> From: Jason Gunthorpe <j...@nvidia.com>
> Sent: Tuesday, January 17, 2023 9:30 PM
>
> On Tue, Jan 17, 2023 at 03:35:08AM +0000, Tian, Kevin wrote:
> > > From: Jason Gunthorpe <j...@nvidia.com>
> > > Sent: Saturday, January 7, 2023 12:43 AM
> > >
> > > @@ -2676,7 +2676,7 @@ static int copy_context_table(struct
> intel_iommu
> > > *iommu,
> > > if (!old_ce)
> > > goto out;
> > >
> > > - new_ce = alloc_pgtable_page(iommu->node);
> > > + new_ce = alloc_pgtable_page(iommu->node,
> > > GFP_KERNEL);
> >
> > GFP_ATOMIC
>
> Can't be:
>
> old_ce = memremap(old_ce_phys, PAGE_SIZE,
> MEMREMAP_WB);
> if (!old_ce)
> goto out;
>
> new_ce = alloc_pgtable_page(iommu->node,
> GFP_KERNEL);
> if (!new_ce)
>
> memremap() is sleeping.
>
> And the only caller is:
>
> ctxt_tbls = kcalloc(ctxt_table_entries, sizeof(void *), GFP_KERNEL);
> if (!ctxt_tbls)
> goto out_unmap;
>
> for (bus = 0; bus < 256; bus++) {
> ret = copy_context_table(iommu, &old_rt[bus],
> ctxt_tbls, bus, ext);
>
Yes, but the patch description says "Push the GFP_ATOMIC to all
callers." implying it's purely a refactoring w/o changing those
semantics.
I'm fine with doing this change in this patch, but it should worth
a clarification in the patch description.
