[dm-devel] [PATCH RESEND v6 4/9] pmem, mm: Implement ->memory_failure in pmem driver

[Dan Williams]

On Fri, Jul 30, 2021 at 3:02 AM Shiyang Ruan <ruansy.fnst at fujitsu.com> wrote:
>
> With dax_holder notify support, we are able to notify the memory failure
> from pmem driver to upper layers.  If there is something not support in
> the notify routine, memory_failure will fall back to the generic hanlder.

How about:

"Any layer can return -EOPNOTSUPP to force memory_failure() to fall
back to its generic implementation."


>
> Signed-off-by: Shiyang Ruan <ruansy.fnst at fujitsu.com>
> ---
>  drivers/nvdimm/pmem.c | 13 +++++++++++++
>  mm/memory-failure.c   | 14 ++++++++++++++
>  2 files changed, 27 insertions(+)
>
> diff --git a/drivers/nvdimm/pmem.c b/drivers/nvdimm/pmem.c
> index 1e0615b8565e..fea4ffc333b8 100644
> --- a/drivers/nvdimm/pmem.c
> +++ b/drivers/nvdimm/pmem.c
> @@ -362,9 +362,22 @@ static void pmem_release_disk(void *__pmem)
>         del_gendisk(pmem->disk);
>  }
>
> +static int pmem_pagemap_memory_failure(struct dev_pagemap *pgmap,
> +               unsigned long pfn, unsigned long nr_pfns, int flags)
> +{
> +       struct pmem_device *pmem =
> +                       container_of(pgmap, struct pmem_device, pgmap);
> +       loff_t offset = PFN_PHYS(pfn) - pmem->phys_addr - pmem->data_offset;
> +
> +       return dax_holder_notify_failure(pmem->dax_dev, offset,
> +                                        page_size(pfn_to_page(pfn)) * nr_pfns,

I do not understand the usage of page_size() here? memory_failure()
assumes PAGE_SIZE pages. DAX pages also do not populate the compound
metadata yet, but even if they did I would expect memory_failure() to
be responsible for doing something like:

    pgmap->ops->memory_failure(pgmap, pfn, size >> PAGE_SHIFT, flags);

...where @size is calculated from dev_pagemap_mapping_shift().

> +                                        &flags);

Why is the local flags variable passed by reference? At a minimum the
memory_failure() flags should be translated to a new set dax-notify
flags, because memory_failure() will not be the only user of this
notification interface. See NVDIMM_REVALIDATE_POISON, and the
discussion Dave and I had about using this notification to signal
unsafe hot-removal of a memory device.


> +}
> +
>  static const struct dev_pagemap_ops fsdax_pagemap_ops = {
>         .kill                   = pmem_pagemap_kill,
>         .cleanup                = pmem_pagemap_cleanup,
> +       .memory_failure         = pmem_pagemap_memory_failure,
>  };
>
>  static int pmem_attach_disk(struct device *dev,
> diff --git a/mm/memory-failure.c b/mm/memory-failure.c
> index 3bdfcb45f66e..ab3eda335acd 100644
> --- a/mm/memory-failure.c
> +++ b/mm/memory-failure.c
> @@ -1600,6 +1600,20 @@ static int memory_failure_dev_pagemap(unsigned long pfn, int flags,
>          */
>         SetPageHWPoison(page);
>
> +       /*
> +        * Call driver's implementation to handle the memory failure, otherwise
> +        * fall back to generic handler.
> +        */
> +       if (pgmap->ops->memory_failure) {
> +               rc = pgmap->ops->memory_failure(pgmap, pfn, 1, flags);
> +               /*
> +                * Fall back to generic handler too if operation is not
> +                * supported inside the driver/device/filesystem.
> +                */
> +               if (rc != EOPNOTSUPP)
> +                       goto out;
> +       }
> +
>         mf_generic_kill_procs(pfn, flags);
>  out:
>         /* drop pgmap ref acquired in caller */
> --
> 2.32.0
>
>
>