On Fri, Jun 20, 2025 at 04:42:02PM +0200, Benjamin Gaignard wrote:

Le 20/06/2025 à 15:52, Benjamin Gaignard a écrit :

quoted

Le 20/06/2025 à 14:05, Jason Gunthorpe a écrit :

quoted

On Fri, Jun 20, 2025 at 10:57:49AM +0200, Benjamin Gaignard wrote:

quoted

Le 19/06/2025 à 18:59, Jason Gunthorpe a écrit :

quoted

On Thu, Jun 19, 2025 at 06:27:52PM +0200, Benjamin Gaignard wrote:

quoted

Le 19/06/2025 à 15:47, Jason Gunthorpe a écrit :

quoted

Ugh. This ignores the gfp flags that are passed into map because you
have to force atomic due to the spinlock that shouldn't be there :(
This means it does not set GFP_KERNEL_ACCOUNT when
required. It would
be better to continue to use the passed in GFP flags
but override them
to atomic mode.

I will add a gfp_t parameter and use it like that:
page_table = iommu_alloc_pages_sz(gfp | GFP_ATOMIC |
GFP_DMA32, SPAGE_SIZE);

This will or together GFP_ATOMIC and GFP_KERNEL, I don't think that
works..

I have test it, I don't see conflicts or errors. What worries you ?

Just looking at the bitmaps I'm not sure? Did you run with lockdep?

Yes and it complains about that.
I see that sun50i driver have more less the same struct than my driver
but doesn't use lock. I will try see if I can remove the lock.

I have replace the two spinlock by a mutex in vsi_iommu structure.
It seems it works well and lockdep doesn't complain anymore.

You cannot use a sleeping lock within the map/unmap
functions. Removing the lock is hard for your case because you have
the cache flushing problem.

Maybe mask GFP_KERNEL off and then or back in GFP_ATOMIC.

Jason