| Message ID | 20200204234117.2974687-11-jhubbard@nvidia.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | mm/gup: track FOLL_PIN pages | expand |
On Tue 04-02-20 15:41:15, John Hubbard wrote: > Now that pages are "DMA-pinned" via pin_user_page*(), and unpinned via > unpin_user_pages*(), we need some visibility into whether all of this is > working correctly. > > Add two new fields to /proc/vmstat: > > nr_foll_pin_acquired > nr_foll_pin_released > > These are documented in Documentation/core-api/pin_user_pages.rst. > They represent the number of pages (since boot time) that have been > pinned ("nr_foll_pin_acquired") and unpinned ("nr_foll_pin_released"), > via pin_user_pages*() and unpin_user_pages*(). > > In the absence of long-running DMA or RDMA operations that hold pages > pinned, the above two fields will normally be equal to each other. > > Also: update Documentation/core-api/pin_user_pages.rst, to remove an > earlier (now confirmed untrue) claim about a performance problem with > /proc/vmstat. > > Also: updated Documentation/core-api/pin_user_pages.rst to rename the > new /proc/vmstat entries, to the names listed here. > > Signed-off-by: John Hubbard <jhubbard@nvidia.com> ... > @@ -104,6 +106,9 @@ static __maybe_unused struct page *try_grab_compound_head(struct page *page, > if (hpage_pincount_available(page)) > hpage_pincount_add(page, refs); > > + mod_node_page_state(page_pgdat(page), NR_FOLL_PIN_ACQUIRED, > + orig_refs); > + > return page; > } > It seems to me you miss mod_node_page_state() in put_compound_head(), don't you? Otherwise I like the new stat names better :). Honza
On Tue, Feb 04, 2020 at 03:41:15PM -0800, John Hubbard wrote: > Now that pages are "DMA-pinned" via pin_user_page*(), and unpinned via > unpin_user_pages*(), we need some visibility into whether all of this is > working correctly. > > Add two new fields to /proc/vmstat: > > nr_foll_pin_acquired > nr_foll_pin_released > > These are documented in Documentation/core-api/pin_user_pages.rst. > They represent the number of pages (since boot time) that have been > pinned ("nr_foll_pin_acquired") and unpinned ("nr_foll_pin_released"), > via pin_user_pages*() and unpin_user_pages*(). > > In the absence of long-running DMA or RDMA operations that hold pages > pinned, the above two fields will normally be equal to each other. > > Also: update Documentation/core-api/pin_user_pages.rst, to remove an > earlier (now confirmed untrue) claim about a performance problem with > /proc/vmstat. > > Also: updated Documentation/core-api/pin_user_pages.rst to rename the > new /proc/vmstat entries, to the names listed here. > > Signed-off-by: John Hubbard <jhubbard@nvidia.com> Please, clarify semantics for huge page. An user may want to know if we count huge page as one pin-acquired or by number of pages. Otherwise looks good (given Jan concern is addressed).
On 2/5/20 1:37 AM, Jan Kara wrote: > ... > >> @@ -104,6 +106,9 @@ static __maybe_unused struct page *try_grab_compound_head(struct page *page, >> if (hpage_pincount_available(page)) >> hpage_pincount_add(page, refs); >> >> + mod_node_page_state(page_pgdat(page), NR_FOLL_PIN_ACQUIRED, >> + orig_refs); >> + >> return page; >> } >> > > It seems to me you miss mod_node_page_state() in put_compound_head(), don't > you? Yes, that was definitely missing. I've added this for the next version: diff --git a/mm/gup.c b/mm/gup.c index 7c543849181b..ae503c51bc7f 100644 --- a/mm/gup.c +++ b/mm/gup.c @@ -2268,6 +2268,8 @@ static int record_subpages(struct page *page, unsigned long addr, static void put_compound_head(struct page *page, int refs, unsigned int flags) { + int orig_refs = refs; + if (flags & FOLL_PIN) { if (hpage_pincount_available(page)) hpage_pincount_sub(page, refs); @@ -2283,6 +2285,8 @@ static void put_compound_head(struct page *page, int refs, unsigned int flags) if (refs > 1) page_ref_sub(page, refs - 1); put_page(page); + + mod_node_page_state(page_pgdat(page), NR_FOLL_PIN_RELEASED, orig_refs); } #ifdef CONFIG_ARCH_HAS_HUGEPD > > Otherwise I like the new stat names better :). > > Honza > Glad to hear that! :) thanks,
On 2/5/20 3:43 AM, Kirill A. Shutemov wrote: > On Tue, Feb 04, 2020 at 03:41:15PM -0800, John Hubbard wrote: >> Now that pages are "DMA-pinned" via pin_user_page*(), and unpinned via >> unpin_user_pages*(), we need some visibility into whether all of this is >> working correctly. >> >> Add two new fields to /proc/vmstat: >> >> nr_foll_pin_acquired >> nr_foll_pin_released >> >> These are documented in Documentation/core-api/pin_user_pages.rst. >> They represent the number of pages (since boot time) that have been >> pinned ("nr_foll_pin_acquired") and unpinned ("nr_foll_pin_released"), >> via pin_user_pages*() and unpin_user_pages*(). >> >> In the absence of long-running DMA or RDMA operations that hold pages >> pinned, the above two fields will normally be equal to each other. >> >> Also: update Documentation/core-api/pin_user_pages.rst, to remove an >> earlier (now confirmed untrue) claim about a performance problem with >> /proc/vmstat. >> >> Also: updated Documentation/core-api/pin_user_pages.rst to rename the >> new /proc/vmstat entries, to the names listed here. >> >> Signed-off-by: John Hubbard <jhubbard@nvidia.com> > > Please, clarify semantics for huge page. An user may want to know if we > count huge page as one pin-acquired or by number of pages. OK, I've added this for the next version: diff --git a/Documentation/core-api/pin_user_pages.rst b/Documentation/core-api/pin_user_pages.rst index 5776ad1ed5e4..2e939ff10b86 100644 --- a/Documentation/core-api/pin_user_pages.rst +++ b/Documentation/core-api/pin_user_pages.rst @@ -211,6 +211,33 @@ since the system was booted, via two new /proc/vmstat entries: :: /proc/vmstat/nr_foll_pin_acquired /proc/vmstat/nr_foll_pin_released +Under normal conditions, these two values will be equal unless there are any +long-term [R]DMA pins in place, or during pin/unpin transitions. + +* nr_foll_pin_acquired: This is the number of logical pins that have been + acquired since the system was powered on. For huge pages, the head page is + pinned once for each page (head page and each tail page) within the huge page. + This follows the same sort of behavior that get_user_pages() uses for huge + pages: the head page is refcounted once for each tail or head page in the huge + page, when get_user_pages() is applied to a huge page. + +* nr_foll_pin_released: The number of logical pins that have been released since + the system was powered on. Note that pages are released (unpinned) on a + PAGE_SIZE granularity, even if the original pin was applied to a huge page. + Becaused of the pin count behavior described above in "nr_foll_pin_acquired", + the accounting balances out, so that after doing this:: + + pin_user_pages(huge_page); + for (each page in huge_page) + unpin_user_page(page); + +...the following is expected:: + + nr_foll_pin_released == nr_foll_pin_acquired + +(...unless it was already out of balance due to a long-term RDMA pin being in +place.) + Other diagnostics ================= thanks,
diff --git a/Documentation/core-api/pin_user_pages.rst b/Documentation/core-api/pin_user_pages.rst index dd21ea140ef4..194204d74a50 100644 --- a/Documentation/core-api/pin_user_pages.rst +++ b/Documentation/core-api/pin_user_pages.rst @@ -208,12 +208,8 @@ has the following new calls to exercise the new pin*() wrapper functions: You can monitor how many total dma-pinned pages have been acquired and released since the system was booted, via two new /proc/vmstat entries: :: - /proc/vmstat/nr_foll_pin_requested - /proc/vmstat/nr_foll_pin_requested - -Those are both going to show zero, unless CONFIG_DEBUG_VM is set. This is -because there is a noticeable performance drop in unpin_user_page(), when they -are activated. + /proc/vmstat/nr_foll_pin_acquired + /proc/vmstat/nr_foll_pin_released Other diagnostics ================= diff --git a/include/linux/mmzone.h b/include/linux/mmzone.h index 462f6873905a..4bca42eeb439 100644 --- a/include/linux/mmzone.h +++ b/include/linux/mmzone.h @@ -243,6 +243,8 @@ enum node_stat_item { NR_DIRTIED, /* page dirtyings since bootup */ NR_WRITTEN, /* page writings since bootup */ NR_KERNEL_MISC_RECLAIMABLE, /* reclaimable non-slab kernel pages */ + NR_FOLL_PIN_ACQUIRED, /* via: pin_user_page(), gup flag: FOLL_PIN */ + NR_FOLL_PIN_RELEASED, /* pages returned via unpin_user_page() */ NR_VM_NODE_STAT_ITEMS }; diff --git a/mm/gup.c b/mm/gup.c index 4d0d94405639..7c543849181b 100644 --- a/mm/gup.c +++ b/mm/gup.c @@ -86,6 +86,8 @@ static __maybe_unused struct page *try_grab_compound_head(struct page *page, if (flags & FOLL_GET) return try_get_compound_head(page, refs); else if (flags & FOLL_PIN) { + int orig_refs = refs; + /* * When pinning a compound page of order > 1 (which is what * hpage_pincount_available() checks for), use an exact count to @@ -104,6 +106,9 @@ static __maybe_unused struct page *try_grab_compound_head(struct page *page, if (hpage_pincount_available(page)) hpage_pincount_add(page, refs); + mod_node_page_state(page_pgdat(page), NR_FOLL_PIN_ACQUIRED, + orig_refs); + return page; } @@ -158,6 +163,8 @@ bool __must_check try_grab_page(struct page *page, unsigned int flags) * once, so that the page really is pinned. */ page_ref_add(page, refs); + + mod_node_page_state(page_pgdat(page), NR_FOLL_PIN_ACQUIRED, 1); } return true; @@ -178,6 +185,7 @@ static bool __unpin_devmap_managed_user_page(struct page *page) count = page_ref_sub_return(page, refs); + mod_node_page_state(page_pgdat(page), NR_FOLL_PIN_RELEASED, 1); /* * devmap page refcounts are 1-based, rather than 0-based: if * refcount is 1, then the page is free and the refcount is @@ -228,6 +236,8 @@ void unpin_user_page(struct page *page) if (page_ref_sub_and_test(page, refs)) __put_page(page); + + mod_node_page_state(page_pgdat(page), NR_FOLL_PIN_RELEASED, 1); } EXPORT_SYMBOL(unpin_user_page); diff --git a/mm/vmstat.c b/mm/vmstat.c index 78d53378db99..c9c0d71f917f 100644 --- a/mm/vmstat.c +++ b/mm/vmstat.c @@ -1168,6 +1168,8 @@ const char * const vmstat_text[] = { "nr_dirtied", "nr_written", "nr_kernel_misc_reclaimable", + "nr_foll_pin_acquired", + "nr_foll_pin_released", /* enum writeback_stat_item counters */ "nr_dirty_threshold",
Now that pages are "DMA-pinned" via pin_user_page*(), and unpinned via unpin_user_pages*(), we need some visibility into whether all of this is working correctly. Add two new fields to /proc/vmstat: nr_foll_pin_acquired nr_foll_pin_released These are documented in Documentation/core-api/pin_user_pages.rst. They represent the number of pages (since boot time) that have been pinned ("nr_foll_pin_acquired") and unpinned ("nr_foll_pin_released"), via pin_user_pages*() and unpin_user_pages*(). In the absence of long-running DMA or RDMA operations that hold pages pinned, the above two fields will normally be equal to each other. Also: update Documentation/core-api/pin_user_pages.rst, to remove an earlier (now confirmed untrue) claim about a performance problem with /proc/vmstat. Also: updated Documentation/core-api/pin_user_pages.rst to rename the new /proc/vmstat entries, to the names listed here. Signed-off-by: John Hubbard <jhubbard@nvidia.com> --- Documentation/core-api/pin_user_pages.rst | 8 ++------ include/linux/mmzone.h | 2 ++ mm/gup.c | 10 ++++++++++ mm/vmstat.c | 2 ++ 4 files changed, 16 insertions(+), 6 deletions(-)