On Wed, Apr 07, 2021 at 03:57:32PM -0700, Arun Easi wrote:
> On Wed, 7 Apr 2021, 3:13pm, Bjorn Helgaas wrote:
>
> > On Wed, Mar 03, 2021 at 02:42:50PM -0800, Arun Easi wrote:
> > > "lspci -vvv" for Qlogic Fibre Channel HBA 1077:2261 displays
> > > "Vital Product Data" as "Not readable" today and thus preventing
> > > customers from getting relevant HBA information. Fix it by removing
> > > the blacklist quirk.
> > >
> > > The VPD quirk was added by [0] to avoid a system NMI; this issue has
> > > been long fixed in the HBA firmware. In addition, PCI also has changes
> > > to check the VPD size [1], so this quirk can be reverted now regardless
> > > of a firmware update.
> >
> > This is not a very convincing argument yet since 104daa71b396 ("PCI:
> > Determine actual VPD size on first access") appeared in v4.6 and
> > 0d5370d1d852 ("PCI: Prevent VPD access for QLogic ISP2722") appeared
> > in v4.11.
> >
> > If 104daa71b396 really fixed the problem, why did we need
> > 0d5370d1d852?
>
> True, 0d5370d1d852 was not really neeeded for 104daa71b396 and newer
> kernels; my theory is that when Ethan Z. ran the tests, he was using an
> older (older than 104daa71b396) kernel, but by the time the blacklisting
> was put in place, the kernel already had the fix that made the
> blacklisting unnecessary.
>
> More of my investigation details explained here:
> https://lore.kernel.org/linux-pci/alpine.LRH.2.21.9999.2012161641230.28924@xxxxxxxxxxxxxxxxxxxxxxxxxxxx/
>
> A quick summary of which is that, when Ethan reported the crash stack, it
> had pci_vpd_pci22* calls which is seen only in older kernels. Though
> 104daa71b396 too had those calls, it was very close to the commit that
> renamed those calls (f1cd93f9aabe) -- and I theorized Ethan probably was
> not running a kernel between 104daa71b396 and f1cd93f9aabe (only 3
> commits (drivers/pci/) away).
We should put the outline of this theory in the commit log for the
benefit of future readers who have the same question I did.
Bjorn
