Re: [PATCH] iommu/arm-smmu-v3-sva: Enable Hardware Access and Hardware Dirty bits

[Pranjal Shrivastava <praan@google.com>]

On Wed, May 13, 2026 at 11:32:13AM -0300, Jason Gunthorpe wrote:
> On Wed, May 13, 2026 at 02:27:48PM +0000, Pranjal Shrivastava wrote:
> 
> > Now, if we're work on an SVA page, with only SMMU supporting HTTU. A DMA
> > writes to the page and the process (CPU) calls fsync(). IIUC, it performs
> > a lookup in the Page Cache specifically for folios tagged as DIRTY.
> > Since, vmscan didn't run yet, this could potentally drop the writes..
> 
> How does it work differently in the MM when the CPU has BBM support?

Hmm... I looked at fsync and I see that it eventually calls 
folio_mkclean() (via writeback [1]), which performs an rmap_walk() [2]
to harvest dirty bits from PTEs into the respective struct folios.
Similarly, the vm_scan path does the same thing via try_to_unmap [3].

Since the MM subsystem actively scans the tables during writeback, it
doesn't matter if the dirty bits were flipped asynchronously by HW (w/o
kernel traps).

I guess that settles it, we don't need to gate this behind cpu_has_hw_af()
and we need not care about the CONFIG_ARM64_HW_AFDBM either.

Reviewed by: Pranjal Shrivastava <praan@google.com>

Thanks,
Praan

[1] https://elixir.bootlin.com/linux/v7.1-rc3/source/mm/page-writeback.c#L2905
[2] https://elixir.bootlin.com/linux/v7.1-rc3/source/mm/rmap.c#L1103
[3] https://elixir.bootlin.com/linux/v7.1-rc3/source/mm/rmap.c#L2164