Re: [GIT PULL v2] PCI changes for v6.17

[Nam Cao <namcao@linutronix.de>]

Ammar Faizi <ammarfaizi2@gnuweeb.org> writes:

> On Fri, Aug 08, 2025 at 05:59:23PM +0700, Ammar Faizi wrote:
>> On Thu, Aug 07, 2025 at 12:13:37PM +0700, Ammar Faizi wrote:
>> > On Thu, Aug 07, 2025 at 07:03:50AM +0200, Nam Cao wrote:
>> > > Does the diff below help?
>> > 
>> > Yes, it works.
>> 
>> So today, I synced with Linus' master branch again:
>> 
>>   37816488247d ("Merge tag 'net-6.17-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net")
>> 
>> and applied your fix on top of it.
>> 
>> I can boot, but I get this splat. Looking at the call trace, it seems
>> it's still related to pci, but different issue. The call trace is also
>> different from the previous one.
>
> It'll be a bit tricky to bisect this one. Because if I step back to
> a bad commit post:
>
>    d7d8ab87e3e ("PCI: vmd: Switch to msi_create_parent_irq_domain()")
>
> I won't be able to boot to reach this new splat :/
>
> I guess I need to apply the fix dirty for each bisection step. But I'll
> also need to make sure the current step has the d7d8ab87e3e commit
> anchestor before applying.

There is no point in bisecting before that commit, because the WARN_ON()
is added by that commit, so you wouldn't see anything before that.

The WARN_ONCE() tells us that some devices down the PCI tree are
allocating MSI, but VMD supports MSI-X only.

From the backtrace:
   msi_create_device_irq_domain+0x1eb/0x290
   __pci_enable_msi_range+0x106/0x300
   pci_alloc_irq_vectors_affinity+0xc5/0x110
   pcie_portdrv_probe+0x24e/0x610

It seems MSI-X are allocated first, but fail for some reason. Then
fallback to MSI, which triggers the WARN_ON().

So we need to figure out why MSI-X allocation fail.

I may need to ask you to insert a bunch of printk() to help me pinpoint
the problem. But let me stare at it first..

Nam