2021-08-21 13:38:56

by Shanker Donthineni

[permalink] [raw]
Subject: [PATCH] PCI/AER: Continue AER recovery of device with NO_BUS_RESET set

In the current implementation, the AER FATAL and NONFTAL recovery will be
terminated for the device that exhibits NO_BUS_RESET quirk. The non-zero
return value from pci_bus_error_reset() is treated as an error condition
in aer_root_reset() which leads to return PCI_ERS_RESULT_DISCONNECT.

aer_recover_work_func()
pcie_do_recovery()
report_frozen_detected()
if (aer_root_reset() == PCI_ERS_RESULT_DISCONNECT)
goto failed # termimates here because of NO_BUS_RESET

...
report_mmio_enabled()
report_resume()
pcie_clear_xxx_status()
...
return 0
failed:
pci_uevent_ers(PCI_ERS_RESULT_DISCONNECT);

The return value -ENOTTY from pci_bus_error_reset() indicates SBR was
skipped but no real errors were encountered. This scenario could be
considered as a non-error case so that the PCI device driver gets the
opportunity to recover the device back to an operational state instead
of keeping it in the DISCONNECT state.

Signed-off-by: Shanker Donthineni <[email protected]>
---
drivers/pci/pcie/aer.c | 8 ++++++--
1 file changed, 6 insertions(+), 2 deletions(-)

diff --git a/drivers/pci/pcie/aer.c b/drivers/pci/pcie/aer.c
index 9784fdcf30061..8cf6bd6a3376d 100644
--- a/drivers/pci/pcie/aer.c
+++ b/drivers/pci/pcie/aer.c
@@ -1414,8 +1414,12 @@ static pci_ers_result_t aer_root_reset(struct pci_dev *dev)
pci_info(dev, "not reset (no FLR support: %d)\n", rc);
} else {
rc = pci_bus_error_reset(dev);
- pci_info(dev, "%s Port link has been reset (%d)\n",
- pci_is_root_bus(dev->bus) ? "Root" : "Downstream", rc);
+ pci_info(dev, "%s Port link has %sbeen reset (%d)\n",
+ pci_is_root_bus(dev->bus) ? "Root" : "Downstream",
+ rc == -ENOTTY ? "not " : "", rc);
+
+ if (rc == -ENOTTY)
+ rc = 0;
}

if ((host->native_aer || pcie_ports_native) && aer) {
--
2.25.1


2021-08-21 15:29:25

by Shanker Donthineni

[permalink] [raw]
Subject: Re: [PATCH] PCI/AER: Continue AER recovery of device with NO_BUS_RESET set



On 8/21/21 8:30 AM, Shanker Donthineni wrote:
> External email: Use caution opening links or attachments
>
>
> In the current implementation, the AER FATAL and NONFTAL recovery will be
> terminated for the device that exhibits NO_BUS_RESET quirk. The non-zero

Correction, this problem happens only for AER_FATAL recovery case.

> return value from pci_bus_error_reset() is treated as an error condition
> in aer_root_reset() which leads to return PCI_ERS_RESULT_DISCONNECT.
>
> aer_recover_work_func()
> pcie_do_recovery()
> report_frozen_detected()
> if (aer_root_reset() == PCI_ERS_RESULT_DISCONNECT)
> goto failed # termimates here because of NO_BUS_RESET
>
> ...
> report_mmio_enabled()
> report_resume()
> pcie_clear_xxx_status()
> ...
> return 0
> failed:
> pci_uevent_ers(PCI_ERS_RESULT_DISCONNECT);
>
> The return value -ENOTTY from pci_bus_error_reset() indicates SBR was
> skipped but no real errors were encountered. This scenario could be
> considered as a non-error case so that the PCI device driver gets the
> opportunity to recover the device back to an operational state instead
> of keeping it in the DISCONNECT state.
>
> Signed-off-by: Shanker Donthineni <[email protected]>
> ---
> drivers/pci/pcie/aer.c | 8 ++++++--
> 1 file changed, 6 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/pci/pcie/aer.c b/drivers/pci/pcie/aer.c
> index 9784fdcf30061..8cf6bd6a3376d 100644
> --- a/drivers/pci/pcie/aer.c
> +++ b/drivers/pci/pcie/aer.c
> @@ -1414,8 +1414,12 @@ static pci_ers_result_t aer_root_reset(struct pci_dev *dev)
> pci_info(dev, "not reset (no FLR support: %d)\n", rc);
> } else {
> rc = pci_bus_error_reset(dev);
> - pci_info(dev, "%s Port link has been reset (%d)\n",
> - pci_is_root_bus(dev->bus) ? "Root" : "Downstream", rc);
> + pci_info(dev, "%s Port link has %sbeen reset (%d)\n",
> + pci_is_root_bus(dev->bus) ? "Root" : "Downstream",
> + rc == -ENOTTY ? "not " : "", rc);
> +
> + if (rc == -ENOTTY)
> + rc = 0;
> }
>
> if ((host->native_aer || pcie_ports_native) && aer) {
> --
> 2.25.1
>