Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S934380AbZFLVvX (ORCPT ); Fri, 12 Jun 2009 17:51:23 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S1762091AbZFLVvL (ORCPT ); Fri, 12 Jun 2009 17:51:11 -0400 Received: from g1t0026.austin.hp.com ([15.216.28.33]:41600 "EHLO g1t0026.austin.hp.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1760049AbZFLVvK (ORCPT ); Fri, 12 Jun 2009 17:51:10 -0400 Subject: Re: [PATCH V4: 2/3] pci: Provide aer support when the error source id is bad From: Andrew Patterson To: "Zhang, Yanmin" Cc: linux-kernel@vger.kernel.org, linux-pci@vger.kernel.org, Jesse Barnes In-Reply-To: <1244776113.2560.320.camel@ymzhang> References: <1244776113.2560.320.camel@ymzhang> Content-Type: text/plain; charset="UTF-8" Date: Fri, 12 Jun 2009 21:51:07 +0000 Message-Id: <1244843467.19708.91.camel@grinch> Mime-Version: 1.0 X-Mailer: Evolution 2.26.1 Content-Transfer-Encoding: 8bit Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 10262 Lines: 319 On Fri, 2009-06-12 at 11:08 +0800, Zhang, Yanmin wrote: > When the bus id part of error source id is equal to 0 or nosourceid=1, > kernel probes the AER status registers of all devices under the root port > to find the initial error reporter. > > Signed-off-by: Zhang Yanmin > > --- > > --- linux-2.6_next_pciwalk/drivers/pci/pcie/aer/aerdrv_core.c 2009-06-12 05:31:53.000000000 +0800 > +++ linux-2.6_next_aernoid/drivers/pci/pcie/aer/aerdrv_core.c 2009-06-12 05:39:24.000000000 +0800 > @@ -26,7 +26,9 @@ > #include "aerdrv.h" > > static int forceload; > +static int nosourceid; > module_param(forceload, bool, 0); > +module_param(nosourceid, bool, 0); > > int pci_enable_pcie_error_reporting(struct pci_dev *dev) > { > @@ -143,34 +145,82 @@ static void set_downstream_devices_error > pci_walk_bus(dev->subordinate, set_device_error_reporting, &enable); > } > > -static int find_device_iter(struct device *device, void *data) > +static int compare_device_id(struct pci_dev *dev, struct aer_err_info *e_info) > { > - struct pci_dev *dev; > - u16 id = *(unsigned long *)data; > - u8 secondary, subordinate, d_bus = id >> 8; > - > - if (device->bus == &pci_bus_type) { > - dev = to_pci_dev(device); > - if (id == ((dev->bus->number << 8) | dev->devfn)) { > - /* > - * Device ID match > - */ > - *(unsigned long*)data = (unsigned long)device; > - return 1; > - } > - > + if (e_info->id == ((dev->bus->number << 8) | dev->devfn)) { > /* > - * If device is P2P, check if it is an upstream? > + * Device ID match > */ > - if (dev->hdr_type & PCI_HEADER_TYPE_BRIDGE) { > - pci_read_config_byte(dev, PCI_SECONDARY_BUS, > - &secondary); > - pci_read_config_byte(dev, PCI_SUBORDINATE_BUS, > - &subordinate); > - if (d_bus >= secondary && d_bus <= subordinate) { > - *(unsigned long*)data = (unsigned long)device; > - return 1; > - } > + e_info->dev = dev; > + return 1; > + } > + > + return 0; > +} > + > +#define PCI_BUS(x) (((x) >> 8) & 0xff) > + > +static int find_device_iter(struct pci_dev *dev, void *data) > +{ > + int pos; > + u32 status; > + u32 mask; > + u16 reg16; > + struct aer_err_info *e_info = (struct aer_err_info *)data; > + > + /* > + * When bus id is equal to 0, it might be a bad id > + * reported by root port. > + */ > + if (!nosourceid && (PCI_BUS(e_info->id) != 0)) > + return compare_device_id(dev, e_info); > + > + /* > + * Next is to check when bus id is equal to 0 or > + * nosourceid==y. Some ports might lose the bus > + * id of error source id. We check AER status > + * registers to find the initial reporter. > + */ > + if (atomic_read(&dev->enable_cnt) == 0) > + return 0; > + pos = pci_find_capability(dev, PCI_CAP_ID_EXP); > + if (!pos) > + return 0; > + /* Check if AER is enabled */ > + pci_read_config_word(dev, pos+PCI_EXP_DEVCTL, ®16); > + if (!(reg16 & ( > + PCI_EXP_DEVCTL_CERE | > + PCI_EXP_DEVCTL_NFERE | > + PCI_EXP_DEVCTL_FERE | > + PCI_EXP_DEVCTL_URRE))) > + return 0; > + pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR); > + if (!pos) > + return 0; > + > + status = 0; > + mask = 0; > + if (e_info->severity == AER_CORRECTABLE) { > + pci_read_config_dword(dev, > + pos + PCI_ERR_COR_STATUS, > + &status); > + pci_read_config_dword(dev, > + pos + PCI_ERR_COR_MASK, > + &mask); > + if (status & ERR_CORRECTABLE_ERROR_MASK & ~mask) { > + e_info->dev = dev; > + return 1; > + } > + } else { > + pci_read_config_dword(dev, > + pos + PCI_ERR_UNCOR_STATUS, > + &status); > + pci_read_config_dword(dev, > + pos + PCI_ERR_UNCOR_MASK, > + &mask); > + if (status & ERR_UNCORRECTABLE_ERROR_MASK & ~mask) { > + e_info->dev = dev; > + return 1; > } > } > > @@ -180,33 +230,23 @@ static int find_device_iter(struct devic > /** > * find_source_device - search through device hierarchy for source device > * @parent: pointer to Root Port pci_dev data structure > - * @id: device ID of agent who sends an error message to this Root Port > + * @err_info: including detailed error information such like id > * > * Invoked when error is detected at the Root Port. > */ > -static struct device* find_source_device(struct pci_dev *parent, u16 id) > +static void find_source_device(struct pci_dev *parent, > + struct aer_err_info *e_info) > { > struct pci_dev *dev = parent; > - struct device *device; > - unsigned long device_addr; > - int status; > + int result; > > /* Is Root Port an agent that sends error message? */ > - if (id == ((dev->bus->number << 8) | dev->devfn)) > - return &dev->dev; > - > - do { > - device_addr = id; > - if ((status = device_for_each_child(&dev->dev, > - &device_addr, find_device_iter))) { > - device = (struct device*)device_addr; > - dev = to_pci_dev(device); > - if (id == ((dev->bus->number << 8) | dev->devfn)) > - return device; > - } > - }while (status); > + result = find_device_iter(dev, e_info); > + if (result) > + return; > > - return NULL; > + pci_walk_bus(parent->subordinate, find_device_iter, e_info); Minor nit. This return is redundant. > + return; > } > Minor. The e_info goes to pointer changes probably belong in a separate patch. > static int report_error_detected(struct pci_dev *dev, void *data) > @@ -501,12 +541,12 @@ static pci_ers_result_t do_recovery(stru > */ > static void handle_error_source(struct pcie_device * aerdev, > struct pci_dev *dev, > - struct aer_err_info info) > + struct aer_err_info *info) > { > pci_ers_result_t status = 0; > int pos; > > - if (info.severity == AER_CORRECTABLE) { > + if (info->severity == AER_CORRECTABLE) { > /* > * Correctable error does not need software intevention. > * No need to go through error recovery process. > @@ -514,9 +554,9 @@ static void handle_error_source(struct p > pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR); > if (pos) > pci_write_config_dword(dev, pos + PCI_ERR_COR_STATUS, > - info.status); > + info->status); > } else { > - status = do_recovery(aerdev, dev, info.severity); > + status = do_recovery(aerdev, dev, info->severity); > if (status == PCI_ERS_RESULT_RECOVERED) { > dev_printk(KERN_DEBUG, &dev->dev, "AER driver " > "successfully recovered\n"); > @@ -673,10 +713,13 @@ static int get_device_error_info(struct > static void aer_isr_one_error(struct pcie_device *p_device, > struct aer_err_source *e_src) > { > - struct device *s_device; > - struct aer_err_info e_info = {0, 0, 0,}; > + struct aer_err_info *e_info; > int i; > - u16 id; > + > + /* struct aer_err_info might be big, so we allocate it with slab */ > + e_info = kmalloc(sizeof(struct aer_err_info), GFP_KERNEL); > + if (e_info == NULL) > + return; > Should we get some sort of warning here? We might be concealing PCIe errors when out of memory. > /* > * There is a possibility that both correctable error and > @@ -688,31 +731,37 @@ static void aer_isr_one_error(struct pci > if (!(e_src->status & i)) > continue; > > + memset(e_info, 0, sizeof(struct aer_err_info)); > + > /* Init comprehensive error information */ > if (i & PCI_ERR_ROOT_COR_RCV) { > - id = ERR_COR_ID(e_src->id); > - e_info.severity = AER_CORRECTABLE; > + e_info->id = ERR_COR_ID(e_src->id); > + e_info->severity = AER_CORRECTABLE; > } else { > - id = ERR_UNCOR_ID(e_src->id); > - e_info.severity = ((e_src->status >> 6) & 1); > + e_info->id = ERR_UNCOR_ID(e_src->id); > + e_info->severity = ((e_src->status >> 6) & 1); > } > if (e_src->status & > (PCI_ERR_ROOT_MULTI_COR_RCV | > PCI_ERR_ROOT_MULTI_UNCOR_RCV)) > - e_info.flags |= AER_MULTI_ERROR_VALID_FLAG; > - if (!(s_device = find_source_device(p_device->port, id))) { > + e_info->flags |= AER_MULTI_ERROR_VALID_FLAG; > + > + find_source_device(p_device->port, e_info); > + if (e_info->dev == NULL) { > printk(KERN_DEBUG "%s->can't find device of ID%04x\n", > - __func__, id); > + __func__, e_info->id); > continue; > } > - if (get_device_error_info(to_pci_dev(s_device), &e_info) == > + if (get_device_error_info(e_info->dev, e_info) == > AER_SUCCESS) { > - aer_print_error(to_pci_dev(s_device), &e_info); > + aer_print_error(e_info->dev, e_info); > handle_error_source(p_device, > - to_pci_dev(s_device), > + e_info->dev, > e_info); > } > } > + > + kfree(e_info); > } > > /** > --- linux-2.6_next_pciwalk/drivers/pci/pcie/aer/aerdrv.h 2009-06-12 05:27:46.000000000 +0800 > +++ linux-2.6_next_aernoid/drivers/pci/pcie/aer/aerdrv.h 2009-06-12 05:39:24.000000000 +0800 > @@ -58,6 +58,8 @@ struct header_log_regs { > }; > > struct aer_err_info { > + struct pci_dev *dev; > + u16 id; > int severity; /* 0:NONFATAL | 1:FATAL | 2:COR */ > int flags; > unsigned int status; /* COR/UNCOR Error Status */ > --- linux-2.6_next_pciwalk/Documentation/PCI/pcieaer-howto.txt 2009-06-12 05:27:40.000000000 +0800 > +++ linux-2.6_next_aernoid/Documentation/PCI/pcieaer-howto.txt 2009-06-12 05:39:24.000000000 +0800 > @@ -60,6 +60,8 @@ forceload, a parameter of type bool, cou > be initiated although firmwares have no _OSC support. To enable the > walkaround, pls. add aerdriver.forceload=y to kernel boot parameter line > when booting kernel. Note that forceload=n by default. We need a new paragraph here. > +nosourceid, another parameter of type bool, is used when hardware (mostly > +chipset) has errata that root ports couldn't get error reporting source id. > Perhaps ".. can be used when broken hardware (mostly chipsets) has root ports that cannot obtain the reporting source ID." > 2.3 AER error output > When a PCI-E AER error is captured, an error message will be outputed to > > > -- > To unsubscribe from this list: send the line "unsubscribe linux-pci" in > the body of a message to majordomo@vger.kernel.org > More majordomo info at http://vger.kernel.org/majordomo-info.html > -- Andrew Patterson Hewlett-Packard -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/