Re: [PATCH v2] iommu/vt-d: fix system hang on reboot -f

[Ethan Zhao]

On 2/28/2025 4:38 AM, Jason Gunthorpe wrote:
> On Thu, Feb 27, 2025 at 08:40:31AM +0800, Ethan Zhao wrote:
> > On 2/26/2025 9:04 PM, Jason Gunthorpe wrote:
> > > On Wed, Feb 26, 2025 at 01:55:28PM +0800, Ethan Zhao wrote:
> > > > > Provided the system does not respond to those events when this function
> > > > > is called, it's fine to remove the lock.
> > > > I agree.
> > > I think it is running the destruction of the iommu far too late in the
> > > process. IMHO it should be done after all the drivers have been
> > > shutdown, before the CPUs go single threaded.
> > Hmm... so far it is fine, the iommu_shutdown only has a little work to
> > do, disable the translation, the PMR disabling is just backward compatible,
> > was deprecated already. if we move it to one position where all CPUs are
> > cycling, we don't know what kind of user-land tasks left there (i.e. reboot -f
> > case), it would be hard to full-fill the requirement of Intel VT-d, no ongoing
> > transaction there on hardware when issue the translation disabling command.
> There is no guarentee device dma is halted anyhow at this point either.

The -f option to reboot command, suggests data corruption possibility.although

the IOMMU strives not to cross the transaction boundaries of its address 
translation.

over all, we should make the reboot -f function works. not to hang 
there. meanwhile

it doesn't break anything else so far.

Thanks,

Ethan

> Jason