Re: [REGRESSION] amdgpu with Thunderbolt eGPU bracket fails since new bridge window alignment calculation code

["Ilpo Järvinen" <ilpo.jarvinen@linux.intel.com>]

[-- Attachment #1: Type: text/plain, Size: 3447 bytes --]

On Thu, 2 Apr 2026, Jonas Höglund wrote:

> On Mon, 30 Mar 2026, at 16:32, Ilpo Järvinen wrote:
> > On Mon, 30 Mar 2026, Jonas Höglund wrote:
> >> 
> >> Unless I misread something, they should both be included in the recently
> >> tagged 7.0.0-rc6--I'll try building it and see if the issue is resolved.
> >> 
> >> I'll reply once I've tested 7.0.0-rc6.
> >
> > Hi again,
> >
> > Now that I look more into the logs that probably won't help. For some 
> > reason, it seems that resize is not even attempted and the errno is 
> > -EINVAL which is a bit unexpected.
> >
> > I'm starting to wonder that the problem fixed by this patch once again is 
> > showing its ugly head (it's currently in pci/resource branch, so it won't 
> > appear until 7.1-rc1):
> >
> > https://lore.kernel.org/linux-pci/20260326200427.GA1340256@bhelgaas/
> 
> Seems your hunch was right--I've now gotten around to testing with
> 7.0-rc6 as well as the pci/resource branch.  The problem persists in the
> former whereas it seems docking succeeds in the latter.
> 
> 
> > I still don't understand why pbus_select_window() would return NULL in 
> > this case but it looks the most likely candidate where -EINVAL could come 
> > from (I still don't understand what cleared resource's flags if that's the 
> > case but it still seems the best explanation).
> >
> > Please take logs from this point on with dyndbg="file drivers/pci/*.c +p" 
> > on the kernel's command line so there's little bit of extra info (and 
> > check you are building with CONFIG_DYNAMIC_DEBUG).
> 
> Here are dmesg logs (with the appropriate dyndbg cmdline flag) for both
> cases, in case it's useful:
> 
> https://up.firefly.nu/pub/amdgpu-egpu-crash-7.0.0-rc6.dmesg.txt
> https://up.firefly.nu/pub/amdgpu-egpu-good-pci-resource.dmesg.txt
> 
> 
> That's good enough on my end, knowing the issue is addressed already
> upstream and slated for 7.1.  I'm happy to test anything else if it'd
> be useful (for eventual backports or so), but otherwise I think I'll
> just pick thoes patches from the pci/resource tree for now.

Hi,

Thanks. It certainly looks the commit dc4b4d04e1ca ("PCI: Prevent 
shrinking bridge window from its required size") I referred to above might 
indeed help here (For Thorsten's convinience: as mentioned above, it is 
currently in the pci/resource branch slated for 7.1).

With the extra debug enabled, "shrunken by" lines appear in the log which 
indicates the hotplug memory distribution algorithm goes to mess with the 
calculated bridge window sizes in between resource sizing and resource 
assignment and my fix aims to prevent that from happening.

If that fix does not help or does not fully solve the issue, please do 
take a new log with that patch included into the kernel (preferrably with 
all the fixes that are currently in the pci/resource branch so we don't 
hit yet another issue that already has a fix). If you need to take more 
logs, please include also /proc/iomem dump (as figuring the iomem layout 
from dmesg is pretty tedious and error prone).

Also if you see this line, it's worth to posting the log (even if things 
would appear as working):

amdgpu 0000:3e:00.0: Not enough PCI address space for a large BAR.

...I'll see if I can somehow improve that as well (not a guarantee but 
it's still worth taking a look, it appears also in the case you labeled 
"good").

-- 
 i.