2009-06-26 09:40:08

by Nicolas Ferre

[permalink] [raw]
Subject: [PATCH 1/2 v2] dmaengine: at_hdmac: new driver for the Atmel AHB DMA Controller

This AHB DMA Controller (aka HDMA or DMAC on AT91 systems) is availlable on
at91sam9rl chip. It will be used on other products in the future.

This first release covers only the memory-to-memory tranfer type. This is the
only tranfer type supported by this chip. On other products, it will be used
also for peripheral DMA transfer (slave API support to come).

I used dmatest client without problem in different configurations to test it.

Full documentation for this controller can be found in the SAM9RL datasheet:
http://www.atmel.com/dyn/products/product_card.asp?part_id=4243

Signed-off-by: Nicolas Ferre <[email protected]>
---
v1 is here:
http://lkml.org/lkml/2008/10/17/294

v1 -> v2:
- header file moved to processor specific include directory
- ack bit on descriptor in control of client is the last of a descriptor chain
- irq and tasklet init in _probe() function instead of channel resources
allocation

arch/arm/mach-at91/include/mach/at_hdmac.h | 26 +
drivers/dma/Kconfig | 8 +
drivers/dma/Makefile | 1 +
drivers/dma/at_hdmac.c | 998 ++++++++++++++++++++++++++++
drivers/dma/at_hdmac_regs.h | 386 +++++++++++
5 files changed, 1419 insertions(+), 0 deletions(-)
create mode 100644 arch/arm/mach-at91/include/mach/at_hdmac.h
create mode 100644 drivers/dma/at_hdmac.c
create mode 100644 drivers/dma/at_hdmac_regs.h

diff --git a/arch/arm/mach-at91/include/mach/at_hdmac.h b/arch/arm/mach-at91/include/mach/at_hdmac.h
new file mode 100644
index 0000000..21a5554
--- /dev/null
+++ b/arch/arm/mach-at91/include/mach/at_hdmac.h
@@ -0,0 +1,26 @@
+/*
+ * Header file for the Atmel AHB DMA Controller driver
+ *
+ * Copyright (C) 2008 Atmel Corporation
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ */
+#ifndef AT_HDMAC_H
+#define AT_HDMAC_H
+
+#include <linux/dmaengine.h>
+
+/**
+ * struct at_dma_platform_data - Controller configuration parameters
+ * @nr_channels: Number of channels supported by hardware (max 8)
+ * @cap_mask: dma_capability flags supported by the platform
+ */
+struct at_dma_platform_data {
+ unsigned int nr_channels;
+ dma_cap_mask_t cap_mask;
+};
+
+#endif /* AT_HDMAC_H */
diff --git a/drivers/dma/Kconfig b/drivers/dma/Kconfig
index babf214..bc8fb41 100644
--- a/drivers/dma/Kconfig
+++ b/drivers/dma/Kconfig
@@ -46,6 +46,14 @@ config DW_DMAC
Support the Synopsys DesignWare AHB DMA controller. This
can be integrated in chips such as the Atmel AT32ap7000.

+config AT_HDMAC
+ tristate "Atmel AHB DMA support"
+ depends on ARCH_AT91SAM9RL
+ select DMA_ENGINE
+ help
+ Support the Atmel AHB DMA controller. This can be integrated in
+ chips such as the Atmel AT91SAM9RL.
+
config FSL_DMA
tristate "Freescale Elo and Elo Plus DMA support"
depends on FSL_SOC
diff --git a/drivers/dma/Makefile b/drivers/dma/Makefile
index 2e5dc96..d7bc5fd 100644
--- a/drivers/dma/Makefile
+++ b/drivers/dma/Makefile
@@ -7,4 +7,5 @@ obj-$(CONFIG_INTEL_IOP_ADMA) += iop-adma.o
obj-$(CONFIG_FSL_DMA) += fsldma.o
obj-$(CONFIG_MV_XOR) += mv_xor.o
obj-$(CONFIG_DW_DMAC) += dw_dmac.o
+obj-$(CONFIG_AT_HDMAC) += at_hdmac.o
obj-$(CONFIG_MX3_IPU) += ipu/
diff --git a/drivers/dma/at_hdmac.c b/drivers/dma/at_hdmac.c
new file mode 100644
index 0000000..f395e2f
--- /dev/null
+++ b/drivers/dma/at_hdmac.c
@@ -0,0 +1,998 @@
+/*
+ * Driver for the Atmel AHB DMA Controller (aka HDMA or DMAC on AT91 systems)
+ *
+ * Copyright (C) 2008 Atmel Corporation
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ *
+ * This supports the Atmel AHB DMA Controller,
+ *
+ * The driver has currently been tested with the Atmel AT91SAM9RL
+ * and AT91SAM9M10.
+ */
+
+#include <linux/clk.h>
+#include <linux/dmaengine.h>
+#include <linux/dma-mapping.h>
+#include <linux/dmapool.h>
+#include <linux/interrupt.h>
+#include <linux/module.h>
+#include <linux/platform_device.h>
+
+#include "at_hdmac_regs.h"
+
+/*
+ * Glossary
+ * --------
+ *
+ * at_hdmac : Name of the ATmel AHB DMA Controller
+ * at_dma_ / atdma : ATmel DMA controller entity related
+ * atc_ / atchan : ATmel DMA Channel entity related
+ */
+
+#define ATC_DEFAULT_CFG (ATC_FIFOCFG_HALFFIFO)
+#define ATC_DEFAULT_CTRLA (0)
+#define ATC_DEFAULT_CTRLB (ATC_SIF(0) \
+ |ATC_DIF(1))
+
+/*
+ * Initial number of descriptors to allocate for each channel. This could
+ * be increased of this amount during dma usage.
+ */
+#define INIT_NR_DESCS_PER_CHANNEL 16
+
+/* prototypes */
+static dma_cookie_t atc_tx_submit(struct dma_async_tx_descriptor *tx);
+
+
+/*----------------------------------------------------------------------*/
+
+static struct at_desc *atc_first_active(struct at_dma_chan *atchan)
+{
+ return list_first_entry(&atchan->active_list,
+ struct at_desc, desc_node);
+}
+
+static struct at_desc *atc_first_queued(struct at_dma_chan *atchan)
+{
+ return list_first_entry(&atchan->queue,
+ struct at_desc, desc_node);
+}
+
+/**
+ * atc_alloc_descriptor - allocate and return an initilized descriptor
+ * @chan: the channel to allocate descriptors for
+ * @gfp_flags: GFP allocation flags
+ *
+ * Note: The ack-bit is positionned in the descriptor flag at creation time
+ * to make initial allocation more convenient. This bit will be cleared
+ * and control will be given to clien at usage time (during
+ * prep_dma_memcpy() for instance).
+ */
+static struct at_desc *atc_alloc_descriptor(struct dma_chan *chan,
+ gfp_t gfp_flags)
+{
+ struct at_desc *desc = NULL;
+ struct at_dma *atdma = to_at_dma(chan->device);
+ dma_addr_t phys;
+
+ desc = dma_pool_alloc(atdma->dma_desc_pool, gfp_flags, &phys);
+ if (desc) {
+ memset(desc, 0, sizeof(struct at_desc));
+ dma_async_tx_descriptor_init(&desc->txd, chan);
+ async_tx_ack(&desc->txd);
+ desc->txd.tx_submit = atc_tx_submit;
+ desc->txd.phys = phys;
+ }
+
+ return desc;
+}
+
+/**
+ * atc_desc_get - get a unsused descriptor from free_list
+ * @atchan: channel we want a new descriptor for
+ */
+static struct at_desc *atc_desc_get(struct at_dma_chan *atchan)
+{
+ struct at_desc *desc, *_desc;
+ struct at_desc *ret = NULL;
+ unsigned int i = 0;
+ LIST_HEAD(tmp_list);
+
+ spin_lock_bh(&atchan->lock);
+ list_for_each_entry_safe(desc, _desc, &atchan->free_list, desc_node) {
+ i++;
+ if (async_tx_test_ack(&desc->txd)) {
+ list_del(&desc->desc_node);
+ ret = desc;
+ break;
+ }
+ dev_dbg(chan2dev(&atchan->chan_common),
+ "desc %p not ACKed\n", desc);
+ }
+ spin_unlock_bh(&atchan->lock);
+ dev_vdbg(chan2dev(&atchan->chan_common),
+ "scanned %u descriptors on freelist\n", i);
+
+ /* no more descriptor available in initial pool : create some more */
+ if (!ret) {
+ for (i = 0; i < INIT_NR_DESCS_PER_CHANNEL; i++) {
+ desc = atc_alloc_descriptor(&atchan->chan_common,
+ GFP_KERNEL);
+ if (!desc)
+ break;
+ /* return first descriptor, queue others in free_list */
+ if (i)
+ list_add_tail(&desc->desc_node, &tmp_list);
+ else
+ ret = desc;
+ }
+
+ spin_lock_bh(&atchan->lock);
+ atchan->descs_allocated += i;
+ list_splice(&tmp_list, &atchan->free_list);
+ spin_unlock_bh(&atchan->lock);
+ }
+
+ return ret;
+}
+
+/**
+ * atc_desc_put - move a descriptor, including any children, to the free list
+ * @atchan: channel we work on
+ * @desc: descriptor, at the head of a chain, to move to free list
+ */
+static void atc_desc_put(struct at_dma_chan *atchan, struct at_desc *desc)
+{
+ if (desc) {
+ struct at_desc *child;
+
+ spin_lock_bh(&atchan->lock);
+ list_for_each_entry(child, &desc->txd.tx_list, desc_node)
+ dev_vdbg(chan2dev(&atchan->chan_common),
+ "moving child desc %p to freelist\n",
+ child);
+ list_splice_init(&desc->txd.tx_list, &atchan->free_list);
+ dev_vdbg(chan2dev(&atchan->chan_common),
+ "moving desc %p to freelist\n", desc);
+ list_add(&desc->desc_node, &atchan->free_list);
+ spin_unlock_bh(&atchan->lock);
+ }
+}
+
+/**
+ * atc_assign_cookie - compute and assign new cookie
+ * @atchan: channel we work on
+ * @desc: descriptor to asign cookie for
+ *
+ * Called with atchan->lock held and bh disabled
+ */
+static dma_cookie_t
+atc_assign_cookie(struct at_dma_chan *atchan, struct at_desc *desc)
+{
+ dma_cookie_t cookie = atchan->chan_common.cookie;
+
+ if (++cookie < 0)
+ cookie = 1;
+
+ atchan->chan_common.cookie = cookie;
+ desc->txd.cookie = cookie;
+
+ return cookie;
+}
+
+/**
+ * atc_dostart - starts the DMA engine for real
+ * @atchan: the channel we want to start
+ * @first: first descriptor in the list we want to begin with
+ *
+ * Called with atchan->lock held and bh disabled
+ */
+static void atc_dostart(struct at_dma_chan *atchan, struct at_desc *first)
+{
+ struct at_dma *atdma = to_at_dma(atchan->chan_common.device);
+
+ /* ASSERT: channel is idle */
+ if (atc_chan_is_enabled(atchan)) {
+ dev_err(chan2dev(&atchan->chan_common),
+ "BUG: Attempted to start non-idle channel\n");
+ dev_err(chan2dev(&atchan->chan_common),
+ " channel: s0x%x d0x%x ctrl0x%x:0x%x l0x%x\n",
+ channel_readl(atchan, SADDR),
+ channel_readl(atchan, DADDR),
+ channel_readl(atchan, CTRLA),
+ channel_readl(atchan, CTRLB),
+ channel_readl(atchan, DSCR));
+
+ /* The tasklet will hopefully advance the queue... */
+ return;
+ }
+
+ vdbg_dump_regs(atchan);
+
+ /* clear any pending interrupt */
+ while (dma_readl(atdma, EBCISR))
+ cpu_relax();
+
+ channel_writel(atchan, SADDR, 0);
+ channel_writel(atchan, DADDR, 0);
+ channel_writel(atchan, CTRLA, 0);
+ channel_writel(atchan, CTRLB, 0);
+ channel_writel(atchan, DSCR, first->txd.phys);
+ dma_writel(atdma, CHER, atchan->mask);
+
+ vdbg_dump_regs(atchan);
+}
+
+/**
+ * atc_chain_complete - finish work for one transaction chain
+ * @atchan: channel we work on
+ * @desc: descriptor at the head of the chain we want do complete
+ *
+ * Called with atchan->lock held and bh disabled */
+static void
+atc_chain_complete(struct at_dma_chan *atchan, struct at_desc *desc)
+{
+ dma_async_tx_callback callback;
+ void *param;
+ struct dma_async_tx_descriptor *txd = &desc->txd;
+
+ dev_vdbg(chan2dev(&atchan->chan_common),
+ "descriptor %u complete\n", txd->cookie);
+
+ atchan->completed_cookie = txd->cookie;
+ callback = txd->callback;
+ param = txd->callback_param;
+
+ /* move children to free_list */
+ list_splice_init(&txd->tx_list, &atchan->free_list);
+ /* move myself to free_list */
+ list_move(&desc->desc_node, &atchan->free_list);
+
+ /*
+ * We use dma_unmap_page() regardless of how the buffers were
+ * mapped before they were submitted...
+ */
+ if (!(txd->flags & DMA_COMPL_SKIP_DEST_UNMAP))
+ dma_unmap_page(chan2parent(&atchan->chan_common), desc->lli.daddr,
+ desc->len, DMA_FROM_DEVICE);
+ if (!(txd->flags & DMA_COMPL_SKIP_SRC_UNMAP))
+ dma_unmap_page(chan2parent(&atchan->chan_common), desc->lli.saddr,
+ desc->len, DMA_TO_DEVICE);
+
+ /*
+ * The API requires that no submissions are done from a
+ * callback, so we don't need to drop the lock here
+ */
+ if (callback)
+ callback(param);
+}
+
+/**
+ * atc_complete_all - finish work for all transactions
+ * @atchan: channel to complete transactions for
+ *
+ * Eventually submit queued descriptors if any
+ *
+ * Assume channel is idle while calling this function
+ * Called with atchan->lock held and bh disabled
+ */
+static void atc_complete_all(struct at_dma_chan *atchan)
+{
+ struct at_desc *desc, *_desc;
+ LIST_HEAD(list);
+
+ dev_vdbg(chan2dev(&atchan->chan_common), "complete all\n");
+
+ BUG_ON(atc_chan_is_enabled(atchan));
+
+ /*
+ * Submit queued descriptors ASAP, i.e. before we go through
+ * the completed ones.
+ */
+ if (!list_empty(&atchan->queue))
+ atc_dostart(atchan, atc_first_queued(atchan));
+ /* empty active_list now it is completed */
+ list_splice_init(&atchan->active_list, &list);
+ /* empty queue list by moving descriptors (if any) to active_list */
+ list_splice_init(&atchan->queue, &atchan->active_list);
+
+ list_for_each_entry_safe(desc, _desc, &list, desc_node)
+ atc_chain_complete(atchan, desc);
+}
+
+/**
+ * atc_cleanup_descriptors - cleanup up finished descriptors in active_list
+ * @atchan: channel to be cleaned up
+ *
+ * Called with atchan->lock held and bh disabled
+ */
+static void atc_cleanup_descriptors(struct at_dma_chan *atchan)
+{
+ struct at_desc *desc, *_desc;
+ struct at_desc *child;
+
+ dev_vdbg(chan2dev(&atchan->chan_common), "cleanup descriptors\n");
+
+ list_for_each_entry_safe(desc, _desc, &atchan->active_list, desc_node) {
+ if (!(desc->lli.ctrla & ATC_DONE))
+ /* This one is currently in progress */
+ return;
+
+ list_for_each_entry(child, &desc->txd.tx_list, desc_node)
+ if (!(child->lli.ctrla & ATC_DONE))
+ /* Currently in progress */
+ return;
+
+ /*
+ * No descriptors so far seem to be in progress, i.e.
+ * this chain must be done.
+ */
+ atc_chain_complete(atchan, desc);
+ }
+}
+
+/**
+ * atc_advance_work - at the end of a transaction, move forward
+ * @atchan: channel where the transaction ended
+ *
+ * Called with atchan->lock held and bh disabled
+ */
+static void atc_advance_work(struct at_dma_chan *atchan)
+{
+ dev_vdbg(chan2dev(&atchan->chan_common), "advance_work\n");
+
+ if ( list_empty(&atchan->active_list)
+ || list_is_singular(&atchan->active_list)) {
+ atc_complete_all(atchan);
+ } else {
+ atc_chain_complete(atchan, atc_first_active(atchan));
+ /* advance work */
+ atc_dostart(atchan, atc_first_active(atchan));
+ }
+}
+
+
+/**
+ * atc_handle_error - handle errors reported by DMA controller
+ * @atchan: channel where error occurs
+ *
+ * Called with atchan->lock held and bh disabled
+ */
+static void atc_handle_error(struct at_dma_chan *atchan)
+{
+ struct at_desc *bad_desc;
+ struct at_desc *child;
+
+ /*
+ * The descriptor currently at the head of the active list is
+ * broked. Since we don't have any way to report errors, we'll
+ * just have to scream loudly and try to carry on.
+ */
+ bad_desc = atc_first_active(atchan);
+ list_del_init(&bad_desc->desc_node);
+
+ /* As we are stopped, take advantage to push queued descriptors
+ * in active_list */
+ list_splice_init(&atchan->queue, atchan->active_list.prev);
+
+ /* Try to restart the controller */
+ if (!list_empty(&atchan->active_list))
+ atc_dostart(atchan, atc_first_active(atchan));
+
+ /*
+ * KERN_CRITICAL may seem harsh, but since this only happens
+ * when someone submits a bad physical address in a
+ * descriptor, we should consider ourselves lucky that the
+ * controller flagged an error instead of scribbling over
+ * random memory locations.
+ */
+ dev_crit(chan2dev(&atchan->chan_common),
+ "Bad descriptor submitted for DMA!\n");
+ dev_crit(chan2dev(&atchan->chan_common),
+ " cookie: %d\n", bad_desc->txd.cookie);
+ atc_dump_lli(atchan, &bad_desc->lli);
+ list_for_each_entry(child, &bad_desc->txd.tx_list, desc_node)
+ atc_dump_lli(atchan, &child->lli);
+
+ /* Pretend the descriptor completed successfully */
+ atc_chain_complete(atchan, bad_desc);
+}
+
+
+/*-- IRQ & Tasklet ---------------------------------------------------*/
+
+static void atc_tasklet(unsigned long data)
+{
+ struct at_dma_chan *atchan = (struct at_dma_chan *)data;
+
+ /* Channel cannot be enabled here */
+ if (atc_chan_is_enabled(atchan)) {
+ dev_err(chan2dev(&atchan->chan_common),
+ "BUG: channel enabled in tasklet\n");
+ return;
+ }
+
+ spin_lock(&atchan->lock);
+ if (test_and_clear_bit(0, &atchan->error_status))
+ atc_handle_error(atchan);
+ else
+ atc_advance_work(atchan);
+
+ spin_unlock(&atchan->lock);
+}
+
+static irqreturn_t at_dma_interrupt(int irq, void *dev_id)
+{
+ struct at_dma *atdma = (struct at_dma *)dev_id;
+ struct at_dma_chan *atchan;
+ int i;
+ u32 status, pending, imr;
+ int ret = IRQ_NONE;
+
+ do {
+ imr = dma_readl(atdma, EBCIMR);
+ status = dma_readl(atdma, EBCISR);
+ pending = status & imr;
+
+ if (!pending)
+ break;
+
+ dev_vdbg(atdma->dma_common.dev,
+ "interrupt: status = 0x%08x, 0x%08x, 0x%08x\n",
+ status, imr, pending);
+
+ for (i = 0; i < atdma->dma_common.chancnt; i++) {
+ atchan = &atdma->chan[i];
+ if (pending & (AT_DMA_CBTC(i) | AT_DMA_ERR(i))) {
+ if (pending & AT_DMA_ERR(i)) {
+ /* Disable channel on AHB error */
+ dma_writel(atdma, CHDR, atchan->mask);
+ /* Give information to tasklet */
+ set_bit(0, &atchan->error_status);
+ }
+ tasklet_schedule(&atchan->tasklet);
+ ret = IRQ_HANDLED;
+ }
+ }
+
+ } while (pending);
+
+ return ret;
+}
+
+
+/*-- DMA Engine API --------------------------------------------------*/
+
+/**
+ * atc_tx_submit - set the prepared descriptor(s) to be executed by the engine
+ * @desc: descriptor at the head of the transaction chain
+ *
+ * Queue chain if DMA engine is working already
+ *
+ * Cookie increment and adding to active_list or queue must be atomic
+ */
+static dma_cookie_t atc_tx_submit(struct dma_async_tx_descriptor *tx)
+{
+ struct at_desc *desc = txd_to_at_desc(tx);
+ struct at_dma_chan *atchan = to_at_dma_chan(tx->chan);
+ dma_cookie_t cookie;
+
+ spin_lock_bh(&atchan->lock);
+ cookie = atc_assign_cookie(atchan, desc);
+
+ if (list_empty(&atchan->active_list)) {
+ dev_vdbg(chan2dev(tx->chan), "tx_submit: started %u\n",
+ desc->txd.cookie);
+ atc_dostart(atchan, desc);
+ list_add_tail(&desc->desc_node, &atchan->active_list);
+ } else {
+ dev_vdbg(chan2dev(tx->chan), "tx_submit: queued %u\n",
+ desc->txd.cookie);
+ list_add_tail(&desc->desc_node, &atchan->queue);
+ }
+
+ spin_unlock_bh(&atchan->lock);
+
+ return cookie;
+}
+
+/**
+ * atc_prep_dma_memcpy - prepare a memcpy operation
+ * @chan: the channel to prepare operation on
+ * @dest: operation virtual destination address
+ * @src: operation virtual source address
+ * @len: operation length
+ * @flags: tx descriptor status flags
+ */
+static struct dma_async_tx_descriptor *
+atc_prep_dma_memcpy(struct dma_chan *chan, dma_addr_t dest, dma_addr_t src,
+ size_t len, unsigned long flags)
+{
+ struct at_dma_chan *atchan = to_at_dma_chan(chan);
+ struct at_desc *desc = NULL;
+ struct at_desc *first = NULL;
+ struct at_desc *prev = NULL;
+ size_t xfer_count;
+ size_t offset;
+ unsigned int src_width;
+ unsigned int dst_width;
+ u32 ctrla;
+ u32 ctrlb;
+
+ dev_vdbg(chan2dev(chan), "prep_dma_memcpy: d0x%x s0x%x l0x%zx f0x%lx\n",
+ dest, src, len, flags);
+
+ if (unlikely(!len)) {
+ dev_dbg(chan2dev(chan), "prep_dma_memcpy: length is zero!\n");
+ return NULL;
+ }
+
+ ctrla = ATC_DEFAULT_CTRLA;
+ ctrlb = ATC_DEFAULT_CTRLB
+ | ATC_SRC_ADDR_MODE_INCR
+ | ATC_DST_ADDR_MODE_INCR
+ | ATC_FC_MEM2MEM;
+
+ /*
+ * We can be a lot more clever here, but this should take care
+ * of the most common optimization.
+ */
+ if (!((src | dest | len) & 3)) {
+ ctrla |= ATC_SRC_WIDTH_WORD | ATC_DST_WIDTH_WORD;
+ src_width = dst_width = 2;
+ } else if (!((src | dest | len) & 1)) {
+ ctrla |= ATC_SRC_WIDTH_HALFWORD | ATC_DST_WIDTH_HALFWORD;
+ src_width = dst_width = 1;
+ } else {
+ ctrla |= ATC_SRC_WIDTH_BYTE | ATC_DST_WIDTH_BYTE;
+ src_width = dst_width = 0;
+ }
+
+ for (offset = 0; offset < len; offset += xfer_count << src_width) {
+ xfer_count = min_t(size_t, (len - offset) >> src_width,
+ ATC_BTSIZE_MAX);
+
+ desc = atc_desc_get(atchan);
+ if (!desc)
+ goto err_desc_get;
+
+ desc->lli.saddr = src + offset;
+ desc->lli.daddr = dest + offset;
+ desc->lli.ctrla = ctrla | xfer_count;
+ desc->lli.ctrlb = ctrlb;
+
+ desc->txd.cookie = 0;
+ async_tx_ack(&desc->txd);
+
+ if (!first) {
+ first = desc;
+ } else {
+ /* inform the HW lli about chaining */
+ prev->lli.dscr = desc->txd.phys;
+ /* insert the link descriptor to the LD ring */
+ list_add_tail(&desc->desc_node,
+ &first->txd.tx_list);
+ }
+ prev = desc;
+ }
+
+ /* First descriptor of the chain embedds additional information */
+ first->txd.cookie = -EBUSY;
+ first->len = len;
+
+ /* set end-of-link to the last link descriptor of list*/
+ set_desc_eol(desc);
+
+ desc->txd.flags = flags; /* client is in control of this ack */
+
+ return &first->txd;
+
+err_desc_get:
+ atc_desc_put(atchan, first);
+ return NULL;
+}
+
+/**
+ * atc_is_tx_complete - poll for transaction completion
+ * @chan: DMA channel
+ * @cookie: transaction identifier to check status of
+ * @done: if not %NULL, updated with last completed transaction
+ * @used: if not %NULL, updated with last used transaction
+ *
+ * If @done and @used are passed in, upon return they reflect the driver
+ * internal state and can be used with dma_async_is_complete() to check
+ * the status of multiple cookies without re-checking hardware state.
+ */
+static enum dma_status
+atc_is_tx_complete(struct dma_chan *chan,
+ dma_cookie_t cookie,
+ dma_cookie_t *done, dma_cookie_t *used)
+{
+ struct at_dma_chan *atchan = to_at_dma_chan(chan);
+ dma_cookie_t last_used;
+ dma_cookie_t last_complete;
+ enum dma_status ret;
+
+ dev_vdbg(chan2dev(chan), "is_tx_complete: %d (d%d, u%d)\n",
+ cookie, done ? *done : 0, used ? *used : 0);
+
+ spin_lock_bh(atchan->lock);
+
+ last_complete = atchan->completed_cookie;
+ last_used = chan->cookie;
+
+ ret = dma_async_is_complete(cookie, last_complete, last_used);
+ if (ret != DMA_SUCCESS) {
+ atc_cleanup_descriptors(atchan);
+
+ last_complete = atchan->completed_cookie;
+ last_used = chan->cookie;
+
+ ret = dma_async_is_complete(cookie, last_complete, last_used);
+ }
+
+ spin_unlock_bh(atchan->lock);
+
+ if (done)
+ *done = last_complete;
+ if (used)
+ *used = last_used;
+
+ return ret;
+}
+
+/**
+ * atc_issue_pending - try to finish work
+ * @chan: target DMA channel
+ */
+static void atc_issue_pending(struct dma_chan *chan)
+{
+ struct at_dma_chan *atchan = to_at_dma_chan(chan);
+
+ dev_vdbg(chan2dev(chan), "issue_pending\n");
+
+ if (!atc_chan_is_enabled(atchan)) {
+ spin_lock_bh(&atchan->lock);
+ atc_advance_work(atchan);
+ spin_unlock_bh(&atchan->lock);
+ }
+}
+
+/**
+ * atc_alloc_chan_resources - allocate resources for DMA channel
+ * @chan: allocate descriptor resources for this channel
+ * @client: current client requesting the channel be ready for requests
+ *
+ * return - the number of allocated descriptors
+ */
+static int atc_alloc_chan_resources(struct dma_chan *chan)
+{
+ struct at_dma_chan *atchan = to_at_dma_chan(chan);
+ struct at_dma *atdma = to_at_dma(chan->device);
+ struct at_desc *desc;
+ int i;
+ LIST_HEAD(tmp_list);
+
+ dev_vdbg(chan2dev(chan), "alloc_chan_resources\n");
+
+ /* ASSERT: channel is idle */
+ if (atc_chan_is_enabled(atchan)) {
+ dev_dbg(chan2dev(chan), "DMA channel not idle ?\n");
+ return -EIO;
+ }
+
+ /* have we already been set up? */
+ if (!list_empty(&atchan->free_list))
+ return atchan->descs_allocated;
+
+ /* Allocate initial pool of descriptors */
+ for (i = 0; i < INIT_NR_DESCS_PER_CHANNEL; i++) {
+ desc = atc_alloc_descriptor(chan, GFP_KERNEL);
+ if (!desc) {
+ dev_err(atdma->dma_common.dev,
+ "Only %d initial descriptors\n", i);
+ break;
+ }
+ list_add_tail(&desc->desc_node, &tmp_list);
+ }
+
+ spin_lock_bh(&atchan->lock);
+ atchan->descs_allocated = i;
+ list_splice(&tmp_list, &atchan->free_list);
+ atchan->completed_cookie = chan->cookie = 1;
+ spin_unlock_bh(&atchan->lock);
+
+ /* channel parameters */
+ channel_writel(atchan, CFG, ATC_DEFAULT_CFG);
+
+ dev_dbg(chan2dev(chan),
+ "alloc_chan_resources: allocated %d descriptors\n",
+ atchan->descs_allocated);
+
+ return atchan->descs_allocated;
+}
+
+/**
+ * atc_free_chan_resources - free all channel resources
+ * @chan: DMA channel
+ */
+static void atc_free_chan_resources(struct dma_chan *chan)
+{
+ struct at_dma_chan *atchan = to_at_dma_chan(chan);
+ struct at_dma *atdma = to_at_dma(chan->device);
+ struct at_desc *desc, *_desc;
+ LIST_HEAD(list);
+
+ dev_dbg(chan2dev(chan), "free_chan_resources: (descs allocated=%u)\n",
+ atchan->descs_allocated);
+
+ /* ASSERT: channel is idle */
+ BUG_ON(!list_empty(&atchan->active_list));
+ BUG_ON(!list_empty(&atchan->queue));
+ BUG_ON(atc_chan_is_enabled(atchan));
+
+ list_for_each_entry_safe(desc, _desc, &atchan->free_list, desc_node) {
+ dev_vdbg(chan2dev(chan), " freeing descriptor %p\n", desc);
+ list_del(&desc->desc_node);
+ /* free link descriptor */
+ dma_pool_free(atdma->dma_desc_pool, desc, desc->txd.phys);
+ }
+ list_splice_init(&atchan->free_list, &list);
+ atchan->descs_allocated = 0;
+
+ dev_vdbg(chan2dev(chan), "free_chan_resources: done\n");
+}
+
+
+/*-- Module Management -----------------------------------------------*/
+
+/**
+ * at_dma_off - disable DMA controller
+ * @atdma: the Atmel HDAMC device
+ */
+static void at_dma_off(struct at_dma *atdma)
+{
+ dma_writel(atdma, EN, 0);
+
+ /* disable all interrupts */
+ dma_writel(atdma, EBCIDR, -1L);
+
+ /* confirm that all channels are disabled */
+ while (dma_readl(atdma, CHSR) & atdma->all_chan_mask)
+ cpu_relax();
+}
+
+static int __init at_dma_probe(struct platform_device *pdev)
+{
+ struct at_dma_platform_data *pdata;
+ struct resource *io;
+ struct at_dma *atdma;
+ size_t size;
+ int irq;
+ int err;
+ int i;
+
+ /* get DMA Controller parameters from platform */
+ pdata = pdev->dev.platform_data;
+ if (!pdata || pdata->nr_channels > AT_DMA_MAX_NR_CHANNELS)
+ return -EINVAL;
+
+ io = platform_get_resource(pdev, IORESOURCE_MEM, 0);
+ if (!io)
+ return -EINVAL;
+
+ irq = platform_get_irq(pdev, 0);
+ if (irq < 0)
+ return irq;
+
+ size = sizeof(struct at_dma);
+ size += pdata->nr_channels * sizeof(struct at_dma_chan);
+ atdma = kzalloc(size, GFP_KERNEL);
+ if (!atdma)
+ return -ENOMEM;
+
+ /* discover transaction capabilites from the platform data */
+ atdma->dma_common.cap_mask = pdata->cap_mask;
+ atdma->all_chan_mask = (1 << pdata->nr_channels) - 1;
+
+ size = io->end - io->start + 1;
+ if (!request_mem_region(io->start, size, pdev->dev.driver->name)) {
+ err = -EBUSY;
+ goto err_kfree;
+ }
+
+ atdma->regs = ioremap(io->start, size);
+ if (!atdma->regs) {
+ err = -ENOMEM;
+ goto err_release_r;
+ }
+
+ atdma->clk = clk_get(&pdev->dev, "dma_clk");
+ if (IS_ERR(atdma->clk)) {
+ err = PTR_ERR(atdma->clk);
+ goto err_clk;
+ }
+ clk_enable(atdma->clk);
+
+ /* force dma off, just in case */
+ at_dma_off(atdma);
+
+ err = request_irq(irq, at_dma_interrupt, 0, "at_hdmac", atdma);
+ if (err)
+ goto err_irq;
+
+ platform_set_drvdata(pdev, atdma);
+
+ /* create a pool of consistent memory blocks for hardware descriptors */
+ atdma->dma_desc_pool = dma_pool_create("at_hdmac_desc_pool",
+ &pdev->dev, sizeof(struct at_desc),
+ 4 /* word alignment */, 0);
+ if (!atdma->dma_desc_pool) {
+ dev_err(&pdev->dev, "No memory for descriptors dma pool\n");
+ err = -ENOMEM;
+ goto err_pool_create;
+ }
+
+ /* clear any pending interrupt */
+ while (dma_readl(atdma, EBCISR))
+ cpu_relax();
+
+ /* initialize channels related values */
+ INIT_LIST_HEAD(&atdma->dma_common.channels);
+ for (i = 0; i < pdata->nr_channels; i++, atdma->dma_common.chancnt++) {
+ struct at_dma_chan *atchan = &atdma->chan[i];
+
+ atchan->chan_common.device = &atdma->dma_common;
+ atchan->chan_common.cookie = atchan->completed_cookie = 1;
+ atchan->chan_common.chan_id = i;
+ list_add_tail(&atchan->chan_common.device_node,
+ &atdma->dma_common.channels);
+
+ atchan->ch_regs = atdma->regs + ch_regs(i);
+ spin_lock_init(&atchan->lock);
+ atchan->mask = 1 << i;
+
+ INIT_LIST_HEAD(&atchan->active_list);
+ INIT_LIST_HEAD(&atchan->queue);
+ INIT_LIST_HEAD(&atchan->free_list);
+
+ tasklet_init(&atchan->tasklet, atc_tasklet,
+ (unsigned long)atchan);
+ atc_enable_irq(atchan);
+ }
+
+ /* set base routines */
+ atdma->dma_common.device_alloc_chan_resources = atc_alloc_chan_resources;
+ atdma->dma_common.device_free_chan_resources = atc_free_chan_resources;
+ atdma->dma_common.device_is_tx_complete = atc_is_tx_complete;
+ atdma->dma_common.device_issue_pending = atc_issue_pending;
+ atdma->dma_common.dev = &pdev->dev;
+
+ /* set prep routines based on capability */
+ if (dma_has_cap(DMA_MEMCPY, atdma->dma_common.cap_mask))
+ atdma->dma_common.device_prep_dma_memcpy = atc_prep_dma_memcpy;
+
+ dma_writel(atdma, EN, AT_DMA_ENABLE);
+
+ dev_info(&pdev->dev, "Atmel AHB DMA Controller ( %s%s), %d channels\n",
+ dma_has_cap(DMA_MEMCPY, atdma->dma_common.cap_mask) ? "cpy " : "",
+ dma_has_cap(DMA_SLAVE, atdma->dma_common.cap_mask) ? "slave " : "",
+ atdma->dma_common.chancnt);
+
+ dma_async_device_register(&atdma->dma_common);
+
+ return 0;
+
+err_pool_create:
+ platform_set_drvdata(pdev, NULL);
+ free_irq(platform_get_irq(pdev, 0), atdma);
+err_irq:
+ clk_disable(atdma->clk);
+ clk_put(atdma->clk);
+err_clk:
+ iounmap(atdma->regs);
+ atdma->regs = NULL;
+err_release_r:
+ release_mem_region(io->start, size);
+err_kfree:
+ kfree(atdma);
+ return err;
+}
+
+static int __exit at_dma_remove(struct platform_device *pdev)
+{
+ struct at_dma *atdma = platform_get_drvdata(pdev);
+ struct dma_chan *chan, *_chan;
+ struct resource *io;
+
+ at_dma_off(atdma);
+ dma_async_device_unregister(&atdma->dma_common);
+
+ dma_pool_destroy(atdma->dma_desc_pool);
+ platform_set_drvdata(pdev, NULL);
+ free_irq(platform_get_irq(pdev, 0), atdma);
+
+ list_for_each_entry_safe(chan, _chan, &atdma->dma_common.channels,
+ device_node) {
+ struct at_dma_chan *atchan = to_at_dma_chan(chan);
+
+ /* Disable interrupts */
+ atc_disable_irq(atchan);
+ tasklet_disable(&atchan->tasklet);
+
+ tasklet_kill(&atchan->tasklet);
+ list_del(&chan->device_node);
+ }
+
+ clk_disable(atdma->clk);
+ clk_put(atdma->clk);
+
+ iounmap(atdma->regs);
+ atdma->regs = NULL;
+
+ io = platform_get_resource(pdev, IORESOURCE_MEM, 0);
+ release_mem_region(io->start, io->end - io->start + 1);
+
+ kfree(atdma);
+
+ return 0;
+}
+
+static void at_dma_shutdown(struct platform_device *pdev)
+{
+ struct at_dma *atdma = platform_get_drvdata(pdev);
+
+ at_dma_off(platform_get_drvdata(pdev));
+ clk_disable(atdma->clk);
+}
+
+static int at_dma_suspend_late(struct platform_device *pdev, pm_message_t mesg)
+{
+ struct at_dma *atdma = platform_get_drvdata(pdev);
+
+ at_dma_off(platform_get_drvdata(pdev));
+ clk_disable(atdma->clk);
+ return 0;
+}
+
+static int at_dma_resume_early(struct platform_device *pdev)
+{
+ struct at_dma *atdma = platform_get_drvdata(pdev);
+
+ clk_enable(atdma->clk);
+ dma_writel(atdma, EN, AT_DMA_ENABLE);
+ return 0;
+
+}
+
+static struct platform_driver at_dma_driver = {
+ .remove = __exit_p(at_dma_remove),
+ .shutdown = at_dma_shutdown,
+ .suspend_late = at_dma_suspend_late,
+ .resume_early = at_dma_resume_early,
+ .driver = {
+ .name = "at_hdmac",
+ },
+};
+
+static int __init at_dma_init(void)
+{
+ return platform_driver_probe(&at_dma_driver, at_dma_probe);
+}
+module_init(at_dma_init);
+
+static void __exit at_dma_exit(void)
+{
+ platform_driver_unregister(&at_dma_driver);
+}
+module_exit(at_dma_exit);
+
+MODULE_DESCRIPTION("Atmel AHB DMA Controller driver");
+MODULE_AUTHOR("Nicolas Ferre <[email protected]>");
+MODULE_LICENSE("GPL");
+MODULE_ALIAS("platform:at_hdmac");
diff --git a/drivers/dma/at_hdmac_regs.h b/drivers/dma/at_hdmac_regs.h
new file mode 100644
index 0000000..ad2d4f4
--- /dev/null
+++ b/drivers/dma/at_hdmac_regs.h
@@ -0,0 +1,386 @@
+/*
+ * Header file for the Atmel AHB DMA Controller driver
+ *
+ * Copyright (C) 2008 Atmel Corporation
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ */
+#ifndef AT_HDMAC_REGS_H
+#define AT_HDMAC_REGS_H
+
+#include <mach/at_hdmac.h>
+
+#define AT_DMA_MAX_NR_CHANNELS 8
+
+
+#define AT_DMA_GCFG 0x00 /* Global Configuration Register */
+#define AT_DMA_IF_BIGEND(i) (0x1 << (i)) /* AHB-Lite Interface i in Big-endian mode */
+#define AT_DMA_ARB_CFG (0x1 << 4) /* Arbiter mode. */
+#define AT_DMA_ARB_CFG_FIXED (0x0 << 4)
+#define AT_DMA_ARB_CFG_ROUND_ROBIN (0x1 << 4)
+
+#define AT_DMA_EN 0x04 /* Controller Enable Register */
+#define AT_DMA_ENABLE (0x1 << 0)
+
+#define AT_DMA_SREQ 0x08 /* Software Single Request Register */
+#define AT_DMA_SSREQ(x) (0x1 << ((x) << 1)) /* Request a source single transfer on channel x */
+#define AT_DMA_DSREQ(x) (0x1 << (1 + ((x) << 1))) /* Request a destination single transfer on channel x */
+
+#define AT_DMA_CREQ 0x0C /* Software Chunk Transfer Request Register */
+#define AT_DMA_SCREQ(x) (0x1 << ((x) << 1)) /* Request a source chunk transfer on channel x */
+#define AT_DMA_DCREQ(x) (0x1 << (1 + ((x) << 1))) /* Request a destination chunk transfer on channel x */
+
+#define AT_DMA_LAST 0x10 /* Software Last Transfer Flag Register */
+#define AT_DMA_SLAST(x) (0x1 << ((x) << 1)) /* This src rq is last tx of buffer on channel x */
+#define AT_DMA_DLAST(x) (0x1 << (1 + ((x) << 1))) /* This dst rq is last tx of buffer on channel x */
+
+#define AT_DMA_SYNC 0x14 /* Request Synchronization Register */
+#define AT_DMA_SYR(h) (0x1 << (h)) /* Synchronize handshake line h */
+
+/* Error, Chained Buffer transfer completed and Buffer transfer completed Interrupt registers */
+#define AT_DMA_EBCIER 0x18 /* Enable register */
+#define AT_DMA_EBCIDR 0x1C /* Disable register */
+#define AT_DMA_EBCIMR 0x20 /* Mask Register */
+#define AT_DMA_EBCISR 0x24 /* Status Register */
+#define AT_DMA_CBTC_OFFSET 8
+#define AT_DMA_ERR_OFFSET 16
+#define AT_DMA_BTC(x) (0x1 << (x))
+#define AT_DMA_CBTC(x) (0x1 << (AT_DMA_CBTC_OFFSET + (x)))
+#define AT_DMA_ERR(x) (0x1 << (AT_DMA_ERR_OFFSET + (x)))
+
+#define AT_DMA_CHER 0x28 /* Channel Handler Enable Register */
+#define AT_DMA_ENA(x) (0x1 << (x))
+#define AT_DMA_SUSP(x) (0x1 << ( 8 + (x)))
+#define AT_DMA_KEEP(x) (0x1 << (24 + (x)))
+
+#define AT_DMA_CHDR 0x2C /* Channel Handler Disable Register */
+#define AT_DMA_DIS(x) (0x1 << (x))
+#define AT_DMA_RES(x) (0x1 << ( 8 + (x)))
+
+#define AT_DMA_CHSR 0x30 /* Channel Handler Status Register */
+#define AT_DMA_EMPT(x) (0x1 << (16 + (x)))
+#define AT_DMA_STAL(x) (0x1 << (24 + (x)))
+
+
+#define AT_DMA_CH_REGS_BASE 0x3C /* Channel registers base address */
+#define ch_regs(x) (AT_DMA_CH_REGS_BASE + (x) * 0x28) /* Channel x base addr */
+
+/* Hardware register offset for each channel */
+#define ATC_SADDR_OFFSET 0x00 /* Source Address Register */
+#define ATC_DADDR_OFFSET 0x04 /* Destination Address Register */
+#define ATC_DSCR_OFFSET 0x08 /* Descriptor Address Register */
+#define ATC_CTRLA_OFFSET 0x0C /* Control A Register */
+#define ATC_CTRLB_OFFSET 0x10 /* Control B Register */
+#define ATC_CFG_OFFSET 0x14 /* Configuration Register */
+#define ATC_SPIP_OFFSET 0x18 /* Src PIP Configuration Register */
+#define ATC_DPIP_OFFSET 0x1C /* Dst PIP Configuration Register */
+
+
+/* Bitfield definitions */
+
+/* Bitfields in DSCR */
+#define ATC_DSCR_IF(i) (0x3 & (i)) /* Dsc feched via AHB-Lite Interface i */
+
+/* Bitfields in CTRLA */
+#define ATC_BTSIZE_MAX 0xFFFFUL /* Maximum Buffer Transfer Size */
+#define ATC_BTSIZE(x) (ATC_BTSIZE_MAX & (x)) /* Buffer Transfer Size */
+#define ATC_SCSIZE_MASK (0x7 << 16) /* Source Chunk Transfer Size */
+#define ATC_SCSIZE_1 (0x0 << 16)
+#define ATC_SCSIZE_4 (0x1 << 16)
+#define ATC_SCSIZE_8 (0x2 << 16)
+#define ATC_SCSIZE_16 (0x3 << 16)
+#define ATC_SCSIZE_32 (0x4 << 16)
+#define ATC_SCSIZE_64 (0x5 << 16)
+#define ATC_SCSIZE_128 (0x6 << 16)
+#define ATC_SCSIZE_256 (0x7 << 16)
+#define ATC_DCSIZE_MASK (0x7 << 20) /* Destination Chunk Transfer Size */
+#define ATC_DCSIZE_1 (0x0 << 20)
+#define ATC_DCSIZE_4 (0x1 << 20)
+#define ATC_DCSIZE_8 (0x2 << 20)
+#define ATC_DCSIZE_16 (0x3 << 20)
+#define ATC_DCSIZE_32 (0x4 << 20)
+#define ATC_DCSIZE_64 (0x5 << 20)
+#define ATC_DCSIZE_128 (0x6 << 20)
+#define ATC_DCSIZE_256 (0x7 << 20)
+#define ATC_SRC_WIDTH_MASK (0x3 << 24) /* Source Single Transfer Size */
+#define ATC_SRC_WIDTH_BYTE (0x0 << 24)
+#define ATC_SRC_WIDTH_HALFWORD (0x1 << 24)
+#define ATC_SRC_WIDTH_WORD (0x2 << 24)
+#define ATC_DST_WIDTH_MASK (0x3 << 28) /* Destination Single Transfer Size */
+#define ATC_DST_WIDTH_BYTE (0x0 << 28)
+#define ATC_DST_WIDTH_HALFWORD (0x1 << 28)
+#define ATC_DST_WIDTH_WORD (0x2 << 28)
+#define ATC_DONE (0x1 << 31) /* Tx Done (only written back in descriptor) */
+
+/* Bitfields in CTRLB */
+#define ATC_SIF(i) (0x3 & (i)) /* Src tx done via AHB-Lite Interface i */
+#define ATC_DIF(i) ((0x3 & (i)) << 4) /* Dst tx done via AHB-Lite Interface i */
+#define ATC_SRC_PIP (0x1 << 8) /* Source Picture-in-Picture enabled */
+#define ATC_DST_PIP (0x1 << 12) /* Destination Picture-in-Picture enabled */
+#define ATC_SRC_DSCR_DIS (0x1 << 16) /* Src Descriptor fetch disable */
+#define ATC_DST_DSCR_DIS (0x1 << 20) /* Dst Descriptor fetch disable */
+#define ATC_FC_MASK (0x7 << 21) /* Choose Flow Controller */
+#define ATC_FC_MEM2MEM (0x0 << 21) /* Mem-to-Mem (DMA) */
+#define ATC_FC_MEM2PER (0x1 << 21) /* Mem-to-Periph (DMA) */
+#define ATC_FC_PER2MEM (0x2 << 21) /* Periph-to-Mem (DMA) */
+#define ATC_FC_PER2PER (0x3 << 21) /* Periph-to-Periph (DMA) */
+#define ATC_FC_PER2MEM_PER (0x4 << 21) /* Periph-to-Mem (Peripheral) */
+#define ATC_FC_MEM2PER_PER (0x5 << 21) /* Mem-to-Periph (Peripheral) */
+#define ATC_FC_PER2PER_PER (0x6 << 21) /* Periph-to-Periph (Src Peripheral) */
+#define ATC_SRC_ADDR_MODE_MASK (0x3 << 24)
+#define ATC_SRC_ADDR_MODE_INCR (0x0 << 24) /* Incrementing Mode */
+#define ATC_SRC_ADDR_MODE_DECR (0x1 << 24) /* Decrementing Mode */
+#define ATC_SRC_ADDR_MODE_FIXED (0x2 << 24) /* Fixed Mode */
+#define ATC_DST_ADDR_MODE_MASK (0x3 << 28)
+#define ATC_DST_ADDR_MODE_INCR (0x0 << 28) /* Incrementing Mode */
+#define ATC_DST_ADDR_MODE_DECR (0x1 << 28) /* Decrementing Mode */
+#define ATC_DST_ADDR_MODE_FIXED (0x2 << 28) /* Fixed Mode */
+#define ATC_IEN (0x1 << 30) /* BTC interrupt enable (active low) */
+#define ATC_AUTO (0x1 << 31) /* Auto multiple buffer tx enable */
+
+/* Bitfields in CFG */
+#define ATC_SRC_PER(h) (0xFU & (h)) /* Channel src rq associated with periph handshaking ifc h */
+#define ATC_DST_PER(h) ((0xFU & (h)) << 4) /* Channel dst rq associated with periph handshaking ifc h */
+#define ATC_SRC_REP (0x1 << 8) /* Source Replay Mod */
+#define ATC_SRC_H2SEL (0x1 << 9) /* Source Handshaking Mod */
+#define ATC_SRC_H2SEL_SW (0x0 << 9)
+#define ATC_SRC_H2SEL_HW (0x1 << 9)
+#define ATC_DST_REP (0x1 << 12) /* Destination Replay Mod */
+#define ATC_DST_H2SEL (0x1 << 13) /* Destination Handshaking Mod */
+#define ATC_DST_H2SEL_SW (0x0 << 13)
+#define ATC_DST_H2SEL_HW (0x1 << 13)
+#define ATC_SOD (0x1 << 16) /* Stop On Done */
+#define ATC_LOCK_IF (0x1 << 20) /* Interface Lock */
+#define ATC_LOCK_B (0x1 << 21) /* AHB Bus Lock */
+#define ATC_LOCK_IF_L (0x1 << 22) /* Master Interface Arbiter Lock */
+#define ATC_LOCK_IF_L_CHUNK (0x0 << 22)
+#define ATC_LOCK_IF_L_BUFFER (0x1 << 22)
+#define ATC_AHB_PROT_MASK (0x7 << 24) /* AHB Protection */
+#define ATC_FIFOCFG_MASK (0x3 << 28) /* FIFO Request Configuration */
+#define ATC_FIFOCFG_LARGESTBURST (0x0 << 28)
+#define ATC_FIFOCFG_HALFFIFO (0x1 << 28)
+#define ATC_FIFOCFG_ENOUGHSPACE (0x2 << 28)
+
+/* Bitfields in SPIP */
+#define ATC_SPIP_HOLE(x) (0xFFFFU & (x))
+#define ATC_SPIP_BOUNDARY(x) ((0x3FF & (x)) << 16)
+
+/* Bitfields in DPIP */
+#define ATC_DPIP_HOLE(x) (0xFFFFU & (x))
+#define ATC_DPIP_BOUNDARY(x) ((0x3FF & (x)) << 16)
+
+
+/*-- descriptors -----------------------------------------------------*/
+
+/* LLI == Linked List Item; aka DMA buffer descriptor */
+struct at_lli {
+ /* values that are not changed by hardware */
+ dma_addr_t saddr;
+ dma_addr_t daddr;
+ /* value that may get written back: */
+ u32 ctrla;
+ /* more values that are not changed by hardware */
+ u32 ctrlb;
+ dma_addr_t dscr; /* chain to next lli */
+};
+
+/**
+ * struct at_desc - software descriptor
+ * @at_lli: hardware lli structure
+ * @txd: support for the async_tx api
+ * @desc_node: node on the channed descriptors list
+ * @len: total transaction bytecount
+ */
+struct at_desc {
+ /* FIRST values the hardware uses */
+ struct at_lli lli;
+
+ /* THEN values for driver housekeeping */
+ struct dma_async_tx_descriptor txd;
+ struct list_head desc_node;
+ size_t len;
+};
+
+static inline struct at_desc *
+txd_to_at_desc(struct dma_async_tx_descriptor *txd)
+{
+ return container_of(txd, struct at_desc, txd);
+}
+
+
+/*-- Channels --------------------------------------------------------*/
+
+/**
+ * struct at_dma_chan - internal representation of an Atmel HDMAC channel
+ * @chan_common: common dmaengine channel object members
+ * @device: parent device
+ * @ch_regs: memory mapped register base
+ * @mask: channel index in a mask
+ * @error_status: transmit error status information from irq handler
+ * to tasklet (use atomic operations)
+ * @tasklet: bottom half to finish transaction work
+ * @lock: serializes enqueue/dequeue operations to descriptors lists
+ * @completed_cookie: identifier for the most recently completed operation
+ * @active_list: list of descriptors dmaengine is being running on
+ * @queue: list of descriptors ready to be submitted to engine
+ * @free_list: list of descriptors usable by the channel
+ * @descs_allocated: records the actual size of the descriptor pool
+ */
+struct at_dma_chan {
+ struct dma_chan chan_common;
+ struct at_dma *device;
+ void __iomem *ch_regs;
+ u8 mask;
+ unsigned long error_status;
+ struct tasklet_struct tasklet;
+
+ spinlock_t lock;
+
+ /* these other elements are all protected by lock */
+ dma_cookie_t completed_cookie;
+ struct list_head active_list;
+ struct list_head queue;
+ struct list_head free_list;
+ unsigned int descs_allocated;
+};
+
+#define channel_readl(atchan, name) \
+ __raw_readl((atchan)->ch_regs + ATC_##name##_OFFSET)
+
+#define channel_writel(atchan, name, val) \
+ __raw_writel((val), (atchan)->ch_regs + ATC_##name##_OFFSET)
+
+static inline struct at_dma_chan *to_at_dma_chan(struct dma_chan *dchan)
+{
+ return container_of(dchan, struct at_dma_chan, chan_common);
+}
+
+
+/*-- Controller ------------------------------------------------------*/
+
+/**
+ * struct at_dma - internal representation of an Atmel HDMA Controller
+ * @chan_common: common dmaengine dma_device object members
+ * @ch_regs: memory mapped register base
+ * @clk: dma controller clock
+ * @all_chan_mask: all channels availlable in a mask
+ * @dma_desc_pool: base of DMA descriptor region (DMA address)
+ * @chan: channels table to store at_dma_chan structures
+ */
+struct at_dma {
+ struct dma_device dma_common;
+ void __iomem *regs;
+ struct clk *clk;
+
+ u8 all_chan_mask;
+
+ struct dma_pool *dma_desc_pool;
+ /* AT THE END channels table */
+ struct at_dma_chan chan[0];
+};
+
+#define dma_readl(atdma, name) \
+ __raw_readl((atdma)->regs + AT_DMA_##name)
+#define dma_writel(atdma, name, val) \
+ __raw_writel((val), (atdma)->regs + AT_DMA_##name)
+
+static inline struct at_dma *to_at_dma(struct dma_device *ddev)
+{
+ return container_of(ddev, struct at_dma, dma_common);
+}
+
+
+/*-- Helper functions ------------------------------------------------*/
+
+static struct device *chan2dev(struct dma_chan *chan)
+{
+ return &chan->dev->device;
+}
+static struct device *chan2parent(struct dma_chan *chan)
+{
+ return chan->dev->device.parent;
+}
+
+#if defined(VERBOSE_DEBUG)
+static void vdbg_dump_regs(struct at_dma_chan *atchan)
+{
+ struct at_dma *atdma = to_at_dma(atchan->chan_common.device);
+
+ dev_err(chan2dev(&atchan->chan_common),
+ " channel %d : imr = 0x%x, chsr = 0x%x\n",
+ atchan->chan_common.chan_id,
+ dma_readl(atdma, EBCIMR),
+ dma_readl(atdma, CHSR));
+
+ dev_err(chan2dev(&atchan->chan_common),
+ " channel: s0x%x d0x%x ctrl0x%x:0x%x l0x%x\n",
+ channel_readl(atchan, SADDR),
+ channel_readl(atchan, DADDR),
+ channel_readl(atchan, CTRLA),
+ channel_readl(atchan, CTRLB),
+ channel_readl(atchan, DSCR));
+}
+#else
+static void vdbg_dump_regs(struct at_dma_chan *atchan) {}
+#endif
+
+static void atc_dump_lli(struct at_dma_chan *atchan, struct at_lli *lli)
+{
+ dev_printk(KERN_CRIT, chan2dev(&atchan->chan_common),
+ " desc: s0x%x d0x%x ctrl0x%x:0x%x l0x%x\n",
+ lli->saddr, lli->daddr,
+ lli->ctrla, lli->ctrlb, lli->dscr);
+}
+
+
+static void atc_setup_irq(struct at_dma_chan *atchan, int on)
+{
+ struct at_dma *atdma = to_at_dma(atchan->chan_common.device);
+ u32 ebci;
+
+ /* enable interrupts on buffer chain completion & error */
+ ebci = AT_DMA_CBTC(atchan->chan_common.chan_id)
+ | AT_DMA_ERR(atchan->chan_common.chan_id);
+ if (on)
+ dma_writel(atdma, EBCIER, ebci);
+ else
+ dma_writel(atdma, EBCIDR, ebci);
+}
+
+static inline void atc_enable_irq(struct at_dma_chan *atchan)
+{
+ atc_setup_irq(atchan, 1);
+}
+
+static inline void atc_disable_irq(struct at_dma_chan *atchan)
+{
+ atc_setup_irq(atchan, 0);
+}
+
+
+/**
+ * atc_chan_is_enabled - test if given channel is enabled
+ * @atchan: channel we want to test status
+ */
+static inline int atc_chan_is_enabled(struct at_dma_chan *atchan)
+{
+ struct at_dma *atdma = to_at_dma(atchan->chan_common.device);
+
+ return !!(dma_readl(atdma, CHSR) & atchan->mask);
+}
+
+
+/**
+ * set_desc_eol - set end-of-link to descriptor so it will end transfer
+ * @desc: descriptor, signle or at the end of a chain, to end chain on
+ */
+static void set_desc_eol(struct at_desc *desc)
+{
+ desc->lli.ctrlb |= ATC_SRC_DSCR_DIS | ATC_DST_DSCR_DIS;
+ desc->lli.dscr = 0;
+}
+
+#endif /* AT_HDMAC_REGS_H */
--
1.5.3.7


2009-06-26 09:39:53

by Nicolas Ferre

[permalink] [raw]
Subject: [PATCH 2/2 v2] at91/dmaengine: integration of at_hdmac driver in at91sam9rl

This is the integration of DMA engine driver into at91sam9rl device file. The
associated driver is at_hdmac.

Signed-off-by: Nicolas Ferre <[email protected]>
---
This patch is based on async_tx.git/next and may need modification for
inclusion in the at91 patch submission flow.

arch/arm/mach-at91/at91sam9rl_devices.c | 49 ++++++++++++++++++++++++++++++-
1 files changed, 48 insertions(+), 1 deletions(-)

diff --git a/arch/arm/mach-at91/at91sam9rl_devices.c b/arch/arm/mach-at91/at91sam9rl_devices.c
index 7281865..65a4997 100644
--- a/arch/arm/mach-at91/at91sam9rl_devices.c
+++ b/arch/arm/mach-at91/at91sam9rl_devices.c
@@ -21,11 +21,57 @@
#include <mach/at91sam9rl.h>
#include <mach/at91sam9rl_matrix.h>
#include <mach/at91sam9_smc.h>
+#include <mach/at_hdmac.h>

#include "generic.h"


/* --------------------------------------------------------------------
+ * HDMAC - AHB DMA Controller
+ * -------------------------------------------------------------------- */
+
+#if defined(CONFIG_AT_HDMAC) || defined(CONFIG_AT_HDMAC_MODULE)
+static u64 hdmac_dmamask = DMA_BIT_MASK(32);
+
+static struct at_dma_platform_data atdma_pdata = {
+ .nr_channels = 2,
+};
+
+static struct resource hdmac_resources[] = {
+ [0] = {
+ .start = AT91_BASE_SYS + AT91_DMA,
+ .end = AT91_BASE_SYS + AT91_DMA + SZ_512 - 1,
+ .flags = IORESOURCE_MEM,
+ },
+ [2] = {
+ .start = AT91SAM9RL_ID_DMA,
+ .end = AT91SAM9RL_ID_DMA,
+ .flags = IORESOURCE_IRQ,
+ },
+};
+
+static struct platform_device at_hdmac_device = {
+ .name = "at_hdmac",
+ .id = -1,
+ .dev = {
+ .dma_mask = &hdmac_dmamask,
+ .coherent_dma_mask = DMA_BIT_MASK(32),
+ .platform_data = &atdma_pdata,
+ },
+ .resource = hdmac_resources,
+ .num_resources = ARRAY_SIZE(hdmac_resources),
+};
+
+void __init at91_add_device_hdmac(void)
+{
+ dma_cap_set(DMA_MEMCPY, atdma_pdata.cap_mask);
+ platform_device_register(&at_hdmac_device);
+}
+#else
+void __init at91_add_device_hdmac(void) {}
+#endif
+
+/* --------------------------------------------------------------------
* USB HS Device (Gadget)
* -------------------------------------------------------------------- */

@@ -823,7 +869,7 @@ static struct resource dbgu_resources[] = {
};

static struct atmel_uart_data dbgu_data = {
- .use_dma_tx = 0,
+ .use_dma_tx = 1,
.use_dma_rx = 0, /* DBGU not capable of receive DMA */
.regs = (void __iomem *)(AT91_VA_BASE_SYS + AT91_DBGU),
};
@@ -1103,6 +1149,7 @@ void __init at91_add_device_serial(void) {}
*/
static int __init at91_add_standard_devices(void)
{
+ at91_add_device_hdmac();
at91_add_device_rtc();
at91_add_device_rtt();
at91_add_device_watchdog();
--
1.5.3.7

2009-06-27 15:06:33

by Atsushi Nemoto

[permalink] [raw]
Subject: Re: [PATCH 1/2 v2] dmaengine: at_hdmac: new driver for the Atmel AHB DMA Controller

On Fri, 26 Jun 2009 12:42:15 +0200, Nicolas Ferre <[email protected]> wrote:
> This AHB DMA Controller (aka HDMA or DMAC on AT91 systems) is availlable on
> at91sam9rl chip. It will be used on other products in the future.

It seems this driver was written based on dw_dmac driver. A while ago
I had some investigation of that driver.
(http://lkml.org/lkml/2008/12/29/172)

Some of issues reported at that time could be applied on your driver
too. With a quick look, the queue list management issue is a
candidate. Here is an excerpt from the thread:

--- --- ---
3. dwc->queue list management

The function dwc_tx_submit() add the descriptor to dwc->queue list if
active list was not empty. But it does not manage lli.llp list. And
all descriptors in the queue list will be moved to active list at a
time. So it seems non-first descriptors in queue list will never
processed by the hardware.

The dwc_tx_submit() should rewrite lli.llp of the last descriptor in
queue list (it it had children, the last children of it) by txd.phys
of newly queued descriptor. Or, only first elements of queue list
should be moved to active list at a time.

Is my analysis correct?
--- --- ---

And one more comment.

> + /*
> + * We use dma_unmap_page() regardless of how the buffers were
> + * mapped before they were submitted...
> + */

You can use DMA_COMPL_{SRC,DEST}_UNMAP_SINGLE since 2.6.30.

---
Atsushi Nemoto

2009-07-01 13:59:17

by Nicolas Ferre

[permalink] [raw]
Subject: Re: [PATCH 1/2 v2] dmaengine: at_hdmac: new driver for the Atmel AHB DMA Controller

Hi,

Just one question Dan (see below).

Nicolas Ferre :
> This AHB DMA Controller (aka HDMA or DMAC on AT91 systems) is availlable on
> at91sam9rl chip. It will be used on other products in the future.

[..]

> +/**
> + * atc_chain_complete - finish work for one transaction chain
> + * @atchan: channel we work on
> + * @desc: descriptor at the head of the chain we want do complete
> + *
> + * Called with atchan->lock held and bh disabled */
> +static void
> +atc_chain_complete(struct at_dma_chan *atchan, struct at_desc *desc)
> +{
> + dma_async_tx_callback callback;
> + void *param;
> + struct dma_async_tx_descriptor *txd = &desc->txd;
> +
> + dev_vdbg(chan2dev(&atchan->chan_common),
> + "descriptor %u complete\n", txd->cookie);
> +
> + atchan->completed_cookie = txd->cookie;
> + callback = txd->callback;
> + param = txd->callback_param;
> +
> + /* move children to free_list */
> + list_splice_init(&txd->tx_list, &atchan->free_list);
> + /* move myself to free_list */
> + list_move(&desc->desc_node, &atchan->free_list);
> +
> + /*
> + * We use dma_unmap_page() regardless of how the buffers were
> + * mapped before they were submitted...
> + */
> + if (!(txd->flags & DMA_COMPL_SKIP_DEST_UNMAP))
> + dma_unmap_page(chan2parent(&atchan->chan_common), desc->lli.daddr,
> + desc->len, DMA_FROM_DEVICE);
> + if (!(txd->flags & DMA_COMPL_SKIP_SRC_UNMAP))
> + dma_unmap_page(chan2parent(&atchan->chan_common), desc->lli.saddr,
> + desc->len, DMA_TO_DEVICE);
> +
> + /*
> + * The API requires that no submissions are done from a
> + * callback, so we don't need to drop the lock here
> + */
> + if (callback)
> + callback(param);

Even if our dma controller does not seems to take advantage of
dependency requests...

...should I add the
dma_run_dependencies(txd);
call here to improve conformance to dmaengine standards ?

As you advice Atsushi Nemoto to do so, I was wondering if I have to do
the same.

Thanks for your help.
--
Nicolas Ferre

2009-07-01 15:31:35

by Nicolas Ferre

[permalink] [raw]
Subject: Re: [PATCH 1/2 v2] dmaengine: at_hdmac: new driver for the Atmel AHB DMA Controller

Hi,

Atsushi Nemoto :
> On Fri, 26 Jun 2009 12:42:15 +0200, Nicolas Ferre <[email protected]> wrote:
>> This AHB DMA Controller (aka HDMA or DMAC on AT91 systems) is availlable on
>> at91sam9rl chip. It will be used on other products in the future.
>
> It seems this driver was written based on dw_dmac driver. A while ago
> I had some investigation of that driver.
> (http://lkml.org/lkml/2008/12/29/172)

Thank you for recall me this thread.

> Some of issues reported at that time could be applied on your driver
> too. With a quick look, the queue list management issue is a
> candidate. Here is an excerpt from the thread:
>
> --- --- ---
> 3. dwc->queue list management
>
> The function dwc_tx_submit() add the descriptor to dwc->queue list if
> active list was not empty. But it does not manage lli.llp list. And
> all descriptors in the queue list will be moved to active list at a
> time. So it seems non-first descriptors in queue list will never
> processed by the hardware.
>
> The dwc_tx_submit() should rewrite lli.llp of the last descriptor in
> queue list (it it had children, the last children of it) by txd.phys
> of newly queued descriptor. Or, only first elements of queue list
> should be moved to active list at a time.
>
> Is my analysis correct?

I try to replay the life of a descriptor chain in "queue" list:
- it is queued by atc_tx_submit()
- tasklet or atc_issue_pendig() will "advance_work" and so run
atc_complete_all() at some point.
- atc_complete_all() will issue an atc_dostart() on the first chain in
queue list and move all to active_list
- then all chains will be managed as active_list members:
- tasklet or atc_issue_pendig() will "advance_work"
- first member will be managed by chain_complete()
- next member will be started by dostart()
- and so on...
- last chain in active_list will run complete_all() and may move again
queued descriptors to active_list.

=> non-first descriptor moved from queue to active_list will be
proceeded by act_dostart() in atc_advance_work() function.

In this way of addressing descriptors, I try to keep descriptor chains
as they are built by the prep_dma_memcpy function. I am not trying to
rewrite the internal arrangement of a descriptor chain (not touching
lli.dscr).
It may be not optimal for the descriptor management speed but it tries
to split problems in a kind of layered way (we build chains and then
manage their flow in dma engine and associated lists).

I hope that there is no hole in this management as I am sure it is
difficult to debug...
I hope that my response is solid enough. Please do not hesitate to break
my demonstration ;-).

> --- --- ---
>
> And one more comment.
>
>> + /*
>> + * We use dma_unmap_page() regardless of how the buffers were
>> + * mapped before they were submitted...
>> + */
>
> You can use DMA_COMPL_{SRC,DEST}_UNMAP_SINGLE since 2.6.30.

Ok, even if it is exactly the same for ARM implementation, I do the
modification for my driver (taking iop_dma.c as an example).


Thanks a lot for your help and your deep review.

Best regards,
--
Nicolas Ferre

2009-07-02 00:58:19

by Dan Williams

[permalink] [raw]
Subject: Re: [PATCH 1/2 v2] dmaengine: at_hdmac: new driver for the Atmel AHB DMA Controller

On Wed, Jul 1, 2009 at 6:58 AM, Nicolas Ferre<[email protected]> wrote:
> Hi,
>
> Just one question Dan (see below).
>
> Nicolas Ferre :
>> This AHB DMA Controller (aka HDMA or DMAC on AT91 systems) is availlable on
>> at91sam9rl chip. It will be used on other products in the future.
>
> [..]
>> + ? ? /*
>> + ? ? ?* The API requires that no submissions are done from a
>> + ? ? ?* callback, so we don't need to drop the lock here
>> + ? ? ?*/
>> + ? ? if (callback)
>> + ? ? ? ? ? ? callback(param);
>
> Even if our dma controller does not seems to take advantage of
> dependency requests...
>
> ...should I add the
> dma_run_dependencies(txd);
> call here to improve conformance to dmaengine standards ?

That would be nice but it is not a strict requirement.

To me it just seems like leaving something for some future developer
to trip over when debugging why channel switching does not work with
this driver. If you can guarantee that this driver will always be the
only dma_device driver in the system and that future products will not
add additional operation types then maybe it can be omitted.
Otherwise, the overhead of including the call should be negligible.

Regards,
Dan

2009-07-02 01:18:29

by Dan Williams

[permalink] [raw]
Subject: Re: [PATCH 1/2 v2] dmaengine: at_hdmac: new driver for the Atmel AHB DMA Controller

On Fri, Jun 26, 2009 at 3:42 AM, Nicolas Ferre<[email protected]> wrote:
> This AHB DMA Controller (aka HDMA or DMAC on AT91 systems) is availlable on
> at91sam9rl chip. It will be used on other products in the future.
>
> This first release covers only the memory-to-memory tranfer type. This is the
> only tranfer type supported by this chip. ?On other products, it will be used
> also for peripheral DMA transfer (slave API support to come).
>
> I used dmatest client without problem in different configurations to test it.
>
> Full documentation for this controller can be found in the SAM9RL datasheet:
> http://www.atmel.com/dyn/products/product_card.asp?part_id=4243
>
> Signed-off-by: Nicolas Ferre <[email protected]>
> ---
[..]

A quick review comment:

> +/**
> + * atc_desc_get - get a unsused descriptor from free_list
> + * @atchan: channel we want a new descriptor for
> + */
> +static struct at_desc *atc_desc_get(struct at_dma_chan *atchan)
> +{
> + ? ? ? struct at_desc *desc, *_desc;
> + ? ? ? struct at_desc *ret = NULL;
> + ? ? ? unsigned int i = 0;
> + ? ? ? LIST_HEAD(tmp_list);
> +
> + ? ? ? spin_lock_bh(&atchan->lock);
> + ? ? ? list_for_each_entry_safe(desc, _desc, &atchan->free_list, desc_node) {
> + ? ? ? ? ? ? ? i++;
> + ? ? ? ? ? ? ? if (async_tx_test_ack(&desc->txd)) {
> + ? ? ? ? ? ? ? ? ? ? ? list_del(&desc->desc_node);
> + ? ? ? ? ? ? ? ? ? ? ? ret = desc;
> + ? ? ? ? ? ? ? ? ? ? ? break;
> + ? ? ? ? ? ? ? }
> + ? ? ? ? ? ? ? dev_dbg(chan2dev(&atchan->chan_common),
> + ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? "desc %p not ACKed\n", desc);
> + ? ? ? }
> + ? ? ? spin_unlock_bh(&atchan->lock);
> + ? ? ? dev_vdbg(chan2dev(&atchan->chan_common),
> + ? ? ? ? ? ? ? "scanned %u descriptors on freelist\n", i);
> +
> + ? ? ? /* no more descriptor available in initial pool : create some more */
> + ? ? ? if (!ret) {
> + ? ? ? ? ? ? ? for (i = 0; i < INIT_NR_DESCS_PER_CHANNEL; i++) {
> + ? ? ? ? ? ? ? ? ? ? ? desc = atc_alloc_descriptor(&atchan->chan_common,
> + ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? GFP_KERNEL);

This cannot be GFP_KERNEL as ->prep_dma_memcpy may be called from an
atomic context. Given that this should only be done in emergency
situations this allocation should probably happen one descriptor at a
time, if it must happen at all. My recommendation is that if you find
that the driver runs out of descriptors on a regular basis increase
INIT_NR_DESCS_PER_CHANNEL and only operate with the descriptors
allocated from atc_alloc_chan_resources(). As it stands this allows
the descriptor list to grow without bounds.

Regards,
Dan

2009-07-03 12:59:32

by Sosnowski, Maciej

[permalink] [raw]
Subject: RE: [PATCH 1/2 v2] dmaengine: at_hdmac: new driver for the Atmel AHB DMA Controller

Nicolas Ferre wrote:
> This AHB DMA Controller (aka HDMA or DMAC on AT91 systems) is availlable on
> at91sam9rl chip. It will be used on other products in the future.
>
> This first release covers only the memory-to-memory tranfer type. This is the
> only tranfer type supported by this chip. On other products, it will be used
> also for peripheral DMA transfer (slave API support to come).
>
> I used dmatest client without problem in different configurations to test it.
>
> Full documentation for this controller can be found in the SAM9RL datasheet:
> http://www.atmel.com/dyn/products/product_card.asp?part_id=4243
>
> Signed-off-by: Nicolas Ferre <[email protected]>
> ---
[...]
> + /*
> + * We use dma_unmap_page() regardless of how the buffers were
> + * mapped before they were submitted...
> + */
> + if (!(txd->flags & DMA_COMPL_SKIP_DEST_UNMAP))
> + dma_unmap_page(chan2parent(&atchan->chan_common), desc->lli.daddr,
> + desc->len, DMA_FROM_DEVICE);
> + if (!(txd->flags & DMA_COMPL_SKIP_SRC_UNMAP))
> + dma_unmap_page(chan2parent(&atchan->chan_common), desc->lli.saddr,
> + desc->len, DMA_TO_DEVICE);

It is similar issue as the one fixed in 2.6.30 by 'ioatdma: fix "ioatdma frees DMA memory with wrong function"' patch:
commit 4f005dbe5584fe54c9f6d6d4f0acd3fb29be84da.
Note that this patch introduces dma control flags: DMA_COMPL_SRC_UNMAP_SINGLE and DMA_COMPL_DEST_UNMAP_SINGLE,
that could be used by the driver to choose proper unmapping.

Regards,
Maciej-

2009-07-03 13:04:57

by Nicolas Ferre

[permalink] [raw]
Subject: Re: [PATCH 1/2 v2] dmaengine: at_hdmac: new driver for the Atmel AHB DMA Controller

Sosnowski, Maciej :
> Nicolas Ferre wrote:
>> This AHB DMA Controller (aka HDMA or DMAC on AT91 systems) is availlable on
>> at91sam9rl chip. It will be used on other products in the future.
>>
>> This first release covers only the memory-to-memory tranfer type. This is the
>> only tranfer type supported by this chip. On other products, it will be used
>> also for peripheral DMA transfer (slave API support to come).
>>
>> I used dmatest client without problem in different configurations to test it.
>>
>> Full documentation for this controller can be found in the SAM9RL datasheet:
>> http://www.atmel.com/dyn/products/product_card.asp?part_id=4243
>>
>> Signed-off-by: Nicolas Ferre <[email protected]>
>> ---
> [...]
>> + /*
>> + * We use dma_unmap_page() regardless of how the buffers were
>> + * mapped before they were submitted...
>> + */
>> + if (!(txd->flags & DMA_COMPL_SKIP_DEST_UNMAP))
>> + dma_unmap_page(chan2parent(&atchan->chan_common), desc->lli.daddr,
>> + desc->len, DMA_FROM_DEVICE);
>> + if (!(txd->flags & DMA_COMPL_SKIP_SRC_UNMAP))
>> + dma_unmap_page(chan2parent(&atchan->chan_common), desc->lli.saddr,
>> + desc->len, DMA_TO_DEVICE);
>
> It is similar issue as the one fixed in 2.6.30 by 'ioatdma: fix "ioatdma frees DMA memory with wrong function"' patch:
> commit 4f005dbe5584fe54c9f6d6d4f0acd3fb29be84da.
> Note that this patch introduces dma control flags: DMA_COMPL_SRC_UNMAP_SINGLE and DMA_COMPL_DEST_UNMAP_SINGLE,
> that could be used by the driver to choose proper unmapping.

Sure Maciej, I correct this in my v3 patch (already done locally BTW).

Kind regards,
--
Nicolas Ferre

2009-07-03 15:12:57

by Atsushi Nemoto

[permalink] [raw]
Subject: Re: [PATCH 1/2 v2] dmaengine: at_hdmac: new driver for the Atmel AHB DMA Controller

On Wed, 01 Jul 2009 17:30:54 +0200, Nicolas Ferre <[email protected]> wrote:
> > Some of issues reported at that time could be applied on your driver
> > too. With a quick look, the queue list management issue is a
> > candidate. Here is an excerpt from the thread:
....
>
> I try to replay the life of a descriptor chain in "queue" list:
> - it is queued by atc_tx_submit()
> - tasklet or atc_issue_pendig() will "advance_work" and so run
> atc_complete_all() at some point.
> - atc_complete_all() will issue an atc_dostart() on the first chain in
> queue list and move all to active_list
> - then all chains will be managed as active_list members:
> - tasklet or atc_issue_pendig() will "advance_work"
> - first member will be managed by chain_complete()
> - next member will be started by dostart()
> - and so on...
> - last chain in active_list will run complete_all() and may move again
> queued descriptors to active_list.
>
> => non-first descriptor moved from queue to active_list will be
> proceeded by act_dostart() in atc_advance_work() function.

Oh yes. I was wrong. I missed act_dostart() in atc_advance_work().
Sorry for false warning.

> In this way of addressing descriptors, I try to keep descriptor chains
> as they are built by the prep_dma_memcpy function. I am not trying to
> rewrite the internal arrangement of a descriptor chain (not touching
> lli.dscr).
> It may be not optimal for the descriptor management speed but it tries
> to split problems in a kind of layered way (we build chains and then
> manage their flow in dma engine and associated lists).
>
> I hope that there is no hole in this management as I am sure it is
> difficult to debug...
> I hope that my response is solid enough. Please do not hesitate to break
> my demonstration ;-).

Thank you for elaboration, I cannot find any hole now. Indeed these
kind of problem is hard to debug. Running dmatest with
threads_per_chan=N will catch such issues, hopefully :-)

---
Atsushi Nemoto

2009-07-03 16:22:52

by Nicolas Ferre

[permalink] [raw]
Subject: [PATCH 1/2 v3] dmaengine: at_hdmac: new driver for the Atmel AHB DMA Controller

This AHB DMA Controller (aka HDMA or DMAC on AT91 systems) is availlable on
at91sam9rl chip. It will be used on other products in the future.

This first release covers only the memory-to-memory tranfer type. This is the
only tranfer type supported by this chip. On other products, it will be used
also for peripheral DMA transfer (slave API support to come).

I used dmatest client without problem in different configurations to test it.

Full documentation for this controller can be found in the SAM9RL datasheet:
http://www.atmel.com/dyn/products/product_card.asp?part_id=4243

Signed-off-by: Nicolas Ferre <[email protected]>
---
v2 is here:
http://lkml.org/lkml/2009/6/26/104

v2 -> v3:
- initial number of descriptors to allocate for each channel raised to 64 and
is now a parameter
- ack-bit in descriptor flag comment synchronized with TXx9 dma driver
- atc_desc_get() when short on descriptors in pool: create one at a time
- allocation flag changed to GFP_ATOMIC in atc_desc_get()
- call to proper funtion while unmapping: use of new
DMA_COMPL_{SRC,DEST}_UNMAP_SINGLE flags
- call dma_run_dependencies() at the end of atc_chain_complete()

arch/arm/mach-at91/include/mach/at_hdmac.h | 26 +
drivers/dma/Kconfig | 8 +
drivers/dma/Makefile | 1 +
drivers/dma/at_hdmac.c | 1009 ++++++++++++++++++++++++++++
drivers/dma/at_hdmac_regs.h | 386 +++++++++++
5 files changed, 1430 insertions(+), 0 deletions(-)
create mode 100644 arch/arm/mach-at91/include/mach/at_hdmac.h
create mode 100644 drivers/dma/at_hdmac.c
create mode 100644 drivers/dma/at_hdmac_regs.h

diff --git a/arch/arm/mach-at91/include/mach/at_hdmac.h b/arch/arm/mach-at91/include/mach/at_hdmac.h
new file mode 100644
index 0000000..21a5554
--- /dev/null
+++ b/arch/arm/mach-at91/include/mach/at_hdmac.h
@@ -0,0 +1,26 @@
+/*
+ * Header file for the Atmel AHB DMA Controller driver
+ *
+ * Copyright (C) 2008 Atmel Corporation
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ */
+#ifndef AT_HDMAC_H
+#define AT_HDMAC_H
+
+#include <linux/dmaengine.h>
+
+/**
+ * struct at_dma_platform_data - Controller configuration parameters
+ * @nr_channels: Number of channels supported by hardware (max 8)
+ * @cap_mask: dma_capability flags supported by the platform
+ */
+struct at_dma_platform_data {
+ unsigned int nr_channels;
+ dma_cap_mask_t cap_mask;
+};
+
+#endif /* AT_HDMAC_H */
diff --git a/drivers/dma/Kconfig b/drivers/dma/Kconfig
index babf214..bc8fb41 100644
--- a/drivers/dma/Kconfig
+++ b/drivers/dma/Kconfig
@@ -46,6 +46,14 @@ config DW_DMAC
Support the Synopsys DesignWare AHB DMA controller. This
can be integrated in chips such as the Atmel AT32ap7000.

+config AT_HDMAC
+ tristate "Atmel AHB DMA support"
+ depends on ARCH_AT91SAM9RL
+ select DMA_ENGINE
+ help
+ Support the Atmel AHB DMA controller. This can be integrated in
+ chips such as the Atmel AT91SAM9RL.
+
config FSL_DMA
tristate "Freescale Elo and Elo Plus DMA support"
depends on FSL_SOC
diff --git a/drivers/dma/Makefile b/drivers/dma/Makefile
index 2e5dc96..d7bc5fd 100644
--- a/drivers/dma/Makefile
+++ b/drivers/dma/Makefile
@@ -7,4 +7,5 @@ obj-$(CONFIG_INTEL_IOP_ADMA) += iop-adma.o
obj-$(CONFIG_FSL_DMA) += fsldma.o
obj-$(CONFIG_MV_XOR) += mv_xor.o
obj-$(CONFIG_DW_DMAC) += dw_dmac.o
+obj-$(CONFIG_AT_HDMAC) += at_hdmac.o
obj-$(CONFIG_MX3_IPU) += ipu/
diff --git a/drivers/dma/at_hdmac.c b/drivers/dma/at_hdmac.c
new file mode 100644
index 0000000..9a4abf1
--- /dev/null
+++ b/drivers/dma/at_hdmac.c
@@ -0,0 +1,1009 @@
+/*
+ * Driver for the Atmel AHB DMA Controller (aka HDMA or DMAC on AT91 systems)
+ *
+ * Copyright (C) 2008 Atmel Corporation
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ *
+ * This supports the Atmel AHB DMA Controller,
+ *
+ * The driver has currently been tested with the Atmel AT91SAM9RL
+ * and AT91SAM9G45 series.
+ */
+
+#include <linux/clk.h>
+#include <linux/dmaengine.h>
+#include <linux/dma-mapping.h>
+#include <linux/dmapool.h>
+#include <linux/interrupt.h>
+#include <linux/module.h>
+#include <linux/platform_device.h>
+
+#include "at_hdmac_regs.h"
+
+/*
+ * Glossary
+ * --------
+ *
+ * at_hdmac : Name of the ATmel AHB DMA Controller
+ * at_dma_ / atdma : ATmel DMA controller entity related
+ * atc_ / atchan : ATmel DMA Channel entity related
+ */
+
+#define ATC_DEFAULT_CFG (ATC_FIFOCFG_HALFFIFO)
+#define ATC_DEFAULT_CTRLA (0)
+#define ATC_DEFAULT_CTRLB (ATC_SIF(0) \
+ |ATC_DIF(1))
+
+/*
+ * Initial number of descriptors to allocate for each channel. This could
+ * be increased during dma usage.
+ */
+static unsigned int init_nr_desc_per_channel = 64;
+module_param(init_nr_desc_per_channel, uint, 0644);
+MODULE_PARM_DESC(init_nr_desc_per_channel,
+ "initial descriptors per channel (default: 64)");
+
+
+/* prototypes */
+static dma_cookie_t atc_tx_submit(struct dma_async_tx_descriptor *tx);
+
+
+/*----------------------------------------------------------------------*/
+
+static struct at_desc *atc_first_active(struct at_dma_chan *atchan)
+{
+ return list_first_entry(&atchan->active_list,
+ struct at_desc, desc_node);
+}
+
+static struct at_desc *atc_first_queued(struct at_dma_chan *atchan)
+{
+ return list_first_entry(&atchan->queue,
+ struct at_desc, desc_node);
+}
+
+/**
+ * atc_alloc_descriptor - allocate and return an initilized descriptor
+ * @chan: the channel to allocate descriptors for
+ * @gfp_flags: GFP allocation flags
+ *
+ * Note: The ack-bit is positioned in the descriptor flag at creation time
+ * to make initial allocation more convenient. This bit will be cleared
+ * and control will be given to client at usage time (during
+ * preparation functions).
+ */
+static struct at_desc *atc_alloc_descriptor(struct dma_chan *chan,
+ gfp_t gfp_flags)
+{
+ struct at_desc *desc = NULL;
+ struct at_dma *atdma = to_at_dma(chan->device);
+ dma_addr_t phys;
+
+ desc = dma_pool_alloc(atdma->dma_desc_pool, gfp_flags, &phys);
+ if (desc) {
+ memset(desc, 0, sizeof(struct at_desc));
+ dma_async_tx_descriptor_init(&desc->txd, chan);
+ /* txd.flags will be overwritten in prep functions */
+ desc->txd.flags = DMA_CTRL_ACK;
+ desc->txd.tx_submit = atc_tx_submit;
+ desc->txd.phys = phys;
+ }
+
+ return desc;
+}
+
+/**
+ * atc_desc_get - get a unsused descriptor from free_list
+ * @atchan: channel we want a new descriptor for
+ */
+static struct at_desc *atc_desc_get(struct at_dma_chan *atchan)
+{
+ struct at_desc *desc, *_desc;
+ struct at_desc *ret = NULL;
+ unsigned int i = 0;
+ LIST_HEAD(tmp_list);
+
+ spin_lock_bh(&atchan->lock);
+ list_for_each_entry_safe(desc, _desc, &atchan->free_list, desc_node) {
+ i++;
+ if (async_tx_test_ack(&desc->txd)) {
+ list_del(&desc->desc_node);
+ ret = desc;
+ break;
+ }
+ dev_dbg(chan2dev(&atchan->chan_common),
+ "desc %p not ACKed\n", desc);
+ }
+ spin_unlock_bh(&atchan->lock);
+ dev_vdbg(chan2dev(&atchan->chan_common),
+ "scanned %u descriptors on freelist\n", i);
+
+ /* no more descriptor available in initial pool: create one more */
+ if (!ret) {
+ ret = atc_alloc_descriptor(&atchan->chan_common, GFP_ATOMIC);
+ if (ret) {
+ spin_lock_bh(&atchan->lock);
+ atchan->descs_allocated++;
+ spin_unlock_bh(&atchan->lock);
+ } else {
+ dev_err(chan2dev(&atchan->chan_common),
+ "not enough descriptors available\n");
+ }
+ }
+
+ return ret;
+}
+
+/**
+ * atc_desc_put - move a descriptor, including any children, to the free list
+ * @atchan: channel we work on
+ * @desc: descriptor, at the head of a chain, to move to free list
+ */
+static void atc_desc_put(struct at_dma_chan *atchan, struct at_desc *desc)
+{
+ if (desc) {
+ struct at_desc *child;
+
+ spin_lock_bh(&atchan->lock);
+ list_for_each_entry(child, &desc->txd.tx_list, desc_node)
+ dev_vdbg(chan2dev(&atchan->chan_common),
+ "moving child desc %p to freelist\n",
+ child);
+ list_splice_init(&desc->txd.tx_list, &atchan->free_list);
+ dev_vdbg(chan2dev(&atchan->chan_common),
+ "moving desc %p to freelist\n", desc);
+ list_add(&desc->desc_node, &atchan->free_list);
+ spin_unlock_bh(&atchan->lock);
+ }
+}
+
+/**
+ * atc_assign_cookie - compute and assign new cookie
+ * @atchan: channel we work on
+ * @desc: descriptor to asign cookie for
+ *
+ * Called with atchan->lock held and bh disabled
+ */
+static dma_cookie_t
+atc_assign_cookie(struct at_dma_chan *atchan, struct at_desc *desc)
+{
+ dma_cookie_t cookie = atchan->chan_common.cookie;
+
+ if (++cookie < 0)
+ cookie = 1;
+
+ atchan->chan_common.cookie = cookie;
+ desc->txd.cookie = cookie;
+
+ return cookie;
+}
+
+/**
+ * atc_dostart - starts the DMA engine for real
+ * @atchan: the channel we want to start
+ * @first: first descriptor in the list we want to begin with
+ *
+ * Called with atchan->lock held and bh disabled
+ */
+static void atc_dostart(struct at_dma_chan *atchan, struct at_desc *first)
+{
+ struct at_dma *atdma = to_at_dma(atchan->chan_common.device);
+
+ /* ASSERT: channel is idle */
+ if (atc_chan_is_enabled(atchan)) {
+ dev_err(chan2dev(&atchan->chan_common),
+ "BUG: Attempted to start non-idle channel\n");
+ dev_err(chan2dev(&atchan->chan_common),
+ " channel: s0x%x d0x%x ctrl0x%x:0x%x l0x%x\n",
+ channel_readl(atchan, SADDR),
+ channel_readl(atchan, DADDR),
+ channel_readl(atchan, CTRLA),
+ channel_readl(atchan, CTRLB),
+ channel_readl(atchan, DSCR));
+
+ /* The tasklet will hopefully advance the queue... */
+ return;
+ }
+
+ vdbg_dump_regs(atchan);
+
+ /* clear any pending interrupt */
+ while (dma_readl(atdma, EBCISR))
+ cpu_relax();
+
+ channel_writel(atchan, SADDR, 0);
+ channel_writel(atchan, DADDR, 0);
+ channel_writel(atchan, CTRLA, 0);
+ channel_writel(atchan, CTRLB, 0);
+ channel_writel(atchan, DSCR, first->txd.phys);
+ dma_writel(atdma, CHER, atchan->mask);
+
+ vdbg_dump_regs(atchan);
+}
+
+/**
+ * atc_chain_complete - finish work for one transaction chain
+ * @atchan: channel we work on
+ * @desc: descriptor at the head of the chain we want do complete
+ *
+ * Called with atchan->lock held and bh disabled */
+static void
+atc_chain_complete(struct at_dma_chan *atchan, struct at_desc *desc)
+{
+ dma_async_tx_callback callback;
+ void *param;
+ struct dma_async_tx_descriptor *txd = &desc->txd;
+
+ dev_vdbg(chan2dev(&atchan->chan_common),
+ "descriptor %u complete\n", txd->cookie);
+
+ atchan->completed_cookie = txd->cookie;
+ callback = txd->callback;
+ param = txd->callback_param;
+
+ /* move children to free_list */
+ list_splice_init(&txd->tx_list, &atchan->free_list);
+ /* move myself to free_list */
+ list_move(&desc->desc_node, &atchan->free_list);
+
+ /* unmap dma addresses */
+ if (!(txd->flags & DMA_COMPL_SKIP_DEST_UNMAP)) {
+ if (txd->flags & DMA_COMPL_DEST_UNMAP_SINGLE)
+ dma_unmap_single(chan2parent(&atchan->chan_common),
+ desc->lli.daddr,
+ desc->len, DMA_FROM_DEVICE);
+ else
+ dma_unmap_page(chan2parent(&atchan->chan_common),
+ desc->lli.daddr,
+ desc->len, DMA_FROM_DEVICE);
+ }
+ if (!(txd->flags & DMA_COMPL_SKIP_SRC_UNMAP)) {
+ if (txd->flags & DMA_COMPL_SRC_UNMAP_SINGLE)
+ dma_unmap_single(chan2parent(&atchan->chan_common),
+ desc->lli.saddr,
+ desc->len, DMA_TO_DEVICE);
+ else
+ dma_unmap_page(chan2parent(&atchan->chan_common),
+ desc->lli.saddr,
+ desc->len, DMA_TO_DEVICE);
+ }
+
+ /*
+ * The API requires that no submissions are done from a
+ * callback, so we don't need to drop the lock here
+ */
+ if (callback)
+ callback(param);
+
+ dma_run_dependencies(txd);
+}
+
+/**
+ * atc_complete_all - finish work for all transactions
+ * @atchan: channel to complete transactions for
+ *
+ * Eventually submit queued descriptors if any
+ *
+ * Assume channel is idle while calling this function
+ * Called with atchan->lock held and bh disabled
+ */
+static void atc_complete_all(struct at_dma_chan *atchan)
+{
+ struct at_desc *desc, *_desc;
+ LIST_HEAD(list);
+
+ dev_vdbg(chan2dev(&atchan->chan_common), "complete all\n");
+
+ BUG_ON(atc_chan_is_enabled(atchan));
+
+ /*
+ * Submit queued descriptors ASAP, i.e. before we go through
+ * the completed ones.
+ */
+ if (!list_empty(&atchan->queue))
+ atc_dostart(atchan, atc_first_queued(atchan));
+ /* empty active_list now it is completed */
+ list_splice_init(&atchan->active_list, &list);
+ /* empty queue list by moving descriptors (if any) to active_list */
+ list_splice_init(&atchan->queue, &atchan->active_list);
+
+ list_for_each_entry_safe(desc, _desc, &list, desc_node)
+ atc_chain_complete(atchan, desc);
+}
+
+/**
+ * atc_cleanup_descriptors - cleanup up finished descriptors in active_list
+ * @atchan: channel to be cleaned up
+ *
+ * Called with atchan->lock held and bh disabled
+ */
+static void atc_cleanup_descriptors(struct at_dma_chan *atchan)
+{
+ struct at_desc *desc, *_desc;
+ struct at_desc *child;
+
+ dev_vdbg(chan2dev(&atchan->chan_common), "cleanup descriptors\n");
+
+ list_for_each_entry_safe(desc, _desc, &atchan->active_list, desc_node) {
+ if (!(desc->lli.ctrla & ATC_DONE))
+ /* This one is currently in progress */
+ return;
+
+ list_for_each_entry(child, &desc->txd.tx_list, desc_node)
+ if (!(child->lli.ctrla & ATC_DONE))
+ /* Currently in progress */
+ return;
+
+ /*
+ * No descriptors so far seem to be in progress, i.e.
+ * this chain must be done.
+ */
+ atc_chain_complete(atchan, desc);
+ }
+}
+
+/**
+ * atc_advance_work - at the end of a transaction, move forward
+ * @atchan: channel where the transaction ended
+ *
+ * Called with atchan->lock held and bh disabled
+ */
+static void atc_advance_work(struct at_dma_chan *atchan)
+{
+ dev_vdbg(chan2dev(&atchan->chan_common), "advance_work\n");
+
+ if ( list_empty(&atchan->active_list)
+ || list_is_singular(&atchan->active_list)) {
+ atc_complete_all(atchan);
+ } else {
+ atc_chain_complete(atchan, atc_first_active(atchan));
+ /* advance work */
+ atc_dostart(atchan, atc_first_active(atchan));
+ }
+}
+
+
+/**
+ * atc_handle_error - handle errors reported by DMA controller
+ * @atchan: channel where error occurs
+ *
+ * Called with atchan->lock held and bh disabled
+ */
+static void atc_handle_error(struct at_dma_chan *atchan)
+{
+ struct at_desc *bad_desc;
+ struct at_desc *child;
+
+ /*
+ * The descriptor currently at the head of the active list is
+ * broked. Since we don't have any way to report errors, we'll
+ * just have to scream loudly and try to carry on.
+ */
+ bad_desc = atc_first_active(atchan);
+ list_del_init(&bad_desc->desc_node);
+
+ /* As we are stopped, take advantage to push queued descriptors
+ * in active_list */
+ list_splice_init(&atchan->queue, atchan->active_list.prev);
+
+ /* Try to restart the controller */
+ if (!list_empty(&atchan->active_list))
+ atc_dostart(atchan, atc_first_active(atchan));
+
+ /*
+ * KERN_CRITICAL may seem harsh, but since this only happens
+ * when someone submits a bad physical address in a
+ * descriptor, we should consider ourselves lucky that the
+ * controller flagged an error instead of scribbling over
+ * random memory locations.
+ */
+ dev_crit(chan2dev(&atchan->chan_common),
+ "Bad descriptor submitted for DMA!\n");
+ dev_crit(chan2dev(&atchan->chan_common),
+ " cookie: %d\n", bad_desc->txd.cookie);
+ atc_dump_lli(atchan, &bad_desc->lli);
+ list_for_each_entry(child, &bad_desc->txd.tx_list, desc_node)
+ atc_dump_lli(atchan, &child->lli);
+
+ /* Pretend the descriptor completed successfully */
+ atc_chain_complete(atchan, bad_desc);
+}
+
+
+/*-- IRQ & Tasklet ---------------------------------------------------*/
+
+static void atc_tasklet(unsigned long data)
+{
+ struct at_dma_chan *atchan = (struct at_dma_chan *)data;
+
+ /* Channel cannot be enabled here */
+ if (atc_chan_is_enabled(atchan)) {
+ dev_err(chan2dev(&atchan->chan_common),
+ "BUG: channel enabled in tasklet\n");
+ return;
+ }
+
+ spin_lock(&atchan->lock);
+ if (test_and_clear_bit(0, &atchan->error_status))
+ atc_handle_error(atchan);
+ else
+ atc_advance_work(atchan);
+
+ spin_unlock(&atchan->lock);
+}
+
+static irqreturn_t at_dma_interrupt(int irq, void *dev_id)
+{
+ struct at_dma *atdma = (struct at_dma *)dev_id;
+ struct at_dma_chan *atchan;
+ int i;
+ u32 status, pending, imr;
+ int ret = IRQ_NONE;
+
+ do {
+ imr = dma_readl(atdma, EBCIMR);
+ status = dma_readl(atdma, EBCISR);
+ pending = status & imr;
+
+ if (!pending)
+ break;
+
+ dev_vdbg(atdma->dma_common.dev,
+ "interrupt: status = 0x%08x, 0x%08x, 0x%08x\n",
+ status, imr, pending);
+
+ for (i = 0; i < atdma->dma_common.chancnt; i++) {
+ atchan = &atdma->chan[i];
+ if (pending & (AT_DMA_CBTC(i) | AT_DMA_ERR(i))) {
+ if (pending & AT_DMA_ERR(i)) {
+ /* Disable channel on AHB error */
+ dma_writel(atdma, CHDR, atchan->mask);
+ /* Give information to tasklet */
+ set_bit(0, &atchan->error_status);
+ }
+ tasklet_schedule(&atchan->tasklet);
+ ret = IRQ_HANDLED;
+ }
+ }
+
+ } while (pending);
+
+ return ret;
+}
+
+
+/*-- DMA Engine API --------------------------------------------------*/
+
+/**
+ * atc_tx_submit - set the prepared descriptor(s) to be executed by the engine
+ * @desc: descriptor at the head of the transaction chain
+ *
+ * Queue chain if DMA engine is working already
+ *
+ * Cookie increment and adding to active_list or queue must be atomic
+ */
+static dma_cookie_t atc_tx_submit(struct dma_async_tx_descriptor *tx)
+{
+ struct at_desc *desc = txd_to_at_desc(tx);
+ struct at_dma_chan *atchan = to_at_dma_chan(tx->chan);
+ dma_cookie_t cookie;
+
+ spin_lock_bh(&atchan->lock);
+ cookie = atc_assign_cookie(atchan, desc);
+
+ if (list_empty(&atchan->active_list)) {
+ dev_vdbg(chan2dev(tx->chan), "tx_submit: started %u\n",
+ desc->txd.cookie);
+ atc_dostart(atchan, desc);
+ list_add_tail(&desc->desc_node, &atchan->active_list);
+ } else {
+ dev_vdbg(chan2dev(tx->chan), "tx_submit: queued %u\n",
+ desc->txd.cookie);
+ list_add_tail(&desc->desc_node, &atchan->queue);
+ }
+
+ spin_unlock_bh(&atchan->lock);
+
+ return cookie;
+}
+
+/**
+ * atc_prep_dma_memcpy - prepare a memcpy operation
+ * @chan: the channel to prepare operation on
+ * @dest: operation virtual destination address
+ * @src: operation virtual source address
+ * @len: operation length
+ * @flags: tx descriptor status flags
+ */
+static struct dma_async_tx_descriptor *
+atc_prep_dma_memcpy(struct dma_chan *chan, dma_addr_t dest, dma_addr_t src,
+ size_t len, unsigned long flags)
+{
+ struct at_dma_chan *atchan = to_at_dma_chan(chan);
+ struct at_desc *desc = NULL;
+ struct at_desc *first = NULL;
+ struct at_desc *prev = NULL;
+ size_t xfer_count;
+ size_t offset;
+ unsigned int src_width;
+ unsigned int dst_width;
+ u32 ctrla;
+ u32 ctrlb;
+
+ dev_vdbg(chan2dev(chan), "prep_dma_memcpy: d0x%x s0x%x l0x%zx f0x%lx\n",
+ dest, src, len, flags);
+
+ if (unlikely(!len)) {
+ dev_dbg(chan2dev(chan), "prep_dma_memcpy: length is zero!\n");
+ return NULL;
+ }
+
+ ctrla = ATC_DEFAULT_CTRLA;
+ ctrlb = ATC_DEFAULT_CTRLB
+ | ATC_SRC_ADDR_MODE_INCR
+ | ATC_DST_ADDR_MODE_INCR
+ | ATC_FC_MEM2MEM;
+
+ /*
+ * We can be a lot more clever here, but this should take care
+ * of the most common optimization.
+ */
+ if (!((src | dest | len) & 3)) {
+ ctrla |= ATC_SRC_WIDTH_WORD | ATC_DST_WIDTH_WORD;
+ src_width = dst_width = 2;
+ } else if (!((src | dest | len) & 1)) {
+ ctrla |= ATC_SRC_WIDTH_HALFWORD | ATC_DST_WIDTH_HALFWORD;
+ src_width = dst_width = 1;
+ } else {
+ ctrla |= ATC_SRC_WIDTH_BYTE | ATC_DST_WIDTH_BYTE;
+ src_width = dst_width = 0;
+ }
+
+ for (offset = 0; offset < len; offset += xfer_count << src_width) {
+ xfer_count = min_t(size_t, (len - offset) >> src_width,
+ ATC_BTSIZE_MAX);
+
+ desc = atc_desc_get(atchan);
+ if (!desc)
+ goto err_desc_get;
+
+ desc->lli.saddr = src + offset;
+ desc->lli.daddr = dest + offset;
+ desc->lli.ctrla = ctrla | xfer_count;
+ desc->lli.ctrlb = ctrlb;
+
+ desc->txd.cookie = 0;
+ async_tx_ack(&desc->txd);
+
+ if (!first) {
+ first = desc;
+ } else {
+ /* inform the HW lli about chaining */
+ prev->lli.dscr = desc->txd.phys;
+ /* insert the link descriptor to the LD ring */
+ list_add_tail(&desc->desc_node,
+ &first->txd.tx_list);
+ }
+ prev = desc;
+ }
+
+ /* First descriptor of the chain embedds additional information */
+ first->txd.cookie = -EBUSY;
+ first->len = len;
+
+ /* set end-of-link to the last link descriptor of list*/
+ set_desc_eol(desc);
+
+ desc->txd.flags = flags; /* client is in control of this ack */
+
+ return &first->txd;
+
+err_desc_get:
+ atc_desc_put(atchan, first);
+ return NULL;
+}
+
+/**
+ * atc_is_tx_complete - poll for transaction completion
+ * @chan: DMA channel
+ * @cookie: transaction identifier to check status of
+ * @done: if not %NULL, updated with last completed transaction
+ * @used: if not %NULL, updated with last used transaction
+ *
+ * If @done and @used are passed in, upon return they reflect the driver
+ * internal state and can be used with dma_async_is_complete() to check
+ * the status of multiple cookies without re-checking hardware state.
+ */
+static enum dma_status
+atc_is_tx_complete(struct dma_chan *chan,
+ dma_cookie_t cookie,
+ dma_cookie_t *done, dma_cookie_t *used)
+{
+ struct at_dma_chan *atchan = to_at_dma_chan(chan);
+ dma_cookie_t last_used;
+ dma_cookie_t last_complete;
+ enum dma_status ret;
+
+ dev_vdbg(chan2dev(chan), "is_tx_complete: %d (d%d, u%d)\n",
+ cookie, done ? *done : 0, used ? *used : 0);
+
+ spin_lock_bh(atchan->lock);
+
+ last_complete = atchan->completed_cookie;
+ last_used = chan->cookie;
+
+ ret = dma_async_is_complete(cookie, last_complete, last_used);
+ if (ret != DMA_SUCCESS) {
+ atc_cleanup_descriptors(atchan);
+
+ last_complete = atchan->completed_cookie;
+ last_used = chan->cookie;
+
+ ret = dma_async_is_complete(cookie, last_complete, last_used);
+ }
+
+ spin_unlock_bh(atchan->lock);
+
+ if (done)
+ *done = last_complete;
+ if (used)
+ *used = last_used;
+
+ return ret;
+}
+
+/**
+ * atc_issue_pending - try to finish work
+ * @chan: target DMA channel
+ */
+static void atc_issue_pending(struct dma_chan *chan)
+{
+ struct at_dma_chan *atchan = to_at_dma_chan(chan);
+
+ dev_vdbg(chan2dev(chan), "issue_pending\n");
+
+ if (!atc_chan_is_enabled(atchan)) {
+ spin_lock_bh(&atchan->lock);
+ atc_advance_work(atchan);
+ spin_unlock_bh(&atchan->lock);
+ }
+}
+
+/**
+ * atc_alloc_chan_resources - allocate resources for DMA channel
+ * @chan: allocate descriptor resources for this channel
+ * @client: current client requesting the channel be ready for requests
+ *
+ * return - the number of allocated descriptors
+ */
+static int atc_alloc_chan_resources(struct dma_chan *chan)
+{
+ struct at_dma_chan *atchan = to_at_dma_chan(chan);
+ struct at_dma *atdma = to_at_dma(chan->device);
+ struct at_desc *desc;
+ int i;
+ LIST_HEAD(tmp_list);
+
+ dev_vdbg(chan2dev(chan), "alloc_chan_resources\n");
+
+ /* ASSERT: channel is idle */
+ if (atc_chan_is_enabled(atchan)) {
+ dev_dbg(chan2dev(chan), "DMA channel not idle ?\n");
+ return -EIO;
+ }
+
+ /* have we already been set up? */
+ if (!list_empty(&atchan->free_list))
+ return atchan->descs_allocated;
+
+ /* Allocate initial pool of descriptors */
+ for (i = 0; i < init_nr_desc_per_channel; i++) {
+ desc = atc_alloc_descriptor(chan, GFP_KERNEL);
+ if (!desc) {
+ dev_err(atdma->dma_common.dev,
+ "Only %d initial descriptors\n", i);
+ break;
+ }
+ list_add_tail(&desc->desc_node, &tmp_list);
+ }
+
+ spin_lock_bh(&atchan->lock);
+ atchan->descs_allocated = i;
+ list_splice(&tmp_list, &atchan->free_list);
+ atchan->completed_cookie = chan->cookie = 1;
+ spin_unlock_bh(&atchan->lock);
+
+ /* channel parameters */
+ channel_writel(atchan, CFG, ATC_DEFAULT_CFG);
+
+ dev_dbg(chan2dev(chan),
+ "alloc_chan_resources: allocated %d descriptors\n",
+ atchan->descs_allocated);
+
+ return atchan->descs_allocated;
+}
+
+/**
+ * atc_free_chan_resources - free all channel resources
+ * @chan: DMA channel
+ */
+static void atc_free_chan_resources(struct dma_chan *chan)
+{
+ struct at_dma_chan *atchan = to_at_dma_chan(chan);
+ struct at_dma *atdma = to_at_dma(chan->device);
+ struct at_desc *desc, *_desc;
+ LIST_HEAD(list);
+
+ dev_dbg(chan2dev(chan), "free_chan_resources: (descs allocated=%u)\n",
+ atchan->descs_allocated);
+
+ /* ASSERT: channel is idle */
+ BUG_ON(!list_empty(&atchan->active_list));
+ BUG_ON(!list_empty(&atchan->queue));
+ BUG_ON(atc_chan_is_enabled(atchan));
+
+ list_for_each_entry_safe(desc, _desc, &atchan->free_list, desc_node) {
+ dev_vdbg(chan2dev(chan), " freeing descriptor %p\n", desc);
+ list_del(&desc->desc_node);
+ /* free link descriptor */
+ dma_pool_free(atdma->dma_desc_pool, desc, desc->txd.phys);
+ }
+ list_splice_init(&atchan->free_list, &list);
+ atchan->descs_allocated = 0;
+
+ dev_vdbg(chan2dev(chan), "free_chan_resources: done\n");
+}
+
+
+/*-- Module Management -----------------------------------------------*/
+
+/**
+ * at_dma_off - disable DMA controller
+ * @atdma: the Atmel HDAMC device
+ */
+static void at_dma_off(struct at_dma *atdma)
+{
+ dma_writel(atdma, EN, 0);
+
+ /* disable all interrupts */
+ dma_writel(atdma, EBCIDR, -1L);
+
+ /* confirm that all channels are disabled */
+ while (dma_readl(atdma, CHSR) & atdma->all_chan_mask)
+ cpu_relax();
+}
+
+static int __init at_dma_probe(struct platform_device *pdev)
+{
+ struct at_dma_platform_data *pdata;
+ struct resource *io;
+ struct at_dma *atdma;
+ size_t size;
+ int irq;
+ int err;
+ int i;
+
+ /* get DMA Controller parameters from platform */
+ pdata = pdev->dev.platform_data;
+ if (!pdata || pdata->nr_channels > AT_DMA_MAX_NR_CHANNELS)
+ return -EINVAL;
+
+ io = platform_get_resource(pdev, IORESOURCE_MEM, 0);
+ if (!io)
+ return -EINVAL;
+
+ irq = platform_get_irq(pdev, 0);
+ if (irq < 0)
+ return irq;
+
+ size = sizeof(struct at_dma);
+ size += pdata->nr_channels * sizeof(struct at_dma_chan);
+ atdma = kzalloc(size, GFP_KERNEL);
+ if (!atdma)
+ return -ENOMEM;
+
+ /* discover transaction capabilites from the platform data */
+ atdma->dma_common.cap_mask = pdata->cap_mask;
+ atdma->all_chan_mask = (1 << pdata->nr_channels) - 1;
+
+ size = io->end - io->start + 1;
+ if (!request_mem_region(io->start, size, pdev->dev.driver->name)) {
+ err = -EBUSY;
+ goto err_kfree;
+ }
+
+ atdma->regs = ioremap(io->start, size);
+ if (!atdma->regs) {
+ err = -ENOMEM;
+ goto err_release_r;
+ }
+
+ atdma->clk = clk_get(&pdev->dev, "dma_clk");
+ if (IS_ERR(atdma->clk)) {
+ err = PTR_ERR(atdma->clk);
+ goto err_clk;
+ }
+ clk_enable(atdma->clk);
+
+ /* force dma off, just in case */
+ at_dma_off(atdma);
+
+ err = request_irq(irq, at_dma_interrupt, 0, "at_hdmac", atdma);
+ if (err)
+ goto err_irq;
+
+ platform_set_drvdata(pdev, atdma);
+
+ /* create a pool of consistent memory blocks for hardware descriptors */
+ atdma->dma_desc_pool = dma_pool_create("at_hdmac_desc_pool",
+ &pdev->dev, sizeof(struct at_desc),
+ 4 /* word alignment */, 0);
+ if (!atdma->dma_desc_pool) {
+ dev_err(&pdev->dev, "No memory for descriptors dma pool\n");
+ err = -ENOMEM;
+ goto err_pool_create;
+ }
+
+ /* clear any pending interrupt */
+ while (dma_readl(atdma, EBCISR))
+ cpu_relax();
+
+ /* initialize channels related values */
+ INIT_LIST_HEAD(&atdma->dma_common.channels);
+ for (i = 0; i < pdata->nr_channels; i++, atdma->dma_common.chancnt++) {
+ struct at_dma_chan *atchan = &atdma->chan[i];
+
+ atchan->chan_common.device = &atdma->dma_common;
+ atchan->chan_common.cookie = atchan->completed_cookie = 1;
+ atchan->chan_common.chan_id = i;
+ list_add_tail(&atchan->chan_common.device_node,
+ &atdma->dma_common.channels);
+
+ atchan->ch_regs = atdma->regs + ch_regs(i);
+ spin_lock_init(&atchan->lock);
+ atchan->mask = 1 << i;
+
+ INIT_LIST_HEAD(&atchan->active_list);
+ INIT_LIST_HEAD(&atchan->queue);
+ INIT_LIST_HEAD(&atchan->free_list);
+
+ tasklet_init(&atchan->tasklet, atc_tasklet,
+ (unsigned long)atchan);
+ atc_enable_irq(atchan);
+ }
+
+ /* set base routines */
+ atdma->dma_common.device_alloc_chan_resources = atc_alloc_chan_resources;
+ atdma->dma_common.device_free_chan_resources = atc_free_chan_resources;
+ atdma->dma_common.device_is_tx_complete = atc_is_tx_complete;
+ atdma->dma_common.device_issue_pending = atc_issue_pending;
+ atdma->dma_common.dev = &pdev->dev;
+
+ /* set prep routines based on capability */
+ if (dma_has_cap(DMA_MEMCPY, atdma->dma_common.cap_mask))
+ atdma->dma_common.device_prep_dma_memcpy = atc_prep_dma_memcpy;
+
+ dma_writel(atdma, EN, AT_DMA_ENABLE);
+
+ dev_info(&pdev->dev, "Atmel AHB DMA Controller ( %s%s), %d channels\n",
+ dma_has_cap(DMA_MEMCPY, atdma->dma_common.cap_mask) ? "cpy " : "",
+ dma_has_cap(DMA_SLAVE, atdma->dma_common.cap_mask) ? "slave " : "",
+ atdma->dma_common.chancnt);
+
+ dma_async_device_register(&atdma->dma_common);
+
+ return 0;
+
+err_pool_create:
+ platform_set_drvdata(pdev, NULL);
+ free_irq(platform_get_irq(pdev, 0), atdma);
+err_irq:
+ clk_disable(atdma->clk);
+ clk_put(atdma->clk);
+err_clk:
+ iounmap(atdma->regs);
+ atdma->regs = NULL;
+err_release_r:
+ release_mem_region(io->start, size);
+err_kfree:
+ kfree(atdma);
+ return err;
+}
+
+static int __exit at_dma_remove(struct platform_device *pdev)
+{
+ struct at_dma *atdma = platform_get_drvdata(pdev);
+ struct dma_chan *chan, *_chan;
+ struct resource *io;
+
+ at_dma_off(atdma);
+ dma_async_device_unregister(&atdma->dma_common);
+
+ dma_pool_destroy(atdma->dma_desc_pool);
+ platform_set_drvdata(pdev, NULL);
+ free_irq(platform_get_irq(pdev, 0), atdma);
+
+ list_for_each_entry_safe(chan, _chan, &atdma->dma_common.channels,
+ device_node) {
+ struct at_dma_chan *atchan = to_at_dma_chan(chan);
+
+ /* Disable interrupts */
+ atc_disable_irq(atchan);
+ tasklet_disable(&atchan->tasklet);
+
+ tasklet_kill(&atchan->tasklet);
+ list_del(&chan->device_node);
+ }
+
+ clk_disable(atdma->clk);
+ clk_put(atdma->clk);
+
+ iounmap(atdma->regs);
+ atdma->regs = NULL;
+
+ io = platform_get_resource(pdev, IORESOURCE_MEM, 0);
+ release_mem_region(io->start, io->end - io->start + 1);
+
+ kfree(atdma);
+
+ return 0;
+}
+
+static void at_dma_shutdown(struct platform_device *pdev)
+{
+ struct at_dma *atdma = platform_get_drvdata(pdev);
+
+ at_dma_off(platform_get_drvdata(pdev));
+ clk_disable(atdma->clk);
+}
+
+static int at_dma_suspend_late(struct platform_device *pdev, pm_message_t mesg)
+{
+ struct at_dma *atdma = platform_get_drvdata(pdev);
+
+ at_dma_off(platform_get_drvdata(pdev));
+ clk_disable(atdma->clk);
+ return 0;
+}
+
+static int at_dma_resume_early(struct platform_device *pdev)
+{
+ struct at_dma *atdma = platform_get_drvdata(pdev);
+
+ clk_enable(atdma->clk);
+ dma_writel(atdma, EN, AT_DMA_ENABLE);
+ return 0;
+
+}
+
+static struct platform_driver at_dma_driver = {
+ .remove = __exit_p(at_dma_remove),
+ .shutdown = at_dma_shutdown,
+ .suspend_late = at_dma_suspend_late,
+ .resume_early = at_dma_resume_early,
+ .driver = {
+ .name = "at_hdmac",
+ },
+};
+
+static int __init at_dma_init(void)
+{
+ return platform_driver_probe(&at_dma_driver, at_dma_probe);
+}
+module_init(at_dma_init);
+
+static void __exit at_dma_exit(void)
+{
+ platform_driver_unregister(&at_dma_driver);
+}
+module_exit(at_dma_exit);
+
+MODULE_DESCRIPTION("Atmel AHB DMA Controller driver");
+MODULE_AUTHOR("Nicolas Ferre <[email protected]>");
+MODULE_LICENSE("GPL");
+MODULE_ALIAS("platform:at_hdmac");
diff --git a/drivers/dma/at_hdmac_regs.h b/drivers/dma/at_hdmac_regs.h
new file mode 100644
index 0000000..ad2d4f4
--- /dev/null
+++ b/drivers/dma/at_hdmac_regs.h
@@ -0,0 +1,386 @@
+/*
+ * Header file for the Atmel AHB DMA Controller driver
+ *
+ * Copyright (C) 2008 Atmel Corporation
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ */
+#ifndef AT_HDMAC_REGS_H
+#define AT_HDMAC_REGS_H
+
+#include <mach/at_hdmac.h>
+
+#define AT_DMA_MAX_NR_CHANNELS 8
+
+
+#define AT_DMA_GCFG 0x00 /* Global Configuration Register */
+#define AT_DMA_IF_BIGEND(i) (0x1 << (i)) /* AHB-Lite Interface i in Big-endian mode */
+#define AT_DMA_ARB_CFG (0x1 << 4) /* Arbiter mode. */
+#define AT_DMA_ARB_CFG_FIXED (0x0 << 4)
+#define AT_DMA_ARB_CFG_ROUND_ROBIN (0x1 << 4)
+
+#define AT_DMA_EN 0x04 /* Controller Enable Register */
+#define AT_DMA_ENABLE (0x1 << 0)
+
+#define AT_DMA_SREQ 0x08 /* Software Single Request Register */
+#define AT_DMA_SSREQ(x) (0x1 << ((x) << 1)) /* Request a source single transfer on channel x */
+#define AT_DMA_DSREQ(x) (0x1 << (1 + ((x) << 1))) /* Request a destination single transfer on channel x */
+
+#define AT_DMA_CREQ 0x0C /* Software Chunk Transfer Request Register */
+#define AT_DMA_SCREQ(x) (0x1 << ((x) << 1)) /* Request a source chunk transfer on channel x */
+#define AT_DMA_DCREQ(x) (0x1 << (1 + ((x) << 1))) /* Request a destination chunk transfer on channel x */
+
+#define AT_DMA_LAST 0x10 /* Software Last Transfer Flag Register */
+#define AT_DMA_SLAST(x) (0x1 << ((x) << 1)) /* This src rq is last tx of buffer on channel x */
+#define AT_DMA_DLAST(x) (0x1 << (1 + ((x) << 1))) /* This dst rq is last tx of buffer on channel x */
+
+#define AT_DMA_SYNC 0x14 /* Request Synchronization Register */
+#define AT_DMA_SYR(h) (0x1 << (h)) /* Synchronize handshake line h */
+
+/* Error, Chained Buffer transfer completed and Buffer transfer completed Interrupt registers */
+#define AT_DMA_EBCIER 0x18 /* Enable register */
+#define AT_DMA_EBCIDR 0x1C /* Disable register */
+#define AT_DMA_EBCIMR 0x20 /* Mask Register */
+#define AT_DMA_EBCISR 0x24 /* Status Register */
+#define AT_DMA_CBTC_OFFSET 8
+#define AT_DMA_ERR_OFFSET 16
+#define AT_DMA_BTC(x) (0x1 << (x))
+#define AT_DMA_CBTC(x) (0x1 << (AT_DMA_CBTC_OFFSET + (x)))
+#define AT_DMA_ERR(x) (0x1 << (AT_DMA_ERR_OFFSET + (x)))
+
+#define AT_DMA_CHER 0x28 /* Channel Handler Enable Register */
+#define AT_DMA_ENA(x) (0x1 << (x))
+#define AT_DMA_SUSP(x) (0x1 << ( 8 + (x)))
+#define AT_DMA_KEEP(x) (0x1 << (24 + (x)))
+
+#define AT_DMA_CHDR 0x2C /* Channel Handler Disable Register */
+#define AT_DMA_DIS(x) (0x1 << (x))
+#define AT_DMA_RES(x) (0x1 << ( 8 + (x)))
+
+#define AT_DMA_CHSR 0x30 /* Channel Handler Status Register */
+#define AT_DMA_EMPT(x) (0x1 << (16 + (x)))
+#define AT_DMA_STAL(x) (0x1 << (24 + (x)))
+
+
+#define AT_DMA_CH_REGS_BASE 0x3C /* Channel registers base address */
+#define ch_regs(x) (AT_DMA_CH_REGS_BASE + (x) * 0x28) /* Channel x base addr */
+
+/* Hardware register offset for each channel */
+#define ATC_SADDR_OFFSET 0x00 /* Source Address Register */
+#define ATC_DADDR_OFFSET 0x04 /* Destination Address Register */
+#define ATC_DSCR_OFFSET 0x08 /* Descriptor Address Register */
+#define ATC_CTRLA_OFFSET 0x0C /* Control A Register */
+#define ATC_CTRLB_OFFSET 0x10 /* Control B Register */
+#define ATC_CFG_OFFSET 0x14 /* Configuration Register */
+#define ATC_SPIP_OFFSET 0x18 /* Src PIP Configuration Register */
+#define ATC_DPIP_OFFSET 0x1C /* Dst PIP Configuration Register */
+
+
+/* Bitfield definitions */
+
+/* Bitfields in DSCR */
+#define ATC_DSCR_IF(i) (0x3 & (i)) /* Dsc feched via AHB-Lite Interface i */
+
+/* Bitfields in CTRLA */
+#define ATC_BTSIZE_MAX 0xFFFFUL /* Maximum Buffer Transfer Size */
+#define ATC_BTSIZE(x) (ATC_BTSIZE_MAX & (x)) /* Buffer Transfer Size */
+#define ATC_SCSIZE_MASK (0x7 << 16) /* Source Chunk Transfer Size */
+#define ATC_SCSIZE_1 (0x0 << 16)
+#define ATC_SCSIZE_4 (0x1 << 16)
+#define ATC_SCSIZE_8 (0x2 << 16)
+#define ATC_SCSIZE_16 (0x3 << 16)
+#define ATC_SCSIZE_32 (0x4 << 16)
+#define ATC_SCSIZE_64 (0x5 << 16)
+#define ATC_SCSIZE_128 (0x6 << 16)
+#define ATC_SCSIZE_256 (0x7 << 16)
+#define ATC_DCSIZE_MASK (0x7 << 20) /* Destination Chunk Transfer Size */
+#define ATC_DCSIZE_1 (0x0 << 20)
+#define ATC_DCSIZE_4 (0x1 << 20)
+#define ATC_DCSIZE_8 (0x2 << 20)
+#define ATC_DCSIZE_16 (0x3 << 20)
+#define ATC_DCSIZE_32 (0x4 << 20)
+#define ATC_DCSIZE_64 (0x5 << 20)
+#define ATC_DCSIZE_128 (0x6 << 20)
+#define ATC_DCSIZE_256 (0x7 << 20)
+#define ATC_SRC_WIDTH_MASK (0x3 << 24) /* Source Single Transfer Size */
+#define ATC_SRC_WIDTH_BYTE (0x0 << 24)
+#define ATC_SRC_WIDTH_HALFWORD (0x1 << 24)
+#define ATC_SRC_WIDTH_WORD (0x2 << 24)
+#define ATC_DST_WIDTH_MASK (0x3 << 28) /* Destination Single Transfer Size */
+#define ATC_DST_WIDTH_BYTE (0x0 << 28)
+#define ATC_DST_WIDTH_HALFWORD (0x1 << 28)
+#define ATC_DST_WIDTH_WORD (0x2 << 28)
+#define ATC_DONE (0x1 << 31) /* Tx Done (only written back in descriptor) */
+
+/* Bitfields in CTRLB */
+#define ATC_SIF(i) (0x3 & (i)) /* Src tx done via AHB-Lite Interface i */
+#define ATC_DIF(i) ((0x3 & (i)) << 4) /* Dst tx done via AHB-Lite Interface i */
+#define ATC_SRC_PIP (0x1 << 8) /* Source Picture-in-Picture enabled */
+#define ATC_DST_PIP (0x1 << 12) /* Destination Picture-in-Picture enabled */
+#define ATC_SRC_DSCR_DIS (0x1 << 16) /* Src Descriptor fetch disable */
+#define ATC_DST_DSCR_DIS (0x1 << 20) /* Dst Descriptor fetch disable */
+#define ATC_FC_MASK (0x7 << 21) /* Choose Flow Controller */
+#define ATC_FC_MEM2MEM (0x0 << 21) /* Mem-to-Mem (DMA) */
+#define ATC_FC_MEM2PER (0x1 << 21) /* Mem-to-Periph (DMA) */
+#define ATC_FC_PER2MEM (0x2 << 21) /* Periph-to-Mem (DMA) */
+#define ATC_FC_PER2PER (0x3 << 21) /* Periph-to-Periph (DMA) */
+#define ATC_FC_PER2MEM_PER (0x4 << 21) /* Periph-to-Mem (Peripheral) */
+#define ATC_FC_MEM2PER_PER (0x5 << 21) /* Mem-to-Periph (Peripheral) */
+#define ATC_FC_PER2PER_PER (0x6 << 21) /* Periph-to-Periph (Src Peripheral) */
+#define ATC_SRC_ADDR_MODE_MASK (0x3 << 24)
+#define ATC_SRC_ADDR_MODE_INCR (0x0 << 24) /* Incrementing Mode */
+#define ATC_SRC_ADDR_MODE_DECR (0x1 << 24) /* Decrementing Mode */
+#define ATC_SRC_ADDR_MODE_FIXED (0x2 << 24) /* Fixed Mode */
+#define ATC_DST_ADDR_MODE_MASK (0x3 << 28)
+#define ATC_DST_ADDR_MODE_INCR (0x0 << 28) /* Incrementing Mode */
+#define ATC_DST_ADDR_MODE_DECR (0x1 << 28) /* Decrementing Mode */
+#define ATC_DST_ADDR_MODE_FIXED (0x2 << 28) /* Fixed Mode */
+#define ATC_IEN (0x1 << 30) /* BTC interrupt enable (active low) */
+#define ATC_AUTO (0x1 << 31) /* Auto multiple buffer tx enable */
+
+/* Bitfields in CFG */
+#define ATC_SRC_PER(h) (0xFU & (h)) /* Channel src rq associated with periph handshaking ifc h */
+#define ATC_DST_PER(h) ((0xFU & (h)) << 4) /* Channel dst rq associated with periph handshaking ifc h */
+#define ATC_SRC_REP (0x1 << 8) /* Source Replay Mod */
+#define ATC_SRC_H2SEL (0x1 << 9) /* Source Handshaking Mod */
+#define ATC_SRC_H2SEL_SW (0x0 << 9)
+#define ATC_SRC_H2SEL_HW (0x1 << 9)
+#define ATC_DST_REP (0x1 << 12) /* Destination Replay Mod */
+#define ATC_DST_H2SEL (0x1 << 13) /* Destination Handshaking Mod */
+#define ATC_DST_H2SEL_SW (0x0 << 13)
+#define ATC_DST_H2SEL_HW (0x1 << 13)
+#define ATC_SOD (0x1 << 16) /* Stop On Done */
+#define ATC_LOCK_IF (0x1 << 20) /* Interface Lock */
+#define ATC_LOCK_B (0x1 << 21) /* AHB Bus Lock */
+#define ATC_LOCK_IF_L (0x1 << 22) /* Master Interface Arbiter Lock */
+#define ATC_LOCK_IF_L_CHUNK (0x0 << 22)
+#define ATC_LOCK_IF_L_BUFFER (0x1 << 22)
+#define ATC_AHB_PROT_MASK (0x7 << 24) /* AHB Protection */
+#define ATC_FIFOCFG_MASK (0x3 << 28) /* FIFO Request Configuration */
+#define ATC_FIFOCFG_LARGESTBURST (0x0 << 28)
+#define ATC_FIFOCFG_HALFFIFO (0x1 << 28)
+#define ATC_FIFOCFG_ENOUGHSPACE (0x2 << 28)
+
+/* Bitfields in SPIP */
+#define ATC_SPIP_HOLE(x) (0xFFFFU & (x))
+#define ATC_SPIP_BOUNDARY(x) ((0x3FF & (x)) << 16)
+
+/* Bitfields in DPIP */
+#define ATC_DPIP_HOLE(x) (0xFFFFU & (x))
+#define ATC_DPIP_BOUNDARY(x) ((0x3FF & (x)) << 16)
+
+
+/*-- descriptors -----------------------------------------------------*/
+
+/* LLI == Linked List Item; aka DMA buffer descriptor */
+struct at_lli {
+ /* values that are not changed by hardware */
+ dma_addr_t saddr;
+ dma_addr_t daddr;
+ /* value that may get written back: */
+ u32 ctrla;
+ /* more values that are not changed by hardware */
+ u32 ctrlb;
+ dma_addr_t dscr; /* chain to next lli */
+};
+
+/**
+ * struct at_desc - software descriptor
+ * @at_lli: hardware lli structure
+ * @txd: support for the async_tx api
+ * @desc_node: node on the channed descriptors list
+ * @len: total transaction bytecount
+ */
+struct at_desc {
+ /* FIRST values the hardware uses */
+ struct at_lli lli;
+
+ /* THEN values for driver housekeeping */
+ struct dma_async_tx_descriptor txd;
+ struct list_head desc_node;
+ size_t len;
+};
+
+static inline struct at_desc *
+txd_to_at_desc(struct dma_async_tx_descriptor *txd)
+{
+ return container_of(txd, struct at_desc, txd);
+}
+
+
+/*-- Channels --------------------------------------------------------*/
+
+/**
+ * struct at_dma_chan - internal representation of an Atmel HDMAC channel
+ * @chan_common: common dmaengine channel object members
+ * @device: parent device
+ * @ch_regs: memory mapped register base
+ * @mask: channel index in a mask
+ * @error_status: transmit error status information from irq handler
+ * to tasklet (use atomic operations)
+ * @tasklet: bottom half to finish transaction work
+ * @lock: serializes enqueue/dequeue operations to descriptors lists
+ * @completed_cookie: identifier for the most recently completed operation
+ * @active_list: list of descriptors dmaengine is being running on
+ * @queue: list of descriptors ready to be submitted to engine
+ * @free_list: list of descriptors usable by the channel
+ * @descs_allocated: records the actual size of the descriptor pool
+ */
+struct at_dma_chan {
+ struct dma_chan chan_common;
+ struct at_dma *device;
+ void __iomem *ch_regs;
+ u8 mask;
+ unsigned long error_status;
+ struct tasklet_struct tasklet;
+
+ spinlock_t lock;
+
+ /* these other elements are all protected by lock */
+ dma_cookie_t completed_cookie;
+ struct list_head active_list;
+ struct list_head queue;
+ struct list_head free_list;
+ unsigned int descs_allocated;
+};
+
+#define channel_readl(atchan, name) \
+ __raw_readl((atchan)->ch_regs + ATC_##name##_OFFSET)
+
+#define channel_writel(atchan, name, val) \
+ __raw_writel((val), (atchan)->ch_regs + ATC_##name##_OFFSET)
+
+static inline struct at_dma_chan *to_at_dma_chan(struct dma_chan *dchan)
+{
+ return container_of(dchan, struct at_dma_chan, chan_common);
+}
+
+
+/*-- Controller ------------------------------------------------------*/
+
+/**
+ * struct at_dma - internal representation of an Atmel HDMA Controller
+ * @chan_common: common dmaengine dma_device object members
+ * @ch_regs: memory mapped register base
+ * @clk: dma controller clock
+ * @all_chan_mask: all channels availlable in a mask
+ * @dma_desc_pool: base of DMA descriptor region (DMA address)
+ * @chan: channels table to store at_dma_chan structures
+ */
+struct at_dma {
+ struct dma_device dma_common;
+ void __iomem *regs;
+ struct clk *clk;
+
+ u8 all_chan_mask;
+
+ struct dma_pool *dma_desc_pool;
+ /* AT THE END channels table */
+ struct at_dma_chan chan[0];
+};
+
+#define dma_readl(atdma, name) \
+ __raw_readl((atdma)->regs + AT_DMA_##name)
+#define dma_writel(atdma, name, val) \
+ __raw_writel((val), (atdma)->regs + AT_DMA_##name)
+
+static inline struct at_dma *to_at_dma(struct dma_device *ddev)
+{
+ return container_of(ddev, struct at_dma, dma_common);
+}
+
+
+/*-- Helper functions ------------------------------------------------*/
+
+static struct device *chan2dev(struct dma_chan *chan)
+{
+ return &chan->dev->device;
+}
+static struct device *chan2parent(struct dma_chan *chan)
+{
+ return chan->dev->device.parent;
+}
+
+#if defined(VERBOSE_DEBUG)
+static void vdbg_dump_regs(struct at_dma_chan *atchan)
+{
+ struct at_dma *atdma = to_at_dma(atchan->chan_common.device);
+
+ dev_err(chan2dev(&atchan->chan_common),
+ " channel %d : imr = 0x%x, chsr = 0x%x\n",
+ atchan->chan_common.chan_id,
+ dma_readl(atdma, EBCIMR),
+ dma_readl(atdma, CHSR));
+
+ dev_err(chan2dev(&atchan->chan_common),
+ " channel: s0x%x d0x%x ctrl0x%x:0x%x l0x%x\n",
+ channel_readl(atchan, SADDR),
+ channel_readl(atchan, DADDR),
+ channel_readl(atchan, CTRLA),
+ channel_readl(atchan, CTRLB),
+ channel_readl(atchan, DSCR));
+}
+#else
+static void vdbg_dump_regs(struct at_dma_chan *atchan) {}
+#endif
+
+static void atc_dump_lli(struct at_dma_chan *atchan, struct at_lli *lli)
+{
+ dev_printk(KERN_CRIT, chan2dev(&atchan->chan_common),
+ " desc: s0x%x d0x%x ctrl0x%x:0x%x l0x%x\n",
+ lli->saddr, lli->daddr,
+ lli->ctrla, lli->ctrlb, lli->dscr);
+}
+
+
+static void atc_setup_irq(struct at_dma_chan *atchan, int on)
+{
+ struct at_dma *atdma = to_at_dma(atchan->chan_common.device);
+ u32 ebci;
+
+ /* enable interrupts on buffer chain completion & error */
+ ebci = AT_DMA_CBTC(atchan->chan_common.chan_id)
+ | AT_DMA_ERR(atchan->chan_common.chan_id);
+ if (on)
+ dma_writel(atdma, EBCIER, ebci);
+ else
+ dma_writel(atdma, EBCIDR, ebci);
+}
+
+static inline void atc_enable_irq(struct at_dma_chan *atchan)
+{
+ atc_setup_irq(atchan, 1);
+}
+
+static inline void atc_disable_irq(struct at_dma_chan *atchan)
+{
+ atc_setup_irq(atchan, 0);
+}
+
+
+/**
+ * atc_chan_is_enabled - test if given channel is enabled
+ * @atchan: channel we want to test status
+ */
+static inline int atc_chan_is_enabled(struct at_dma_chan *atchan)
+{
+ struct at_dma *atdma = to_at_dma(atchan->chan_common.device);
+
+ return !!(dma_readl(atdma, CHSR) & atchan->mask);
+}
+
+
+/**
+ * set_desc_eol - set end-of-link to descriptor so it will end transfer
+ * @desc: descriptor, signle or at the end of a chain, to end chain on
+ */
+static void set_desc_eol(struct at_desc *desc)
+{
+ desc->lli.ctrlb |= ATC_SRC_DSCR_DIS | ATC_DST_DSCR_DIS;
+ desc->lli.dscr = 0;
+}
+
+#endif /* AT_HDMAC_REGS_H */
--
1.5.3.7

2009-07-17 10:38:37

by Nicolas Ferre

[permalink] [raw]
Subject: Re: [PATCH 1/2 v3] dmaengine: at_hdmac: new driver for the Atmel AHB DMA Controller

Nicolas Ferre :
> This AHB DMA Controller (aka HDMA or DMAC on AT91 systems) is availlable on
> at91sam9rl chip. It will be used on other products in the future.
>
> This first release covers only the memory-to-memory tranfer type. This is the
> only tranfer type supported by this chip. On other products, it will be used
> also for peripheral DMA transfer (slave API support to come).
>
> I used dmatest client without problem in different configurations to test it.
>
> Full documentation for this controller can be found in the SAM9RL datasheet:
> http://www.atmel.com/dyn/products/product_card.asp?part_id=4243
>
> Signed-off-by: Nicolas Ferre <[email protected]>
> ---
> v2 is here:
> http://lkml.org/lkml/2009/6/26/104
>
> v2 -> v3:
> - initial number of descriptors to allocate for each channel raised to 64 and
> is now a parameter
> - ack-bit in descriptor flag comment synchronized with TXx9 dma driver
> - atc_desc_get() when short on descriptors in pool: create one at a time
> - allocation flag changed to GFP_ATOMIC in atc_desc_get()
> - call to proper funtion while unmapping: use of new
> DMA_COMPL_{SRC,DEST}_UNMAP_SINGLE flags
> - call dma_run_dependencies() at the end of atc_chain_complete()

Dan, Maciej: Ping ?

I have a "slave" support on top of this patch and would like the initial
work to be accepted before going further...

Thanks, Bye,
--
Nicolas Ferre

2009-07-18 16:01:33

by Dan Williams

[permalink] [raw]
Subject: Re: [PATCH 1/2 v3] dmaengine: at_hdmac: new driver for the Atmel AHB DMA Controller

On Fri, Jul 3, 2009 at 10:24 AM, Nicolas Ferre<[email protected]> wrote:
> This AHB DMA Controller (aka HDMA or DMAC on AT91 systems) is availlable on
> at91sam9rl chip. It will be used on other products in the future.
>
> This first release covers only the memory-to-memory tranfer type. This is the
> only tranfer type supported by this chip. ?On other products, it will be used
> also for peripheral DMA transfer (slave API support to come).
>
> I used dmatest client without problem in different configurations to test it.
>
> Full documentation for this controller can be found in the SAM9RL datasheet:
> http://www.atmel.com/dyn/products/product_card.asp?part_id=4243
>
> Signed-off-by: Nicolas Ferre <[email protected]>
> ---
> v2 is here:
> http://lkml.org/lkml/2009/6/26/104
>
> v2 -> v3:
> ?- initial number of descriptors to allocate for each channel raised to 64 and
> ? is now a parameter
> ?- ack-bit in descriptor flag comment synchronized with TXx9 dma driver
> ?- atc_desc_get() when short on descriptors in pool: create one at a time
> ?- allocation flag changed to GFP_ATOMIC in atc_desc_get()
> ?- call to proper funtion while unmapping: use of new
> ? DMA_COMPL_{SRC,DEST}_UNMAP_SINGLE flags
> ?- call dma_run_dependencies() at the end of atc_chain_complete()
>

Looks good, but now I belatedly wonder if that GFP_ATOMIC should be
GFP_NOWAIT instead? Do we really want to consume from the system
emergency pools for these allocations (a similar fix is need for
ioatdma and fsldma)?

Thanks,
Dan

2009-07-21 08:34:49

by Nicolas Ferre

[permalink] [raw]
Subject: Re: [PATCH 1/2 v3] dmaengine: at_hdmac: new driver for the Atmel AHB DMA Controller

Dan Williams :
> On Fri, Jul 3, 2009 at 10:24 AM, Nicolas Ferre<[email protected]> wrote:
>> This AHB DMA Controller (aka HDMA or DMAC on AT91 systems) is availlable on
>> at91sam9rl chip. It will be used on other products in the future.
>>
>> This first release covers only the memory-to-memory tranfer type. This is the
>> only tranfer type supported by this chip. On other products, it will be used
>> also for peripheral DMA transfer (slave API support to come).
>>
>> I used dmatest client without problem in different configurations to test it.
>>
>> Full documentation for this controller can be found in the SAM9RL datasheet:
>> http://www.atmel.com/dyn/products/product_card.asp?part_id=4243
>>
>> Signed-off-by: Nicolas Ferre <[email protected]>
>> ---
>> v2 is here:
>> http://lkml.org/lkml/2009/6/26/104
>>
>> v2 -> v3:
>> - initial number of descriptors to allocate for each channel raised to 64 and
>> is now a parameter
>> - ack-bit in descriptor flag comment synchronized with TXx9 dma driver
>> - atc_desc_get() when short on descriptors in pool: create one at a time
>> - allocation flag changed to GFP_ATOMIC in atc_desc_get()
>> - call to proper funtion while unmapping: use of new
>> DMA_COMPL_{SRC,DEST}_UNMAP_SINGLE flags
>> - call dma_run_dependencies() at the end of atc_chain_complete()
>>
>
> Looks good, but now I belatedly wonder if that GFP_ATOMIC should be
> GFP_NOWAIT instead? Do we really want to consume from the system
> emergency pools for these allocations (a similar fix is need for
> ioatdma and fsldma)?

Seems sensible indeed but I know little about allocation flags.

What do you think about including the driver and then building a patch
that fixes this flag in all allocation functions at a time.
It may add exposure to this modification and maybe encourage people to
react...

Regards,
--
Nicolas Ferre

2009-07-22 13:35:18

by Sosnowski, Maciej

[permalink] [raw]
Subject: RE: [PATCH 1/2 v3] dmaengine: at_hdmac: new driver for the Atmel AHB DMA Controller

Nicolas Ferre wrote:
> Dan Williams :
>> On Fri, Jul 3, 2009 at 10:24 AM, Nicolas Ferre<[email protected]> wrote:
>>> This AHB DMA Controller (aka HDMA or DMAC on AT91 systems) is availlable on
>>> at91sam9rl chip. It will be used on other products in the future.
>>>
>>> This first release covers only the memory-to-memory tranfer type. This is the
>>> only tranfer type supported by this chip. On other products, it will be used
>>> also for peripheral DMA transfer (slave API support to come).
>>>
>>> I used dmatest client without problem in different configurations to test it.
>>>
>>> Full documentation for this controller can be found in the SAM9RL datasheet:
>>> http://www.atmel.com/dyn/products/product_card.asp?part_id=4243
>>>
>>> Signed-off-by: Nicolas Ferre <[email protected]> ---
>>> v2 is here:
>>> http://lkml.org/lkml/2009/6/26/104
>>>
>>> v2 -> v3:
>>> - initial number of descriptors to allocate for each channel raised to 64 and is now a
>>> parameter
>>> - ack-bit in descriptor flag comment synchronized with TXx9 dma driver
>>> - atc_desc_get() when short on descriptors in pool: create one at a time
>>> - allocation flag changed to GFP_ATOMIC in atc_desc_get()
>>> - call to proper funtion while unmapping: use of new
>>> DMA_COMPL_{SRC,DEST}_UNMAP_SINGLE flags
>>> - call dma_run_dependencies() at the end of atc_chain_complete()
>>>
>>
>> Looks good, but now I belatedly wonder if that GFP_ATOMIC should be
>> GFP_NOWAIT instead? Do we really want to consume from the system
>> emergency pools for these allocations (a similar fix is need for
>> ioatdma and fsldma)?
>
> Seems sensible indeed but I know little about allocation flags.
>
> What do you think about including the driver and then building a patch
> that fixes this flag in all allocation functions at a time.
> It may add exposure to this modification and maybe encourage people to
> react...

I think it's a good plan.
I will prepare similar patch on allocation flags for ioatdma.

For this v3 patch:

Acked-by: Maciej Sosnowski <[email protected]>

Regards,
Maciej-

2009-07-22 17:00:44

by Nicolas Ferre

[permalink] [raw]
Subject: [PATCH] dmaengine: at_hdmac: add DMA slave transfers

This patch for at_hdmac adds the slave transfers capability to the Atmel DMA
controller available on some AT91 SOCs. This allow peripheral to memory and
memory to peripheral transfers with hardware handshaking.

Slave structure for controller specific information is passed through channel
private data. This at_dma_slave structure is defined in at_hdmac.h header file
and relative hardware definition are moved to this file from at_hdmac_regs.h.
Doing this we allow the channel configuration from platform definition code.

This work is intensively based on dw_dmac and several slave implementations.

Signed-off-by: Nicolas Ferre <[email protected]>
---
This patch needs the previous at_hdmac driver inclusion:
"[PATCH 1/2 v3] dmaengine: at_hdmac: new driver for the Atmel AHB DMA Controller"
http://lkml.org/lkml/2009/7/3/332

arch/arm/mach-at91/include/mach/at_hdmac.h | 76 ++++++++++
drivers/dma/at_hdmac.c | 208 +++++++++++++++++++++++++++-
drivers/dma/at_hdmac_regs.h | 49 +------
3 files changed, 290 insertions(+), 43 deletions(-)

diff --git a/arch/arm/mach-at91/include/mach/at_hdmac.h b/arch/arm/mach-at91/include/mach/at_hdmac.h
index 21a5554..187cb58 100644
--- a/arch/arm/mach-at91/include/mach/at_hdmac.h
+++ b/arch/arm/mach-at91/include/mach/at_hdmac.h
@@ -23,4 +23,80 @@ struct at_dma_platform_data {
dma_cap_mask_t cap_mask;
};

+/**
+ * enum at_dma_slave_width - DMA slave register access width.
+ * @AT_DMA_SLAVE_WIDTH_8BIT: Do 8-bit slave register accesses
+ * @AT_DMA_SLAVE_WIDTH_16BIT: Do 16-bit slave register accesses
+ * @AT_DMA_SLAVE_WIDTH_32BIT: Do 32-bit slave register accesses
+ */
+enum at_dma_slave_width {
+ AT_DMA_SLAVE_WIDTH_8BIT = 0,
+ AT_DMA_SLAVE_WIDTH_16BIT,
+ AT_DMA_SLAVE_WIDTH_32BIT,
+};
+
+/**
+ * struct at_dma_slave - Controller-specific information about a slave
+ * @dma_dev: required DMA master device
+ * @tx_reg: physical address of data register used for
+ * memory-to-peripheral transfers
+ * @rx_reg: physical address of data register used for
+ * peripheral-to-memory transfers
+ * @reg_width: peripheral register width
+ * @cfg: Platform-specific initializer for the CFG register
+ * @ctrla: Platform-specific initializer for the CTRLA register
+ */
+struct at_dma_slave {
+ struct device *dma_dev;
+ dma_addr_t tx_reg;
+ dma_addr_t rx_reg;
+ enum at_dma_slave_width reg_width;
+ u32 cfg;
+ u32 ctrla;
+};
+
+
+/* Platform-configurable bits in CFG */
+#define ATC_SRC_PER(h) (0xFU & (h)) /* Channel src rq associated with periph handshaking ifc h */
+#define ATC_DST_PER(h) ((0xFU & (h)) << 4) /* Channel dst rq associated with periph handshaking ifc h */
+#define ATC_SRC_REP (0x1 << 8) /* Source Replay Mod */
+#define ATC_SRC_H2SEL (0x1 << 9) /* Source Handshaking Mod */
+#define ATC_SRC_H2SEL_SW (0x0 << 9)
+#define ATC_SRC_H2SEL_HW (0x1 << 9)
+#define ATC_DST_REP (0x1 << 12) /* Destination Replay Mod */
+#define ATC_DST_H2SEL (0x1 << 13) /* Destination Handshaking Mod */
+#define ATC_DST_H2SEL_SW (0x0 << 13)
+#define ATC_DST_H2SEL_HW (0x1 << 13)
+#define ATC_SOD (0x1 << 16) /* Stop On Done */
+#define ATC_LOCK_IF (0x1 << 20) /* Interface Lock */
+#define ATC_LOCK_B (0x1 << 21) /* AHB Bus Lock */
+#define ATC_LOCK_IF_L (0x1 << 22) /* Master Interface Arbiter Lock */
+#define ATC_LOCK_IF_L_CHUNK (0x0 << 22)
+#define ATC_LOCK_IF_L_BUFFER (0x1 << 22)
+#define ATC_AHB_PROT_MASK (0x7 << 24) /* AHB Protection */
+#define ATC_FIFOCFG_MASK (0x3 << 28) /* FIFO Request Configuration */
+#define ATC_FIFOCFG_LARGESTBURST (0x0 << 28)
+#define ATC_FIFOCFG_HALFFIFO (0x1 << 28)
+#define ATC_FIFOCFG_ENOUGHSPACE (0x2 << 28)
+
+/* Platform-configurable bits in CTRLA */
+#define ATC_SCSIZE_MASK (0x7 << 16) /* Source Chunk Transfer Size */
+#define ATC_SCSIZE_1 (0x0 << 16)
+#define ATC_SCSIZE_4 (0x1 << 16)
+#define ATC_SCSIZE_8 (0x2 << 16)
+#define ATC_SCSIZE_16 (0x3 << 16)
+#define ATC_SCSIZE_32 (0x4 << 16)
+#define ATC_SCSIZE_64 (0x5 << 16)
+#define ATC_SCSIZE_128 (0x6 << 16)
+#define ATC_SCSIZE_256 (0x7 << 16)
+#define ATC_DCSIZE_MASK (0x7 << 20) /* Destination Chunk Transfer Size */
+#define ATC_DCSIZE_1 (0x0 << 20)
+#define ATC_DCSIZE_4 (0x1 << 20)
+#define ATC_DCSIZE_8 (0x2 << 20)
+#define ATC_DCSIZE_16 (0x3 << 20)
+#define ATC_DCSIZE_32 (0x4 << 20)
+#define ATC_DCSIZE_64 (0x5 << 20)
+#define ATC_DCSIZE_128 (0x6 << 20)
+#define ATC_DCSIZE_256 (0x7 << 20)
+
#endif /* AT_HDMAC_H */
diff --git a/drivers/dma/at_hdmac.c b/drivers/dma/at_hdmac.c
index 9a4abf1..4435c82 100644
--- a/drivers/dma/at_hdmac.c
+++ b/drivers/dma/at_hdmac.c
@@ -608,6 +608,187 @@ err_desc_get:
return NULL;
}

+
+/**
+ * atc_prep_slave_sg - prepare descriptors for a DMA_SLAVE transaction
+ * @chan: DMA channel
+ * @sgl: scatterlist to transfer to/from
+ * @sg_len: number of entries in @scatterlist
+ * @direction: DMA direction
+ * @flags: tx descriptor status flags
+ */
+static struct dma_async_tx_descriptor *
+atc_prep_slave_sg(struct dma_chan *chan, struct scatterlist *sgl,
+ unsigned int sg_len, enum dma_data_direction direction,
+ unsigned long flags)
+{
+ struct at_dma_chan *atchan = to_at_dma_chan(chan);
+ struct at_dma_slave *atslave = chan->private;
+ struct at_desc *first = NULL;
+ struct at_desc *prev = NULL;
+ u32 ctrla;
+ u32 ctrlb;
+ dma_addr_t reg;
+ unsigned int reg_width;
+ unsigned int mem_width;
+ unsigned int i;
+ struct scatterlist *sg;
+ size_t total_len = 0;
+
+ dev_vdbg(chan2dev(chan), "prep_slave_sg: %s f0x%lx\n",
+ direction == DMA_TO_DEVICE ? "TO DEVICE" : "FROM DEVICE",
+ flags);
+
+ if (unlikely(!atslave || !sg_len)) {
+ dev_dbg(chan2dev(chan), "prep_dma_memcpy: length is zero!\n");
+ return NULL;
+ }
+
+ reg_width = atslave->reg_width;
+
+ sg_len = dma_map_sg(chan2parent(chan), sgl, sg_len, direction);
+
+ ctrla = ATC_DEFAULT_CTRLA | atslave->ctrla;
+ ctrlb = ATC_DEFAULT_CTRLB | ATC_IEN;
+
+ switch (direction) {
+ case DMA_TO_DEVICE:
+ ctrla |= ATC_DST_WIDTH(reg_width);
+ ctrlb |= ATC_DST_ADDR_MODE_FIXED
+ | ATC_SRC_ADDR_MODE_INCR
+ | ATC_FC_MEM2PER;
+ reg = atslave->tx_reg;
+ for_each_sg(sgl, sg, sg_len, i) {
+ struct at_desc *desc;
+ u32 len;
+ u32 mem;
+
+ desc = atc_desc_get(atchan);
+ if (!desc)
+ goto err_desc_get;
+
+ mem = sg_phys(sg);
+ len = sg_dma_len(sg);
+ mem_width = 2;
+ if (unlikely(mem & 3 || len & 3))
+ mem_width = 0;
+
+ desc->lli.saddr = mem;
+ desc->lli.daddr = reg;
+ desc->lli.ctrla = ctrla
+ | ATC_SRC_WIDTH(mem_width)
+ | len >> mem_width;
+ desc->lli.ctrlb = ctrlb;
+
+ if (!first) {
+ first = desc;
+ } else {
+ /* inform the HW lli about chaining */
+ prev->lli.dscr = desc->txd.phys;
+ /* insert the link descriptor to the LD ring */
+ list_add_tail(&desc->desc_node,
+ &first->txd.tx_list);
+ }
+ prev = desc;
+ total_len += len;
+ }
+ break;
+ case DMA_FROM_DEVICE:
+ ctrla |= ATC_SRC_WIDTH(reg_width);
+ ctrlb |= ATC_DST_ADDR_MODE_INCR
+ | ATC_SRC_ADDR_MODE_FIXED
+ | ATC_FC_PER2MEM;
+
+ reg = atslave->rx_reg;
+ for_each_sg(sgl, sg, sg_len, i) {
+ struct at_desc *desc;
+ u32 len;
+ u32 mem;
+
+ desc = atc_desc_get(atchan);
+ if (!desc)
+ goto err_desc_get;
+
+ mem = sg_phys(sg);
+ len = sg_dma_len(sg);
+ mem_width = 2;
+ if (unlikely(mem & 3 || len & 3))
+ mem_width = 0;
+
+ desc->lli.saddr = reg;
+ desc->lli.daddr = mem;
+ desc->lli.ctrla = ctrla
+ | ATC_DST_WIDTH(mem_width)
+ | len >> mem_width;
+ desc->lli.ctrlb = ctrlb;
+
+ if (!first) {
+ first = desc;
+ } else {
+ /* inform the HW lli about chaining */
+ prev->lli.dscr = desc->txd.phys;
+ /* insert the link descriptor to the LD ring */
+ list_add_tail(&desc->desc_node,
+ &first->txd.tx_list);
+ }
+ prev = desc;
+ total_len += len;
+ }
+ break;
+ default:
+ return NULL;
+ }
+
+ /* set end-of-link to the last link descriptor of list*/
+ set_desc_eol(prev);
+
+ /* First descriptor of the chain embedds additional information */
+ first->txd.cookie = -EBUSY;
+ first->len = total_len;
+
+ /* last link descriptor of list is responsible of flags */
+ prev->txd.flags = flags; /* client is in control of this ack */
+
+ return &first->txd;
+
+err_desc_get:
+ dev_err(chan2dev(chan), "not enough descriptors available\n");
+ atc_desc_put(atchan, first);
+ return NULL;
+}
+
+static void atc_terminate_all(struct dma_chan *chan)
+{
+ struct at_dma_chan *atchan = to_at_dma_chan(chan);
+ struct at_dma *atdma = to_at_dma(chan->device);
+ struct at_desc *desc, *_desc;
+ LIST_HEAD(list);
+
+ /*
+ * This is only called when something went wrong elsewhere, so
+ * we don't really care about the data. Just disable the
+ * channel. We still have to poll the channel enable bit due
+ * to AHB/HSB limitations.
+ */
+ spin_lock_bh(&atchan->lock);
+
+ dma_writel(atdma, CHDR, atchan->mask);
+
+ /* confirm that this channel is disabled */
+ while (dma_readl(atdma, CHSR) & atchan->mask)
+ cpu_relax();
+
+ /* active_list entries will end up before queued entries */
+ list_splice_init(&atchan->queue, &list);
+ list_splice_init(&atchan->active_list, &list);
+
+ spin_unlock_bh(&atchan->lock);
+
+ /* Flush all pending and queued descriptors */
+ list_for_each_entry_safe(desc, _desc, &list, desc_node)
+ atc_chain_complete(atchan, desc);
+}
+
/**
* atc_is_tx_complete - poll for transaction completion
* @chan: DMA channel
@@ -686,7 +867,9 @@ static int atc_alloc_chan_resources(struct dma_chan *chan)
struct at_dma_chan *atchan = to_at_dma_chan(chan);
struct at_dma *atdma = to_at_dma(chan->device);
struct at_desc *desc;
+ struct at_dma_slave *atslave;
int i;
+ u32 cfg;
LIST_HEAD(tmp_list);

dev_vdbg(chan2dev(chan), "alloc_chan_resources\n");
@@ -697,7 +880,23 @@ static int atc_alloc_chan_resources(struct dma_chan *chan)
return -EIO;
}

- /* have we already been set up? */
+ cfg = ATC_DEFAULT_CFG;
+
+ atslave = chan->private;
+ if (atslave) {
+ /*
+ * We need controller-specific data to set up slave
+ * transfers.
+ */
+ BUG_ON(!atslave->dma_dev || atslave->dma_dev != atdma->dma_common.dev);
+
+ /* if cfg configuration specified take it instad of default */
+ if (atslave->cfg)
+ cfg = atslave->cfg;
+ }
+
+ /* have we already been set up?
+ * reconfigure channel but no need to reallocate descriptors */
if (!list_empty(&atchan->free_list))
return atchan->descs_allocated;

@@ -719,7 +918,7 @@ static int atc_alloc_chan_resources(struct dma_chan *chan)
spin_unlock_bh(&atchan->lock);

/* channel parameters */
- channel_writel(atchan, CFG, ATC_DEFAULT_CFG);
+ channel_writel(atchan, CFG, cfg);

dev_dbg(chan2dev(chan),
"alloc_chan_resources: allocated %d descriptors\n",
@@ -888,6 +1087,11 @@ static int __init at_dma_probe(struct platform_device *pdev)
if (dma_has_cap(DMA_MEMCPY, atdma->dma_common.cap_mask))
atdma->dma_common.device_prep_dma_memcpy = atc_prep_dma_memcpy;

+ if (dma_has_cap(DMA_SLAVE, atdma->dma_common.cap_mask)) {
+ atdma->dma_common.device_prep_slave_sg = atc_prep_slave_sg;
+ atdma->dma_common.device_terminate_all = atc_terminate_all;
+ }
+
dma_writel(atdma, EN, AT_DMA_ENABLE);

dev_info(&pdev->dev, "Atmel AHB DMA Controller ( %s%s), %d channels\n",
diff --git a/drivers/dma/at_hdmac_regs.h b/drivers/dma/at_hdmac_regs.h
index ad2d4f4..4c972af 100644
--- a/drivers/dma/at_hdmac_regs.h
+++ b/drivers/dma/at_hdmac_regs.h
@@ -87,29 +87,14 @@
/* Bitfields in CTRLA */
#define ATC_BTSIZE_MAX 0xFFFFUL /* Maximum Buffer Transfer Size */
#define ATC_BTSIZE(x) (ATC_BTSIZE_MAX & (x)) /* Buffer Transfer Size */
-#define ATC_SCSIZE_MASK (0x7 << 16) /* Source Chunk Transfer Size */
-#define ATC_SCSIZE_1 (0x0 << 16)
-#define ATC_SCSIZE_4 (0x1 << 16)
-#define ATC_SCSIZE_8 (0x2 << 16)
-#define ATC_SCSIZE_16 (0x3 << 16)
-#define ATC_SCSIZE_32 (0x4 << 16)
-#define ATC_SCSIZE_64 (0x5 << 16)
-#define ATC_SCSIZE_128 (0x6 << 16)
-#define ATC_SCSIZE_256 (0x7 << 16)
-#define ATC_DCSIZE_MASK (0x7 << 20) /* Destination Chunk Transfer Size */
-#define ATC_DCSIZE_1 (0x0 << 20)
-#define ATC_DCSIZE_4 (0x1 << 20)
-#define ATC_DCSIZE_8 (0x2 << 20)
-#define ATC_DCSIZE_16 (0x3 << 20)
-#define ATC_DCSIZE_32 (0x4 << 20)
-#define ATC_DCSIZE_64 (0x5 << 20)
-#define ATC_DCSIZE_128 (0x6 << 20)
-#define ATC_DCSIZE_256 (0x7 << 20)
+/* Chunck Tranfer size definitions are in at_hdmac.h */
#define ATC_SRC_WIDTH_MASK (0x3 << 24) /* Source Single Transfer Size */
+#define ATC_SRC_WIDTH(x) ((x) << 24)
#define ATC_SRC_WIDTH_BYTE (0x0 << 24)
#define ATC_SRC_WIDTH_HALFWORD (0x1 << 24)
#define ATC_SRC_WIDTH_WORD (0x2 << 24)
#define ATC_DST_WIDTH_MASK (0x3 << 28) /* Destination Single Transfer Size */
+#define ATC_DST_WIDTH(x) ((x) << 28)
#define ATC_DST_WIDTH_BYTE (0x0 << 28)
#define ATC_DST_WIDTH_HALFWORD (0x1 << 28)
#define ATC_DST_WIDTH_WORD (0x2 << 28)
@@ -129,7 +114,8 @@
#define ATC_FC_PER2PER (0x3 << 21) /* Periph-to-Periph (DMA) */
#define ATC_FC_PER2MEM_PER (0x4 << 21) /* Periph-to-Mem (Peripheral) */
#define ATC_FC_MEM2PER_PER (0x5 << 21) /* Mem-to-Periph (Peripheral) */
-#define ATC_FC_PER2PER_PER (0x6 << 21) /* Periph-to-Periph (Src Peripheral) */
+#define ATC_FC_PER2PER_SRCPER (0x6 << 21) /* Periph-to-Periph (Src Peripheral) */
+#define ATC_FC_PER2PER_DSTPER (0x7 << 21) /* Periph-to-Periph (Dst Peripheral) */
#define ATC_SRC_ADDR_MODE_MASK (0x3 << 24)
#define ATC_SRC_ADDR_MODE_INCR (0x0 << 24) /* Incrementing Mode */
#define ATC_SRC_ADDR_MODE_DECR (0x1 << 24) /* Decrementing Mode */
@@ -142,27 +128,7 @@
#define ATC_AUTO (0x1 << 31) /* Auto multiple buffer tx enable */

/* Bitfields in CFG */
-#define ATC_SRC_PER(h) (0xFU & (h)) /* Channel src rq associated with periph handshaking ifc h */
-#define ATC_DST_PER(h) ((0xFU & (h)) << 4) /* Channel dst rq associated with periph handshaking ifc h */
-#define ATC_SRC_REP (0x1 << 8) /* Source Replay Mod */
-#define ATC_SRC_H2SEL (0x1 << 9) /* Source Handshaking Mod */
-#define ATC_SRC_H2SEL_SW (0x0 << 9)
-#define ATC_SRC_H2SEL_HW (0x1 << 9)
-#define ATC_DST_REP (0x1 << 12) /* Destination Replay Mod */
-#define ATC_DST_H2SEL (0x1 << 13) /* Destination Handshaking Mod */
-#define ATC_DST_H2SEL_SW (0x0 << 13)
-#define ATC_DST_H2SEL_HW (0x1 << 13)
-#define ATC_SOD (0x1 << 16) /* Stop On Done */
-#define ATC_LOCK_IF (0x1 << 20) /* Interface Lock */
-#define ATC_LOCK_B (0x1 << 21) /* AHB Bus Lock */
-#define ATC_LOCK_IF_L (0x1 << 22) /* Master Interface Arbiter Lock */
-#define ATC_LOCK_IF_L_CHUNK (0x0 << 22)
-#define ATC_LOCK_IF_L_BUFFER (0x1 << 22)
-#define ATC_AHB_PROT_MASK (0x7 << 24) /* AHB Protection */
-#define ATC_FIFOCFG_MASK (0x3 << 28) /* FIFO Request Configuration */
-#define ATC_FIFOCFG_LARGESTBURST (0x0 << 28)
-#define ATC_FIFOCFG_HALFFIFO (0x1 << 28)
-#define ATC_FIFOCFG_ENOUGHSPACE (0x2 << 28)
+/* are in at_hdmac.h */

/* Bitfields in SPIP */
#define ATC_SPIP_HOLE(x) (0xFFFFU & (x))
@@ -316,11 +282,12 @@ static void vdbg_dump_regs(struct at_dma_chan *atchan)
dma_readl(atdma, CHSR));

dev_err(chan2dev(&atchan->chan_common),
- " channel: s0x%x d0x%x ctrl0x%x:0x%x l0x%x\n",
+ " channel: s0x%x d0x%x ctrl0x%x:0x%x cfg0x%x l0x%x\n",
channel_readl(atchan, SADDR),
channel_readl(atchan, DADDR),
channel_readl(atchan, CTRLA),
channel_readl(atchan, CTRLB),
+ channel_readl(atchan, CFG),
channel_readl(atchan, DSCR));
}
#else
--
1.5.3.7

2009-07-23 17:13:34

by Dan Williams

[permalink] [raw]
Subject: Re: [PATCH] dmaengine: at_hdmac: add DMA slave transfers

On Wed, Jul 22, 2009 at 11:04 AM, Nicolas Ferre<[email protected]> wrote:
> This patch for at_hdmac adds the slave transfers capability to the Atmel DMA
> controller available on some AT91 SOCs. This allow peripheral to memory and
> memory to peripheral transfers with hardware handshaking.
>
> Slave structure for controller specific information is passed through channel
> private data. This at_dma_slave structure is defined in at_hdmac.h header file
> and relative hardware definition are moved to this file from at_hdmac_regs.h.
> Doing this we allow the channel configuration from platform definition code.
>
> This work is intensively based on dw_dmac and several slave implementations.
>
> Signed-off-by: Nicolas Ferre <[email protected]>
> ---

Seems straightforward and self contained, applied. Just curious which
drivers will take advantage of this slave capability? As it stands I
do not see any in-tree users of at_hdmac beyond dmatest?

Thanks,
Dan

2009-07-24 07:35:57

by Nicolas Ferre

[permalink] [raw]
Subject: Re: [PATCH] dmaengine: at_hdmac: add DMA slave transfers

Dan Williams :
> On Wed, Jul 22, 2009 at 11:04 AM, Nicolas Ferre<[email protected]> wrote:
>> This patch for at_hdmac adds the slave transfers capability to the Atmel DMA
>> controller available on some AT91 SOCs. This allow peripheral to memory and
>> memory to peripheral transfers with hardware handshaking.
>>
>> Slave structure for controller specific information is passed through channel
>> private data. This at_dma_slave structure is defined in at_hdmac.h header file
>> and relative hardware definition are moved to this file from at_hdmac_regs.h.
>> Doing this we allow the channel configuration from platform definition code.
>>
>> This work is intensively based on dw_dmac and several slave implementations.
>>
>> Signed-off-by: Nicolas Ferre <[email protected]>
>> ---
>
> Seems straightforward and self contained, applied. Just curious which
> drivers will take advantage of this slave capability? As it stands I
> do not see any in-tree users of at_hdmac beyond dmatest?

Well, in fact the same as dw_dmac: atmel-mci. Adaptation of other
drivers are coming: probably ac97 and spi interfaces.

The patches for adaptation of atmel-mci to at_hdmac are coming...

Best regards,
--
Nicolas Ferre

2009-07-24 13:50:16

by Atsushi Nemoto

[permalink] [raw]
Subject: Re: [PATCH] dmaengine: at_hdmac: add DMA slave transfers

On Fri, 24 Jul 2009 09:35:37 +0200, Nicolas Ferre <[email protected]> wrote:
> > Seems straightforward and self contained, applied. Just curious which
> > drivers will take advantage of this slave capability? As it stands I
> > do not see any in-tree users of at_hdmac beyond dmatest?
>
> Well, in fact the same as dw_dmac: atmel-mci. Adaptation of other
> drivers are coming: probably ac97 and spi interfaces.
>
> The patches for adaptation of atmel-mci to at_hdmac are coming...

Your atc_chain_complete() calls dma_unmap_xxx unless
DMA_COMPL_SKIP_XXX_UNMAP specified. But atmel-mci driver does not set
the flag on dma_async_tx_descriptor. I suppose one of them should be
fixed.

It is not clear for me how dma driver should handle unmapping on
completion of slave DMA. Are there any consensus on it?

I enclosed unmapping code of txx9dmac driver by "if (!slave)" and did
not use DMA_COMPL_SKIP_XXX_UNMAP flag in its client driver (txx9aclc
ASoC driver). If consensus made on how to handle unmapping on slave
DMA, I will follow it.

---
Atsushi Nemoto

2009-07-24 14:10:24

by Haavard Skinnemoen

[permalink] [raw]
Subject: Re: [PATCH] dmaengine: at_hdmac: add DMA slave transfers

On Fri, 24 Jul 2009 22:29:19 +0900 (JST)
Atsushi Nemoto <[email protected]> wrote:

> > The patches for adaptation of atmel-mci to at_hdmac are coming...
>
> Your atc_chain_complete() calls dma_unmap_xxx unless
> DMA_COMPL_SKIP_XXX_UNMAP specified. But atmel-mci driver does not set
> the flag on dma_async_tx_descriptor. I suppose one of them should be
> fixed.

atmel-mci should definitely set that flag.

Haavard

2009-07-27 13:23:44

by Sosnowski, Maciej

[permalink] [raw]
Subject: RE: [PATCH] dmaengine: at_hdmac: add DMA slave transfers

Nicolas Ferre wrote:
> This patch for at_hdmac adds the slave transfers capability to the Atmel DMA
> controller available on some AT91 SOCs. This allow peripheral to memory and
> memory to peripheral transfers with hardware handshaking.
>
> Slave structure for controller specific information is passed through channel
> private data. This at_dma_slave structure is defined in at_hdmac.h header file
> and relative hardware definition are moved to this file from at_hdmac_regs.h.
> Doing this we allow the channel configuration from platform definition code.
>
> This work is intensively based on dw_dmac and several slave implementations.
>
> Signed-off-by: Nicolas Ferre <[email protected]>
> ---

Acked-by: Maciej Sosnowski <[email protected]>-

2009-07-27 13:25:00

by Sosnowski, Maciej

[permalink] [raw]
Subject: RE: [PATCH] dmaengine: at_hdmac: add DMA slave transfers

Haavard Skinnemoen wrote:
> On Fri, 24 Jul 2009 22:29:19 +0900 (JST)
> Atsushi Nemoto <[email protected]> wrote:
>
>>> The patches for adaptation of atmel-mci to at_hdmac are coming...
>>
>> Your atc_chain_complete() calls dma_unmap_xxx unless
>> DMA_COMPL_SKIP_XXX_UNMAP specified. But atmel-mci driver does not set
>> the flag on dma_async_tx_descriptor. I suppose one of them should be
>> fixed.
>
> atmel-mci should definitely set that flag.
>
> Haavard

I agree with Haavard.

Regards,
Maciej-

2009-07-28 15:18:34

by Atsushi Nemoto

[permalink] [raw]
Subject: Re: [PATCH] dmaengine: at_hdmac: add DMA slave transfers

On Mon, 27 Jul 2009 14:24:26 +0100, "Sosnowski, Maciej" <[email protected]> wrote:
> >> Your atc_chain_complete() calls dma_unmap_xxx unless
> >> DMA_COMPL_SKIP_XXX_UNMAP specified. But atmel-mci driver does not set
> >> the flag on dma_async_tx_descriptor. I suppose one of them should be
> >> fixed.
> >
> > atmel-mci should definitely set that flag.
> >
> > Haavard
>
> I agree with Haavard.

Then, what should dma driver do when client driver did not set these
flags? If it should call dma_unmap_sg(), the dma driver should keep
sg and direction somewhere...

Also, calling dma_map_sg() in its prep_slave_sg function will not fit
for sound drivers, which use DMA buffers prepared in its framework.

For slave DMA, doing all mapping/unmapping in DMA client is better,
isn't it?

---
Atsushi Nemoto

2009-07-29 13:05:55

by Sosnowski, Maciej

[permalink] [raw]
Subject: RE: [PATCH] dmaengine: at_hdmac: add DMA slave transfers

Atsushi Nemoto wrote:
> On Mon, 27 Jul 2009 14:24:26 +0100, "Sosnowski, Maciej" <[email protected]> wrote:
>>>> Your atc_chain_complete() calls dma_unmap_xxx unless
>>>> DMA_COMPL_SKIP_XXX_UNMAP specified. But atmel-mci driver does not set
>>>> the flag on dma_async_tx_descriptor. I suppose one of them should be
>>>> fixed.
>>>
>>> atmel-mci should definitely set that flag.
>>>
>>> Haavard
>>
>> I agree with Haavard.
>
> Then, what should dma driver do when client driver did not set these
> flags? If it should call dma_unmap_sg(), the dma driver should keep
> sg and direction somewhere...

Well, what about BUG_ON?

>
> Also, calling dma_map_sg() in its prep_slave_sg function will not fit
> for sound drivers, which use DMA buffers prepared in its framework.

It looks like a need for one more flag: DMA_SKIP_MAP_SG.
What do you think?

Regards,
Maciej

2009-07-29 14:28:03

by Dan Williams

[permalink] [raw]
Subject: Re: [PATCH] dmaengine: at_hdmac: add DMA slave transfers

Atsushi Nemoto wrote:
> On Mon, 27 Jul 2009 14:24:26 +0100, "Sosnowski, Maciej" <[email protected]> wrote:
>>>> Your atc_chain_complete() calls dma_unmap_xxx unless
>>>> DMA_COMPL_SKIP_XXX_UNMAP specified. But atmel-mci driver does not set
>>>> the flag on dma_async_tx_descriptor. I suppose one of them should be
>>>> fixed.
>>> atmel-mci should definitely set that flag.
>>>
>>> Haavard
>> I agree with Haavard.
>
> Then, what should dma driver do when client driver did not set these
> flags? If it should call dma_unmap_sg(), the dma driver should keep
> sg and direction somewhere...
>
> Also, calling dma_map_sg() in its prep_slave_sg function will not fit
> for sound drivers, which use DMA buffers prepared in its framework.
>
> For slave DMA, doing all mapping/unmapping in DMA client is better,
> isn't it?
>

Yes it is. The whole point of the dma driver doing its own unmapping is
specifically for clients that use the async_tx api, i.e. where they do
not know if the transaction is carried out in hardware and have no way
of tracking the details necessary to perform the unmap. DMA-slave
clients request specific channels and know the hardware details at a low
level, so it should not be too high an expectation to push dma mapping
responsibility to the client.

--
Dan

2009-07-29 15:21:36

by Atsushi Nemoto

[permalink] [raw]
Subject: Re: [PATCH] dmaengine: at_hdmac: add DMA slave transfers

On Wed, 29 Jul 2009 14:05:17 +0100, "Sosnowski, Maciej" <[email protected]> wrote:
> Atsushi Nemoto wrote:
> > On Mon, 27 Jul 2009 14:24:26 +0100, "Sosnowski, Maciej" <[email protected]> wrote:
> >>>> Your atc_chain_complete() calls dma_unmap_xxx unless
> >>>> DMA_COMPL_SKIP_XXX_UNMAP specified. But atmel-mci driver does not set
> >>>> the flag on dma_async_tx_descriptor. I suppose one of them should be
> >>>> fixed.
> >>>
> >>> atmel-mci should definitely set that flag.
> >>>
> >>> Haavard
> >>
> >> I agree with Haavard.
> >
> > Then, what should dma driver do when client driver did not set these
> > flags? If it should call dma_unmap_sg(), the dma driver should keep
> > sg and direction somewhere...
>
> Well, what about BUG_ON?

Or prep_slave_sg can preset both DMA_COMPL_SKIP_XXX_UNMAP flags, or do
not unmap in dma driver for slave channel. I suppose latter is safe
and simple.

> > Also, calling dma_map_sg() in its prep_slave_sg function will not fit
> > for sound drivers, which use DMA buffers prepared in its framework.
>
> It looks like a need for one more flag: DMA_SKIP_MAP_SG.
> What do you think?

I do not think it is worth to introduce new flag.

As Dan said, pushing dma mapping responsibility to the client looks
simple and clean way for me.

---
Atsushi Nemoto

2009-07-31 04:03:56

by Atsushi Nemoto

[permalink] [raw]
Subject: Re: [PATCH] dmaengine: at_hdmac: add DMA slave transfers

On Wed, 29 Jul 2009 07:27:59 -0700, Dan Williams <[email protected]> wrote:
> > Then, what should dma driver do when client driver did not set these
> > flags? If it should call dma_unmap_sg(), the dma driver should keep
> > sg and direction somewhere...
> >
> > Also, calling dma_map_sg() in its prep_slave_sg function will not fit
> > for sound drivers, which use DMA buffers prepared in its framework.
> >
> > For slave DMA, doing all mapping/unmapping in DMA client is better,
> > isn't it?
>
> Yes it is. The whole point of the dma driver doing its own unmapping is
> specifically for clients that use the async_tx api, i.e. where they do
> not know if the transaction is carried out in hardware and have no way
> of tracking the details necessary to perform the unmap. DMA-slave
> clients request specific channels and know the hardware details at a low
> level, so it should not be too high an expectation to push dma mapping
> responsibility to the client.

OK, I will send a patch to move all map_sg/unmap_sg for slave channel
to its client. Affected drivers are at_hdmac, dw_dmac and atmel-mci.

I do not have any of them, I hope someone who have them can test it.
Thank you.

---
Atsushi Nemoto