Re: [PATCH 1/1] Revert "iommu/vt-d: Fix possible recursive locking in intel_iommu_init()"

[Baolu Lu]

On 2022/9/20 20:16, Robin Murphy wrote:
> On 2022-09-20 12:58, Thorsten Leemhuis wrote:
> > On 20.09.22 10:17, Lu Baolu wrote:
> > > This reverts commit 9cd4f1434479f1ac25c440c421fbf52069079914.
> >
> > Thx for taking care of this.
> >
> > > Some issues were reported on the original commit. Some thunderbolt 
> > > devices
> > > don't work anymore due to the following DMA fault.
> > >
> > > DMAR: DRHD: handling fault status reg 2
> > > DMAR: [INTR-REMAP] Request device [09:00.0] fault index 0x8080
> > >        [fault reason 0x25]
> > >        Blocked a compatibility format interrupt request
> > >
> > > Bring it back for now to avoid functional regression.
> > >
> > > Fixes: 9cd4f1434479f ("iommu/vt-d: Fix possible recursive locking in 
> > > intel_iommu_init()")
> > > Link: 
> > > https://lore.kernel.org/linux-iommu/485A6EA5-6D58-42EA-B298-8571E97422DE@xxxxxxxxxxxxxxxxx/
> > > Link: https://bugzilla.kernel.org/show_bug.cgi?id=216497
> >
> > Both those reports were against 5.19.y, so this afaics should have a
> >
> > Cc: <stable@xxxxxxxxxxxxxxx> # 5.19.x
> >
> > to ensure it's backported.
> >
> > Speaking of which: Joerg/Will/Robin, it seems quite a few people are
> > running into this, it hence would be great to get this quickly mainlined
> > (maybe by letting Linus pick it up straight from the list once ready?)
> > so stable can pick it up.
>
> As a heads-up, a straight revert is likely to lead to people reporting 
> lockdep warnings against -next, for the patches queued there which 
> exposed this dodgy locking in the first place.

I plan to fix that lockdep warning with below patch:

https://github.com/LuBaolu/intel-iommu/commit/dff18af627a2a76651b74cd6531f3e9357a97072

It works on my test machines. I am about to test it with more hardware.

> Does it work to just move the dmar_register_bus_notifier() call back to 
> where it was, without undoing the rest of the patch? That seems like the 
> change that's overwhelmingly likely to have broken IRQ remapping, and 
> TBH it wasn't clear to me why the original patch moved it to begin with.

The callbacks of dmar_register_bus_notifier() possibly races with
intel_iommu_init(). So the offending commit had to move it down until
the Intel IOMMU initialization is done.

Best regards,
baolu