| Message ID | 20210318015959.2986837-1-shakeelb@google.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | memcg: set page->private before calling swap_readpage | expand |
On Wed, Mar 17, 2021 at 06:59:59PM -0700, Shakeel Butt wrote: > The function swap_readpage() (and other functions it call) extracts swap > entry from page->private. However for SWP_SYNCHRONOUS_IO, the kernel > skips the swapcache and thus we need to manually set the page->private > with the swap entry before calling swap_readpage(). > > Signed-off-by: Shakeel Butt <shakeelb@google.com> > Reported-by: Heiko Carstens <hca@linux.ibm.com> > --- > > Andrew, please squash this into "memcg: charge before adding to > swapcache on swapin" patch. > > mm/memory.c | 4 ++++ > 1 file changed, 4 insertions(+) > > diff --git a/mm/memory.c b/mm/memory.c > index aefd158ae1ea..b6f3410b5902 100644 > --- a/mm/memory.c > +++ b/mm/memory.c > @@ -3324,7 +3324,11 @@ vm_fault_t do_swap_page(struct vm_fault *vmf) > workingset_refault(page, shadow); > > lru_cache_add(page); > + > + /* To provide entry to swap_readpage() */ > + set_page_private(page, entry.val); > swap_readpage(page, true); > + set_page_private(page, 0); Yes, this seems to fix it. Thanks a lot! Tested-by: Heiko Carstens <hca@linux.ibm.com>
On Wed 17-03-21 18:59:59, Shakeel Butt wrote: > The function swap_readpage() (and other functions it call) extracts swap > entry from page->private. However for SWP_SYNCHRONOUS_IO, the kernel > skips the swapcache and thus we need to manually set the page->private > with the swap entry before calling swap_readpage(). One thing that is not really clear to me is whether/why this is only needed with your patch. Can you expand a bit on that please? Maybe I am just missing something obvious but I just do not see any connection. > Signed-off-by: Shakeel Butt <shakeelb@google.com> > Reported-by: Heiko Carstens <hca@linux.ibm.com> > --- > > Andrew, please squash this into "memcg: charge before adding to > swapcache on swapin" patch. > > mm/memory.c | 4 ++++ > 1 file changed, 4 insertions(+) > > diff --git a/mm/memory.c b/mm/memory.c > index aefd158ae1ea..b6f3410b5902 100644 > --- a/mm/memory.c > +++ b/mm/memory.c > @@ -3324,7 +3324,11 @@ vm_fault_t do_swap_page(struct vm_fault *vmf) > workingset_refault(page, shadow); > > lru_cache_add(page); > + > + /* To provide entry to swap_readpage() */ > + set_page_private(page, entry.val); > swap_readpage(page, true); > + set_page_private(page, 0); > } > } else { > page = swapin_readahead(entry, GFP_HIGHUSER_MOVABLE, > -- > 2.31.0.rc2.261.g7f71774620-goog >
On Thu, Mar 18, 2021 at 03:01:25PM +0100, Michal Hocko wrote: > On Wed 17-03-21 18:59:59, Shakeel Butt wrote: > > The function swap_readpage() (and other functions it call) extracts swap > > entry from page->private. However for SWP_SYNCHRONOUS_IO, the kernel > > skips the swapcache and thus we need to manually set the page->private > > with the swap entry before calling swap_readpage(). > > One thing that is not really clear to me is whether/why this is only > needed with your patch. Can you expand a bit on that please? Maybe I am > just missing something obvious but I just do not see any connection. It was always needed, his original patch erroneously removed it.
On Wed, Mar 17, 2021 at 06:59:59PM -0700, Shakeel Butt wrote: > The function swap_readpage() (and other functions it call) extracts swap > entry from page->private. However for SWP_SYNCHRONOUS_IO, the kernel > skips the swapcache and thus we need to manually set the page->private > with the swap entry before calling swap_readpage(). > > Signed-off-by: Shakeel Butt <shakeelb@google.com> > Reported-by: Heiko Carstens <hca@linux.ibm.com> LGTM
On Thu 18-03-21 11:02:17, Johannes Weiner wrote: > On Thu, Mar 18, 2021 at 03:01:25PM +0100, Michal Hocko wrote: > > On Wed 17-03-21 18:59:59, Shakeel Butt wrote: > > > The function swap_readpage() (and other functions it call) extracts swap > > > entry from page->private. However for SWP_SYNCHRONOUS_IO, the kernel > > > skips the swapcache and thus we need to manually set the page->private > > > with the swap entry before calling swap_readpage(). > > > > One thing that is not really clear to me is whether/why this is only > > needed with your patch. Can you expand a bit on that please? Maybe I am > > just missing something obvious but I just do not see any connection. > > It was always needed, his original patch erroneously removed it. Ahh, I can see it now. I must have been blind. Thanks!
diff --git a/mm/memory.c b/mm/memory.c index aefd158ae1ea..b6f3410b5902 100644 --- a/mm/memory.c +++ b/mm/memory.c @@ -3324,7 +3324,11 @@ vm_fault_t do_swap_page(struct vm_fault *vmf) workingset_refault(page, shadow); lru_cache_add(page); + + /* To provide entry to swap_readpage() */ + set_page_private(page, entry.val); swap_readpage(page, true); + set_page_private(page, 0); } } else { page = swapin_readahead(entry, GFP_HIGHUSER_MOVABLE,
The function swap_readpage() (and other functions it call) extracts swap entry from page->private. However for SWP_SYNCHRONOUS_IO, the kernel skips the swapcache and thus we need to manually set the page->private with the swap entry before calling swap_readpage(). Signed-off-by: Shakeel Butt <shakeelb@google.com> Reported-by: Heiko Carstens <hca@linux.ibm.com> --- Andrew, please squash this into "memcg: charge before adding to swapcache on swapin" patch. mm/memory.c | 4 ++++ 1 file changed, 4 insertions(+)