Let's mention the vendor and device name in the subject to make the log more useful. On Mon, Mar 06, 2023 at 12:53:40PM +0530, Basavaraj Natikar wrote: > One of the AMD USB controllers fails to maintain internal functional > context when transitioning from D3 to D0, desynchronizing MSI-X bits. > As a result, add a quirk to this controller to clear the MSI-X bits > on suspend. Is this a documented erratum? Please include a citation if so. Are there any other AMD USB devices with the same defect? The quick clears the Function Mask bit, so the MSI-X vectors may be *unmasked* depending on the state of each vectors Mask bit. I assume the potential unmasking is safe because you also clear the MSI-X Enable bit, so the function can't use MSI-X at all. All state is lost in D3cold, so I guess this problem must occur during a D3hot to D0 transition, right? I assume this device sets No_Soft_Reset, so the function is supposed to return to D0active with all internal state intact. But this device returns to D0active with the MSI-X internal state corrupted? I assume this relies on pci_restore_state() to restore the MSI-X state. Seems like that might be enough to restore the internal state even without this quirk, but I guess it must not be. > Note: This quirk works in all scenarios, regardless of whether the > integrated GPU is disabled in the BIOS. I don't know how the integrated GPU is related to this USB controller, but I assume this fact is important somehow? > Cc: Mario Limonciello <mario.limonciello@xxxxxxx> > Reported-by: Thomas Glanzmann <thomas@xxxxxxxxxxxx> > Link: https://lore.kernel.org/linux-usb/Y%2Fz9GdHjPyF2rNG3@xxxxxxxxxxxx/T/#u Apparently the symptom is one of these: xhci_hcd 0000:0c:00.0: Error while assigning device slot ID: Command Aborted xhci_hcd 0000:0c:00.0: Max number of devices this xHCI host supports is 64. usb usb1-port1: couldn't allocate usb_device xhci_hcd 0000:0c:00.0: WARN: xHC save state timeout xhci_hcd 0000:0c:00.0: PM: suspend_common(): xhci_pci_suspend+0x0/0x150 [xhci_pci] returns -110 xhci_hcd 0000:0c:00.0: can't suspend (hcd_pci_runtime_suspend [usbcore] returned -110) We should include the critical line or two in the commit log to help users find the fix. I see this must be xhci_suspend() returning -ETIMEDOUT after xhci_save_registers(), but I don't see the connection from there to a PCI_FIXUP_SUSPEND. Can you connect the dots for me? > Signed-off-by: Basavaraj Natikar <Basavaraj.Natikar@xxxxxxx> > --- > drivers/pci/quirks.c | 10 ++++++++++ > 1 file changed, 10 insertions(+) > > diff --git a/drivers/pci/quirks.c b/drivers/pci/quirks.c > index 44cab813bf95..ddf7100227d3 100644 > --- a/drivers/pci/quirks.c > +++ b/drivers/pci/quirks.c > @@ -6023,3 +6023,13 @@ DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_INTEL, 0x9a2d, dpc_log_size); > DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_INTEL, 0x9a2f, dpc_log_size); > DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_INTEL, 0x9a31, dpc_log_size); > #endif > + > +static void quirk_clear_msix(struct pci_dev *dev) > +{ > + u16 ctrl; > + > + pci_read_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, &ctrl); > + ctrl &= ~(PCI_MSIX_FLAGS_MASKALL | PCI_MSIX_FLAGS_ENABLE); > + pci_write_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, ctrl); > +} > +DECLARE_PCI_FIXUP_SUSPEND(PCI_VENDOR_ID_AMD, 0x15b8, quirk_clear_msix); > -- > 2.25.1 >
