RE: [PATCH v4 04/10] mm, fsdax: Refactor memory-failure handler for dax mapping

[ruansy.fnst@xxxxxxxxxxx]

> -----Original Message-----
> From: Dan Williams <dan.j.williams@xxxxxxxxx>
> Subject: Re: [PATCH v4 04/10] mm, fsdax: Refactor memory-failure handler for
> dax mapping
> 
> [ drop old nvdimm list, add the new one ]
> 
> On Thu, Jun 3, 2021 at 6:19 PM Shiyang Ruan <ruansy.fnst@xxxxxxxxxxx> wrote:
> >
> > The current memory_failure_dev_pagemap() can only handle single-mapped
> > dax page for fsdax mode.  The dax page could be mapped by multiple
> > files and offsets if we let reflink feature & fsdax mode work
> > together.  So, we refactor current implementation to support handle
> > memory failure on each file and offset.
> 
> I don't understand this organization, perhaps because this patch introduces
> mf_dax_kill_procs() without a user. 

Yes, I think I made it a mess... I should reorganize this whole patchset.

The mf_dax_kill_procs() is used by xfs in patch 9.  I was mean to refactor these code for the next patches usage.

> However, my expectation is that
> memory_failure() is mostly untouched save for an early check via
> pgmap->notify_memory_failure(). If pgmap->notify_memory_failure() indicates
> that the memory failure was handled by the pgmap->owner / dax_dev holder
> stack, then the typical memory failure path is short-circuited. Otherwise, for
> non-reflink filesystems where
> page->mapping() is valid the legacy / existing memory_failure()
> operates as it does currently.

You can find this logic in patch 5.

When it comes to memory-failure() and memory_failure_dev_pagemap(), after some check, it will try to call pgmap->ops->memory_failure().  If this interface is implemented, for example pgmap is pmem device, it will call the dax_dev holder stack. And finally, it comes to mf_dax_kill_procs().
However, if something wrong happens in this stack, such as feature not support or interface not implemented, it will roll back to normal memory-failure hanlder which is refactored as mf_generic_kill_porcs().

So, I think we are in agreement on this.

Let me reorganize these code.

> If reflink capable filesystems want to share a
> common implementation to map pfns to files they can, but I don't think that
> common code belongs in mm/memory-failure.c.
> 
> >
> > Signed-off-by: Shiyang Ruan <ruansy.fnst@xxxxxxxxxxx>
> > ---
> >  fs/dax.c            |  21 ++++++++
> >  include/linux/dax.h |   1 +
> >  include/linux/mm.h  |   9 ++++
> >  mm/memory-failure.c | 114
> > ++++++++++++++++++++++++++++----------------
> >  4 files changed, 105 insertions(+), 40 deletions(-)
> >
> > diff --git a/fs/dax.c b/fs/dax.c
> > index 62352cbcf0f4..58faca85455a 100644
> > --- a/fs/dax.c
> > +++ b/fs/dax.c
> > @@ -389,6 +389,27 @@ static struct page *dax_busy_page(void *entry)
> >         return NULL;
> >  }
> >
> > +/*
> > + * dax_load_pfn - Load pfn of the DAX entry corresponding to a page
> > + * @mapping: The file whose entry we want to load
> > + * @index:   The offset where the DAX entry located in
> > + *
> > + * Return:   pfn of the DAX entry
> > + */
> > +unsigned long dax_load_pfn(struct address_space *mapping, unsigned
> > +long index) {
> > +       XA_STATE(xas, &mapping->i_pages, index);
> > +       void *entry;
> > +       unsigned long pfn;
> > +
> > +       xas_lock_irq(&xas);
> > +       entry = xas_load(&xas);
> > +       pfn = dax_to_pfn(entry);
> > +       xas_unlock_irq(&xas);
> 
> This looks racy, what happened to the locking afforded by dax_lock_page()?
> 
> > +
> > +       return pfn;
> > +}
> > +
> >  /*
> >   * dax_lock_mapping_entry - Lock the DAX entry corresponding to a page
> >   * @page: The page whose entry we want to lock diff --git
> > a/include/linux/dax.h b/include/linux/dax.h index
> > 1ce343a960ab..6e758daa5004 100644
> > --- a/include/linux/dax.h
> > +++ b/include/linux/dax.h
> > @@ -158,6 +158,7 @@ int dax_writeback_mapping_range(struct
> > address_space *mapping,
> >
> >  struct page *dax_layout_busy_page(struct address_space *mapping);
> > struct page *dax_layout_busy_page_range(struct address_space *mapping,
> > loff_t start, loff_t end);
> > +unsigned long dax_load_pfn(struct address_space *mapping, unsigned
> > +long index);
> >  dax_entry_t dax_lock_page(struct page *page);  void
> > dax_unlock_page(struct page *page, dax_entry_t cookie);  #else diff
> > --git a/include/linux/mm.h b/include/linux/mm.h index
> > c274f75efcf9..2b7527e93c77 100644
> > --- a/include/linux/mm.h
> > +++ b/include/linux/mm.h
> > @@ -1187,6 +1187,14 @@ static inline bool is_device_private_page(const
> struct page *page)
> >                 page->pgmap->type == MEMORY_DEVICE_PRIVATE;  }
> >
> > +static inline bool is_device_fsdax_page(const struct page *page) {
> > +       return IS_ENABLED(CONFIG_DEV_PAGEMAP_OPS) &&
> > +               IS_ENABLED(CONFIG_FS_DAX) &&
> > +               is_zone_device_page(page) &&
> > +               page->pgmap->type == MEMORY_DEVICE_FS_DAX;
> 
> Why is this necessary? The dax_dev holder is the one that knows the nature of
> the pfn. The common memory_failure() code should not care about fsdax vs
> devdax.

add_to_kill() in collect_procs() needs this. Please see explanation at below.

> 
> > +}
> > +
> >  static inline bool is_pci_p2pdma_page(const struct page *page)  {
> >         return IS_ENABLED(CONFIG_DEV_PAGEMAP_OPS) && @@
> -3078,6
> > +3086,7 @@ enum mf_flags {
> >         MF_MUST_KILL = 1 << 2,
> >         MF_SOFT_OFFLINE = 1 << 3,
> >  };
> > +int mf_dax_kill_procs(struct address_space *mapping, pgoff_t index,
> > +int flags);
> >  extern int memory_failure(unsigned long pfn, int flags);  extern void
> > memory_failure_queue(unsigned long pfn, int flags);  extern void
> > memory_failure_queue_kick(int cpu); diff --git a/mm/memory-failure.c
> > b/mm/memory-failure.c index 85ad98c00fd9..4377e727d478 100644
> > --- a/mm/memory-failure.c
> > +++ b/mm/memory-failure.c
> > @@ -56,6 +56,7 @@
> >  #include <linux/kfifo.h>
> >  #include <linux/ratelimit.h>
> >  #include <linux/page-isolation.h>
> > +#include <linux/dax.h>
> >  #include "internal.h"
> >  #include "ras/ras_event.h"
> >
> > @@ -120,6 +121,13 @@ static int hwpoison_filter_dev(struct page *p)
> >         if (PageSlab(p))
> >                 return -EINVAL;
> >
> > +       if (pfn_valid(page_to_pfn(p))) {
> > +               if (is_device_fsdax_page(p))
> 
> This is racy unless the page is pinned. Also, not clear why this is needed?
> 
> > +                       return 0;
> > +               else
> > +                       return -EINVAL;
> > +       }
> > +
> >         mapping = page_mapping(p);
> >         if (mapping == NULL || mapping->host == NULL)
> >                 return -EINVAL;
> > @@ -290,10 +298,9 @@ void shake_page(struct page *p, int access)  }
> > EXPORT_SYMBOL_GPL(shake_page);
> >
> > -static unsigned long dev_pagemap_mapping_shift(struct page *page,
> > -               struct vm_area_struct *vma)
> > +static unsigned long dev_pagemap_mapping_shift(struct vm_area_struct
> *vma,
> > +                                              unsigned long
> address)
> >  {
> > -       unsigned long address = vma_address(page, vma);
> >         pgd_t *pgd;
> >         p4d_t *p4d;
> >         pud_t *pud;
> > @@ -333,9 +340,8 @@ static unsigned long
> dev_pagemap_mapping_shift(struct page *page,
> >   * Schedule a process for later kill.
> >   * Uses GFP_ATOMIC allocations to avoid potential recursions in the VM.
> >   */
> > -static void add_to_kill(struct task_struct *tsk, struct page *p,
> > -                      struct vm_area_struct *vma,
> > -                      struct list_head *to_kill)
> > +static void add_to_kill(struct task_struct *tsk, struct page *p, pgoff_t pgoff,
> > +                       struct vm_area_struct *vma, struct list_head
> > +*to_kill)
> >  {
> >         struct to_kill *tk;
> >
> > @@ -346,9 +352,12 @@ static void add_to_kill(struct task_struct *tsk, struct
> page *p,
> >         }
> >
> >         tk->addr = page_address_in_vma(p, vma);
> > -       if (is_zone_device_page(p))
> > -               tk->size_shift = dev_pagemap_mapping_shift(p, vma);
> > -       else
> > +       if (is_zone_device_page(p)) {
> > +               if (is_device_fsdax_page(p))
> > +                       tk->addr = vma->vm_start +
> > +                                       ((pgoff - vma->vm_pgoff) <<
> PAGE_SHIFT);
> > +               tk->size_shift = dev_pagemap_mapping_shift(vma,
> > + tk->addr);
> 
> What was wrong with the original code?

Here is the explanation: For normal page, it associate file's mapping and offset to page->mapping, page->index for rmap tracking.  But for fsdax, in order to support reflink, we no longer use this mechanism, using dax_device holder stack instead.  Thus, this dax page->mapping is NULL.  As a result, we need is_device_fsdax_page(p) to distinguish if a page is a fsdax page and calculate this tk->addr manually.

> 
> > +       } else
> >                 tk->size_shift = page_shift(compound_head(p));
> >
> >         /*
> > @@ -496,7 +505,7 @@ static void collect_procs_anon(struct page *page,
> struct list_head *to_kill,
> >                         if (!page_mapped_in_vma(page, vma))
> >                                 continue;
> >                         if (vma->vm_mm == t->mm)
> > -                               add_to_kill(t, page, vma, to_kill);
> > +                               add_to_kill(t, page, 0, vma, to_kill);
> >                 }
> >         }
> >         read_unlock(&tasklist_lock);
> > @@ -506,24 +515,19 @@ static void collect_procs_anon(struct page
> > *page, struct list_head *to_kill,
> >  /*
> >   * Collect processes when the error hit a file mapped page.
> >   */
> > -static void collect_procs_file(struct page *page, struct list_head *to_kill,
> > -                               int force_early)
> > +static void collect_procs_file(struct page *page, struct address_space
> *mapping,
> > +               pgoff_t pgoff, struct list_head *to_kill, int
> > +force_early)
> >  {
> 
> collect_procs() and kill_procs() are the only core memory_failure() helpers I
> expect would be exported for a fileystem dax_dev holder to call when it is trying
> to cleanup a memory_failure() on a reflink'd mapping.

Yes, they are the core we need.  But there are some small different when dealing with normal page and dax page.  So, I factor these two core functions into two helpers.  One is mf_generic_kill_procs() for normal page. Another is mf_dax_kill_procs() for dax page.

> 
> >         struct vm_area_struct *vma;
> >         struct task_struct *tsk;
> > -       struct address_space *mapping = page->mapping;
> > -       pgoff_t pgoff;
> >
> >         i_mmap_lock_read(mapping);
> >         read_lock(&tasklist_lock);
> > -       pgoff = page_to_pgoff(page);
> >         for_each_process(tsk) {
> >                 struct task_struct *t = task_early_kill(tsk,
> > force_early);
> > -
> >                 if (!t)
> >                         continue;
> > -               vma_interval_tree_foreach(vma, &mapping->i_mmap,
> pgoff,
> > -                                     pgoff) {
> > +               vma_interval_tree_foreach(vma, &mapping->i_mmap,
> > + pgoff, pgoff) {
> >                         /*
> >                          * Send early kill signal to tasks where a vma
> covers
> >                          * the page but the corrupted page is not
> > necessarily @@ -532,7 +536,7 @@ static void collect_procs_file(struct page
> *page, struct list_head *to_kill,
> >                          * to be informed of all such data corruptions.
> >                          */
> >                         if (vma->vm_mm == t->mm)
> > -                               add_to_kill(t, page, vma, to_kill);
> > +                               add_to_kill(t, page, pgoff, vma,
> > + to_kill);
> >                 }
> >         }
> >         read_unlock(&tasklist_lock);
> > @@ -551,7 +555,8 @@ static void collect_procs(struct page *page, struct
> list_head *tokill,
> >         if (PageAnon(page))
> >                 collect_procs_anon(page, tokill, force_early);
> >         else
> > -               collect_procs_file(page, tokill, force_early);
> > +               collect_procs_file(page, page_mapping(page),
> page_to_pgoff(page),
> > +                                  tokill, force_early);
> >  }
> >
> >  static const char *action_name[] = {
> > @@ -1218,6 +1223,51 @@ static int try_to_split_thp_page(struct page *page,
> const char *msg)
> >         return 0;
> >  }
> >
> > +static void unmap_and_kill(struct list_head *to_kill, unsigned long pfn,
> > +               struct address_space *mapping, pgoff_t index, int
> > +flags) {
> > +       struct to_kill *tk;
> > +       unsigned long size = 0;
> > +       loff_t start;
> > +
> > +       list_for_each_entry(tk, to_kill, nd)
> > +               if (tk->size_shift)
> > +                       size = max(size, 1UL << tk->size_shift);
> > +       if (size) {
> > +               /*
> > +                * Unmap the largest mapping to avoid breaking up
> > +                * device-dax mappings which are constant size. The
> > +                * actual size of the mapping being torn down is
> > +                * communicated in siginfo, see kill_proc()
> > +                */
> > +               start = (index << PAGE_SHIFT) & ~(size - 1);
> > +               unmap_mapping_range(mapping, start, size, 0);
> > +       }
> > +
> > +       kill_procs(to_kill, flags & MF_MUST_KILL, false, pfn, flags);
> > +}
> > +
> > +int mf_dax_kill_procs(struct address_space *mapping, pgoff_t index,
> > +int flags) {
> > +       LIST_HEAD(to_kill);
> > +       /* load the pfn of the dax mapping file */
> > +       unsigned long pfn = dax_load_pfn(mapping, index);
> > +
> > +       /*
> > +        * Unlike System-RAM there is no possibility to swap in a
> > +        * different physical page at a given virtual address, so all
> > +        * userspace consumption of ZONE_DEVICE memory necessitates
> > +        * SIGBUS (i.e. MF_MUST_KILL)
> > +        */
> > +       flags |= MF_ACTION_REQUIRED | MF_MUST_KILL;
> > +       collect_procs_file(pfn_to_page(pfn), mapping, index, &to_kill,
> > +                          flags & MF_ACTION_REQUIRED);
> > +
> > +       unmap_and_kill(&to_kill, pfn, mapping, index, flags);
> > +       return 0;
> > +}
> > +EXPORT_SYMBOL_GPL(mf_dax_kill_procs);
> > +
> >  static int memory_failure_hugetlb(unsigned long pfn, int flags)  {
> >         struct page *p = pfn_to_page(pfn); @@ -1298,12 +1348,8 @@
> > static int memory_failure_dev_pagemap(unsigned long pfn, int flags,
> >                 struct dev_pagemap *pgmap)  {
> >         struct page *page = pfn_to_page(pfn);
> > -       const bool unmap_success = true;
> > -       unsigned long size = 0;
> > -       struct to_kill *tk;
> > -       LIST_HEAD(tokill);
> > +       LIST_HEAD(to_kill);
> >         int rc = -EBUSY;
> > -       loff_t start;
> >         dax_entry_t cookie;
> >
> >         if (flags & MF_COUNT_INCREASED) @@ -1355,22 +1401,10 @@
> static
> > int memory_failure_dev_pagemap(unsigned long pfn, int flags,
> >          * SIGBUS (i.e. MF_MUST_KILL)
> >          */
> >         flags |= MF_ACTION_REQUIRED | MF_MUST_KILL;
> > -       collect_procs(page, &tokill, flags & MF_ACTION_REQUIRED);
> > +       collect_procs_file(page, page->mapping, page->index, &to_kill,
> > +                          flags & MF_ACTION_REQUIRED);
> >
> > -       list_for_each_entry(tk, &tokill, nd)
> > -               if (tk->size_shift)
> > -                       size = max(size, 1UL << tk->size_shift);
> > -       if (size) {
> > -               /*
> > -                * Unmap the largest mapping to avoid breaking up
> > -                * device-dax mappings which are constant size. The
> > -                * actual size of the mapping being torn down is
> > -                * communicated in siginfo, see kill_proc()
> > -                */
> > -               start = (page->index << PAGE_SHIFT) & ~(size - 1);
> > -               unmap_mapping_range(page->mapping, start, size, 0);
> > -       }
> > -       kill_procs(&tokill, flags & MF_MUST_KILL, !unmap_success, pfn,
> flags);
> > +       unmap_and_kill(&to_kill, pfn, page->mapping, page->index,
> > + flags);
> 
> There's just too much change in this patch and not enough justification of what
> is being refactored and why.

Sorry again for the mess...


--
Thanks,
Ruan.