On Wed, Jul 24, 2019 at 01:45:48PM -0500, Bjorn Helgaas wrote:
> On Wed, Jul 10, 2019 at 10:36:16AM -0700, sathyanarayanan kuppuswamy wrote:
> > On 7/10/19 10:22 AM, Patel, Mayurkumar wrote:
> > > > On 7/9/19 1:00 AM, Patel, Mayurkumar wrote:
> > > > > After system suspend/resume cycle AER registers settings are
> > > > > lost. Not restoring Root Error Command Register bits if it were
> > > > > set, keeps AER interrupts disabled after system resume.
> > > > > Moreover, AER mask and severity registers are also required
> > > > > to be restored back to AER settings prior to system suspend.
> > > > >
> > > > > Signed-off-by: Mayurkumar Patel <mayurkumar.patel@xxxxxxxxx>
> > > > > ---
> > > > > drivers/pci/pci.c | 2 ++
> > > > > drivers/pci/pcie/aer.c | 49 +++++++++++++++++++++++++++++++++++++++++++++++++
> > > > > include/linux/aer.h | 4 ++++
> > > > > 3 files changed, 55 insertions(+)
> > > > >
> > > > > diff --git a/drivers/pci/pci.c b/drivers/pci/pci.c
> > > > > index 8abc843..40d5507 100644
> > > > > --- a/drivers/pci/pci.c
> > > > > +++ b/drivers/pci/pci.c
> > > > > @@ -1340,6 +1340,7 @@ int pci_save_state(struct pci_dev *dev)
> > > > >
> > > > > pci_save_ltr_state(dev);
> > > > > pci_save_dpc_state(dev);
> > > > > + pci_save_aer_state(dev);
> > > > > return pci_save_vc_state(dev);
> > > > > }
> > > > > EXPORT_SYMBOL(pci_save_state);
> > > > > @@ -1453,6 +1454,7 @@ void pci_restore_state(struct pci_dev *dev)
> > > > > pci_restore_dpc_state(dev);
> > > > >
> > > > > pci_cleanup_aer_error_status_regs(dev);
> > > > > + pci_restore_aer_state(dev);
> > > > >
> > > > > pci_restore_config_space(dev);
> > > > >
> > > > > diff --git a/drivers/pci/pcie/aer.c b/drivers/pci/pcie/aer.c
> > > > > index b45bc47..1acc641 100644
> > > > > --- a/drivers/pci/pcie/aer.c
> > > > > +++ b/drivers/pci/pcie/aer.c
> > > > > @@ -448,6 +448,54 @@ int pci_cleanup_aer_error_status_regs(struct pci_dev *dev)
> > > > > return 0;
> > > > > }
> > > > >
> > > > > +void pci_save_aer_state(struct pci_dev *dev)
> > > > > +{
> > > > > + int pos = 0;
> > > > > + struct pci_cap_saved_state *save_state;
> > > > > + u32 *cap;
> > > > > +
> > > > > + if (!pci_is_pcie(dev))
> > > > > + return;
> > > > > +
> > > > > + save_state = pci_find_saved_ext_cap(dev, PCI_EXT_CAP_ID_ERR);
> > > > > + if (!save_state)
> > > > > + return;
> > > > > +
> > > > > + pos = dev->aer_cap;
> > > > > + if (!pos)
> > > > > + return;
> > > > > +
> > > > > + cap = &save_state->cap.data[0];
> > > > > + pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_MASK, cap++);
> > > > > + pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_SEVER, cap++);
> > > > > + pci_read_config_dword(dev, pos + PCI_ERR_COR_MASK, cap++);
> > > > > + pci_read_config_dword(dev, pos + PCI_ERR_ROOT_COMMAND, cap++);
>
> > > > I don't see AER driver modifying UNCOR_MASK/SEVER/COR_MASK register. If
> > > > all it has is default value then why do you want to preserve it ?
> > > >
> > > Thanks for reply.
> > > You are right about UNCOR_MASK/SEVER/COR_MASK mask AER driver
> > > leaves untouched, But IMHO users for example can use "setpci" to
> > > unmask specific errors and its severity based on their debugging
> > > requirement on Root port and/or Endpoint. So during resume, if
> > > PCIe endpoint fails due to any error which by default stays masked
> > > then it can't be catched and/or debugged. Moreover, Endpoint
> > > driver may also unmask during "driver probe" certain specific
> > > errors for endpoint, which needs to be restored while resume.
> >
> > Isn't these registers configuration usually done by firmware ? I
> > think user application rarely touch them. Also, IMO,
> > Caching/Restoring registers under the assumption that it might be
> > useful for user if they modified it is not a convincing argument.
> > But I will let Bjorn and others decide whether its alright to cache
> > these registers.
>
> I think the ideal user experience would be "suspend/resume has no
> effect on any configuration". That would argue for saving/restoring
> registers even if we think it's unlikely the user would change them.
The call to pci_save_state most likely occurs long before a user has an
opportunity to alter these regsiters, though. Won't this just restore
what was previously there, and not the state you changed it to?
> > > @Bjorn/Anybody else has any opinion to cache/restore
> > > UNCOR_MASK/SEVER/COR_MASK registers? Please help to comment.
> > >
> > > > Also, any reason for not preserving ECRC settings ?
> > > No specific reason. I can incorporte that with v2 of this patchset
> > > but I don’t have HW on which I can validate that.
>
> I think we should preserve ECRC settings as well for the same reason.
>
> > > > > +}
> > > > > +
> > > > > +void pci_restore_aer_state(struct pci_dev *dev)
> > > > > +{
> > > > > + int pos = 0;
> > > > > + struct pci_cap_saved_state *save_state;
> > > > > + u32 *cap;
> > > > > +
> > > > > + if (!pci_is_pcie(dev))
> > > > > + return;
> > > > > +
> > > > > + save_state = pci_find_saved_ext_cap(dev, PCI_EXT_CAP_ID_ERR);
> > > > > + if (!save_state)
> > > > > + return;
> > > > > +
> > > > > + pos = dev->aer_cap;
> > > > > + if (!pos)
> > > > > + return;
> > > > > +
> > > > > + cap = &save_state->cap.data[0];
> > > > > + pci_write_config_dword(dev, pos + PCI_ERR_UNCOR_MASK, *cap++);
> > > > > + pci_write_config_dword(dev, pos + PCI_ERR_UNCOR_SEVER, *cap++);
> > > > > + pci_write_config_dword(dev, pos + PCI_ERR_COR_MASK, *cap++);
> > > > > + pci_write_config_dword(dev, pos + PCI_ERR_ROOT_COMMAND, *cap++);
> > > > > +}
> > > > > +
> > > > > void pci_aer_init(struct pci_dev *dev)
> > > > > {
> > > > > dev->aer_cap = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
> > > > > @@ -1396,6 +1444,7 @@ static int aer_probe(struct pcie_device *dev)
> > > > > return status;
> > > > > }
> > > > >
> > > > > + pci_add_ext_cap_save_buffer(port, PCI_EXT_CAP_ID_ERR, sizeof(u32) * 4);
> > > > > aer_enable_rootport(rpc);
> > > > > pci_info(port, "enabled with IRQ %d\n", dev->irq);
> > > > > return 0;
You are allocating the capability save buffer in aer_probe(), so this
save/restore applies only to root ports. But you mention above that you
want to restore end devices, right?
