On 9/2/20 11:42 AM, Andrey Grodzovsky wrote:
This reverts commit 6d2c89441571ea534d6240f7724f518936c44f8d.
In the code bellow
pci_walk_bus(bus, report_frozen_detected, &status);
- if (reset_link(dev, service) != PCI_ERS_RESULT_RECOVERED)
+ status = reset_link(dev, service);
status returned from report_frozen_detected is unconditionally masked
by status returned from reset_link which is wrong.
This breaks error recovery implementation for AMDGPU driver
by masking PCI_ERS_RESULT_NEED_RESET returned from amdgpu_pci_error_detected
and hence skiping slot reset callback which is necessary for proper
ASIC recovery. Effectively no other callback besides resume callback will
be called after link reset the way it is implemented now regardless of what
value error_detected callback returns.
}
Instead of reverting this change, can you try following patch ?
https://lore.kernel.org/linux-pci/56ad4901-725f-7b88-2117-b124b28b027f@xxxxxxxxxxxxxxx/T/#me8029c04f63c21f9d1cb3b1ba2aeffbca3a60df5
--
Sathyanarayanan Kuppuswamy
Linux Kernel Developer
_______________________________________________
amd-gfx mailing list
amd-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
