define QM_CL_DQRR_CI_CENA 0x3300 +#define QM_CL_MR_PI_CENA 0x3400 +#define QM_CL_MR_CI_CENA 0x3500 +#define QM_CL_CR 0x3800 +#define QM_CL_RR0 0x3900 +#define QM_CL_RR1 0x3940 + +/* BTW, the drivers (and h/w programming model) already obtain the required + * synchronisation for portal accesses via lwsync(), hwsync(), and + * data-dependencies. Use of barrier()s or other order-preserving primitives + * simply degrade performance. Hence the use of the __raw_*() interfaces, which + * simply ensure that the compiler treats the portal registers as volatile (ie. + * non-coherent). */ + +/* Cache-inhibited register access. */ +#define __qm_in(qm, o) __raw_readl((qm)->addr_ci + (o)) +#define __qm_out(qm, o, val) __raw_writel((val), (qm)->addr_ci + (o)) +#define qm_in(reg) __qm_in(&portal->addr, QM_REG_##reg) +#define qm_out(reg, val) __qm_out(&portal->addr, QM_REG_##reg, val) + +/* Cache-enabled (index) register access */ +#define __qm_cl_touch_ro(qm, o) dcbt_ro((qm)->addr_ce + (o)) +#define __qm_cl_touch_rw(qm, o) dcbt_rw((qm)->addr_ce + (o)) +#define __qm_cl_in(qm, o) __raw_readl((qm)->addr_ce + (o)) +#define __qm_cl_out(qm, o, val) \ + do { \ + u32 *__tmpclout = (qm)->addr_ce + (o); \ + __raw_writel((val), __tmpclout); \ + dcbf(__tmpclout); \ + } while (0) +#define __qm_cl_invalidate(qm, o) dcbi((qm)->addr_ce + (o)) +#define qm_cl_touch_ro(reg) __qm_cl_touch_ro(&portal->addr, QM_CL_##reg##_CENA) +#define qm_cl_touch_rw(reg) __qm_cl_touch_rw(&portal->addr, QM_CL_##reg##_CENA) +#define qm_cl_in(reg) __qm_cl_in(&portal->addr, QM_CL_##reg##_CENA) +#define qm_cl_out(reg, val) __qm_cl_out(&portal->addr, QM_CL_##reg##_CENA, val) +#define qm_cl_invalidate(reg)\ + __qm_cl_invalidate(&portal->addr, QM_CL_##reg##_CENA) + +/* Cache-enabled ring access */ +#define qm_cl(base, idx) ((void *)base + ((idx) << 6)) + +/* Cyclic helper for rings. FIXME: once we are able to do fine-grain perf + * analysis, look at using the "extra" bit in the ring index registers to avoid + * cyclic issues. */ +static inline u8 qm_cyc_diff(u8 ringsize, u8 first, u8 last) +{ + /* 'first' is included, 'last' is excluded */ + if (first <= last) + return last - first; + return ringsize + last - first; +} + +/* Portal modes. + * Enum types; + * pmode == production mode + * cmode == consumption mode, + * dmode == h/w dequeue mode. + * Enum values use 3 letter codes. First letter matches the portal mode, + * remaining two letters indicate; + * ci == cache-inhibited portal register + * ce == cache-enabled portal register + * vb == in-band valid-bit (cache-enabled) + * dc == DCA (Discrete Consumption Acknowledgement), DQRR-only + * As for "enum qm_dqrr_dmode", it should be self-explanatory. + */ +enum qm_eqcr_pmode { /* matches QCSP_CFG::EPM */ + qm_eqcr_pci = 0, /* PI index, cache-inhibited */ + qm_eqcr_pce = 1, /* PI index, cache-enabled */ + qm_eqcr_pvb = 2 /* valid-bit */ +}; +enum qm_dqrr_dmode { /* matches QCSP_CFG::DP */ + qm_dqrr_dpush = 0, /* SDQCR + VDQCR */ + qm_dqrr_dpull = 1 /* PDQCR */ +}; +enum qm_dqrr_pmode { /* s/w-only */ + qm_dqrr_pci, /* reads DQRR_PI_CINH */ + qm_dqrr_pce, /* reads DQRR_PI_CENA */ + qm_dqrr_pvb /* reads valid-bit */ +}; +enum qm_dqrr_cmode { /* matches QCSP_CFG::DCM */ + qm_dqrr_cci = 0, /* CI index, cache-inhibited */ + qm_dqrr_cce = 1, /* CI index, cache-enabled */ + qm_dqrr_cdc = 2 /* Discrete Consumption Acknowledgement */ +}; +enum qm_mr_pmode { /* s/w-only */ + qm_mr_pci, /* reads MR_PI_CINH */ + qm_mr_pce, /* reads MR_PI_CENA */ + qm_mr_pvb /* reads valid-bit */ +}; +enum qm_mr_cmode { /* matches QCSP_CFG::MM */ + qm_mr_cci = 0, /* CI index, cache-inhibited */ + qm_mr_cce = 1 /* CI index, cache-enabled */ +}; + + +/* ------------------------- */ +/* --- Portal structures --- */ + +#define QM_EQCR_SIZE 8 +#define QM_DQRR_SIZE 16 +#define QM_MR_SIZE 8 + +struct qm_eqcr { + struct qm_eqcr_entry *ring, *cursor; + u8 ci, available, ithresh, vbit; +#ifdef CONFIG_FSL_DPA_CHECKING + u32 busy; + enum qm_eqcr_pmode pmode; +#endif +}; + +struct qm_dqrr { + const struct qm_dqrr_entry *ring, *cursor; + u8 pi, ci, fill, ithresh, vbit; +#ifdef CONFIG_FSL_DPA_CHECKING + enum qm_dqrr_dmode dmode; + enum qm_dqrr_pmode pmode; + enum qm_dqrr_cmode cmode; +#endif +}; + +struct qm_mr { + const struct qm_mr_entry *ring, *cursor; + u8 pi, ci, fill, ithresh, vbit; +#ifdef CONFIG_FSL_DPA_CHECKING + enum qm_mr_pmode pmode; + enum qm_mr_cmode cmode; +#endif +}; + +struct qm_mc { + struct qm_mc_command *cr; + struct qm_mc_result *rr; + u8 rridx, vbit; +#ifdef CONFIG_FSL_DPA_CHECKING + enum { + /* Can be _mc_start()ed */ + qman_mc_idle, + /* Can be _mc_commit()ed or _mc_abort()ed */ + qman_mc_user, + /* Can only be _mc_retry()ed */ + qman_mc_hw + } state; +#endif +}; + +#define QM_PORTAL_ALIGNMENT ____cacheline_aligned + +struct qm_addr { + void __iomem *addr_ce; /* cache-enabled */ + void __iomem *addr_ci; /* cache-inhibited */ +}; + +struct qm_portal { + /* In the non-CONFIG_FSL_DPA_CHECKING case, the following stuff up to + * and including 'mc' fits within a cacheline (yay!). The 'config' part + * is setup-only, so isn't a cause for a concern. In other words, don't + * rearrange this structure on a whim, there be dragons ... */ + struct qm_addr addr; + struct qm_eqcr eqcr; + struct qm_dqrr dqrr; + struct qm_mr mr; + struct qm_mc mc; +} QM_PORTAL_ALIGNMENT; + + +/* ---------------- */ +/* --- EQCR API --- */ + +/* Bit-wise logic to wrap a ring pointer by clearing the "carry bit" */ +#define EQCR_CARRYCLEAR(p) \ + (void *)((unsigned long)(p) & (~(unsigned long)(QM_EQCR_SIZE << 6))) + +/* Bit-wise logic to convert a ring pointer to a ring index */ +static inline u8 EQCR_PTR2IDX(struct qm_eqcr_entry *e) +{ + return ((uintptr_t)e >> 6) & (QM_EQCR_SIZE - 1); +} + +/* Increment the 'cursor' ring pointer, taking 'vbit' into account */ +static inline void EQCR_INC(struct qm_eqcr *eqcr) +{ + /* NB: this is odd-looking, but experiments show that it generates fast + * code with essentially no branching overheads. We increment to the + * next EQCR pointer and handle overflow and 'vbit'. */ + struct qm_eqcr_entry *partial = eqcr->cursor + 1; + eqcr->cursor = EQCR_CARRYCLEAR(partial); + if (partial != eqcr->cursor) + eqcr->vbit ^= QM_EQCR_VERB_VBIT; +} + +static inline int qm_eqcr_init(struct qm_portal *portal, + enum qm_eqcr_pmode pmode, + unsigned int eq_stash_thresh, + int eq_stash_prio) +{ + /* This use of 'register', as well as all other occurrences, is because + * it has been observed to generate much faster code with gcc than is + * otherwise the case. */ + register struct qm_eqcr *eqcr = &portal->eqcr; + u32 cfg; + u8 pi; + + eqcr->ring = portal->addr.addr_ce + QM_CL_EQCR; + eqcr->ci = qm_in(EQCR_CI_CINH) & (QM_EQCR_SIZE - 1); + qm_cl_invalidate(EQCR_CI); + pi = qm_in(EQCR_PI_CINH) & (QM_EQCR_SIZE - 1); + eqcr->cursor = eqcr->ring + pi; + eqcr->vbit = (qm_in(EQCR_PI_CINH) & QM_EQCR_SIZE) ? + QM_EQCR_VERB_VBIT : 0; + eqcr->available = QM_EQCR_SIZE - 1 - + qm_cyc_diff(QM_EQCR_SIZE, eqcr->ci, pi); + eqcr->ithresh = qm_in(EQCR_ITR); +#ifdef CONFIG_FSL_DPA_CHECKING + eqcr->busy = 0; + eqcr->pmode = pmode; +#endif + cfg = (qm_in(CFG) & 0x00ffffff) | + (eq_stash_thresh << 28) | /* QCSP_CFG: EST */ + (eq_stash_prio << 26) | /* QCSP_CFG: EP */ + ((pmode & 0x3) << 24); /* QCSP_CFG::EPM */ + qm_out(CFG, cfg); + return 0; +} + +static inline unsigned int qm_eqcr_get_ci_stashing(struct qm_portal *portal) +{ + return (qm_in(CFG) >> 28) & 0x7; +} + +static inline void qm_eqcr_finish(struct qm_portal *portal) +{ + register struct qm_eqcr *eqcr = &portal->eqcr; + u8 pi = qm_in(EQCR_PI_CINH) & (QM_EQCR_SIZE - 1); + u8 ci = qm_in(EQCR_CI_CINH) & (QM_EQCR_SIZE - 1); + + DPA_ASSERT(!eqcr->busy); + if (pi != EQCR_PTR2IDX(eqcr->cursor)) + pr_crit("losing uncommited EQCR entries\n"); + if (ci != eqcr->ci) + pr_crit("missing existing EQCR completions\n"); + if (eqcr->ci != EQCR_PTR2IDX(eqcr->cursor)) + pr_crit("EQCR destroyed unquiesced\n"); +} + +static inline struct qm_eqcr_entry *qm_eqcr_start_no_stash(struct qm_portal + *portal) +{ + register struct qm_eqcr *eqcr = &portal->eqcr; + DPA_ASSERT(!eqcr->busy); + if (!eqcr->available) + return NULL; + + +#ifdef CONFIG_FSL_DPA_CHECKING + eqcr->busy = 1; +#endif + dcbz_64(eqcr->cursor); + return eqcr->cursor; +} + +static inline struct qm_eqcr_entry *qm_eqcr_start_stash(struct qm_portal + *portal) +{ + register struct qm_eqcr *eqcr = &portal->eqcr; + u8 diff, old_ci; + + DPA_ASSERT(!eqcr->busy); + if (!eqcr->available) { + old_ci = eqcr->ci; + eqcr->ci = qm_cl_in(EQCR_CI) & (QM_EQCR_SIZE - 1); + diff = qm_cyc_diff(QM_EQCR_SIZE, old_ci, eqcr->ci); + eqcr->available += diff; + if (!diff) + return NULL; + } +#ifdef CONFIG_FSL_DPA_CHECKING + eqcr->busy = 1; +#endif + dcbz_64(eqcr->cursor); + return eqcr->cursor; +} + +static inline void qm_eqcr_abort(struct qm_portal *portal) +{ + __maybe_unused register struct qm_eqcr *eqcr = &portal->eqcr; + DPA_ASSERT(eqcr->busy); +#ifdef CONFIG_FSL_DPA_CHECKING + eqcr->busy = 0; +#endif +} + +static inline struct qm_eqcr_entry *qm_eqcr_pend_and_next( + struct qm_portal *portal, u8 myverb) +{ + register struct qm_eqcr *eqcr = &portal->eqcr; + DPA_ASSERT(eqcr->busy); + DPA_ASSERT(eqcr->pmode != qm_eqcr_pvb); + if (eqcr->available == 1) + return NULL; + eqcr->cursor->__dont_write_directly__verb = myverb | eqcr->vbit; + dcbf(eqcr->cursor); + EQCR_INC(eqcr); + eqcr->available--; + dcbz_64(eqcr->cursor); + return eqcr->cursor; +} + +#define EQCR_COMMIT_CHECKS(eqcr) \ +do { \ + DPA_ASSERT(eqcr->busy); \ + DPA_ASSERT(eqcr->cursor->orp == (eqcr->cursor->orp & 0x00ffffff)); \ + DPA_ASSERT(eqcr->cursor->fqid == (eqcr->cursor->fqid & 0x00ffffff)); \ +} while (0) + +static inline void qm_eqcr_pci_commit(struct qm_portal *portal, u8 myverb) +{ + register struct qm_eqcr *eqcr = &portal->eqcr; + EQCR_COMMIT_CHECKS(eqcr); + DPA_ASSERT(eqcr->pmode == qm_eqcr_pci); + eqcr->cursor->__dont_write_directly__verb = myverb | eqcr->vbit; + EQCR_INC(eqcr); + eqcr->available--; + dcbf(eqcr->cursor); + hwsync(); + qm_out(EQCR_PI_CINH, EQCR_PTR2IDX(eqcr->cursor)); +#ifdef CONFIG_FSL_DPA_CHECKING + eqcr->busy = 0; +#endif +} + +static inline void qm_eqcr_pce_prefetch(struct qm_portal *portal) +{ + __maybe_unused register struct qm_eqcr *eqcr = &portal->eqcr; + DPA_ASSERT(eqcr->pmode == qm_eqcr_pce); + qm_cl_invalidate(EQCR_PI); + qm_cl_touch_rw(EQCR_PI); +} + +static inline void qm_eqcr_pce_commit(struct qm_portal *portal, u8 myverb) +{ + register struct qm_eqcr *eqcr = &portal->eqcr; + EQCR_COMMIT_CHECKS(eqcr); + DPA_ASSERT(eqcr->pmode == qm_eqcr_pce); + eqcr->cursor->__dont_write_directly__verb = myverb | eqcr->vbit; + EQCR_INC(eqcr); + eqcr->available--; + dcbf(eqcr->cursor); + lwsync(); + qm_cl_out(EQCR_PI, EQCR_PTR2IDX(eqcr->cursor)); +#ifdef CONFIG_FSL_DPA_CHECKING + eqcr->busy = 0; +#endif +} + +static inline void qm_eqcr_pvb_commit(struct qm_portal *portal, u8 myverb) +{ + register struct qm_eqcr *eqcr = &portal->eqcr; + struct qm_eqcr_entry *eqcursor; + EQCR_COMMIT_CHECKS(eqcr); + DPA_ASSERT(eqcr->pmode == qm_eqcr_pvb); + lwsync(); + eqcursor = eqcr->cursor; + eqcursor->__dont_write_directly__verb = myverb | eqcr->vbit; + dcbf(eqcursor); + EQCR_INC(eqcr); + eqcr->available--; +#ifdef CONFIG_FSL_DPA_CHECKING + eqcr->busy = 0; +#endif +} + +static inline u8 qm_eqcr_cci_update(struct qm_portal *portal) +{ + register struct qm_eqcr *eqcr = &portal->eqcr; + u8 diff, old_ci = eqcr->ci; + eqcr->ci = qm_in(EQCR_CI_CINH) & (QM_EQCR_SIZE - 1); + diff = qm_cyc_diff(QM_EQCR_SIZE, old_ci, eqcr->ci); + eqcr->available += diff; + return diff; +} + +static inline void qm_eqcr_cce_prefetch(struct qm_portal *portal) +{ + __maybe_unused register struct qm_eqcr *eqcr = &portal->eqcr; + qm_cl_touch_ro(EQCR_CI); +} + +static inline u8 qm_eqcr_cce_update(struct qm_portal *portal) +{ + register struct qm_eqcr *eqcr = &portal->eqcr; + u8 diff, old_ci = eqcr->ci; + eqcr->ci = qm_cl_in(EQCR_CI) & (QM_EQCR_SIZE - 1); + qm_cl_invalidate(EQCR_CI); + diff = qm_cyc_diff(QM_EQCR_SIZE, old_ci, eqcr->ci); + eqcr->available += diff; + return diff; +} + +static inline u8 qm_eqcr_get_ithresh(struct qm_portal *portal) +{ + register struct qm_eqcr *eqcr = &portal->eqcr; + return eqcr->ithresh; +} + +static inline void qm_eqcr_set_ithresh(struct qm_portal *portal, u8 ithresh) +{ + register struct qm_eqcr *eqcr = &portal->eqcr; + eqcr->ithresh = ithresh; + qm_out(EQCR_ITR, ithresh); +} + +static inline u8 qm_eqcr_get_avail(struct qm_portal *portal) +{ + register struct qm_eqcr *eqcr = &portal->eqcr; + return eqcr->available; +} + +static inline u8 qm_eqcr_get_fill(struct qm_portal *portal) +{ + register struct qm_eqcr *eqcr = &portal->eqcr; + return QM_EQCR_SIZE - 1 - eqcr->available; +} + + +/* ---------------- */ +/* --- DQRR API --- */ + +/* FIXME: many possible improvements; + * - look at changing the API to use pointer rather than index parameters now + * that 'cursor' is a pointer, + * - consider moving other parameters to pointer if it could help (ci) + */ + +#define DQRR_CARRYCLEAR(p) \ + (void *)((unsigned long)(p) & (~(unsigned long)(QM_DQRR_SIZE << 6))) + +static inline u8 DQRR_PTR2IDX(const struct qm_dqrr_entry *e) +{ + return ((uintptr_t)e >> 6) & (QM_DQRR_SIZE - 1); +} + +static inline const struct qm_dqrr_entry *DQRR_INC( + const struct qm_dqrr_entry *e) +{ + return DQRR_CARRYCLEAR(e + 1); +} + +static inline void qm_dqrr_set_maxfill(struct qm_portal *portal, u8 mf) +{ + qm_out(CFG, (qm_in(CFG) & 0xff0fffff) | + ((mf & (QM_DQRR_SIZE - 1)) << 20)); +} + +static inline int qm_dqrr_init(struct qm_portal *portal, + const struct qm_portal_config *config, + enum qm_dqrr_dmode dmode, + __maybe_unused enum qm_dqrr_pmode pmode, + enum qm_dqrr_cmode cmode, u8 max_fill) +{ + register struct qm_dqrr *dqrr = &portal->dqrr; + u32 cfg; + + /* Make sure the DQRR will be idle when we enable */ + qm_out(DQRR_SDQCR, 0); + qm_out(DQRR_VDQCR, 0); + qm_out(DQRR_PDQCR, 0); + dqrr->ring = portal->addr.addr_ce + QM_CL_DQRR; + dqrr->pi = qm_in(DQRR_PI_CINH) & (QM_DQRR_SIZE - 1); + dqrr->ci = qm_in(DQRR_CI_CINH) & (QM_DQRR_SIZE - 1); + dqrr->cursor = dqrr->ring + dqrr->ci; + dqrr->fill = qm_cyc_diff(QM_DQRR_SIZE, dqrr->ci, dqrr->pi); + dqrr->vbit = (qm_in(DQRR_PI_CINH) & QM_DQRR_SIZE) ? + QM_DQRR_VERB_VBIT : 0; + dqrr->ithresh = qm_in(DQRR_ITR); +#ifdef CONFIG_FSL_DPA_CHECKING + dqrr->dmode = dmode; + dqrr->pmode = pmode; + dqrr->cmode = cmode; +#endif + /* Invalidate every ring entry before beginning */ + for (cfg = 0; cfg < QM_DQRR_SIZE; cfg++) + dcbi(qm_cl(dqrr->ring, cfg)); + cfg = (qm_in(CFG) & 0xff000f00) | + ((max_fill & (QM_DQRR_SIZE - 1)) << 20) | /* DQRR_MF */ + ((dmode & 1) << 18) | /* DP */ + ((cmode & 3) << 16) | /* DCM */ + 0xa0 | /* RE+SE */ + (0 ? 0x40 : 0) | /* Ignore RP */ + (0 ? 0x10 : 0); /* Ignore SP */ + qm_out(CFG, cfg); + qm_dqrr_set_maxfill(portal, max_fill); + return 0; +} + +static inline void qm_dqrr_finish(struct qm_portal *portal) +{ + __maybe_unused register struct qm_dqrr *dqrr = &portal->dqrr; +#ifdef CONFIG_FSL_DPA_CHECKING + if ((dqrr->cmode != qm_dqrr_cdc) && + (dqrr->ci != DQRR_PTR2IDX(dqrr->cursor))) + pr_crit("Ignoring completed DQRR entries\n"); +#endif +} + +static inline const struct qm_dqrr_entry *qm_dqrr_current( + struct qm_portal *portal) +{ + register struct qm_dqrr *dqrr = &portal->dqrr; + if (!dqrr->fill) + return NULL; + return dqrr->cursor; +} + +static inline u8 qm_dqrr_cursor(struct qm_portal *portal) +{ + register struct qm_dqrr *dqrr = &portal->dqrr; + return DQRR_PTR2IDX(dqrr->cursor); +} + +static inline u8 qm_dqrr_next(struct qm_portal *portal) +{ + register struct qm_dqrr *dqrr = &portal->dqrr; + DPA_ASSERT(dqrr->fill); + dqrr->cursor = DQRR_INC(dqrr->cursor); + return --dqrr->fill; +} + +static inline u8 qm_dqrr_pci_update(struct qm_portal *portal) +{ + register struct qm_dqrr *dqrr = &portal->dqrr; + u8 diff, old_pi = dqrr->pi; + DPA_ASSERT(dqrr->pmode == qm_dqrr_pci); + dqrr->pi = qm_in(DQRR_PI_CINH) & (QM_DQRR_SIZE - 1); + diff = qm_cyc_diff(QM_DQRR_SIZE, old_pi, dqrr->pi); + dqrr->fill += diff; + return diff; +} + +static inline void qm_dqrr_pce_prefetch(struct qm_portal *portal) +{ + __maybe_unused register struct qm_dqrr *dqrr = &portal->dqrr; + DPA_ASSERT(dqrr->pmode == qm_dqrr_pce); + qm_cl_invalidate(DQRR_PI); + qm_cl_touch_ro(DQRR_PI); +} + +static inline u8 qm_dqrr_pce_update(struct qm_portal *portal) +{ + register struct qm_dqrr *dqrr = &portal->dqrr; + u8 diff, old_pi = dqrr->pi; + DPA_ASSERT(dqrr->pmode == qm_dqrr_pce); + dqrr->pi = qm_cl_in(DQRR_PI) & (QM_DQRR_SIZE - 1); + diff = qm_cyc_diff(QM_DQRR_SIZE, old_pi, dqrr->pi); + dqrr->fill += diff; + return diff; +} + +static inline void qm_dqrr_pvb_update(struct qm_portal *portal) +{ + register struct qm_dqrr *dqrr = &portal->dqrr; + const struct qm_dqrr_entry *res = qm_cl(dqrr->ring, dqrr->pi); + DPA_ASSERT(dqrr->pmode == qm_dqrr_pvb); + /* when accessing 'verb', use __raw_readb() to ensure that compiler + * inlining doesn't try to optimise out "excess reads". */ + if ((__raw_readb(&res->verb) & QM_DQRR_VERB_VBIT) == dqrr->vbit) { + dqrr->pi = (dqrr->pi + 1) & (QM_DQRR_SIZE - 1); + if (!dqrr->pi) + dqrr->vbit ^= QM_DQRR_VERB_VBIT; + dqrr->fill++; + } +} + +static inline void qm_dqrr_cci_consume(struct qm_portal *portal, u8 num) +{ + register struct qm_dqrr *dqrr = &portal->dqrr; + DPA_ASSERT(dqrr->cmode == qm_dqrr_cci); + dqrr->ci = (dqrr->ci + num) & (QM_DQRR_SIZE - 1); + qm_out(DQRR_CI_CINH, dqrr->ci); +} + +static inline void qm_dqrr_cci_consume_to_current(struct qm_portal *portal) +{ + register struct qm_dqrr *dqrr = &portal->dqrr; + DPA_ASSERT(dqrr->cmode == qm_dqrr_cci); + dqrr->ci = DQRR_PTR2IDX(dqrr->cursor); + qm_out(DQRR_CI_CINH, dqrr->ci); +} + +static inline void qm_dqrr_cce_prefetch(struct qm_portal *portal) +{ + __maybe_unused register struct qm_dqrr *dqrr = &portal->dqrr; + DPA_ASSERT(dqrr->cmode == qm_dqrr_cce); + qm_cl_invalidate(DQRR_CI); + qm_cl_touch_rw(DQRR_CI); +} + +static inline void qm_dqrr_cce_consume(struct qm_portal *portal, u8 num) +{ + register struct qm_dqrr *dqrr = &portal->dqrr; + DPA_ASSERT(dqrr->cmode == qm_dqrr_cce); + dqrr->ci = (dqrr->ci + num) & (QM_DQRR_SIZE - 1); + qm_cl_out(DQRR_CI, dqrr->ci); +} + +static inline void qm_dqrr_cce_consume_to_current(struct qm_portal *portal) +{ + register struct qm_dqrr *dqrr = &portal->dqrr; + DPA_ASSERT(dqrr->cmode == qm_dqrr_cce); + dqrr->ci = DQRR_PTR2IDX(dqrr->cursor); + qm_cl_out(DQRR_CI, dqrr->ci); +} + +static inline void qm_dqrr_cdc_consume_1(struct qm_portal *portal, u8 idx, + int park) +{ + __maybe_unused register struct qm_dqrr *dqrr = &portal->dqrr; + DPA_ASSERT(dqrr->cmode == qm_dqrr_cdc); + DPA_ASSERT(idx < QM_DQRR_SIZE); + qm_out(DQRR_DCAP, (0 << 8) | /* S */ + ((park ? 1 : 0) << 6) | /* PK */ + idx); /* DCAP_CI */ +} + +static inline void qm_dqrr_cdc_consume_1ptr(struct qm_portal *portal, + const struct qm_dqrr_entry *dq, + int park) +{ + __maybe_unused register struct qm_dqrr *dqrr = &portal->dqrr; + u8 idx = DQRR_PTR2IDX(dq); + DPA_ASSERT(dqrr->cmode == qm_dqrr_cdc); + DPA_ASSERT((dqrr->ring + idx) == dq); + DPA_ASSERT(idx < QM_DQRR_SIZE); + qm_out(DQRR_DCAP, (0 << 8) | /* DQRR_DCAP::S */ + ((park ? 1 : 0) << 6) | /* DQRR_DCAP::PK */ + idx); /* DQRR_DCAP::DCAP_CI */ +} + +static inline void qm_dqrr_cdc_consume_n(struct qm_portal *portal, u16 bitmask) +{ + __maybe_unused register struct qm_dqrr *dqrr = &portal->dqrr; + DPA_ASSERT(dqrr->cmode == qm_dqrr_cdc); + qm_out(DQRR_DCAP, (1 << 8) | /* DQRR_DCAP::S */ + ((u32)bitmask << 16)); /* DQRR_DCAP::DCAP_CI */ +} + +static inline u8 qm_dqrr_cdc_cci(struct qm_portal *portal) +{ + __maybe_unused register struct qm_dqrr *dqrr = &portal->dqrr; + DPA_ASSERT(dqrr->cmode == qm_dqrr_cdc); + return qm_in(DQRR_CI_CINH) & (QM_DQRR_SIZE - 1); +} + +static inline void qm_dqrr_cdc_cce_prefetch(struct qm_portal *portal) +{ + __maybe_unused register struct qm_dqrr *dqrr = &portal->dqrr; + DPA_ASSERT(dqrr->cmode == qm_dqrr_cdc); + qm_cl_invalidate(DQRR_CI); + qm_cl_touch_ro(DQRR_CI); +} + +static inline u8 qm_dqrr_cdc_cce(struct qm_portal *portal) +{ + __maybe_unused register struct qm_dqrr *dqrr = &portal->dqrr; + DPA_ASSERT(dqrr->cmode == qm_dqrr_cdc); + return qm_cl_in(DQRR_CI) & (QM_DQRR_SIZE - 1); +} + +static inline u8 qm_dqrr_get_ci(struct qm_portal *portal) +{ + register struct qm_dqrr *dqrr = &portal->dqrr; + DPA_ASSERT(dqrr->cmode != qm_dqrr_cdc); + return dqrr->ci; +} + +static inline void qm_dqrr_park(struct qm_portal *portal, u8 idx) +{ + __maybe_unused register struct qm_dqrr *dqrr = &portal->dqrr; + DPA_ASSERT(dqrr->cmode != qm_dqrr_cdc); + qm_out(DQRR_DCAP, (0 << 8) | /* S */ + (1 << 6) | /* PK */ + (idx & (QM_DQRR_SIZE - 1))); /* DCAP_CI */ +} + +static inline void qm_dqrr_park_current(struct qm_portal *portal) +{ + register struct qm_dqrr *dqrr = &portal->dqrr; + DPA_ASSERT(dqrr->cmode != qm_dqrr_cdc); + qm_out(DQRR_DCAP, (0 << 8) | /* S */ + (1 << 6) | /* PK */ + DQRR_PTR2IDX(dqrr->cursor)); /* DCAP_CI */ +} + +static inline void qm_dqrr_sdqcr_set(struct qm_portal *portal, u32 sdqcr) +{ + qm_out(DQRR_SDQCR, sdqcr); +} + +static inline u32 qm_dqrr_sdqcr_get(struct qm_portal *portal) +{ + return qm_in(DQRR_SDQCR); +} + +static inline void qm_dqrr_vdqcr_set(struct qm_portal *portal, u32 vdqcr) +{ + qm_out(DQRR_VDQCR, vdqcr); +} + +static inline u32 qm_dqrr_vdqcr_get(struct qm_portal *portal) +{ + return qm_in(DQRR_VDQCR); +} + +static inline void qm_dqrr_pdqcr_set(struct qm_portal *portal, u32 pdqcr) +{ + qm_out(DQRR_PDQCR, pdqcr); +} + +static inline u32 qm_dqrr_pdqcr_get(struct qm_portal *portal) +{ + return qm_in(DQRR_PDQCR); +} + +static inline u8 qm_dqrr_get_ithresh(struct qm_portal *portal) +{ + register struct qm_dqrr *dqrr = &portal->dqrr; + return dqrr->ithresh; +} + +static inline void qm_dqrr_set_ithresh(struct qm_portal *portal, u8 ithresh) +{ + qm_out(DQRR_ITR, ithresh); +} + +static inline u8 qm_dqrr_get_maxfill(struct qm_portal *portal) +{ + return (qm_in(CFG) & 0x00f00000) >> 20; +} + + +/* -------------- */ +/* --- MR API --- */ + +#define MR_CARRYCLEAR(p) \ + (void *)((unsigned long)(p) & (~(unsigned long)(QM_MR_SIZE << 6))) + +static inline u8 MR_PTR2IDX(const struct qm_mr_entry *e) +{ + return ((uintptr_t)e >> 6) & (QM_MR_SIZE - 1); +} + +static inline const struct qm_mr_entry *MR_INC(const struct qm_mr_entry *e) +{ + return MR_CARRYCLEAR(e + 1); +} + +static inline int qm_mr_init(struct qm_portal *portal, enum qm_mr_pmode pmode, + enum qm_mr_cmode cmode) +{ + register struct qm_mr *mr = &portal->mr; + u32 cfg; + + mr->ring = portal->addr.addr_ce + QM_CL_MR; + mr->pi = qm_in(MR_PI_CINH) & (QM_MR_SIZE - 1); + mr->ci = qm_in(MR_CI_CINH) & (QM_MR_SIZE - 1); + mr->cursor = mr->ring + mr->ci; + mr->fill = qm_cyc_diff(QM_MR_SIZE, mr->ci, mr->pi); + mr->vbit = (qm_in(MR_PI_CINH) & QM_MR_SIZE) ? QM_MR_VERB_VBIT : 0; + mr->ithresh = qm_in(MR_ITR); +#ifdef CONFIG_FSL_DPA_CHECKING + mr->pmode = pmode; + mr->cmode = cmode; +#endif + cfg = (qm_in(CFG) & 0xfffff0ff) | + ((cmode & 1) << 8); /* QCSP_CFG:MM */ + qm_out(CFG, cfg); + return 0; +} + +static inline void qm_mr_finish(struct qm_portal *portal) +{ + register struct qm_mr *mr = &portal->mr; + if (mr->ci != MR_PTR2IDX(mr->cursor)) + pr_crit("Ignoring completed MR entries\n"); +} + +static inline const struct qm_mr_entry *qm_mr_current(struct qm_portal *portal) +{ + register struct qm_mr *mr = &portal->mr; + if (!mr->fill) + return NULL; + return mr->cursor; +} + +static inline u8 qm_mr_cursor(struct qm_portal *portal) +{ + register struct qm_mr *mr = &portal->mr; + return MR_PTR2IDX(mr->cursor); +} + +static inline u8 qm_mr_next(struct qm_portal *portal) +{ + register struct qm_mr *mr = &portal->mr; + DPA_ASSERT(mr->fill); + mr->cursor = MR_INC(mr->cursor); + return --mr->fill; +} + +static inline u8 qm_mr_pci_update(struct qm_portal *portal) +{ + register struct qm_mr *mr = &portal->mr; + u8 diff, old_pi = mr->pi; + DPA_ASSERT(mr->pmode == qm_mr_pci); + mr->pi = qm_in(MR_PI_CINH); + diff = qm_cyc_diff(QM_MR_SIZE, old_pi, mr->pi); + mr->fill += diff; + return diff; +} + +static inline void qm_mr_pce_prefetch(struct qm_portal *portal) +{ + __maybe_unused register struct qm_mr *mr = &portal->mr; + DPA_ASSERT(mr->pmode == qm_mr_pce); + qm_cl_invalidate(MR_PI); + qm_cl_touch_ro(MR_PI); +} + +static inline u8 qm_mr_pce_update(struct qm_portal *portal) +{ + register struct qm_mr *mr = &portal->mr; + u8 diff, old_pi = mr->pi; + DPA_ASSERT(mr->pmode == qm_mr_pce); + mr->pi = qm_cl_in(MR_PI) & (QM_MR_SIZE - 1); + diff = qm_cyc_diff(QM_MR_SIZE, old_pi, mr->pi); + mr->fill += diff; + return diff; +} + +static inline void qm_mr_pvb_update(struct qm_portal *portal) +{ + register struct qm_mr *mr = &portal->mr; + const struct qm_mr_entry *res = qm_cl(mr->ring, mr->pi); + DPA_ASSERT(mr->pmode == qm_mr_pvb); + /* when accessing 'verb', use __raw_readb() to ensure that compiler + * inlining doesn't try to optimise out "excess reads". */ + if ((__raw_readb(&res->verb) & QM_MR_VERB_VBIT) == mr->vbit) { + mr->pi = (mr->pi + 1) & (QM_MR_SIZE - 1); + if (!mr->pi) + mr->vbit ^= QM_MR_VERB_VBIT; + mr->fill++; + res = MR_INC(res); + } + dcbit_ro(res); +} + +static inline void qm_mr_cci_consume(struct qm_portal *portal, u8 num) +{ + register struct qm_mr *mr = &portal->mr; + DPA_ASSERT(mr->cmode == qm_mr_cci); + mr->ci = (mr->ci + num) & (QM_MR_SIZE - 1); + qm_out(MR_CI_CINH, mr->ci); +} + +static inline void qm_mr_cci_consume_to_current(struct qm_portal *portal) +{ + register struct qm_mr *mr = &portal->mr; + DPA_ASSERT(mr->cmode == qm_mr_cci); + mr->ci = MR_PTR2IDX(mr->cursor); + qm_out(MR_CI_CINH, mr->ci); +} + +static inline void qm_mr_cce_prefetch(struct qm_portal *portal) +{ + __maybe_unused register struct qm_mr *mr = &portal->mr; + DPA_ASSERT(mr->cmode == qm_mr_cce); + qm_cl_invalidate(MR_CI); + qm_cl_touch_rw(MR_CI); +} + +static inline void qm_mr_cce_consume(struct qm_portal *portal, u8 num) +{ + register struct qm_mr *mr = &portal->mr; + DPA_ASSERT(mr->cmode == qm_mr_cce); + mr->ci = (mr->ci + num) & (QM_MR_SIZE - 1); + qm_cl_out(MR_CI, mr->ci); +} + +static inline void qm_mr_cce_consume_to_current(struct qm_portal *portal) +{ + register struct qm_mr *mr = &portal->mr; + DPA_ASSERT(mr->cmode == qm_mr_cce); + mr->ci = MR_PTR2IDX(mr->cursor); + qm_cl_out(MR_CI, mr->ci); +} + +static inline u8 qm_mr_get_ci(struct qm_portal *portal) +{ + register struct qm_mr *mr = &portal->mr; + return mr->ci; +} + +static inline u8 qm_mr_get_ithresh(struct qm_portal *portal) +{ + register struct qm_mr *mr = &portal->mr; + return mr->ithresh; +} + +static inline void qm_mr_set_ithresh(struct qm_portal *portal, u8 ithresh) +{ + qm_out(MR_ITR, ithresh); +} + + +/* ------------------------------ */ +/* --- Management command API --- */ + +static inline int qm_mc_init(struct qm_portal *portal) +{ + register struct qm_mc *mc = &portal->mc; + mc->cr = portal->addr.addr_ce + QM_CL_CR; + mc->rr = portal->addr.addr_ce + QM_CL_RR0; + mc->rridx = (__raw_readb(&mc->cr->__dont_write_directly__verb) & + QM_MCC_VERB_VBIT) ? 0 : 1; + mc->vbit = mc->rridx ? QM_MCC_VERB_VBIT : 0; +#ifdef CONFIG_FSL_DPA_CHECKING + mc->state = qman_mc_idle; +#endif + return 0; +} + +static inline void qm_mc_finish(struct qm_portal *portal) +{ + __maybe_unused register struct qm_mc *mc = &portal->mc; + DPA_ASSERT(mc->state == qman_mc_idle); +#ifdef CONFIG_FSL_DPA_CHECKING + if (mc->state != qman_mc_idle) + pr_crit("Losing incomplete MC command\n"); +#endif +} + +static inline struct qm_mc_command *qm_mc_start(struct qm_portal *portal) +{ + register struct qm_mc *mc = &portal->mc; + DPA_ASSERT(mc->state == qman_mc_idle); +#ifdef CONFIG_FSL_DPA_CHECKING + mc->state = qman_mc_user; +#endif + dcbz_64(mc->cr); + return mc->cr; +} + +static inline void qm_mc_abort(struct qm_portal *portal) +{ + __maybe_unused register struct qm_mc *mc = &portal->mc; + DPA_ASSERT(mc->state == qman_mc_user); +#ifdef CONFIG_FSL_DPA_CHECKING + mc->state = qman_mc_idle; +#endif +} + +static inline void qm_mc_commit(struct qm_portal *portal, u8 myverb) +{ + register struct qm_mc *mc = &portal->mc; + struct qm_mc_result *rr = mc->rr + mc->rridx; + DPA_ASSERT(mc->state == qman_mc_user); + lwsync(); + mc->cr->__dont_write_directly__verb = myverb | mc->vbit; + dcbf(mc->cr); + dcbit_ro(rr); +#ifdef CONFIG_FSL_DPA_CHECKING + mc->state = qman_mc_hw; +#endif +} + +static inline struct qm_mc_result *qm_mc_result(struct qm_portal *portal) +{ + register struct qm_mc *mc = &portal->mc; + struct qm_mc_result *rr = mc->rr + mc->rridx; + DPA_ASSERT(mc->state == qman_mc_hw); + /* The inactive response register's verb byte always returns zero until + * its command is submitted and completed. This includes the valid-bit, + * in case you were wondering... */ + if (!__raw_readb(&rr->verb)) { + dcbit_ro(rr); + return NULL; + } + mc->rridx ^= 1; + mc->vbit ^= QM_MCC_VERB_VBIT; +#ifdef CONFIG_FSL_DPA_CHECKING + mc->state = qman_mc_idle; +#endif + return rr; +} + + +/* ------------------------------------- */ +/* --- Portal interrupt register API --- */ + +static inline int qm_isr_init(__always_unused struct qm_portal *portal) +{ + return 0; +} + +static inline void qm_isr_finish(__always_unused struct qm_portal *portal) +{ +} + +static inline void qm_isr_set_iperiod(struct qm_portal *portal, u16 iperiod) +{ + qm_out(ITPR, iperiod); +} + +static inline u32 __qm_isr_read(struct qm_portal *portal, enum qm_isr_reg n) +{ + return __qm_in(&portal->addr, QM_REG_ISR + (n << 2)); +} + +static inline void __qm_isr_write(struct qm_portal *portal, enum qm_isr_reg n, + u32 val) +{ + __qm_out(&portal->addr, QM_REG_ISR + (n << 2), val); +} + +/* Cleanup FQs */ +static inline int qm_shutdown_fq(struct qm_portal **portal, int portal_count, + u32 fqid) +{ + + struct qm_mc_command *mcc; + struct qm_mc_result *mcr; + u8 state; + int orl_empty, fq_empty, i, drain = 0; + u32 result; + u32 channel, wq; + + /* Determine the state of the FQID */ + mcc = qm_mc_start(portal[0]); + mcc->queryfq_np.fqid = fqid; + qm_mc_commit(portal[0], QM_MCC_VERB_QUERYFQ_NP); + while (!(mcr = qm_mc_result(portal[0]))) + cpu_relax(); + DPA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == QM_MCR_VERB_QUERYFQ_NP); + state = mcr->queryfq_np.state & QM_MCR_NP_STATE_MASK; + if (state == QM_MCR_NP_STATE_OOS) + return 0; /* Already OOS, no need to do anymore checks */ + + /* Query which channel the FQ is using */ + mcc = qm_mc_start(portal[0]); + mcc->queryfq.fqid = fqid; + qm_mc_commit(portal[0], QM_MCC_VERB_QUERYFQ); + while (!(mcr = qm_mc_result(portal[0]))) + cpu_relax(); + DPA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == QM_MCR_VERB_QUERYFQ); + + /* Need to store these since the MCR gets reused */ + channel = mcr->queryfq.fqd.dest.channel; + wq = mcr->queryfq.fqd.dest.wq; + + switch (state) { + case QM_MCR_NP_STATE_TEN_SCHED: + case QM_MCR_NP_STATE_TRU_SCHED: + case QM_MCR_NP_STATE_ACTIVE: + case QM_MCR_NP_STATE_PARKED: + orl_empty = 0; + mcc = qm_mc_start(portal[0]); + mcc->alterfq.fqid = fqid; + qm_mc_commit(portal[0], QM_MCC_VERB_ALTER_RETIRE); + while (!(mcr = qm_mc_result(portal[0]))) + cpu_relax(); + DPA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == + QM_MCR_VERB_ALTER_RETIRE); + result = mcr->result; /* Make a copy as we reuse MCR below */ + + if (result == QM_MCR_RESULT_PENDING) { + /* Need to wait for the FQRN in the message ring, which + will only occur once the FQ has been drained. In + order for the FQ to drain the portal needs to be set + to dequeue from the channel the FQ is scheduled on */ + const struct qm_mr_entry *msg; + const struct qm_dqrr_entry *dqrr = NULL; + int found_fqrn = 0; + u16 dequeue_wq = 0; + + /* Flag that we need to drain FQ */ + drain = 1; + + if (channel >= qm_channel_pool1 && + channel < (qm_channel_pool1 + 15)) { + /* Pool channel, enable the bit in the portal */ + dequeue_wq = (channel - + qm_channel_pool1 + 1)<<4 | wq; + } else if (channel < qm_channel_pool1) { + /* Dedicated channel */ + dequeue_wq = wq; + } else { + pr_info("Cannot recover FQ 0x%x, it is " + "scheduled on channel 0x%x", + fqid, channel); + return -EBUSY; + } + /* Set the sdqcr to drain this channel */ + if (channel < qm_channel_pool1) + for (i = 0; i < portal_count; i++) + qm_dqrr_sdqcr_set(portal[i], + QM_SDQCR_TYPE_ACTIVE | + QM_SDQCR_CHANNELS_DEDICATED); + else + for (i = 0; i < portal_count; i++) + qm_dqrr_sdqcr_set( + portal[i], + QM_SDQCR_TYPE_ACTIVE | + QM_SDQCR_CHANNELS_POOL_CONV + (channel)); + while (!found_fqrn) { + /* Keep draining DQRR while checking the MR*/ + for (i = 0; i < portal_count; i++) { + qm_dqrr_pvb_update(portal[i]); + dqrr = qm_dqrr_current(portal[i]); + while (dqrr) { + qm_dqrr_cdc_consume_1ptr( + portal[i], dqrr, 0); + qm_dqrr_pvb_update(portal[i]); + qm_dqrr_next(portal[i]); + dqrr = qm_dqrr_current( + portal[i]); + } + /* Process message ring too */ + qm_mr_pvb_update(portal[i]); + msg = qm_mr_current(portal[i]); + while (msg) { + if ((msg->verb & + QM_MR_VERB_TYPE_MASK) + == QM_MR_VERB_FQRN) + found_fqrn = 1; + qm_mr_next(portal[i]); + qm_mr_cci_consume_to_current( + portal[i]); + qm_mr_pvb_update(portal[i]); + msg = qm_mr_current(portal[i]); + } + cpu_relax(); + } + } + } + if (result != QM_MCR_RESULT_OK && + result != QM_MCR_RESULT_PENDING) { + /* error */ + pr_err("qman_retire_fq failed on FQ 0x%x, result=0x%x\n", + fqid, result); + return -1; + } + if (!(mcr->alterfq.fqs & QM_MCR_FQS_ORLPRESENT)) { + /* ORL had no entries, no need to wait until the + ERNs come in */ + orl_empty = 1; + } + /* Retirement succeeded, check to see if FQ needs + to be drained */ + if (drain || mcr->alterfq.fqs & QM_MCR_FQS_NOTEMPTY) { + /* FQ is Not Empty, drain using volatile DQ commands */ + fq_empty = 0; + do { + const struct qm_dqrr_entry *dqrr = NULL; + u32 vdqcr = fqid | QM_VDQCR_NUMFRAMES_SET(3); + qm_dqrr_vdqcr_set(portal[0], vdqcr); + + /* Wait for a dequeue to occur */ + while (dqrr == NULL) { + qm_dqrr_pvb_update(portal[0]); + dqrr = qm_dqrr_current(portal[0]); + if (!dqrr) + cpu_relax(); + } + /* Process the dequeues, making sure to + empty the ring completely */ + while (dqrr) { + if (dqrr->fqid == fqid && + dqrr->stat & QM_DQRR_STAT_FQ_EMPTY) + fq_empty = 1; + qm_dqrr_cdc_consume_1ptr(portal[0], + dqrr, 0); + qm_dqrr_pvb_update(portal[0]); + qm_dqrr_next(portal[0]); + dqrr = qm_dqrr_current(portal[0]); + } + } while (fq_empty == 0); + } + for (i = 0; i < portal_count; i++) + qm_dqrr_sdqcr_set(portal[i], 0); + + /* Wait for the ORL to have been completely drained */ + while (orl_empty == 0) { + const struct qm_mr_entry *msg; + qm_mr_pvb_update(portal[0]); + msg = qm_mr_current(portal[0]); + while (msg) { + if ((msg->verb & QM_MR_VERB_TYPE_MASK) == + QM_MR_VERB_FQRL) + orl_empty = 1; + qm_mr_next(portal[0]); + qm_mr_cci_consume_to_current(portal[0]); + qm_mr_pvb_update(portal[0]); + msg = qm_mr_current(portal[0]); + } + cpu_relax(); + } + mcc = qm_mc_start(portal[0]); + mcc->alterfq.fqid = fqid; + qm_mc_commit(portal[0], QM_MCC_VERB_ALTER_OOS); + while (!(mcr = qm_mc_result(portal[0]))) + cpu_relax(); + DPA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == + QM_MCR_VERB_ALTER_OOS); + if (mcr->result != QM_MCR_RESULT_OK) { + pr_err("OOS after drain Failed on FQID 0x%x, result 0x%x\n", + fqid, mcr->result); + return -1; + } + return 0; + case QM_MCR_NP_STATE_RETIRED: + /* Send OOS Command */ + mcc = qm_mc_start(portal[0]); + mcc->alterfq.fqid = fqid; + qm_mc_commit(portal[0], QM_MCC_VERB_ALTER_OOS); + while (!(mcr = qm_mc_result(portal[0]))) + cpu_relax(); + DPA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == + QM_MCR_VERB_ALTER_OOS); + if (mcr->result) { + pr_err("OOS Failed on FQID 0x%x\n", fqid); + return -1; + } + return 0; + case QM_MCR_NP_STATE_OOS: + /* Done */ + return 0; + } + return -1; +} diff --git a/drivers/staging/fsl_qbman/qman_private.h b/drivers/staging/fsl_qbman/qman_private.h new file mode 100644 index 0000000..2265daf --- /dev/null +++ b/drivers/staging/fsl_qbman/qman_private.h @@ -0,0 +1,275 @@ +/* Copyright 2008-2012 Freescale Semiconductor, Inc. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions are met: + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * * Neither the name of Freescale Semiconductor nor the + * names of its contributors may be used to endorse or promote products + * derived from this software without specific prior written permission. + * + * + * ALTERNATIVELY, this software may be distributed under the terms of the + * GNU General Public License ("GPL") as published by the Free Software + * Foundation, either version 2 of that License or (at your option) any + * later version. + * + * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED + * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE + * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY + * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; + * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND + * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#include "dpa_sys.h" +#include +#include +#include + +#if !defined(CONFIG_FSL_QMAN_FQ_LOOKUP) && defined(CONFIG_PPC64) +#error "_PPC64 requires _FSL_QMAN_FQ_LOOKUP" +#endif + + /* ----------------- */ + /* Congestion Groups */ + /* ----------------- */ +/* This wrapper represents a bit-array for the state of the 256 Qman congestion + * groups. Is also used as a *mask* for congestion groups, eg. so we ignore + * those that don't concern us. We harness the structure and accessor details + * already used in the management command to query congestion groups. */ +struct qman_cgrs { + struct __qm_mcr_querycongestion q; +}; +static inline void qman_cgrs_init(struct qman_cgrs *c) +{ + memset(c, 0, sizeof(*c)); +} +static inline void qman_cgrs_fill(struct qman_cgrs *c) +{ + memset(c, 0xff, sizeof(*c)); +} +static inline int qman_cgrs_get(struct qman_cgrs *c, int num) +{ + return QM_MCR_QUERYCONGESTION(&c->q, num); +} +static inline void qman_cgrs_set(struct qman_cgrs *c, int num) +{ + c->q.__state[__CGR_WORD(num)] |= (0x80000000 >> __CGR_SHIFT(num)); +} +static inline void qman_cgrs_unset(struct qman_cgrs *c, int num) +{ + c->q.__state[__CGR_WORD(num)] &= ~(0x80000000 >> __CGR_SHIFT(num)); +} +static inline int qman_cgrs_next(struct qman_cgrs *c, int num) +{ + while ((++num < __CGR_NUM) && !qman_cgrs_get(c, num)) + ; + return num; +} +static inline void qman_cgrs_cp(struct qman_cgrs *dest, + const struct qman_cgrs *src) +{ + *dest = *src; +} +static inline void qman_cgrs_and(struct qman_cgrs *dest, + const struct qman_cgrs *a, const struct qman_cgrs *b) +{ + int ret; + u32 *_d = dest->q.__state; + const u32 *_a = a->q.__state; + const u32 *_b = b->q.__state; + for (ret = 0; ret < 8; ret++) + *(_d++) = *(_a++) & *(_b++); +} +static inline void qman_cgrs_xor(struct qman_cgrs *dest, + const struct qman_cgrs *a, const struct qman_cgrs *b) +{ + int ret; + u32 *_d = dest->q.__state; + const u32 *_a = a->q.__state; + const u32 *_b = b->q.__state; + for (ret = 0; ret < 8; ret++) + *(_d++) = *(_a++) ^ *(_b++); +} + +/* used by CCSR and portal interrupt code */ +enum qm_isr_reg { + qm_isr_status = 0, + qm_isr_enable = 1, + qm_isr_disable = 2, + qm_isr_inhibit = 3 +}; + +struct qm_portal_config { + /* Corenet portal addresses; + * [0]==cache-enabled, [1]==cache-inhibited. */ + __iomem void *addr_virt[2]; + struct resource addr_phys[2]; + struct device dev; + struct iommu_domain *iommu_domain; + /* Allow these to be joined in lists */ + struct list_head list; + /* User-visible portal configuration settings */ + struct qman_portal_config public_cfg; +}; + +/* Revision info (for errata and feature handling) */ +#define QMAN_REV11 0x0101 +#define QMAN_REV12 0x0102 +#define QMAN_REV20 0x0200 +#define QMAN_REV30 0x0300 +#define QMAN_REV31 0x0301 +extern u16 qman_ip_rev; /* 0 if uninitialised, otherwise QMAN_REVx */ + +#ifdef CONFIG_FSL_QMAN_CONFIG +/* Hooks from qman_driver.c to qman_config.c */ +int qman_init_ccsr(struct device_node *node); +void qman_liodn_fixup(u16 channel); +int qman_set_sdest(u16 channel, unsigned int cpu_idx); +#endif + +int qm_set_wpm(int wpm); +int qm_get_wpm(int *wpm); + +/* Hooks from qman_driver.c in to qman_high.c */ +struct qman_portal *qman_create_portal( + struct qman_portal *portal, + const struct qm_portal_config *config, + const struct qman_cgrs *cgrs); + +struct qman_portal *qman_create_affine_portal( + const struct qm_portal_config *config, + const struct qman_cgrs *cgrs); +struct qman_portal *qman_create_affine_slave(struct qman_portal *redirect, + int cpu); +const struct qm_portal_config *qman_destroy_affine_portal(void); +void qman_destroy_portal(struct qman_portal *qm); + +#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP +/* If the fq object pointer is greater than the size of context_b field, + * than a lookup table is required. */ +int qman_setup_fq_lookup_table(size_t num_entries); +#endif + + +/*************************************************/ +/* QMan s/w corenet portal, low-level i/face */ +/*************************************************/ + +/* Note: most functions are only used by the high-level interface, so are + * inlined from qman_low.h. The stuff below is for use by other parts of the + * driver. */ + +/* For qm_dqrr_sdqcr_set(); Choose one SOURCE. Choose one COUNT. Choose one + * dequeue TYPE. Choose TOKEN (8-bit). + * If SOURCE == CHANNELS, + * Choose CHANNELS_DEDICATED and/or CHANNELS_POOL(n). + * You can choose DEDICATED_PRECEDENCE if the portal channel should have + * priority. + * If SOURCE == SPECIFICWQ, + * Either select the work-queue ID with SPECIFICWQ_WQ(), or select the + * channel (SPECIFICWQ_DEDICATED or SPECIFICWQ_POOL()) and specify the + * work-queue priority (0-7) with SPECIFICWQ_WQ() - either way, you get the + * same value. + */ +#define QM_SDQCR_SOURCE_CHANNELS 0x0 +#define QM_SDQCR_SOURCE_SPECIFICWQ 0x40000000 +#define QM_SDQCR_COUNT_EXACT1 0x0 +#define QM_SDQCR_COUNT_UPTO3 0x20000000 +#define QM_SDQCR_DEDICATED_PRECEDENCE 0x10000000 +#define QM_SDQCR_TYPE_MASK 0x03000000 +#define QM_SDQCR_TYPE_NULL 0x0 +#define QM_SDQCR_TYPE_PRIO_QOS 0x01000000 +#define QM_SDQCR_TYPE_ACTIVE_QOS 0x02000000 +#define QM_SDQCR_TYPE_ACTIVE 0x03000000 +#define QM_SDQCR_TOKEN_MASK 0x00ff0000 +#define QM_SDQCR_TOKEN_SET(v) (((v) & 0xff) << 16) +#define QM_SDQCR_TOKEN_GET(v) (((v) >> 16) & 0xff) +#define QM_SDQCR_CHANNELS_DEDICATED 0x00008000 +#define QM_SDQCR_SPECIFICWQ_MASK 0x000000f7 +#define QM_SDQCR_SPECIFICWQ_DEDICATED 0x00000000 +#define QM_SDQCR_SPECIFICWQ_POOL(n) ((n) << 4) +#define QM_SDQCR_SPECIFICWQ_WQ(n) (n) + +/* For qm_dqrr_vdqcr_set(): use FQID(n) to fill in the frame queue ID */ +#define QM_VDQCR_FQID_MASK 0x00ffffff +#define QM_VDQCR_FQID(n) ((n) & QM_VDQCR_FQID_MASK) + +/* For qm_dqrr_pdqcr_set(); Choose one MODE. Choose one COUNT. + * If MODE==SCHEDULED + * Choose SCHEDULED_CHANNELS or SCHEDULED_SPECIFICWQ. Choose one dequeue TYPE. + * If CHANNELS, + * Choose CHANNELS_DEDICATED and/or CHANNELS_POOL() channels. + * You can choose DEDICATED_PRECEDENCE if the portal channel should have + * priority. + * If SPECIFICWQ, + * Either select the work-queue ID with SPECIFICWQ_WQ(), or select the + * channel (SPECIFICWQ_DEDICATED or SPECIFICWQ_POOL()) and specify the + * work-queue priority (0-7) with SPECIFICWQ_WQ() - either way, you get the + * same value. + * If MODE==UNSCHEDULED + * Choose FQID(). + */ +#define QM_PDQCR_MODE_SCHEDULED 0x0 +#define QM_PDQCR_MODE_UNSCHEDULED 0x80000000 +#define QM_PDQCR_SCHEDULED_CHANNELS 0x0 +#define QM_PDQCR_SCHEDULED_SPECIFICWQ 0x40000000 +#define QM_PDQCR_COUNT_EXACT1 0x0 +#define QM_PDQCR_COUNT_UPTO3 0x20000000 +#define QM_PDQCR_DEDICATED_PRECEDENCE 0x10000000 +#define QM_PDQCR_TYPE_MASK 0x03000000 +#define QM_PDQCR_TYPE_NULL 0x0 +#define QM_PDQCR_TYPE_PRIO_QOS 0x01000000 +#define QM_PDQCR_TYPE_ACTIVE_QOS 0x02000000 +#define QM_PDQCR_TYPE_ACTIVE 0x03000000 +#define QM_PDQCR_CHANNELS_DEDICATED 0x00008000 +#define QM_PDQCR_CHANNELS_POOL(n) (0x00008000 >> (n)) +#define QM_PDQCR_SPECIFICWQ_MASK 0x000000f7 +#define QM_PDQCR_SPECIFICWQ_DEDICATED 0x00000000 +#define QM_PDQCR_SPECIFICWQ_POOL(n) ((n) << 4) +#define QM_PDQCR_SPECIFICWQ_WQ(n) (n) +#define QM_PDQCR_FQID(n) ((n) & 0xffffff) + +/* Used by all portal interrupt registers except 'inhibit' + * Channels with frame availability + */ +#define QM_PIRQ_DQAVAIL 0x0000ffff + +/* The DQAVAIL interrupt fields break down into these bits; */ +#define QM_DQAVAIL_PORTAL 0x8000 /* Portal channel */ +#define QM_DQAVAIL_POOL(n) (0x8000 >> (n)) /* Pool channel, n==[1..15] */ +#define QM_DQAVAIL_MASK 0xffff +/* This mask contains all the "irqsource" bits visible to API users */ +#define QM_PIRQ_VISIBLE (QM_PIRQ_SLOW | QM_PIRQ_DQRI) + +/* These are qm__(). So for example, qm_disable_write() means "write + * the disable register" rather than "disable the ability to write". */ +#define qm_isr_status_read(qm) __qm_isr_read(qm, qm_isr_status) +#define qm_isr_status_clear(qm, m) __qm_isr_write(qm, qm_isr_status, m) +#define qm_isr_enable_read(qm) __qm_isr_read(qm, qm_isr_enable) +#define qm_isr_enable_write(qm, v) __qm_isr_write(qm, qm_isr_enable, v) +#define qm_isr_disable_read(qm) __qm_isr_read(qm, qm_isr_disable) +#define qm_isr_disable_write(qm, v) __qm_isr_write(qm, qm_isr_disable, v) +/* TODO: unfortunate name-clash here, reword? */ +#define qm_isr_inhibit(qm) __qm_isr_write(qm, qm_isr_inhibit, 1) +#define qm_isr_uninhibit(qm) __qm_isr_write(qm, qm_isr_inhibit, 0) + +#ifdef CONFIG_FSL_QMAN_CONFIG +int qman_have_ccsr(void); +#else +#define qman_have_ccsr 0 +#endif + +__init int qman_init(void); + +extern void *affine_portals[NR_CPUS]; +const struct qm_portal_config *qman_get_qm_portal_config( + struct qman_portal *portal); diff --git a/drivers/staging/fsl_qbman/qman_utility.c b/drivers/staging/fsl_qbman/qman_utility.c new file mode 100644 index 0000000..dbb92fd --- /dev/null +++ b/drivers/staging/fsl_qbman/qman_utility.c @@ -0,0 +1,129 @@ +/* Copyright 2008-2011 Freescale Semiconductor, Inc. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions are met: + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * * Neither the name of Freescale Semiconductor nor the + * names of its contributors may be used to endorse or promote products + * derived from this software without specific prior written permission. + * + * + * ALTERNATIVELY, this software may be distributed under the terms of the + * GNU General Public License ("GPL") as published by the Free Software + * Foundation, either version 2 of that License or (at your option) any + * later version. + * + * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED + * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE + * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY + * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; + * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND + * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#include "qman_private.h" + +/* ----------------- */ +/* --- FQID Pool --- */ + +struct qman_fqid_pool { + /* Base and size of the FQID range */ + u32 fqid_base; + u32 total; + /* Number of FQIDs currently "allocated" */ + u32 used; + /* Allocation optimisation. When 'usedfqid_base = fqid_start; + pool->total = num; + pool->used = 0; + pool->next = 0; + pool->bits = kzalloc(QNUM_BYTES(num), GFP_KERNEL); + if (!pool->bits) { + kfree(pool); + return NULL; + } + /* If num is not an even multiple of QLONG_BITS (or even 8, for + * byte-oriented searching) then we fill the trailing bits with 1, to + * make them look allocated (permanently). */ + for (i = num + 1; i < QNUM_BITS(num); i++) + set_bit(i, pool->bits); + return pool; +} +EXPORT_SYMBOL(qman_fqid_pool_create); + +int qman_fqid_pool_destroy(struct qman_fqid_pool *pool) +{ + int ret = pool->used; + kfree(pool->bits); + kfree(pool); + return ret; +} +EXPORT_SYMBOL(qman_fqid_pool_destroy); + +int qman_fqid_pool_alloc(struct qman_fqid_pool *pool, u32 *fqid) +{ + int ret; + if (pool->used == pool->total) + return -ENOMEM; + *fqid = pool->fqid_base + pool->next; + ret = test_and_set_bit(pool->next, pool->bits); + BUG_ON(ret); + if (++pool->used == pool->total) + return 0; + pool->next = find_next_zero_bit(pool->bits, pool->total, pool->next); + if (pool->next >= pool->total) + pool->next = find_first_zero_bit(pool->bits, pool->total); + BUG_ON(pool->next >= pool->total); + return 0; +} +EXPORT_SYMBOL(qman_fqid_pool_alloc); + +void qman_fqid_pool_free(struct qman_fqid_pool *pool, u32 fqid) +{ + int ret; + + fqid -= pool->fqid_base; + ret = test_and_clear_bit(fqid, pool->bits); + BUG_ON(!ret); + if (pool->used-- == pool->total) + pool->next = fqid; +} +EXPORT_SYMBOL(qman_fqid_pool_free); + +u32 qman_fqid_pool_used(struct qman_fqid_pool *pool) +{ + return pool->used; +} +EXPORT_SYMBOL(qman_fqid_pool_used); diff --git a/include/linux/fsl_bman.h b/include/linux/fsl_bman.h index d9a0dc9..16845d3 100644 --- a/include/linux/fsl_bman.h +++ b/include/linux/fsl_bman.h @@ -323,9 +323,9 @@ const cpumask_t *bman_affine_cpus(void); * NB, unlike the legacy wrapper bman_poll(), this function will * deterministically check for the presence of portal processing work and do it, * which implies some latency even if there's nothing to do. The bman_poll() - * wrapper on the other hand attenuates this by + * wrapper on the other hand (like the qman_poll() wrapper) attenuates this by * checking for (and doing) portal processing infrequently. Ie. such that - * bman_poll() can be called from core-processing loops. Use + * qman_poll() and bman_poll() can be called from core-processing loops. Use * bman_poll_slow() when you yourself are deciding when to incur the overhead of * processing. */ diff --git a/include/linux/fsl_qman.h b/include/linux/fsl_qman.h new file mode 100644 index 0000000..c12cbd3 --- /dev/null +++ b/include/linux/fsl_qman.h @@ -0,0 +1,1949 @@ +/* Copyright 2008-2012 Freescale Semiconductor, Inc. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions are met: + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * * Neither the name of Freescale Semiconductor nor the + * names of its contributors may be used to endorse or promote products + * derived from this software without specific prior written permission. + * + * + * ALTERNATIVELY, this software may be distributed under the terms of the + * GNU General Public License ("GPL") as published by the Free Software + * Foundation, either version 2 of that License or (at your option) any + * later version. + * + * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED + * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE + * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY + * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; + * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND + * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#ifndef FSL_QMAN_H +#define FSL_QMAN_H + +#ifdef __cplusplus +extern "C" { +#endif + +/* Last updated for v00.800 of the BG */ + +/* Hardware constants */ +#define QM_CHANNEL_SWPORTAL0 0 +#define QMAN_CHANNEL_POOL1 0x21 +#define QMAN_CHANNEL_CAAM 0x80 +#define QMAN_CHANNEL_PME 0xa0 +#define QMAN_CHANNEL_POOL1_REV3 0x401 +#define QMAN_CHANNEL_CAAM_REV3 0x840 +#define QMAN_CHANNEL_PME_REV3 0x860 +#define QMAN_CHANNEL_DCE 0x8a0 +extern u16 qm_channel_pool1; +extern u16 qm_channel_caam; +extern u16 qm_channel_pme; +extern u16 qm_channel_dce; +enum qm_dc_portal { + qm_dc_portal_fman0 = 0, + qm_dc_portal_fman1 = 1, + qm_dc_portal_caam = 2, + qm_dc_portal_pme = 3, + qm_dc_portal_rman = 4, + qm_dc_portal_dce = 5 +}; + +/* Portal processing (interrupt) sources */ +#define QM_PIRQ_CSCI 0x00100000 /* Congestion State Change */ +#define QM_PIRQ_EQCI 0x00080000 /* Enqueue Command Committed */ +#define QM_PIRQ_EQRI 0x00040000 /* EQCR Ring (below threshold) */ +#define QM_PIRQ_DQRI 0x00020000 /* DQRR Ring (non-empty) */ +#define QM_PIRQ_MRI 0x00010000 /* MR Ring (non-empty) */ +/* This mask contains all the interrupt sources that need handling except DQRI, + * ie. that if present should trigger slow-path processing. */ +#define QM_PIRQ_SLOW (QM_PIRQ_CSCI | QM_PIRQ_EQCI | QM_PIRQ_EQRI | \ + QM_PIRQ_MRI) + +/* --- Clock speed --- */ +/* A qman driver instance may or may not know the current qman clock speed. + * However, certain CEETM calculations may not be possible if this is not known. + * The 'set' function will only succeed (return zero) if the driver did not + * already know the clock speed. Likewise, the 'get' function will only succeed + * if the driver does know the clock speed (either because it knew when booting, + * or was told via 'set'). In cases where software is running on a driver + * instance that does not know the clock speed (eg. on a hypervised data-plane), + * and the user can obtain the current qman clock speed by other means (eg. from + * a message sent from the control-plane), then the 'set' function can be used + * to enable rate-calculations in a driver where it would otherwise not be + * possible. */ +int qm_get_clock(u64 *clock_hz); +int qm_set_clock(u64 clock_hz); + +/* For qman_static_dequeue_*** APIs */ +#define QM_SDQCR_CHANNELS_POOL_MASK 0x00007fff +/* for n in [1,15] */ +#define QM_SDQCR_CHANNELS_POOL(n) (0x00008000 >> (n)) +/* for conversion from n of qm_channel */ +static inline u32 QM_SDQCR_CHANNELS_POOL_CONV(u16 channel) +{ + return QM_SDQCR_CHANNELS_POOL(channel + 1 - qm_channel_pool1); +} + +/* For qman_volatile_dequeue(); Choose one PRECEDENCE. EXACT is optional. Use + * NUMFRAMES(n) (6-bit) or NUMFRAMES_TILLEMPTY to fill in the frame-count. Use + * FQID(n) to fill in the frame queue ID. */ +#define QM_VDQCR_PRECEDENCE_VDQCR 0x0 +#define QM_VDQCR_PRECEDENCE_SDQCR 0x80000000 +#define QM_VDQCR_EXACT 0x40000000 +#define QM_VDQCR_NUMFRAMES_MASK 0x3f000000 +#define QM_VDQCR_NUMFRAMES_SET(n) (((n) & 0x3f) << 24) +#define QM_VDQCR_NUMFRAMES_GET(n) (((n) >> 24) & 0x3f) +#define QM_VDQCR_NUMFRAMES_TILLEMPTY QM_VDQCR_NUMFRAMES_SET(0) + + +/* ------------------------------------------------------- */ +/* --- Qman data structures (and associated constants) --- */ + +/* Represents s/w corenet portal mapped data structures */ +struct qm_eqcr_entry; /* EQCR (EnQueue Command Ring) entries */ +struct qm_dqrr_entry; /* DQRR (DeQueue Response Ring) entries */ +struct qm_mr_entry; /* MR (Message Ring) entries */ +struct qm_mc_command; /* MC (Management Command) command */ +struct qm_mc_result; /* MC result */ + +/* See David Lapp's "Frame formats" document, "dpateam", Jan 07, 2008 */ +#define QM_FD_FORMAT_SG 0x4 +#define QM_FD_FORMAT_LONG 0x2 +#define QM_FD_FORMAT_COMPOUND 0x1 +enum qm_fd_format { + /* 'contig' implies a contiguous buffer, whereas 'sg' implies a + * scatter-gather table. 'big' implies a 29-bit length with no offset + * field, otherwise length is 20-bit and offset is 9-bit. 'compound' + * implies a s/g-like table, where each entry itself represents a frame + * (contiguous or scatter-gather) and the 29-bit "length" is + * interpreted purely for congestion calculations, ie. a "congestion + * weight". */ + qm_fd_contig = 0, + qm_fd_contig_big = QM_FD_FORMAT_LONG, + qm_fd_sg = QM_FD_FORMAT_SG, + qm_fd_sg_big = QM_FD_FORMAT_SG | QM_FD_FORMAT_LONG, + qm_fd_compound = QM_FD_FORMAT_COMPOUND +}; + +/* Capitalised versions are un-typed but can be used in static expressions */ +#define QM_FD_CONTIG 0 +#define QM_FD_CONTIG_BIG QM_FD_FORMAT_LONG +#define QM_FD_SG QM_FD_FORMAT_SG +#define QM_FD_SG_BIG (QM_FD_FORMAT_SG | QM_FD_FORMAT_LONG) +#define QM_FD_COMPOUND QM_FD_FORMAT_COMPOUND + +/* See 1.5.1.1: "Frame Descriptor (FD)" */ +struct qm_fd { + union { + struct { + u8 dd:2; /* dynamic debug */ + u8 liodn_offset:6; + u8 bpid:8; /* Buffer Pool ID */ + u8 eliodn_offset:4; + u8 __reserved:4; + u8 addr_hi; /* high 8-bits of 40-bit address */ + u32 addr_lo; /* low 32-bits of 40-bit address */ + }; + struct { + u64 __notaddress:24; + /* More efficient address accessor */ + u64 addr:40; + }; + u64 opaque_addr; + }; + /* The 'format' field indicates the interpretation of the remaining 29 + * bits of the 32-bit word. For packing reasons, it is duplicated in the + * other union elements. Note, union'd structs are difficult to use with + * static initialisation under gcc, in which case use the "opaque" form + * with one of the macros. */ + union { + /* For easier/faster copying of this part of the fd (eg. from a + * DQRR entry to an EQCR entry) copy 'opaque' */ + u32 opaque; + /* If 'format' is _contig or _sg, 20b length and 9b offset */ + struct { + enum qm_fd_format format:3; + u16 offset:9; + u32 length20:20; + }; + /* If 'format' is _contig_big or _sg_big, 29b length */ + struct { + enum qm_fd_format _format1:3; + u32 length29:29; + }; + /* If 'format' is _compound, 29b "congestion weight" */ + struct { + enum qm_fd_format _format2:3; + u32 cong_weight:29; + }; + }; + union { + u32 cmd; + u32 status; + }; +} __aligned(8); +#define QM_FD_DD_NULL 0x00 +#define QM_FD_PID_MASK 0x3f +static inline u64 qm_fd_addr_get64(const struct qm_fd *fd) +{ + return fd->addr; +} + +static inline dma_addr_t qm_fd_addr(const struct qm_fd *fd) +{ + return (dma_addr_t)fd->addr; +} +/* Macro, so we compile better if 'v' isn't always 64-bit */ +#define qm_fd_addr_set64(fd, v) \ + do { \ + struct qm_fd *__fd931 = (fd); \ + __fd931->addr = v; \ + } while (0) + +/* For static initialisation of FDs (which is complicated by the use of unions + * in "struct qm_fd"), use the following macros. Note that; + * - 'dd', 'pid' and 'bpid' are ignored because there's no static initialisation + * use-case), + * - use capitalised QM_FD_*** formats for static initialisation. + */ +#define QM_FD_FMT_20(cmd, addr_hi, addr_lo, fmt, off, len) \ + { 0, 0, 0, 0, 0, addr_hi, addr_lo, \ + { (((fmt)&0x7) << 29) | (((off)&0x1ff) << 20) | ((len)&0xfffff) }, \ + { cmd } } +#define QM_FD_FMT_29(cmd, addr_hi, addr_lo, fmt, len) \ + { 0, 0, 0, 0, 0, addr_hi, addr_lo, \ + { (((fmt)&0x7) << 29) | ((len)&0x1fffffff) }, \ + { cmd } } + +/* See 2.2.1.3 Multi-Core Datapath Acceleration Architecture */ +struct qm_sg_entry { + union { + struct { + u8 __reserved1[3]; + u8 addr_hi; /* high 8-bits of 40-bit address */ + u32 addr_lo; /* low 32-bits of 40-bit address */ + }; + struct { + u64 __notaddress:24; + u64 addr:40; + }; + }; + u32 extension:1; /* Extension bit */ + u32 final:1; /* Final bit */ + u32 length:30; + u8 __reserved2; + u8 bpid; + u16 __reserved3:3; + u16 offset:13; +} __packed; +static inline u64 qm_sg_entry_get64(const struct qm_sg_entry *sg) +{ + return sg->addr; +} +static inline dma_addr_t qm_sg_addr(const struct qm_sg_entry *sg) +{ + return (dma_addr_t)sg->addr; +} +/* Macro, so we compile better if 'v' isn't always 64-bit */ +#define qm_sg_entry_set64(sg, v) \ + do { \ + struct qm_sg_entry *__sg931 = (sg); \ + __sg931->addr = v; \ + } while (0) + +/* See 1.5.8.1: "Enqueue Command" */ +struct qm_eqcr_entry { + u8 __dont_write_directly__verb; + u8 dca; + u16 seqnum; + u32 orp; /* 24-bit */ + u32 fqid; /* 24-bit */ + u32 tag; + struct qm_fd fd; + u8 __reserved3[32]; +} __packed; +#define QM_EQCR_VERB_VBIT 0x80 +#define QM_EQCR_VERB_CMD_MASK 0x61 /* but only one value; */ +#define QM_EQCR_VERB_CMD_ENQUEUE 0x01 +#define QM_EQCR_VERB_COLOUR_MASK 0x18 /* 4 possible values; */ +#define QM_EQCR_VERB_COLOUR_GREEN 0x00 +#define QM_EQCR_VERB_COLOUR_YELLOW 0x08 +#define QM_EQCR_VERB_COLOUR_RED 0x10 +#define QM_EQCR_VERB_COLOUR_OVERRIDE 0x18 +#define QM_EQCR_VERB_INTERRUPT 0x04 /* on command consumption */ +#define QM_EQCR_VERB_ORP 0x02 /* enable order restoration */ +#define QM_EQCR_DCA_ENABLE 0x80 +#define QM_EQCR_DCA_PARK 0x40 +#define QM_EQCR_DCA_IDXMASK 0x0f /* "DQRR::idx" goes here */ +#define QM_EQCR_SEQNUM_NESN 0x8000 /* Advance NESN */ +#define QM_EQCR_SEQNUM_NLIS 0x4000 /* More fragments to come */ +#define QM_EQCR_SEQNUM_SEQMASK 0x3fff /* sequence number goes here */ +#define QM_EQCR_FQID_NULL 0 /* eg. for an ORP seqnum hole */ + +/* See 1.5.8.2: "Frame Dequeue Response" */ +struct qm_dqrr_entry { + u8 verb; + u8 stat; + u16 seqnum; /* 15-bit */ + u8 tok; + u8 __reserved2[3]; + u32 fqid; /* 24-bit */ + u32 contextB; + struct qm_fd fd; + u8 __reserved4[32]; +}; +#define QM_DQRR_VERB_VBIT 0x80 +#define QM_DQRR_VERB_MASK 0x7f /* where the verb contains; */ +#define QM_DQRR_VERB_FRAME_DEQUEUE 0x60 /* "this format" */ +#define QM_DQRR_STAT_FQ_EMPTY 0x80 /* FQ empty */ +#define QM_DQRR_STAT_FQ_HELDACTIVE 0x40 /* FQ held active */ +#define QM_DQRR_STAT_FQ_FORCEELIGIBLE 0x20 /* FQ was force-eligible'd */ +#define QM_DQRR_STAT_FD_VALID 0x10 /* has a non-NULL FD */ +#define QM_DQRR_STAT_UNSCHEDULED 0x02 /* Unscheduled dequeue */ +#define QM_DQRR_STAT_DQCR_EXPIRED 0x01 /* VDQCR or PDQCR expired*/ + +/* See 1.5.8.3: "ERN Message Response" */ +/* See 1.5.8.4: "FQ State Change Notification" */ +struct qm_mr_entry { + u8 verb; + union { + struct { + u8 dca; + u16 seqnum; + u8 rc; /* Rejection Code */ + u32 orp:24; + u32 fqid; /* 24-bit */ + u32 tag; + struct qm_fd fd; + } __packed ern; + struct { + u8 colour:2; /* See QM_MR_DCERN_COLOUR_* */ + u8 __reserved1:3; + enum qm_dc_portal portal:3; + u16 __reserved2; + u8 rc; /* Rejection Code */ + u32 __reserved3:24; + u32 fqid; /* 24-bit */ + u32 tag; + struct qm_fd fd; + } __packed dcern; + struct { + u8 fqs; /* Frame Queue Status */ + u8 __reserved1[6]; + u32 fqid; /* 24-bit */ + u32 contextB; + u8 __reserved2[16]; + } __packed fq; /* FQRN/FQRNI/FQRL/FQPN */ + }; + u8 __reserved2[32]; +} __packed; +#define QM_MR_VERB_VBIT 0x80 +/* The "ern" VERB bits match QM_EQCR_VERB_*** so aren't reproduced here. ERNs + * originating from direct-connect portals ("dcern") use 0x20 as a verb which + * would be invalid as a s/w enqueue verb. A s/w ERN can be distinguished from + * the other MR types by noting if the 0x20 bit is unset. */ +#define QM_MR_VERB_TYPE_MASK 0x27 +#define QM_MR_VERB_DC_ERN 0x20 +#define QM_MR_VERB_FQRN 0x21 +#define QM_MR_VERB_FQRNI 0x22 +#define QM_MR_VERB_FQRL 0x23 +#define QM_MR_VERB_FQPN 0x24 +#define QM_MR_RC_MASK 0xf0 /* contains one of; */ +#define QM_MR_RC_CGR_TAILDROP 0x00 +#define QM_MR_RC_WRED 0x10 +#define QM_MR_RC_ERROR 0x20 +#define QM_MR_RC_ORPWINDOW_EARLY 0x30 +#define QM_MR_RC_ORPWINDOW_LATE 0x40 +#define QM_MR_RC_FQ_TAILDROP 0x50 +#define QM_MR_RC_ORPWINDOW_RETIRED 0x60 +#define QM_MR_RC_ORP_ZERO 0x70 +#define QM_MR_FQS_ORLPRESENT 0x02 /* ORL fragments to come */ +#define QM_MR_FQS_NOTEMPTY 0x01 /* FQ has enqueued frames */ +#define QM_MR_DCERN_COLOUR_GREEN 0x00 +#define QM_MR_DCERN_COLOUR_YELLOW 0x01 +#define QM_MR_DCERN_COLOUR_RED 0x02 +#define QM_MR_DCERN_COLOUR_OVERRIDE 0x03 + +/* An identical structure of FQD fields is present in the "Init FQ" command and + * the "Query FQ" result, it's suctioned out into the "struct qm_fqd" type. + * Within that, the 'stashing' and 'taildrop' pieces are also factored out, the + * latter has two inlines to assist with converting to/from the mant+exp + * representation. */ +struct qm_fqd_stashing { + /* See QM_STASHING_EXCL_<...> */ + u8 exclusive; + u8 __reserved1:2; + /* Numbers of cachelines */ + u8 annotation_cl:2; + u8 data_cl:2; + u8 context_cl:2; +} __packed; +struct qm_fqd_taildrop { + u16 __reserved1:3; + u16 mant:8; + u16 exp:5; +} __packed; +struct qm_fqd_oac { + /* See QM_OAC_<...> */ + u8 oac:2; /* "Overhead Accounting Control" */ + u8 __reserved1:6; + /* Two's-complement value (-128 to +127) */ + signed char oal; /* "Overhead Accounting Length" */ +} __packed; +struct qm_fqd { + union { + u8 orpc; + struct { + u8 __reserved1:2; + u8 orprws:3; + u8 oa:1; + u8 olws:2; + } __packed; + }; + u8 cgid; + u16 fq_ctrl; /* See QM_FQCTRL_<...> */ + union { + u16 dest_wq; + struct { + u16 channel:13; /* qm_channel */ + u16 wq:3; + } __packed dest; + }; + u16 __reserved2:1; + u16 ics_cred:15; + /* For "Initialize Frame Queue" commands, the write-enable mask + * determines whether 'td' or 'oac_init' is observed. For query + * commands, this field is always 'td', and 'oac_query' (below) reflects + * the Overhead ACcounting values. */ + union { + struct qm_fqd_taildrop td; + struct qm_fqd_oac oac_init; + }; + u32 context_b; + union { + /* Treat it as 64-bit opaque */ + u64 opaque; + struct { + u32 hi; + u32 lo; + }; + /* Treat it as s/w portal stashing config */ + /* See 1.5.6.7.1: "FQD Context_A field used for [...] */ + struct { + struct qm_fqd_stashing stashing; + /* 48-bit address of FQ context to + * stash, must be cacheline-aligned */ + u16 context_hi; + u32 context_lo; + } __packed; + } context_a; + struct qm_fqd_oac oac_query; +} __packed; +/* 64-bit converters for context_hi/lo */ +static inline u64 qm_fqd_stashing_get64(const struct qm_fqd *fqd) +{ + return ((u64)fqd->context_a.context_hi << 32) | + (u64)fqd->context_a.context_lo; +} +static inline dma_addr_t qm_fqd_stashing_addr(const struct qm_fqd *fqd) +{ + return (dma_addr_t)qm_fqd_stashing_get64(fqd); +} +static inline u64 qm_fqd_context_a_get64(const struct qm_fqd *fqd) +{ + return ((u64)fqd->context_a.hi << 32) | + (u64)fqd->context_a.lo; +} +/* Macro, so we compile better when 'v' isn't necessarily 64-bit */ +#define qm_fqd_stashing_set64(fqd, v) \ + do { \ + struct qm_fqd *__fqd931 = (fqd); \ + __fqd931->context_a.context_hi = upper_32_bits(v); \ + __fqd931->context_a.context_lo = lower_32_bits(v); \ + } while (0) +#define qm_fqd_context_a_set64(fqd, v) \ + do { \ + struct qm_fqd *__fqd931 = (fqd); \ + __fqd931->context_a.hi = upper_32_bits(v); \ + __fqd931->context_a.lo = lower_32_bits(v); \ + } while (0) +/* convert a threshold value into mant+exp representation */ +static inline int qm_fqd_taildrop_set(struct qm_fqd_taildrop *td, u32 val, + int roundup) +{ + u32 e = 0; + int oddbit = 0; + if (val > 0xe0000000) + return -ERANGE; + while (val > 0xff) { + oddbit = val & 1; + val >>= 1; + e++; + if (roundup && oddbit) + val++; + } + td->exp = e; + td->mant = val; + return 0; +} +/* and the other direction */ +static inline u32 qm_fqd_taildrop_get(const struct qm_fqd_taildrop *td) +{ + return (u32)td->mant << td->exp; +} + +/* See 1.5.2.2: "Frame Queue Descriptor (FQD)" */ +/* Frame Queue Descriptor (FQD) field 'fq_ctrl' uses these constants */ +#define QM_FQCTRL_MASK 0x07ff /* 'fq_ctrl' flags; */ +#define QM_FQCTRL_CGE 0x0400 /* Congestion Group Enable */ +#define QM_FQCTRL_TDE 0x0200 /* Tail-Drop Enable */ +#define QM_FQCTRL_ORP 0x0100 /* ORP Enable */ +#define QM_FQCTRL_CTXASTASHING 0x0080 /* Context-A stashing */ +#define QM_FQCTRL_CPCSTASH 0x0040 /* CPC Stash Enable */ +#define QM_FQCTRL_FORCESFDR 0x0008 /* High-priority SFDRs */ +#define QM_FQCTRL_AVOIDBLOCK 0x0004 /* Don't block active */ +#define QM_FQCTRL_HOLDACTIVE 0x0002 /* Hold active in portal */ +#define QM_FQCTRL_PREFERINCACHE 0x0001 /* Aggressively cache FQD */ +#define QM_FQCTRL_LOCKINCACHE QM_FQCTRL_PREFERINCACHE /* older naming */ + +/* See 1.5.6.7.1: "FQD Context_A field used for [...] */ +/* Frame Queue Descriptor (FQD) field 'CONTEXT_A' uses these constants */ +#define QM_STASHING_EXCL_ANNOTATION 0x04 +#define QM_STASHING_EXCL_DATA 0x02 +#define QM_STASHING_EXCL_CTX 0x01 + +/* See 1.5.5.3: "Intra Class Scheduling" */ +/* FQD field 'OAC' (Overhead ACcounting) uses these constants */ +#define QM_OAC_ICS 0x2 /* Accounting for Intra-Class Scheduling */ +#define QM_OAC_CG 0x1 /* Accounting for Congestion Groups */ + +/* See 1.5.8.4: "FQ State Change Notification" */ +/* This struct represents the 32-bit "WR_PARM_[GYR]" parameters in CGR fields + * and associated commands/responses. The WRED parameters are calculated from + * these fields as follows; + * MaxTH = MA * (2 ^ Mn) + * Slope = SA / (2 ^ Sn) + * MaxP = 4 * (Pn + 1) + */ +struct qm_cgr_wr_parm { + union { + u32 word; + struct { + u32 MA:8; + u32 Mn:5; + u32 SA:7; /* must be between 64-127 */ + u32 Sn:6; + u32 Pn:6; + } __packed; + }; +} __packed; +/* This struct represents the 13-bit "CS_THRES" CGR field. In the corresponding + * management commands, this is padded to a 16-bit structure field, so that's + * how we represent it here. The congestion state threshold is calculated from + * these fields as follows; + * CS threshold = TA * (2 ^ Tn) + */ +struct qm_cgr_cs_thres { + u16 __reserved:3; + u16 TA:8; + u16 Tn:5; +} __packed; +/* This identical structure of CGR fields is present in the "Init/Modify CGR" + * commands and the "Query CGR" result. It's suctioned out here into its own + * struct. */ +struct __qm_mc_cgr { + struct qm_cgr_wr_parm wr_parm_g; + struct qm_cgr_wr_parm wr_parm_y; + struct qm_cgr_wr_parm wr_parm_r; + u8 wr_en_g; /* boolean, use QM_CGR_EN */ + u8 wr_en_y; /* boolean, use QM_CGR_EN */ + u8 wr_en_r; /* boolean, use QM_CGR_EN */ + u8 cscn_en; /* boolean, use QM_CGR_EN */ + union { + struct { + u16 cscn_targ_upd_ctrl; /* use QM_CSCN_TARG_UDP_ */ + u16 cscn_targ_dcp_low; /* CSCN_TARG_DCP low-16bits */ + }; + u32 cscn_targ; /* use QM_CGR_TARG_* */ + }; + u8 cstd_en; /* boolean, use QM_CGR_EN */ + u8 cs; /* boolean, only used in query response */ + struct qm_cgr_cs_thres cs_thres; /* use qm_cgr_cs_thres_set64() */ + u8 mode; /* QMAN_CGR_MODE_FRAME not supported in rev1.0 */ +} __packed; +#define QM_CGR_EN 0x01 /* For wr_en_*, cscn_en, cstd_en */ +#define QM_CGR_TARG_UDP_CTRL_WRITE_BIT 0x8000 /* value written to portal bit*/ +#define QM_CGR_TARG_UDP_CTRL_DCP 0x4000 /* 0: SWP, 1: DCP */ +#define QM_CGR_TARG_PORTAL(n) (0x80000000 >> (n)) /* s/w portal, 0-9 */ +#define QM_CGR_TARG_FMAN0 0x00200000 /* direct-connect portal: fman0 */ +#define QM_CGR_TARG_FMAN1 0x00100000 /* : fman1 */ +/* Convert CGR thresholds to/from "cs_thres" format */ +static inline u64 qm_cgr_cs_thres_get64(const struct qm_cgr_cs_thres *th) +{ + return (u64)th->TA << th->Tn; +} +static inline int qm_cgr_cs_thres_set64(struct qm_cgr_cs_thres *th, u64 val, + int roundup) +{ + u32 e = 0; + int oddbit = 0; + while (val > 0xff) { + oddbit = val & 1; + val >>= 1; + e++; + if (roundup && oddbit) + val++; + } + th->Tn = e; + th->TA = val; + return 0; +} + +/* See 1.5.8.5.1: "Initialize FQ" */ +/* See 1.5.8.5.2: "Query FQ" */ +/* See 1.5.8.5.3: "Query FQ Non-Programmable Fields" */ +/* See 1.5.8.5.4: "Alter FQ State Commands " */ +/* See 1.5.8.6.1: "Initialize/Modify CGR" */ +/* See 1.5.8.6.2: "CGR Test Write" */ +/* See 1.5.8.6.3: "Query CGR" */ +/* See 1.5.8.6.4: "Query Congestion Group State" */ +struct qm_mcc_initfq { + u8 __reserved1; + u16 we_mask; /* Write Enable Mask */ + u32 fqid; /* 24-bit */ + u16 count; /* Initialises 'count+1' FQDs */ + struct qm_fqd fqd; /* the FQD fields go here */ + u8 __reserved3[30]; +} __packed; +struct qm_mcc_queryfq { + u8 __reserved1[3]; + u32 fqid; /* 24-bit */ + u8 __reserved2[56]; +} __packed; +struct qm_mcc_queryfq_np { + u8 __reserved1[3]; + u32 fqid; /* 24-bit */ + u8 __reserved2[56]; +} __packed; +struct qm_mcc_alterfq { + u8 __reserved1[3]; + u32 fqid; /* 24-bit */ + u8 __reserved2; + u8 count; /* number of consecutive FQID */ + u8 __reserved3[10]; + u32 context_b; /* frame queue context b */ + u8 __reserved4[40]; +} __packed; +struct qm_mcc_initcgr { + u8 __reserved1; + u16 we_mask; /* Write Enable Mask */ + struct __qm_mc_cgr cgr; /* CGR fields */ + u8 __reserved2[2]; + u8 cgid; + u8 __reserved4[32]; +} __packed; +struct qm_mcc_cgrtestwrite { + u8 __reserved1[2]; + u8 i_bcnt_hi:8;/* high 8-bits of 40-bit "Instant" */ + u32 i_bcnt_lo; /* low 32-bits of 40-bit */ + u8 __reserved2[23]; + u8 cgid; + u8 __reserved3[32]; +} __packed; +struct qm_mcc_querycgr { + u8 __reserved1[30]; + u8 cgid; + u8 __reserved2[32]; +} __packed; +struct qm_mcc_querycongestion { + u8 __reserved[63]; +} __packed; +struct qm_mcc_querywq { + u8 __reserved; + /* select channel if verb != QUERYWQ_DEDICATED */ + union { + u16 channel_wq; /* ignores wq (3 lsbits) */ + struct { + u16 id:13; /* qm_channel */ + u16 __reserved1:3; + } __packed channel; + }; + u8 __reserved2[60]; +} __packed; + +struct qm_mc_command { + u8 __dont_write_directly__verb; + union { + struct qm_mcc_initfq initfq; + struct qm_mcc_queryfq queryfq; + struct qm_mcc_queryfq_np queryfq_np; + struct qm_mcc_alterfq alterfq; + struct qm_mcc_initcgr initcgr; + struct qm_mcc_cgrtestwrite cgrtestwrite; + struct qm_mcc_querycgr querycgr; + struct qm_mcc_querycongestion querycongestion; + struct qm_mcc_querywq querywq; + }; +} __packed; +#define QM_MCC_VERB_VBIT 0x80 +#define QM_MCC_VERB_MASK 0x7f /* where the verb contains; */ +#define QM_MCC_VERB_INITFQ_PARKED 0x40 +#define QM_MCC_VERB_INITFQ_SCHED 0x41 +#define QM_MCC_VERB_QUERYFQ 0x44 +#define QM_MCC_VERB_QUERYFQ_NP 0x45 /* "non-programmable" fields */ +#define QM_MCC_VERB_QUERYWQ 0x46 +#define QM_MCC_VERB_QUERYWQ_DEDICATED 0x47 +#define QM_MCC_VERB_ALTER_SCHED 0x48 /* Schedule FQ */ +#define QM_MCC_VERB_ALTER_FE 0x49 /* Force Eligible FQ */ +#define QM_MCC_VERB_ALTER_RETIRE 0x4a /* Retire FQ */ +#define QM_MCC_VERB_ALTER_OOS 0x4b /* Take FQ out of service */ +#define QM_MCC_VERB_ALTER_FQXON 0x4d /* FQ XON */ +#define QM_MCC_VERB_ALTER_FQXOFF 0x4e /* FQ XOFF */ +#define QM_MCC_VERB_INITCGR 0x50 +#define QM_MCC_VERB_MODIFYCGR 0x51 +#define QM_MCC_VERB_CGRTESTWRITE 0x52 +#define QM_MCC_VERB_QUERYCGR 0x58 +#define QM_MCC_VERB_QUERYCONGESTION 0x59 +/* INITFQ-specific flags */ +#define QM_INITFQ_WE_MASK 0x01ff /* 'Write Enable' flags; */ +#define QM_INITFQ_WE_OAC 0x0100 +#define QM_INITFQ_WE_ORPC 0x0080 +#define QM_INITFQ_WE_CGID 0x0040 +#define QM_INITFQ_WE_FQCTRL 0x0020 +#define QM_INITFQ_WE_DESTWQ 0x0010 +#define QM_INITFQ_WE_ICSCRED 0x0008 +#define QM_INITFQ_WE_TDTHRESH 0x0004 +#define QM_INITFQ_WE_CONTEXTB 0x0002 +#define QM_INITFQ_WE_CONTEXTA 0x0001 +/* INITCGR/MODIFYCGR-specific flags */ +#define QM_CGR_WE_MASK 0x07ff /* 'Write Enable Mask'; */ +#define QM_CGR_WE_WR_PARM_G 0x0400 +#define QM_CGR_WE_WR_PARM_Y 0x0200 +#define QM_CGR_WE_WR_PARM_R 0x0100 +#define QM_CGR_WE_WR_EN_G 0x0080 +#define QM_CGR_WE_WR_EN_Y 0x0040 +#define QM_CGR_WE_WR_EN_R 0x0020 +#define QM_CGR_WE_CSCN_EN 0x0010 +#define QM_CGR_WE_CSCN_TARG 0x0008 +#define QM_CGR_WE_CSTD_EN 0x0004 +#define QM_CGR_WE_CS_THRES 0x0002 +#define QM_CGR_WE_MODE 0x0001 + +/* See 1.5.8.5.1: "Initialize FQ" */ +/* See 1.5.8.5.2: "Query FQ" */ +/* See 1.5.8.5.3: "Query FQ Non-Programmable Fields" */ +/* See 1.5.8.5.4: "Alter FQ State Commands " */ +/* See 1.5.8.6.1: "Initialize/Modify CGR" */ +/* See 1.5.8.6.2: "CGR Test Write" */ +/* See 1.5.8.6.3: "Query CGR" */ +/* See 1.5.8.6.4: "Query Congestion Group State" */ +struct qm_mcr_initfq { + u8 __reserved1[62]; +} __packed; +struct qm_mcr_queryfq { + u8 __reserved1[8]; + struct qm_fqd fqd; /* the FQD fields are here */ + u8 __reserved2[30]; +} __packed; +struct qm_mcr_queryfq_np { + u8 __reserved1; + u8 state; /* QM_MCR_NP_STATE_*** */ + u8 __reserved2; + u32 fqd_link:24; + u16 __reserved3:2; + u16 odp_seq:14; + u16 __reserved4:2; + u16 orp_nesn:14; + u16 __reserved5:1; + u16 orp_ea_hseq:15; + u16 __reserved6:1; + u16 orp_ea_tseq:15; + u8 __reserved7; + u32 orp_ea_hptr:24; + u8 __reserved8; + u32 orp_ea_tptr:24; + u8 __reserved9; + u32 pfdr_hptr:24; + u8 __reserved10; + u32 pfdr_tptr:24; + u8 __reserved11[5]; + u8 __reserved12:7; + u8 is:1; + u16 ics_surp; + u32 byte_cnt; + u8 __reserved13; + u32 frm_cnt:24; + u32 __reserved14; + u16 ra1_sfdr; /* QM_MCR_NP_RA1_*** */ + u16 ra2_sfdr; /* QM_MCR_NP_RA2_*** */ + u16 __reserved15; + u16 od1_sfdr; /* QM_MCR_NP_OD1_*** */ + u16 od2_sfdr; /* QM_MCR_NP_OD2_*** */ + u16 od3_sfdr; /* QM_MCR_NP_OD3_*** */ +} __packed; +struct qm_mcr_alterfq { + u8 fqs; /* Frame Queue Status */ + u8 __reserved1[61]; +} __packed; +struct qm_mcr_initcgr { + u8 __reserved1[62]; +} __packed; +struct qm_mcr_cgrtestwrite { + u16 __reserved1; + struct __qm_mc_cgr cgr; /* CGR fields */ + u8 __reserved2[3]; + u32 __reserved3:24; + u32 i_bcnt_hi:8;/* high 8-bits of 40-bit "Instant" */ + u32 i_bcnt_lo; /* low 32-bits of 40-bit */ + u32 __reserved4:24; + u32 a_bcnt_hi:8;/* high 8-bits of 40-bit "Average" */ + u32 a_bcnt_lo; /* low 32-bits of 40-bit */ + u16 lgt; /* Last Group Tick */ + u16 wr_prob_g; + u16 wr_prob_y; + u16 wr_prob_r; + u8 __reserved5[8]; +} __packed; +struct qm_mcr_querycgr { + u16 __reserved1; + struct __qm_mc_cgr cgr; /* CGR fields */ + u8 __reserved2[3]; + u32 __reserved3:24; + u32 i_bcnt_hi:8;/* high 8-bits of 40-bit "Instant" */ + u32 i_bcnt_lo; /* low 32-bits of 40-bit */ + u32 __reserved4:24; + u32 a_bcnt_hi:8;/* high 8-bits of 40-bit "Average" */ + u32 a_bcnt_lo; /* low 32-bits of 40-bit */ + union { + u32 cscn_targ_swp[4]; + u8 __reserved5[16]; + }; +} __packed; +static inline u64 qm_mcr_querycgr_i_get64(const struct qm_mcr_querycgr *q) +{ + return ((u64)q->i_bcnt_hi << 32) | (u64)q->i_bcnt_lo; +} +static inline u64 qm_mcr_querycgr_a_get64(const struct qm_mcr_querycgr *q) +{ + return ((u64)q->a_bcnt_hi << 32) | (u64)q->a_bcnt_lo; +} +static inline u64 qm_mcr_cgrtestwrite_i_get64( + const struct qm_mcr_cgrtestwrite *q) +{ + return ((u64)q->i_bcnt_hi << 32) | (u64)q->i_bcnt_lo; +} +static inline u64 qm_mcr_cgrtestwrite_a_get64( + const struct qm_mcr_cgrtestwrite *q) +{ + return ((u64)q->a_bcnt_hi << 32) | (u64)q->a_bcnt_lo; +} +/* Macro, so we compile better if 'v' isn't always 64-bit */ +#define qm_mcr_querycgr_i_set64(q, v) \ + do { \ + struct qm_mcr_querycgr *__q931 = (fd); \ + __q931->i_bcnt_hi = upper_32_bits(v); \ + __q931->i_bcnt_lo = lower_32_bits(v); \ + } while (0) +#define qm_mcr_querycgr_a_set64(q, v) \ + do { \ + struct qm_mcr_querycgr *__q931 = (fd); \ + __q931->a_bcnt_hi = upper_32_bits(v); \ + __q931->a_bcnt_lo = lower_32_bits(v); \ + } while (0) +struct __qm_mcr_querycongestion { + u32 __state[8]; +}; +struct qm_mcr_querycongestion { + u8 __reserved[30]; + /* Access this struct using QM_MCR_QUERYCONGESTION() */ + struct __qm_mcr_querycongestion state; +} __packed; +struct qm_mcr_querywq { + union { + u16 channel_wq; /* ignores wq (3 lsbits) */ + struct { + u16 id:13; /* qm_channel */ + u16 __reserved:3; + } __packed channel; + }; + u8 __reserved[28]; + u32 wq_len[8]; +} __packed; + +struct qm_mc_result { + u8 verb; + u8 result; + union { + struct qm_mcr_initfq initfq; + struct qm_mcr_queryfq queryfq; + struct qm_mcr_queryfq_np queryfq_np; + struct qm_mcr_alterfq alterfq; + struct qm_mcr_initcgr initcgr; + struct qm_mcr_cgrtestwrite cgrtestwrite; + struct qm_mcr_querycgr querycgr; + struct qm_mcr_querycongestion querycongestion; + struct qm_mcr_querywq querywq; + }; +} __packed; + +#define QM_MCR_VERB_RRID 0x80 +#define QM_MCR_VERB_MASK QM_MCC_VERB_MASK +#define QM_MCR_VERB_INITFQ_PARKED QM_MCC_VERB_INITFQ_PARKED +#define QM_MCR_VERB_INITFQ_SCHED QM_MCC_VERB_INITFQ_SCHED +#define QM_MCR_VERB_QUERYFQ QM_MCC_VERB_QUERYFQ +#define QM_MCR_VERB_QUERYFQ_NP QM_MCC_VERB_QUERYFQ_NP +#define QM_MCR_VERB_QUERYWQ QM_MCC_VERB_QUERYWQ +#define QM_MCR_VERB_QUERYWQ_DEDICATED QM_MCC_VERB_QUERYWQ_DEDICATED +#define QM_MCR_VERB_ALTER_SCHED QM_MCC_VERB_ALTER_SCHED +#define QM_MCR_VERB_ALTER_FE QM_MCC_VERB_ALTER_FE +#define QM_MCR_VERB_ALTER_RETIRE QM_MCC_VERB_ALTER_RETIRE +#define QM_MCR_VERB_ALTER_OOS QM_MCC_VERB_ALTER_OOS +#define QM_MCR_RESULT_NULL 0x00 +#define QM_MCR_RESULT_OK 0xf0 +#define QM_MCR_RESULT_ERR_FQID 0xf1 +#define QM_MCR_RESULT_ERR_FQSTATE 0xf2 +#define QM_MCR_RESULT_ERR_NOTEMPTY 0xf3 /* OOS fails if FQ is !empty */ +#define QM_MCR_RESULT_ERR_BADCHANNEL 0xf4 +#define QM_MCR_RESULT_PENDING 0xf8 +#define QM_MCR_RESULT_ERR_BADCOMMAND 0xff +#define QM_MCR_NP_STATE_FE 0x10 +#define QM_MCR_NP_STATE_R 0x08 +#define QM_MCR_NP_STATE_MASK 0x07 /* Reads FQD::STATE; */ +#define QM_MCR_NP_STATE_OOS 0x00 +#define QM_MCR_NP_STATE_RETIRED 0x01 +#define QM_MCR_NP_STATE_TEN_SCHED 0x02 +#define QM_MCR_NP_STATE_TRU_SCHED 0x03 +#define QM_MCR_NP_STATE_PARKED 0x04 +#define QM_MCR_NP_STATE_ACTIVE 0x05 +#define QM_MCR_NP_PTR_MASK 0x07ff /* for RA[12] & OD[123] */ +#define QM_MCR_NP_RA1_NRA(v) (((v) >> 14) & 0x3) /* FQD::NRA */ +#define QM_MCR_NP_RA2_IT(v) (((v) >> 14) & 0x1) /* FQD::IT */ +#define QM_MCR_NP_OD1_NOD(v) (((v) >> 14) & 0x3) /* FQD::NOD */ +#define QM_MCR_NP_OD3_NPC(v) (((v) >> 14) & 0x3) /* FQD::NPC */ +#define QM_MCR_FQS_ORLPRESENT 0x02 /* ORL fragments to come */ +#define QM_MCR_FQS_NOTEMPTY 0x01 /* FQ has enqueued frames */ +/* This extracts the state for congestion group 'n' from a query response. + * Eg. + * u8 cgr = [...]; + * struct qm_mc_result *res = [...]; + * printf("congestion group %d congestion state: %d\n", cgr, + * QM_MCR_QUERYCONGESTION(&res->querycongestion.state, cgr)); + */ +#define __CGR_WORD(num) (num >> 5) +#define __CGR_SHIFT(num) (num & 0x1f) +#define __CGR_NUM (sizeof(struct __qm_mcr_querycongestion) << 3) +static inline int QM_MCR_QUERYCONGESTION(struct __qm_mcr_querycongestion *p, + u8 cgr) +{ + return p->__state[__CGR_WORD(cgr)] & (0x80000000 >> __CGR_SHIFT(cgr)); +} + + +/*********************/ +/* Utility interface */ +/*********************/ + +/* Represents an allocator over a range of FQIDs. NB, accesses are not locked, + * spinlock them yourself if needed. */ +struct qman_fqid_pool; + +/* Create/destroy a FQID pool, num must be a multiple of 32. NB, _destroy() + * always succeeds, but returns non-zero if there were "leaked" FQID + * allocations. */ +struct qman_fqid_pool *qman_fqid_pool_create(u32 fqid_start, u32 num); +int qman_fqid_pool_destroy(struct qman_fqid_pool *pool); +/* Alloc/free a FQID from the range. _alloc() returns zero for success. */ +int qman_fqid_pool_alloc(struct qman_fqid_pool *pool, u32 *fqid); +void qman_fqid_pool_free(struct qman_fqid_pool *pool, u32 fqid); +u32 qman_fqid_pool_used(struct qman_fqid_pool *pool); + +/*******************************************************************/ +/* Managed (aka "shared" or "mux/demux") portal, high-level i/face */ +/*******************************************************************/ + + /* Portal and Frame Queues */ + /* ----------------------- */ +/* Represents a managed portal */ +struct qman_portal; + +/* This object type represents Qman frame queue descriptors (FQD), it is + * cacheline-aligned, and initialised by qman_create_fq(). The structure is + * defined further down. */ +struct qman_fq; + +/* This object type represents a Qman congestion group, it is defined further + * down. */ +struct qman_cgr; + +struct qman_portal_config { + /* If the caller enables DQRR stashing (and thus wishes to operate the + * portal from only one cpu), this is the logical CPU that the portal + * will stash to. Whether stashing is enabled or not, this setting is + * also used for any "core-affine" portals, ie. default portals + * associated to the corresponding cpu. -1 implies that there is no core + * affinity configured. */ + int cpu; + /* portal interrupt line */ + int irq; + /* Is this portal shared? (If so, it has coarser locking and demuxes + * processing on behalf of other CPUs.) */ + int is_shared; + /* The portal's dedicated channel id, use this value for initialising + * frame queues to target this portal when scheduled. */ + u16 channel; + /* A mask of which pool channels this portal has dequeue access to + * (using QM_SDQCR_CHANNELS_POOL(n) for the bitmask) */ + u32 pools; +}; + +/* This enum, and the callback type that returns it, are used when handling + * dequeued frames via DQRR. Note that for "null" callbacks registered with the + * portal object (for handling dequeues that do not demux because contextB is + * NULL), the return value *MUST* be qman_cb_dqrr_consume. */ +enum qman_cb_dqrr_result { + /* DQRR entry can be consumed */ + qman_cb_dqrr_consume, + /* Like _consume, but requests parking - FQ must be held-active */ + qman_cb_dqrr_park, + /* Does not consume, for DCA mode only. This allows out-of-order + * consumes by explicit calls to qman_dca() and/or the use of implicit + * DCA via EQCR entries. */ + qman_cb_dqrr_defer, + /* Stop processing without consuming this ring entry. Exits the current + * qman_poll_dqrr() or interrupt-handling, as appropriate. If within an + * interrupt handler, the callback would typically call + * qman_irqsource_remove(QM_PIRQ_DQRI) before returning this value, + * otherwise the interrupt will reassert immediately. */ + qman_cb_dqrr_stop, + /* Like qman_cb_dqrr_stop, but consumes the current entry. */ + qman_cb_dqrr_consume_stop +}; +typedef enum qman_cb_dqrr_result (*qman_cb_dqrr)(struct qman_portal *qm, + struct qman_fq *fq, + const struct qm_dqrr_entry *dqrr); + +/* This callback type is used when handling ERNs, FQRNs and FQRLs via MR. They + * are always consumed after the callback returns. */ +typedef void (*qman_cb_mr)(struct qman_portal *qm, struct qman_fq *fq, + const struct qm_mr_entry *msg); + +/* This callback type is used when handling DCP ERNs */ +typedef void (*qman_cb_dc_ern)(struct qman_portal *qm, + const struct qm_mr_entry *msg); + +/* s/w-visible states. Ie. tentatively scheduled + truly scheduled + active + + * held-active + held-suspended are just "sched". Things like "retired" will not + * be assumed until it is complete (ie. QMAN_FQ_STATE_CHANGING is set until + * then, to indicate it's completing and to gate attempts to retry the retire + * command). Note, park commands do not set QMAN_FQ_STATE_CHANGING because it's + * technically impossible in the case of enqueue DCAs (which refer to DQRR ring + * index rather than the FQ that ring entry corresponds to), so repeated park + * commands are allowed (if you're silly enough to try) but won't change FQ + * state, and the resulting park notifications move FQs from "sched" to + * "parked". */ +enum qman_fq_state { + qman_fq_state_oos, + qman_fq_state_parked, + qman_fq_state_sched, + qman_fq_state_retired +}; + +/* Frame queue objects (struct qman_fq) are stored within memory passed to + * qman_create_fq(), as this allows stashing of caller-provided demux callback + * pointers at no extra cost to stashing of (driver-internal) FQ state. If the + * caller wishes to add per-FQ state and have it benefit from dequeue-stashing, + * they should; + * + * (a) extend the qman_fq structure with their state; eg. + * + * // myfq is allocated and driver_fq callbacks filled in; + * struct my_fq { + * struct qman_fq base; + * int an_extra_field; + * [ ... add other fields to be associated with each FQ ...] + * } *myfq = some_my_fq_allocator(); + * struct qman_fq *fq = qman_create_fq(fqid, flags, &myfq->base); + * + * // in a dequeue callback, access extra fields from 'fq' via a cast; + * struct my_fq *myfq = (struct my_fq *)fq; + * do_something_with(myfq->an_extra_field); + * [...] + * + * (b) when and if configuring the FQ for context stashing, specify how ever + * many cachelines are required to stash 'struct my_fq', to accelerate not + * only the Qman driver but the callback as well. + */ + +struct qman_fq_cb { + qman_cb_dqrr dqrr; /* for dequeued frames */ + qman_cb_mr ern; /* for s/w ERNs */ + qman_cb_mr fqs; /* frame-queue state changes*/ +}; + +struct qman_fq { + /* Caller of qman_create_fq() provides these demux callbacks */ + struct qman_fq_cb cb; + /* These are internal to the driver, don't touch. In particular, they + * may change, be removed, or extended (so you shouldn't rely on + * sizeof(qman_fq) being a constant). */ + spinlock_t fqlock; + u32 fqid; + volatile unsigned long flags; + enum qman_fq_state state; + int cgr_groupid; + struct rb_node node; +#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP + u32 key; +#endif +}; + +/* This callback type is used when handling congestion group entry/exit. + * 'congested' is non-zero on congestion-entry, and zero on congestion-exit. */ +typedef void (*qman_cb_cgr)(struct qman_portal *qm, + struct qman_cgr *cgr, int congested); + +struct qman_cgr { + /* Set these prior to qman_create_cgr() */ + u32 cgrid; /* 0..255, but u32 to allow specials like -1, 256, etc.*/ + qman_cb_cgr cb; + /* These are private to the driver */ + u16 chan; /* portal channel this object is created on */ + struct list_head node; +}; + +/* Flags to qman_create_fq() */ +#define QMAN_FQ_FLAG_NO_ENQUEUE 0x00000001 /* can't enqueue */ +#define QMAN_FQ_FLAG_NO_MODIFY 0x00000002 /* can only enqueue */ +#define QMAN_FQ_FLAG_TO_DCPORTAL 0x00000004 /* consumed by CAAM/PME/Fman */ +#define QMAN_FQ_FLAG_LOCKED 0x00000008 /* multi-core locking */ +#define QMAN_FQ_FLAG_AS_IS 0x00000010 /* query h/w state */ +#define QMAN_FQ_FLAG_DYNAMIC_FQID 0x00000020 /* (de)allocate fqid */ + +/* Flags to qman_destroy_fq() */ +#define QMAN_FQ_DESTROY_PARKED 0x00000001 /* FQ can be parked or OOS */ + +/* Flags from qman_fq_state() */ +#define QMAN_FQ_STATE_CHANGING 0x80000000 /* 'state' is changing */ +#define QMAN_FQ_STATE_NE 0x40000000 /* retired FQ isn't empty */ +#define QMAN_FQ_STATE_ORL 0x20000000 /* retired FQ has ORL */ +#define QMAN_FQ_STATE_BLOCKOOS 0xe0000000 /* if any are set, no OOS */ +#define QMAN_FQ_STATE_CGR_EN 0x10000000 /* CGR enabled */ +#define QMAN_FQ_STATE_VDQCR 0x08000000 /* being volatile dequeued */ + +/* Flags to qman_init_fq() */ +#define QMAN_INITFQ_FLAG_SCHED 0x00000001 /* schedule rather than park */ +#define QMAN_INITFQ_FLAG_LOCAL 0x00000004 /* set dest portal */ + +/* Flags to qman_volatile_dequeue() */ +#ifdef CONFIG_FSL_DPA_CAN_WAIT +#define QMAN_VOLATILE_FLAG_WAIT 0x00000001 /* wait if VDQCR is in use */ +#define QMAN_VOLATILE_FLAG_WAIT_INT 0x00000002 /* if wait, interruptible? */ +#define QMAN_VOLATILE_FLAG_FINISH 0x00000004 /* wait till VDQCR completes */ +#endif + +/* Flags to qman_enqueue(). NB, the strange numbering is to align with hardware, + * bit-wise. (NB: the PME API is sensitive to these precise numberings too, so + * any change here should be audited in PME.) */ +#ifdef CONFIG_FSL_DPA_CAN_WAIT +#define QMAN_ENQUEUE_FLAG_WAIT 0x00010000 /* wait if EQCR is full */ +#define QMAN_ENQUEUE_FLAG_WAIT_INT 0x00020000 /* if wait, interruptible? */ +#ifdef CONFIG_FSL_DPA_CAN_WAIT_SYNC +#define QMAN_ENQUEUE_FLAG_WAIT_SYNC 0x00000004 /* if wait, until consumed? */ +#endif +#endif +#define QMAN_ENQUEUE_FLAG_WATCH_CGR 0x00080000 /* watch congestion state */ +#define QMAN_ENQUEUE_FLAG_DCA 0x00008000 /* perform enqueue-DCA */ +#define QMAN_ENQUEUE_FLAG_DCA_PARK 0x00004000 /* If DCA, requests park */ +#define QMAN_ENQUEUE_FLAG_DCA_PTR(p) /* If DCA, p is DQRR entry */ \ + (((u32)(p) << 2) & 0x00000f00) +#define QMAN_ENQUEUE_FLAG_C_GREEN 0x00000000 /* choose one C_*** flag */ +#define QMAN_ENQUEUE_FLAG_C_YELLOW 0x00000008 +#define QMAN_ENQUEUE_FLAG_C_RED 0x00000010 +#define QMAN_ENQUEUE_FLAG_C_OVERRIDE 0x00000018 +/* For the ORP-specific qman_enqueue_orp() variant; + * - this flag indicates "Not Last In Sequence", ie. all but the final fragment + * of a frame. */ +#define QMAN_ENQUEUE_FLAG_NLIS 0x01000000 +/* - this flag performs no enqueue but fills in an ORP sequence number that + * would otherwise block it (eg. if a frame has been dropped). */ +#define QMAN_ENQUEUE_FLAG_HOLE 0x02000000 +/* - this flag performs no enqueue but advances NESN to the given sequence + * number. */ +#define QMAN_ENQUEUE_FLAG_NESN 0x04000000 + +/* Flags to qman_modify_cgr() */ +#define QMAN_CGR_FLAG_USE_INIT 0x00000001 +#define QMAN_CGR_MODE_FRAME 0x00000001 + + /* Portal Management */ + /* ----------------- */ +/** + * qman_get_portal_config - get portal configuration settings + * + * This returns a read-only view of the current cpu's affine portal settings. + */ +const struct qman_portal_config *qman_get_portal_config(void); + +/** + * qman_irqsource_get - return the portal work that is interrupt-driven + * + * Returns a bitmask of QM_PIRQ_**I processing sources that are currently + * enabled for interrupt handling on the current cpu's affine portal. These + * sources will trigger the portal interrupt and the interrupt handler (or a + * tasklet/bottom-half it defers to) will perform the corresponding processing + * work. The qman_poll_***() functions will only process sources that are not in + * this bitmask. If the current CPU is sharing a portal hosted on another CPU, + * this always returns zero. + */ +u32 qman_irqsource_get(void); + +/** + * qman_irqsource_add - add processing sources to be interrupt-driven + * @bits: bitmask of QM_PIRQ_**I processing sources + * + * Adds processing sources that should be interrupt-driven (rather than + * processed via qman_poll_***() functions). Returns zero for success, or + * -EINVAL if the current CPU is sharing a portal hosted on another CPU. + */ +int qman_irqsource_add(u32 bits); + +/** + * qman_irqsource_remove - remove processing sources from being interrupt-driven + * @bits: bitmask of QM_PIRQ_**I processing sources + * + * Removes processing sources from being interrupt-driven, so that they will + * instead be processed via qman_poll_***() functions. Returns zero for success, + * or -EINVAL if the current CPU is sharing a portal hosted on another CPU. + */ +int qman_irqsource_remove(u32 bits); + +/** + * qman_affine_cpus - return a mask of cpus that have affine portals + */ +const cpumask_t *qman_affine_cpus(void); + +/** + * qman_affine_channel - return the channel ID of an portal + * @cpu: the cpu whose affine portal is the subject of the query + * + * If @cpu is -1, the affine portal for the current CPU will be used. It is a + * bug to call this function for any value of @cpu (other than -1) that is not a + * member of the mask returned from qman_affine_cpus(). + */ +u16 qman_affine_channel(int cpu); + +/** + * qman_get_affine_portal - return the portal pointer affine to cpu + * @cpu: the cpu whose affine portal is the subject of the query + * + */ +void *qman_get_affine_portal(int cpu); + +/** + * qman_poll_dqrr - process DQRR (fast-path) entries + * @limit: the maximum number of DQRR entries to process + * + * Use of this function requires that DQRR processing not be interrupt-driven. + * Ie. the value returned by qman_irqsource_get() should not include + * QM_PIRQ_DQRI. If the current CPU is sharing a portal hosted on another CPU, + * this function will return -EINVAL, otherwise the return value is >=0 and + * represents the number of DQRR entries processed. + */ +int qman_poll_dqrr(unsigned int limit); + +/** + * qman_poll_slow - process anything (except DQRR) that isn't interrupt-driven. + * + * This function does any portal processing that isn't interrupt-driven. If the + * current CPU is sharing a portal hosted on another CPU, this function will + * return (u32)-1, otherwise the return value is a bitmask of QM_PIRQ_* sources + * indicating what interrupt sources were actually processed by the call. + */ +u32 qman_poll_slow(void); + +/** + * qman_poll - legacy wrapper for qman_poll_dqrr() and qman_poll_slow() + * + * Dispatcher logic on a cpu can use this to trigger any maintenance of the + * affine portal. There are two classes of portal processing in question; + * fast-path (which involves demuxing dequeue ring (DQRR) entries and tracking + * enqueue ring (EQCR) consumption), and slow-path (which involves EQCR + * thresholds, congestion state changes, etc). This function does whatever + * processing is not triggered by interrupts. + * + * Note, if DQRR and some slow-path processing are poll-driven (rather than + * interrupt-driven) then this function uses a heuristic to determine how often + * to run slow-path processing - as slow-path processing introduces at least a + * minimum latency each time it is run, whereas fast-path (DQRR) processing is + * close to zero-cost if there is no work to be done. Applications can tune this + * behaviour themselves by using qman_poll_dqrr() and qman_poll_slow() directly + * rather than going via this wrapper. + */ +void qman_poll(void); + +/** + * qman_stop_dequeues - Stop h/w dequeuing to the s/w portal + * + * Disables DQRR processing of the portal. This is reference-counted, so + * qman_start_dequeues() must be called as many times as qman_stop_dequeues() to + * truly re-enable dequeuing. + */ +void qman_stop_dequeues(void); + +/** + * qman_start_dequeues - (Re)start h/w dequeuing to the s/w portal + * + * Enables DQRR processing of the portal. This is reference-counted, so + * qman_start_dequeues() must be called as many times as qman_stop_dequeues() to + * truly re-enable dequeuing. + */ +void qman_start_dequeues(void); + +/** + * qman_static_dequeue_add - Add pool channels to the portal SDQCR + * @pools: bit-mask of pool channels, using QM_SDQCR_CHANNELS_POOL(n) + * + * Adds a set of pool channels to the portal's static dequeue command register + * (SDQCR). The requested pools are limited to those the portal has dequeue + * access to. + */ +void qman_static_dequeue_add(u32 pools); + +/** + * qman_static_dequeue_del - Remove pool channels from the portal SDQCR + * @pools: bit-mask of pool channels, using QM_SDQCR_CHANNELS_POOL(n) + * + * Removes a set of pool channels from the portal's static dequeue command + * register (SDQCR). The requested pools are limited to those the portal has + * dequeue access to. + */ +void qman_static_dequeue_del(u32 pools); + +/** + * qman_static_dequeue_get - return the portal's current SDQCR + * + * Returns the portal's current static dequeue command register (SDQCR). The + * entire register is returned, so if only the currently-enabled pool channels + * are desired, mask the return value with QM_SDQCR_CHANNELS_POOL_MASK. + */ +u32 qman_static_dequeue_get(void); + +/** + * qman_dca - Perform a Discrete Consumption Acknowledgement + * @dq: the DQRR entry to be consumed + * @park_request: indicates whether the held-active @fq should be parked + * + * Only allowed in DCA-mode portals, for DQRR entries whose handler callback had + * previously returned 'qman_cb_dqrr_defer'. NB, as with the other APIs, this + * does not take a 'portal' argument but implies the core affine portal from the + * cpu that is currently executing the function. For reasons of locking, this + * function must be called from the same CPU as that which processed the DQRR + * entry in the first place. + */ +void qman_dca(struct qm_dqrr_entry *dq, int park_request); + +/** + * qman_eqcr_is_empty - Determine if portal's EQCR is empty + * + * For use in situations where a cpu-affine caller needs to determine when all + * enqueues for the local portal have been processed by Qman but can't use the + * QMAN_ENQUEUE_FLAG_WAIT_SYNC flag to do this from the final qman_enqueue(). + * The function forces tracking of EQCR consumption (which normally doesn't + * happen until enqueue processing needs to find space to put new enqueue + * commands), and returns zero if the ring still has unprocessed entries, + * non-zero if it is empty. + */ +int qman_eqcr_is_empty(void); + +/** + * qman_set_dc_ern - Set the handler for DCP enqueue rejection notifications + * @handler: callback for processing DCP ERNs + * @affine: whether this handler is specific to the locally affine portal + * + * If a hardware block's interface to Qman (ie. its direct-connect portal, or + * DCP) is configured not to receive enqueue rejections, then any enqueues + * through that DCP that are rejected will be sent to a given software portal. + * If @affine is non-zero, then this handler will only be used for DCP ERNs + * received on the portal affine to the current CPU. If multiple CPUs share a + * portal and they all call this function, they will be setting the handler for + * the same portal! If @affine is zero, then this handler will be global to all + * portals handled by this instance of the driver. Only those portals that do + * not have their own affine handler will use the global handler. + */ +void qman_set_dc_ern(qman_cb_dc_ern handler, int affine); + + /* FQ management */ + /* ------------- */ +/** + * qman_create_fq - Allocates a FQ + * @fqid: the index of the FQD to encapsulate, must be "Out of Service" + * @flags: bit-mask of QMAN_FQ_FLAG_*** options + * @fq: memory for storing the 'fq', with callbacks filled in + * + * Creates a frame queue object for the given @fqid, unless the + * QMAN_FQ_FLAG_DYNAMIC_FQID flag is set in @flags, in which case a FQID is + * dynamically allocated (or the function fails if none are available). Once + * created, the caller should not touch the memory at 'fq' except as extended to + * adjacent memory for user-defined fields (see the definition of "struct + * qman_fq" for more info). NO_MODIFY is only intended for enqueuing to + * pre-existing frame-queues that aren't to be otherwise interfered with, it + * prevents all other modifications to the frame queue. The TO_DCPORTAL flag + * causes the driver to honour any contextB modifications requested in the + * qm_init_fq() API, as this indicates the frame queue will be consumed by a + * direct-connect portal (PME, CAAM, or Fman). When frame queues are consumed by + * software portals, the contextB field is controlled by the driver and can't be + * modified by the caller. If the AS_IS flag is specified, management commands + * will be used on portal @p to query state for frame queue @fqid and construct + * a frame queue object based on that, rather than assuming/requiring that it be + * Out of Service. + */ +int qman_create_fq(u32 fqid, u32 flags, struct qman_fq *fq); + +/** + * qman_destroy_fq - Deallocates a FQ + * @fq: the frame queue object to release + * @flags: bit-mask of QMAN_FQ_FREE_*** options + * + * The memory for this frame queue object ('fq' provided in qman_create_fq()) is + * not deallocated but the caller regains ownership, to do with as desired. The + * FQ must be in the 'out-of-service' state unless the QMAN_FQ_FREE_PARKED flag + * is specified, in which case it may also be in the 'parked' state. + */ +void qman_destroy_fq(struct qman_fq *fq, u32 flags); + +/** + * qman_fq_fqid - Queries the frame queue ID of a FQ object + * @fq: the frame queue object to query + */ +u32 qman_fq_fqid(struct qman_fq *fq); + +/** + * qman_fq_state - Queries the state of a FQ object + * @fq: the frame queue object to query + * @state: pointer to state enum to return the FQ scheduling state + * @flags: pointer to state flags to receive QMAN_FQ_STATE_*** bitmask + * + * Queries the state of the FQ object, without performing any h/w commands. + * This captures the state, as seen by the driver, at the time the function + * executes. + */ +void qman_fq_state(struct qman_fq *fq, enum qman_fq_state *state, u32 *flags); + +/** + * qman_init_fq - Initialises FQ fields, leaves the FQ "parked" or "scheduled" + * @fq: the frame queue object to modify, must be 'parked' or new. + * @flags: bit-mask of QMAN_INITFQ_FLAG_*** options + * @opts: the FQ-modification settings, as defined in the low-level API + * + * The @opts parameter comes from the low-level portal API. Select + * QMAN_INITFQ_FLAG_SCHED in @flags to cause the frame queue to be scheduled + * rather than parked. NB, @opts can be NULL. + * + * Note that some fields and options within @opts may be ignored or overwritten + * by the driver; + * 1. the 'count' and 'fqid' fields are always ignored (this operation only + * affects one frame queue: @fq). + * 2. the QM_INITFQ_WE_CONTEXTB option of the 'we_mask' field and the associated + * 'fqd' structure's 'context_b' field are sometimes overwritten; + * - if @fq was not created with QMAN_FQ_FLAG_TO_DCPORTAL, then context_b is + * initialised to a value used by the driver for demux. + * - if context_b is initialised for demux, so is context_a in case stashing + * is requested (see item 4). + * (So caller control of context_b is only possible for TO_DCPORTAL frame queue + * objects.) + * 3. if @flags contains QMAN_INITFQ_FLAG_LOCAL, the 'fqd' structure's + * 'dest::channel' field will be overwritten to match the portal used to issue + * the command. If the WE_DESTWQ write-enable bit had already been set by the + * caller, the channel workqueue will be left as-is, otherwise the write-enable + * bit is set and the workqueue is set to a default of 4. If the "LOCAL" flag + * isn't set, the destination channel/workqueue fields and the write-enable bit + * are left as-is. + * 4. if the driver overwrites context_a/b for demux, then if + * QM_INITFQ_WE_CONTEXTA is set, the driver will only overwrite + * context_a.address fields and will leave the stashing fields provided by the + * user alone, otherwise it will zero out the context_a.stashing fields. + */ +int qman_init_fq(struct qman_fq *fq, u32 flags, struct qm_mcc_initfq *opts); + +/** + * qman_schedule_fq - Schedules a FQ + * @fq: the frame queue object to schedule, must be 'parked' + * + * Schedules the frame queue, which must be Parked, which takes it to + * Tentatively-Scheduled or Truly-Scheduled depending on its fill-level. + */ +int qman_schedule_fq(struct qman_fq *fq); + +/** + * qman_retire_fq - Retires a FQ + * @fq: the frame queue object to retire + * @flags: FQ flags (as per qman_fq_state) if retirement completes immediately + * + * Retires the frame queue. This returns zero if it succeeds immediately, +1 if + * the retirement was started asynchronously, otherwise it returns negative for + * failure. When this function returns zero, @flags is set to indicate whether + * the retired FQ is empty and/or whether it has any ORL fragments (to show up + * as ERNs). Otherwise the corresponding flags will be known when a subsequent + * FQRN message shows up on the portal's message ring. + * + * NB, if the retirement is asynchronous (the FQ was in the Truly Scheduled or + * Active state), the completion will be via the message ring as a FQRN - but + * the corresponding callback may occur before this function returns!! Ie. the + * caller should be prepared to accept the callback as the function is called, + * not only once it has returned. + */ +int qman_retire_fq(struct qman_fq *fq, u32 *flags); + +/** + * qman_oos_fq - Puts a FQ "out of service" + * @fq: the frame queue object to be put out-of-service, must be 'retired' + * + * The frame queue must be retired and empty, and if any order restoration list + * was released as ERNs at the time of retirement, they must all be consumed. + */ +int qman_oos_fq(struct qman_fq *fq); + +/** + * qman_fq_flow_control - Set the XON/XOFF state of a FQ + * @fq: the frame queue object to be set to XON/XOFF state, must not be 'oos', + * or 'retired' or 'parked' state + * @xon: boolean to set fq in XON or XOFF state + * + * The frame should be in Tentatively Scheduled state or Truly Schedule sate, + * otherwise the IFSI interrupt will be asserted. + */ +int qman_fq_flow_control(struct qman_fq *fq, int xon); + +/** + * qman_query_fq - Queries FQD fields (via h/w query command) + * @fq: the frame queue object to be queried + * @fqd: storage for the queried FQD fields + */ +int qman_query_fq(struct qman_fq *fq, struct qm_fqd *fqd); + +/** + * qman_query_fq_np - Queries non-programmable FQD fields + * @fq: the frame queue object to be queried + * @np: storage for the queried FQD fields + */ +int qman_query_fq_np(struct qman_fq *fq, struct qm_mcr_queryfq_np *np); + +/** + * qman_query_wq - Queries work queue lengths + * @query_dedicated: If non-zero, query length of WQs in the channel dedicated + * to this software portal. Otherwise, query length of WQs in a + * channel specified in wq. + * @wq: storage for the queried WQs lengths. Also specified the channel to + * to query if query_dedicated is zero. + */ +int qman_query_wq(u8 query_dedicated, struct qm_mcr_querywq *wq); + +/** + * qman_volatile_dequeue - Issue a volatile dequeue command + * @fq: the frame queue object to dequeue from + * @flags: a bit-mask of QMAN_VOLATILE_FLAG_*** options + * @vdqcr: bit mask of QM_VDQCR_*** options, as per qm_dqrr_vdqcr_set() + * + * Attempts to lock access to the portal's VDQCR volatile dequeue functionality. + * The function will block and sleep if QMAN_VOLATILE_FLAG_WAIT is specified and + * the VDQCR is already in use, otherwise returns non-zero for failure. If + * QMAN_VOLATILE_FLAG_FINISH is specified, the function will only return once + * the VDQCR command has finished executing (ie. once the callback for the last + * DQRR entry resulting from the VDQCR command has been called). If not using + * the FINISH flag, completion can be determined either by detecting the + * presence of the QM_DQRR_STAT_UNSCHEDULED and QM_DQRR_STAT_DQCR_EXPIRED bits + * in the "stat" field of the "struct qm_dqrr_entry" passed to the FQ's dequeue + * callback, or by waiting for the QMAN_FQ_STATE_VDQCR bit to disappear from the + * "flags" retrieved from qman_fq_state(). + */ +int qman_volatile_dequeue(struct qman_fq *fq, u32 flags, u32 vdqcr); + +/** + * qman_enqueue - Enqueue a frame to a frame queue + * @fq: the frame queue object to enqueue to + * @fd: a descriptor of the frame to be enqueued + * @flags: bit-mask of QMAN_ENQUEUE_FLAG_*** options + * + * Fills an entry in the EQCR of portal @qm to enqueue the frame described by + * @fd. The descriptor details are copied from @fd to the EQCR entry, the 'pid' + * field is ignored. The return value is non-zero on error, such as ring full + * (and FLAG_WAIT not specified), congestion avoidance (FLAG_WATCH_CGR + * specified), etc. If the ring is full and FLAG_WAIT is specified, this + * function will block. If FLAG_INTERRUPT is set, the EQCI bit of the portal + * interrupt will assert when Qman consumes the EQCR entry (subject to "status + * disable", "enable", and "inhibit" registers). If FLAG_DCA is set, Qman will + * perform an implied "discrete consumption acknowledgement" on the dequeue + * ring's (DQRR) entry, at the ring index specified by the FLAG_DCA_IDX(x) + * macro. (As an alternative to issuing explicit DCA actions on DQRR entries, + * this implicit DCA can delay the release of a "held active" frame queue + * corresponding to a DQRR entry until Qman consumes the EQCR entry - providing + * order-preservation semantics in packet-forwarding scenarios.) If FLAG_DCA is + * set, then FLAG_DCA_PARK can also be set to imply that the DQRR consumption + * acknowledgement should "park request" the "held active" frame queue. Ie. + * when the portal eventually releases that frame queue, it will be left in the + * Parked state rather than Tentatively Scheduled or Truly Scheduled. If the + * portal is watching congestion groups, the QMAN_ENQUEUE_FLAG_WATCH_CGR flag + * is requested, and the FQ is a member of a congestion group, then this + * function returns -EAGAIN if the congestion group is currently congested. + * Note, this does not eliminate ERNs, as the async interface means we can be + * sending enqueue commands to an un-congested FQ that becomes congested before + * the enqueue commands are processed, but it does minimise needless thrashing + * of an already busy hardware resource by throttling many of the to-be-dropped + * enqueues "at the source". + */ +int qman_enqueue(struct qman_fq *fq, const struct qm_fd *fd, u32 flags); + +typedef int (*qman_cb_precommit) (void *arg); +/** + * qman_enqueue_precommit - Enqueue a frame to a frame queue and call cb + * @fq: the frame queue object to enqueue to + * @fd: a descriptor of the frame to be enqueued + * @flags: bit-mask of QMAN_ENQUEUE_FLAG_*** options + * @cb: user supplied callback function to invoke before writing commit verb. + * @cb_arg: callback function argument + * + * This is similar to qman_enqueue except that it will invoke a user supplied + * callback function just before writng the commit verb. This is useful + * when the user want to do something *just before* enqueuing the request and + * the enqueue can't fail. + */ +int qman_enqueue_precommit(struct qman_fq *fq, const struct qm_fd *fd, + u32 flags, qman_cb_precommit cb, void *cb_arg); + +/** + * qman_enqueue_orp - Enqueue a frame to a frame queue using an ORP + * @fq: the frame queue object to enqueue to + * @fd: a descriptor of the frame to be enqueued + * @flags: bit-mask of QMAN_ENQUEUE_FLAG_*** options + * @orp: the frame queue object used as an order restoration point. + * @orp_seqnum: the sequence number of this frame in the order restoration path + * + * Similar to qman_enqueue(), but with the addition of an Order Restoration + * Point (@orp) and corresponding sequence number (@orp_seqnum) for this + * enqueue operation to employ order restoration. Each frame queue object acts + * as an Order Definition Point (ODP) by providing each frame dequeued from it + * with an incrementing sequence number, this value is generally ignored unless + * that sequence of dequeued frames will need order restoration later. Each + * frame queue object also encapsulates an Order Restoration Point (ORP), which + * is a re-assembly context for re-ordering frames relative to their sequence + * numbers as they are enqueued. The ORP does not have to be within the frame + * queue that receives the enqueued frame, in fact it is usually the frame + * queue from which the frames were originally dequeued. For the purposes of + * order restoration, multiple frames (or "fragments") can be enqueued for a + * single sequence number by setting the QMAN_ENQUEUE_FLAG_NLIS flag for all + * enqueues except the final fragment of a given sequence number. Ordering + * between sequence numbers is guaranteed, even if fragments of different + * sequence numbers are interlaced with one another. Fragments of the same + * sequence number will retain the order in which they are enqueued. If no + * enqueue is to performed, QMAN_ENQUEUE_FLAG_HOLE indicates that the given + * sequence number is to be "skipped" by the ORP logic (eg. if a frame has been + * dropped from a sequence), or QMAN_ENQUEUE_FLAG_NESN indicates that the given + * sequence number should become the ORP's "Next Expected Sequence Number". + * + * Side note: a frame queue object can be used purely as an ORP, without + * carrying any frames at all. Care should be taken not to deallocate a frame + * queue object that is being actively used as an ORP, as a future allocation + * of the frame queue object may start using the internal ORP before the + * previous use has finished. + */ +int qman_enqueue_orp(struct qman_fq *fq, const struct qm_fd *fd, u32 flags, + struct qman_fq *orp, u16 orp_seqnum); + +/** + * qman_alloc_fqid_range - Allocate a contiguous range of FQIDs + * @result: is set by the API to the base FQID of the allocated range + * @count: the number of FQIDs required + * @align: required alignment of the allocated range + * @partial: non-zero if the API can return fewer than @count FQIDs + * + * Returns the number of frame queues allocated, or a negative error code. If + * @partial is non zero, the allocation request may return a smaller range of + * FQs than requested (though alignment will be as requested). If @partial is + * zero, the return value will either be 'count' or negative. + */ +int qman_alloc_fqid_range(u32 *result, u32 count, u32 align, int partial); +static inline int qman_alloc_fqid(u32 *result) +{ + int ret = qman_alloc_fqid_range(result, 1, 0, 0); + return (ret > 0) ? 0 : ret; +} + +/** + * qman_release_fqid_range - Release the specified range of frame queue IDs + * @fqid: the base FQID of the range to deallocate + * @count: the number of FQIDs in the range + * + * This function can also be used to seed the allocator with ranges of FQIDs + * that it can subsequently allocate from. + */ +void qman_release_fqid_range(u32 fqid, unsigned int count); +static inline void qman_release_fqid(u32 fqid) +{ + qman_release_fqid_range(fqid, 1); +} + +void qman_seed_fqid_range(u32 fqid, unsigned int count); + + +int qman_shutdown_fq(u32 fqid); + +/** + * qman_reserve_fqid_range - Reserve the specified range of frame queue IDs + * @fqid: the base FQID of the range to deallocate + * @count: the number of FQIDs in the range + */ +int qman_reserve_fqid_range(u32 fqid, unsigned int count); +static inline int qman_reserve_fqid(u32 fqid) +{ + return qman_reserve_fqid_range(fqid, 1); +} + + /* Pool-channel management */ + /* ----------------------- */ +/** + * qman_alloc_pool_range - Allocate a contiguous range of pool-channel IDs + * @result: is set by the API to the base pool-channel ID of the allocated range + * @count: the number of pool-channel IDs required + * @align: required alignment of the allocated range + * @partial: non-zero if the API can return fewer than @count + * + * Returns the number of pool-channel IDs allocated, or a negative error code. + * If @partial is non zero, the allocation request may return a smaller range of + * than requested (though alignment will be as requested). If @partial is zero, + * the return value will either be 'count' or negative. + */ +int qman_alloc_pool_range(u32 *result, u32 count, u32 align, int partial); +static inline int qman_alloc_pool(u32 *result) +{ + int ret = qman_alloc_pool_range(result, 1, 0, 0); + return (ret > 0) ? 0 : ret; +} + +/** + * qman_release_pool_range - Release the specified range of pool-channel IDs + * @id: the base pool-channel ID of the range to deallocate + * @count: the number of pool-channel IDs in the range + */ +void qman_release_pool_range(u32 id, unsigned int count); +static inline void qman_release_pool(u32 id) +{ + qman_release_pool_range(id, 1); +} + +/** + * qman_reserve_pool_range - Reserve the specified range of pool-channel IDs + * @id: the base pool-channel ID of the range to reserve + * @count: the number of pool-channel IDs in the range + */ +int qman_reserve_pool_range(u32 id, unsigned int count); +static inline int qman_reserve_pool(u32 id) +{ + return qman_reserve_pool_range(id, 1); +} + +void qman_seed_pool_range(u32 id, unsigned int count); + + /* CGR management */ + /* -------------- */ +/** + * qman_create_cgr - Register a congestion group object + * @cgr: the 'cgr' object, with fields filled in + * @flags: QMAN_CGR_FLAG_* values + * @opts: optional state of CGR settings + * + * Registers this object to receiving congestion entry/exit callbacks on the + * portal affine to the cpu portal on which this API is executed. If opts is + * NULL then only the callback (cgr->cb) function is registered. If @flags + * contains QMAN_CGR_FLAG_USE_INIT, then an init hw command (which will reset + * any unspecified parameters) will be used rather than a modify hw hardware + * (which only modifies the specified parameters). + */ +int qman_create_cgr(struct qman_cgr *cgr, u32 flags, + struct qm_mcc_initcgr *opts); + +/** + * qman_create_cgr_to_dcp - Register a congestion group object to DCP portal + * @cgr: the 'cgr' object, with fields filled in + * @flags: QMAN_CGR_FLAG_* values + * @dcp_portal: the DCP portal to which the cgr object is registered. + * @opts: optional state of CGR settings + * + */ +int qman_create_cgr_to_dcp(struct qman_cgr *cgr, u32 flags, u16 dcp_portal, + struct qm_mcc_initcgr *opts); + +/** + * qman_delete_cgr - Deregisters a congestion group object + * @cgr: the 'cgr' object to deregister + * + * "Unplugs" this CGR object from the portal affine to the cpu on which this API + * is executed. This must be excuted on the same affine portal on which it was + * created. + */ +int qman_delete_cgr(struct qman_cgr *cgr); + +/** + * qman_modify_cgr - Modify CGR fields + * @cgr: the 'cgr' object to modify + * @flags: QMAN_CGR_FLAG_* values + * @opts: the CGR-modification settings + * + * The @opts parameter comes from the low-level portal API, and can be NULL. + * Note that some fields and options within @opts may be ignored or overwritten + * by the driver, in particular the 'cgrid' field is ignored (this operation + * only affects the given CGR object). If @flags contains + * QMAN_CGR_FLAG_USE_INIT, then an init hw command (which will reset any + * unspecified parameters) will be used rather than a modify hw hardware (which + * only modifies the specified parameters). + */ +int qman_modify_cgr(struct qman_cgr *cgr, u32 flags, + struct qm_mcc_initcgr *opts); + +/** +* qman_query_cgr - Queries CGR fields +* @cgr: the 'cgr' object to query +* @result: storage for the queried congestion group record +*/ +int qman_query_cgr(struct qman_cgr *cgr, struct qm_mcr_querycgr *result); + +/** + * qman_alloc_cgrid_range - Allocate a contiguous range of CGR IDs + * @result: is set by the API to the base CGR ID of the allocated range + * @count: the number of CGR IDs required + * @align: required alignment of the allocated range + * @partial: non-zero if the API can return fewer than @count + * + * Returns the number of CGR IDs allocated, or a negative error code. + * If @partial is non zero, the allocation request may return a smaller range of + * than requested (though alignment will be as requested). If @partial is zero, + * the return value will either be 'count' or negative. + */ +int qman_alloc_cgrid_range(u32 *result, u32 count, u32 align, int partial); +static inline int qman_alloc_cgrid(u32 *result) +{ + int ret = qman_alloc_cgrid_range(result, 1, 0, 0); + return (ret > 0) ? 0 : ret; +} + +/** + * qman_release_cgrid_range - Release the specified range of CGR IDs + * @id: the base CGR ID of the range to deallocate + * @count: the number of CGR IDs in the range + */ +void qman_release_cgrid_range(u32 id, unsigned int count); +static inline void qman_release_cgrid(u32 id) +{ + qman_release_cgrid_range(id, 1); +} + +/** + * qman_reserve_cgrid_range - Reserve the specified range of CGR ID + * @id: the base CGR ID of the range to reserve + * @count: the number of CGR IDs in the range + */ +int qman_reserve_cgrid_range(u32 id, unsigned int count); +static inline int qman_reserve_cgrid(u32 id) +{ + return qman_reserve_cgrid_range(id, 1); +} + +void qman_seed_cgrid_range(u32 id, unsigned int count); + + + /* Helpers */ + /* ------- */ +/** + * qman_poll_fq_for_init - Check if an FQ has been initialised from OOS + * @fqid: the FQID that will be initialised by other s/w + * + * In many situations, a FQID is provided for communication between s/w + * entities, and whilst the consumer is responsible for initialising and + * scheduling the FQ, the producer(s) generally create a wrapper FQ object using + * and only call qman_enqueue() (no FQ initialisation, scheduling, etc). Ie; + * qman_create_fq(..., QMAN_FQ_FLAG_NO_MODIFY, ...); + * However, data can not be enqueued to the FQ until it is initialised out of + * the OOS state - this function polls for that condition. It is particularly + * useful for users of IPC functions - each endpoint's Rx FQ is the other + * endpoint's Tx FQ, so each side can initialise and schedule their Rx FQ object + * and then use this API on the (NO_MODIFY) Tx FQ object in order to + * synchronise. The function returns zero for success, +1 if the FQ is still in + * the OOS state, or negative if there was an error. + */ +static inline int qman_poll_fq_for_init(struct qman_fq *fq) +{ + struct qm_mcr_queryfq_np np; + int err; + err = qman_query_fq_np(fq, &np); + if (err) + return err; + if ((np.state & QM_MCR_NP_STATE_MASK) == QM_MCR_NP_STATE_OOS) + return 1; + return 0; +} + +/** + * qman_set_wpm - Set waterfall power management + * + * @wpm_enable: boolean, 1 = enable wpm, 0 = disable wpm. + * + * Return 0 for success, return -ENODEV if QMan misc_cfg register is not + * accessible. + */ +int qman_set_wpm(int wpm_enable); + +/** + * qman_get_swp - Query the waterfall power management setting + * + * @wpm_enable: boolean, 1 = enable wpm, 0 = disable wpm. + * + * Return 0 for success, return -ENODEV if QMan misc_cfg register is not + * accessible. + */ +int qman_get_wpm(int *wpm_enable); + +/* The below qman_p_***() variants might be called in a migration situation + * (e.g. cpu hotplug). They are used to continue accessing the portal that + * execution was affine to prior to migration. + * @qman_portal specifies which portal the APIs will use. +*/ +const struct qman_portal_config *qman_p_get_portal_config(struct qman_portal + *p); +int qman_p_irqsource_add(struct qman_portal *p, u32 bits); +int qman_p_irqsource_remove(struct qman_portal *p, u32 bits); +int qman_p_poll_dqrr(struct qman_portal *p, unsigned int limit); +u32 qman_p_poll_slow(struct qman_portal *p); +void qman_p_poll(struct qman_portal *p); +void qman_p_stop_dequeues(struct qman_portal *p); +void qman_p_start_dequeues(struct qman_portal *p); +void qman_p_static_dequeue_add(struct qman_portal *p, u32 pools); +void qman_p_static_dequeue_del(struct qman_portal *p, u32 pools); +u32 qman_p_static_dequeue_get(struct qman_portal *p); +void qman_p_dca(struct qman_portal *p, struct qm_dqrr_entry *dq, + int park_request); +int qman_p_volatile_dequeue(struct qman_portal *p, struct qman_fq *fq, + u32 flags __maybe_unused, u32 vdqcr); +int qman_p_enqueue(struct qman_portal *p, struct qman_fq *fq, + const struct qm_fd *fd, u32 flags); +int qman_p_enqueue_orp(struct qman_portal *p, struct qman_fq *fq, + const struct qm_fd *fd, u32 flags, + struct qman_fq *orp, u16 orp_seqnum); +int qman_p_enqueue_precommit(struct qman_portal *p, struct qman_fq *fq, + const struct qm_fd *fd, u32 flags, + qman_cb_precommit cb, void *cb_arg); +#ifdef __cplusplus +} +#endif + +#endif /* FSL_QMAN_H */ -- 2.2.2 -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/