Re: [BUG v7.3-rc1] WARNING: mm/page_alloc.c:5340 at __alloc_frozen_pages_noprof

[Mathias Nyman <mathias.nyman@linux.intel.com>]

On 9/3/26 10:15, Michal Pecio wrote:
> On Wed, 2 Sep 2026 14:33:51 -0400, Steven Rostedt wrote:
>> I'm triggering this warning on boot:
>>
>> [   32.862044] ------------[ cut here ]------------
>> [   32.864573] WARNING: mm/page_alloc.c:5340 at __alloc_frozen_pages_noprof+0xd81/0x30d0, CPU#3: swapper/0/1
> 
> Apparently that's a nonsense allocation size.
> 
>> [   32.881048] Modules linked in:
>> [   32.883735] CPU: 3 UID: 0 PID: 1 Comm: swapper/0 Not tainted 7.3.0-rc1-test #510 PREEMPT(full)
>> [   32.890178] Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 1.17.0-debian-1.17.0-1 04/01/2014
> 
> Hmm, I recall hearing about QEMU recently.
> 
> Could this be the "horrible splat" mentioned by 045b5bef916d ?
> 
> Find it in usb-linus branch at
> git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/usb.git
> 
Yes, this is the case

A patch switching to GENMASK and FIELD_GET has a flaw and changed the parsing of
the max segments value reported by xHCI host controller.

xHC returns _exponent_ of maximum segments supported, QEMU supports one segment
so value read is 0.  Parsing fails and assumes maximum number of supported ring segments
is 0.  Most hardware xHCI controllers (except some Mediatek) support more than one
segments so issue is not visible for those

Issue (7.3-rc1):
6d45e9556d4a  ("usb: xhci: standardize multi bit-field macros") in parsing

Fix: (in Greg's tree, Hopeful it can make make 7.3-rc2)
045b5bef916d usb: xhci: Fix HCS_ERST_MAX conversion

Thanks
Mathias