| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-ID: <09cf1319-619d-4a6b-97f7-188c1a73aea0@linux.alibaba.com>
Date: Wed, 28 Jan 2026 20:27:31 +0800
From: Shuai Xue <xueshuai@...ux.alibaba.com>
To: Jonathan Cameron <jonathan.cameron@...wei.com>
Cc: linux-pci@...r.kernel.org, linux-kernel@...r.kernel.org,
linuxppc-dev@...ts.ozlabs.org, bhelgaas@...gle.com, kbusch@...nel.org,
sathyanarayanan.kuppuswamy@...ux.intel.com, mahesh@...ux.ibm.com,
oohall@...il.com, terry.bowman@....com, tianruidong@...ux.alibaba.com,
lukas@...ner.de
Subject: Re: [PATCH v7 2/5] PCI/DPC: Run recovery on device that detected the
error
On 1/27/26 6:24 PM, Jonathan Cameron wrote:
> On Sat, 24 Jan 2026 15:45:54 +0800
> Shuai Xue <xueshuai@...ux.alibaba.com> wrote:
>
>> The current implementation of pcie_do_recovery() assumes that the
>> recovery process is executed for the device that detected the error.
>> However, the DPC driver currently passes the error port that experienced
>> the DPC event to pcie_do_recovery().
>>
>> Use the SOURCE ID register to correctly identify the device that
>> detected the error. When passing the error device, the
>> pcie_do_recovery() will find the upstream bridge and walk bridges
>> potentially AER affected. And subsequent commits will be able to
>> accurately access AER status of the error device.
>>
>> Should not observe any functional changes.
>>
>> Reviewed-by: Kuppuswamy Sathyanarayanan <sathyanarayanan.kuppuswamy@...ux.intel.com>
>> Signed-off-by: Shuai Xue <xueshuai@...ux.alibaba.com>
>> ---
>> drivers/pci/pci.h | 2 +-
>> drivers/pci/pcie/dpc.c | 25 +++++++++++++++++++++----
>> drivers/pci/pcie/edr.c | 7 ++++---
>> 3 files changed, 26 insertions(+), 8 deletions(-)
>>
>> diff --git a/drivers/pci/pci.h b/drivers/pci/pci.h
>> index 0e67014aa001..58640e656897 100644
>> --- a/drivers/pci/pci.h
>> +++ b/drivers/pci/pci.h
>> @@ -771,7 +771,7 @@ struct rcec_ea {
>> void pci_save_dpc_state(struct pci_dev *dev);
>> void pci_restore_dpc_state(struct pci_dev *dev);
>> void pci_dpc_init(struct pci_dev *pdev);
>> -void dpc_process_error(struct pci_dev *pdev);
>> +struct pci_dev *dpc_process_error(struct pci_dev *pdev);
>> pci_ers_result_t dpc_reset_link(struct pci_dev *pdev);
>> bool pci_dpc_recovered(struct pci_dev *pdev);
>> unsigned int dpc_tlp_log_len(struct pci_dev *dev);
>> diff --git a/drivers/pci/pcie/dpc.c b/drivers/pci/pcie/dpc.c
>> index bff29726c6a5..f6069f621683 100644
>> --- a/drivers/pci/pcie/dpc.c
>> +++ b/drivers/pci/pcie/dpc.c
>> @@ -260,10 +260,20 @@ static int dpc_get_aer_uncorrect_severity(struct pci_dev *dev,
>> return 1;
>> }
>>
>> -void dpc_process_error(struct pci_dev *pdev)
>> +/**
>> + * dpc_process_error - handle the DPC error status
>> + * @pdev: the port that experienced the containment event
>> + *
>> + * Return: the device that detected the error.
>> + *
>> + * NOTE: The device reference count is increased, the caller must decrement
>> + * the reference count by calling pci_dev_put().
>> + */
>> +struct pci_dev *dpc_process_error(struct pci_dev *pdev)
>
> Maybe it makes sense to carry the err_port naming for the pci_dev
> in here as well? Seems stronger than just relying on people
> reading the documentation you've added.
Good point. I think renaming the parameter would improve clarity. However,
I'd prefer to handle it in a separate patch to keep this change focused on
the functional modification. Would that work for you?
>
>> {
>> u16 cap = pdev->dpc_cap, status, source, reason, ext_reason;
>> struct aer_err_info info = {};
>> + struct pci_dev *err_dev;
>>
>> pci_read_config_word(pdev, cap + PCI_EXP_DPC_STATUS, &status);
>>
>> @@ -279,6 +289,7 @@ void dpc_process_error(struct pci_dev *pdev)
>> pci_aer_clear_nonfatal_status(pdev);
>> pci_aer_clear_fatal_status(pdev);
>> }
>> + err_dev = pci_dev_get(pdev);
>> break;
>> case PCI_EXP_DPC_STATUS_TRIGGER_RSN_NFE:
>> case PCI_EXP_DPC_STATUS_TRIGGER_RSN_FE:
>> @@ -290,6 +301,8 @@ void dpc_process_error(struct pci_dev *pdev)
>> "ERR_FATAL" : "ERR_NONFATAL",
>> pci_domain_nr(pdev->bus), PCI_BUS_NUM(source),
>> PCI_SLOT(source), PCI_FUNC(source));
>> + err_dev = pci_get_domain_bus_and_slot(pci_domain_nr(pdev->bus),
>> + PCI_BUS_NUM(source), source & 0xff);
>
> Bunch of replication in her with the pci_warn(). Maybe some local variables?
> Maybe even rebuild the final parameter from PCI_DEVFN(slot, func) just to make the
> association with the print really obvious?
Agreed. Here's the improved version:
--- a/drivers/pci/pcie/dpc.c
+++ b/drivers/pci/pcie/dpc.c
@@ -293,17 +293,28 @@ struct pci_dev *dpc_process_error(struct pci_dev *pdev)
break;
case PCI_EXP_DPC_STATUS_TRIGGER_RSN_NFE:
case PCI_EXP_DPC_STATUS_TRIGGER_RSN_FE:
- pci_read_config_word(pdev, cap + PCI_EXP_DPC_SOURCE_ID,
- &source);
+ {
+ int domain, bus;
+ u8 slot, func, devfn;
+ const char *err_type;
+
+ pci_read_config_word(pdev, cap + PCI_EXP_DPC_SOURCE_ID, &source);
+
+ /* Extract source device location */
+ domain = pci_domain_nr(pdev->bus);
+ bus = PCI_BUS_NUM(source);
+ slot = PCI_SLOT(source);
+ func = PCI_FUNC(source);
+ devfn = PCI_DEVFN(slot, func);
+
+ err_type = (reason == PCI_EXP_DPC_STATUS_TRIGGER_RSN_FE) ?
+ "ERR_FATAL" : "ERR_NONFATAL";
+
pci_warn(pdev, "containment event, status:%#06x, %s received from %04x:%02x:%02x.%d\n",
- status,
- (reason == PCI_EXP_DPC_STATUS_TRIGGER_RSN_FE) ?
- "ERR_FATAL" : "ERR_NONFATAL",
- pci_domain_nr(pdev->bus), PCI_BUS_NUM(source),
- PCI_SLOT(source), PCI_FUNC(source));
- err_dev = pci_get_domain_bus_and_slot(pci_domain_nr(pdev->bus),
- PCI_BUS_NUM(source), source & 0xff);
+ status, err_type, domain, bus, slot, func);
+ err_dev = pci_get_domain_bus_and_slot(domain, bus, devfn);
break;
+ }
case PCI_EXP_DPC_STATUS_TRIGGER_RSN_IN_EXT:
ext_reason = status & PCI_EXP_DPC_STATUS_TRIGGER_RSN_EXT;
pci_warn(pdev, "containment event, status:%#06x: %s detected\n",
>
> Is there any chance that this might return NULL? Feels like maybe that's
> only a possibility on a broken setup, but I'm not sure of all the wonderful
> races around hotplug and DPC occurring before the OS has caught up.
Surprise Down events are handled separately in
dpc_handle_surprise_removal() and won't reach dpc_process_error().
Please correct me if I missed anything.
Thanks for valuable comments.
Best Regards,
Shuai
Powered by blists - more mailing lists