Fix bogus get_page() calls in hugepage code
Andrew, please apply:
On some archs the functions used to implement follow_page() for hugepages do a get_page(). This is unlike the normal-page path for follow_page(), so presumably a bug. This patch fixes it.
arch/ppc64/mm/hugetlbpage.c +1 -3
@@ -360,10 +360,8 @@BUG_ON (!pmd_hugepage (*pmd)); page =hugepte_page (*(hugepte_t *)pmd); -if (page) { +if (page) page += ((address & ~HPAGE_MASK ) >>PAGE_SHIFT ); -get_page (page); - }return page; }
arch/i386/mm/hugetlbpage.c +1 -3
@@ -206,10 +206,8 @@struct page *page; page =pte_page (*(pte_t *)pmd); -if (page) { +if (page) page += ((address & ~HPAGE_MASK ) >>PAGE_SHIFT ); -get_page (page); - } return page; }#endif
arch/ia64/mm/hugetlbpage.c +0 -1
@@ -170,7 +170,6 @@ ptep =huge_pte_offset (mm, addr); page =pte_page (*ptep); page += ((addr & ~HPAGE_MASK ) >>PAGE_SHIFT ); -get_page (page);return page; }int pmd_huge (pmd_t pmd)
Signature · 4 lines
David Gibson | For every complex problem there is a david AT gibson.dropbear.id.au | solution which is simple, neat and | wrong. http://www.ozlabs.org/people/dgibson
Quoted 4 lines · David Gibson
On some archs the functions used to implement
follow_page()for
hugepages do aget_page(). This is unlike the normal-page path forfollow_page(), so presumably a bug. This patch fixes it.
get_user_pages() is supposed to pin the pages which it placed into the callers pages[] array.
And the caller of get_user_pages() is supposed to unpin those pages when they are finished with.
So follow_hugetlb_page() is currently doing the right thing. The asymmetry with follow_page() is awkward, but the overall intent was to minimise the amount of impact which the hugepage code has on core MM.
Aside: note that get_user_pages() doesn't hold page_table_lock while walking the pagetables, whereas it does hold that lock while walking the regular pages' pagetables. This is because the caller of get_user_pages() holds down_read(mmap_sem), whereas hugetlb pagetable setup and teardown always happens under down_write(mmap_sem).
Quoted 9 lines · Andrew Morton
Quoted 4 lines · David Gibson
On some archs the functions used to implement
follow_page()for
hugepages do aget_page(). This is unlike the normal-page path forfollow_page(), so presumably a bug. This patch fixes it.
get_user_pages()is supposed to pin the pages which it placed into the callers pages[] array.
Quoted 6 lines
And the caller of
get_user_pages()is supposed to unpin those pages when they are finished with.So
follow_hugetlb_page()is currently doing the right thing. The asymmetry withfollow_page()is awkward, but the overall intent was to minimise the amount of impact which the hugepage code has on core MM.
Yes, but I'm not talking about follow_hugetlb_page() (my patch doesn't touch it). I'm talking about follow_huge_addr() and follow_huge_pmd(). These are called to implement follow_page(). In get_user_pages(), follow_pages() is bypassed bu the call to follow_hugetlb_page(), but these extra get_pages() are presumably a bug if follow_page() is called on a hugepage from somewhere other than get_user_pages().
So I think my patch is correct.
Quoted 5 lines
Aside: note that
get_user_pages()doesn't holdpage_table_lockwhile walking the pagetables, whereas it does hold that lock while walking the regular pages' pagetables. This is because the caller ofget_user_pages()holdsdown_read(mmap_sem), whereas hugetlb pagetable setup and teardown always happens underdown_write(mmap_sem).
Ah, that's useful to know.
Signature · 4 lines
David Gibson | For every complex problem there is a david AT gibson.dropbear.id.au | solution which is simple, neat and | wrong. http://www.ozlabs.org/people/dgibson
Quoted 18 lines · David Gibson
Quoted 9 lines · Andrew Morton
Quoted 4 lines · David Gibson
On some archs the functions used to implement
follow_page()for
hugepages do aget_page(). This is unlike the normal-page path forfollow_page(), so presumably a bug. This patch fixes it.
get_user_pages()is supposed to pin the pages which it placed into the callers pages[] array.Quoted 6 lines
And the caller of
get_user_pages()is supposed to unpin those pages when they are finished with.So
follow_hugetlb_page()is currently doing the right thing. The asymmetry withfollow_page()is awkward, but the overall intent was to minimise the amount of impact which the hugepage code has on core MM.
Quoted 7 lines
Yes, but I'm not talking about
follow_hugetlb_page()(my patch doesn't touch it). I'm talking aboutfollow_huge_addr()andfollow_huge_pmd(). These are called to implementfollow_page(). Inget_user_pages(),follow_pages()is bypassed bu the call tofollow_hugetlb_page(), but these extraget_pages()are presumably a bug iffollow_page()is called on a hugepage from somewhere other thanget_user_pages().
Oh, OK. IIRC that stuff was added to support futexes-in-large-pages. The caller holds ->page_table_lock, as does zap_hugepage_range(), so that seems fine.
Quoted 1 line
So I think my patch is correct.
yup, thanks.