| Message ID | 20230613215346.1022773-6-peterx@redhat.com (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | mm/gup: Unify hugetlb, speed up thp | expand |
On Tue, Jun 13, 2023 at 05:53:44PM -0400, Peter Xu wrote: > The only path that doesn't use generic "**pages" handling is the gate vma. > Make it use the same path, meanwhile tune the next_page label upper to > cover "**pages" handling. This prepares for THP handling for "**pages". > > Signed-off-by: Peter Xu <peterx@redhat.com> > --- > mm/gup.c | 7 +++---- > 1 file changed, 3 insertions(+), 4 deletions(-) > > diff --git a/mm/gup.c b/mm/gup.c > index 8d59ae4554e7..a2d1b3c4b104 100644 > --- a/mm/gup.c > +++ b/mm/gup.c > @@ -1135,7 +1135,7 @@ static long __get_user_pages(struct mm_struct *mm, > if (!vma && in_gate_area(mm, start)) { > ret = get_gate_page(mm, start & PAGE_MASK, > gup_flags, &vma, > - pages ? &pages[i] : NULL); > + pages ? &page : NULL); Good spot... ugh that we handled this differently. > if (ret) > goto out; > ctx.page_mask = 0; We can drop this line now right? As the new next_page block will duplicate this. > @@ -1205,19 +1205,18 @@ static long __get_user_pages(struct mm_struct *mm, > ret = PTR_ERR(page); > goto out; > } > - > - goto next_page; This is neat, we've already checked if pages != NULL so the if (pages) block at the new next_page label will not be run. > } else if (IS_ERR(page)) { > ret = PTR_ERR(page); > goto out; > } > +next_page: > if (pages) { > pages[i] = page; > flush_anon_page(vma, page, start); > flush_dcache_page(page); I guess there's no harm that we now flush here, though it seems to me to be superfluous, it's not a big deal I don't think. > ctx.page_mask = 0; > } > -next_page: > + > page_increm = 1 + (~(start >> PAGE_SHIFT) & ctx.page_mask); > if (page_increm > nr_pages) > page_increm = nr_pages; > -- > 2.40.1 > Other than that, LGTM, Reviewed-by: Lorenzo Stoakes <lstoakes@gmail.com>
On Sat, Jun 17, 2023 at 08:48:38PM +0100, Lorenzo Stoakes wrote: > On Tue, Jun 13, 2023 at 05:53:44PM -0400, Peter Xu wrote: > > The only path that doesn't use generic "**pages" handling is the gate vma. > > Make it use the same path, meanwhile tune the next_page label upper to > > cover "**pages" handling. This prepares for THP handling for "**pages". > > > > Signed-off-by: Peter Xu <peterx@redhat.com> > > --- > > mm/gup.c | 7 +++---- > > 1 file changed, 3 insertions(+), 4 deletions(-) > > > > diff --git a/mm/gup.c b/mm/gup.c > > index 8d59ae4554e7..a2d1b3c4b104 100644 > > --- a/mm/gup.c > > +++ b/mm/gup.c > > @@ -1135,7 +1135,7 @@ static long __get_user_pages(struct mm_struct *mm, > > if (!vma && in_gate_area(mm, start)) { > > ret = get_gate_page(mm, start & PAGE_MASK, > > gup_flags, &vma, > > - pages ? &pages[i] : NULL); > > + pages ? &page : NULL); > > Good spot... ugh that we handled this differently. > > > if (ret) > > goto out; > > ctx.page_mask = 0; > > We can drop this line now right? As the new next_page block will duplicate > this. OK I can see why you left this in given the last patch in the series :) Please disregard. > > > @@ -1205,19 +1205,18 @@ static long __get_user_pages(struct mm_struct *mm, > > ret = PTR_ERR(page); > > goto out; > > } > > - > > - goto next_page; > > This is neat, we've already checked if pages != NULL so the if (pages) > block at the new next_page label will not be run. > > > } else if (IS_ERR(page)) { > > ret = PTR_ERR(page); > > goto out; > > } > > +next_page: > > if (pages) { > > pages[i] = page; > > flush_anon_page(vma, page, start); > > flush_dcache_page(page); > > I guess there's no harm that we now flush here, though it seems to me to be > superfluous, it's not a big deal I don't think. > > > ctx.page_mask = 0; > > } > > -next_page: > > + > > page_increm = 1 + (~(start >> PAGE_SHIFT) & ctx.page_mask); > > if (page_increm > nr_pages) > > page_increm = nr_pages; > > -- > > 2.40.1 > > > > Other than that, LGTM, > > Reviewed-by: Lorenzo Stoakes <lstoakes@gmail.com>
On Sat, Jun 17, 2023 at 09:00:34PM +0100, Lorenzo Stoakes wrote: > On Sat, Jun 17, 2023 at 08:48:38PM +0100, Lorenzo Stoakes wrote: > > On Tue, Jun 13, 2023 at 05:53:44PM -0400, Peter Xu wrote: > > > The only path that doesn't use generic "**pages" handling is the gate vma. > > > Make it use the same path, meanwhile tune the next_page label upper to > > > cover "**pages" handling. This prepares for THP handling for "**pages". > > > > > > Signed-off-by: Peter Xu <peterx@redhat.com> > > > --- > > > mm/gup.c | 7 +++---- > > > 1 file changed, 3 insertions(+), 4 deletions(-) > > > > > > diff --git a/mm/gup.c b/mm/gup.c > > > index 8d59ae4554e7..a2d1b3c4b104 100644 > > > --- a/mm/gup.c > > > +++ b/mm/gup.c > > > @@ -1135,7 +1135,7 @@ static long __get_user_pages(struct mm_struct *mm, > > > if (!vma && in_gate_area(mm, start)) { > > > ret = get_gate_page(mm, start & PAGE_MASK, > > > gup_flags, &vma, > > > - pages ? &pages[i] : NULL); > > > + pages ? &page : NULL); > > > > Good spot... ugh that we handled this differently. > > > > > if (ret) > > > goto out; > > > ctx.page_mask = 0; > > > > We can drop this line now right? As the new next_page block will duplicate > > this. > > OK I can see why you left this in given the last patch in the series :) > Please disregard. Yes the other "page_mask=0" will be removed in the next (not last) patch. > > > > > > @@ -1205,19 +1205,18 @@ static long __get_user_pages(struct mm_struct *mm, > > > ret = PTR_ERR(page); > > > goto out; > > > } > > > - > > > - goto next_page; > > > > This is neat, we've already checked if pages != NULL so the if (pages) > > block at the new next_page label will not be run. Yes. > > > > > } else if (IS_ERR(page)) { > > > ret = PTR_ERR(page); > > > goto out; > > > } > > > +next_page: > > > if (pages) { > > > pages[i] = page; > > > flush_anon_page(vma, page, start); > > > flush_dcache_page(page); > > > > I guess there's no harm that we now flush here, though it seems to me to be > > superfluous, it's not a big deal I don't think. I'd say GUP on gate vma page should be so rare so yeah I think it shouldn't be a big deal. Even iiuc vsyscall=xonly should be the default, so gup may have already failed on a gate vma page even trying to read-only.. > > > > > ctx.page_mask = 0; > > > } > > > -next_page: > > > + > > > page_increm = 1 + (~(start >> PAGE_SHIFT) & ctx.page_mask); > > > if (page_increm > nr_pages) > > > page_increm = nr_pages; > > > -- > > > 2.40.1 > > > > > > > Other than that, LGTM, > > > > Reviewed-by: Lorenzo Stoakes <lstoakes@gmail.com> Thanks for looking!
diff --git a/mm/gup.c b/mm/gup.c index 8d59ae4554e7..a2d1b3c4b104 100644 --- a/mm/gup.c +++ b/mm/gup.c @@ -1135,7 +1135,7 @@ static long __get_user_pages(struct mm_struct *mm, if (!vma && in_gate_area(mm, start)) { ret = get_gate_page(mm, start & PAGE_MASK, gup_flags, &vma, - pages ? &pages[i] : NULL); + pages ? &page : NULL); if (ret) goto out; ctx.page_mask = 0; @@ -1205,19 +1205,18 @@ static long __get_user_pages(struct mm_struct *mm, ret = PTR_ERR(page); goto out; } - - goto next_page; } else if (IS_ERR(page)) { ret = PTR_ERR(page); goto out; } +next_page: if (pages) { pages[i] = page; flush_anon_page(vma, page, start); flush_dcache_page(page); ctx.page_mask = 0; } -next_page: + page_increm = 1 + (~(start >> PAGE_SHIFT) & ctx.page_mask); if (page_increm > nr_pages) page_increm = nr_pages;
The only path that doesn't use generic "**pages" handling is the gate vma. Make it use the same path, meanwhile tune the next_page label upper to cover "**pages" handling. This prepares for THP handling for "**pages". Signed-off-by: Peter Xu <peterx@redhat.com> --- mm/gup.c | 7 +++---- 1 file changed, 3 insertions(+), 4 deletions(-)