Message ID | 20190607095213.13372-1-liwang@redhat.com |
---|---|
State | Superseded |
Headers | show |
Series | [RFC] move_pages12: handle errno EBUSY for madvise(..., MADV_SOFT_OFFLINE) | expand |
Hi Li Wang, Thank you for maintaining the testcase. Recently (since 4.19) we have a semantics change on the return value of madvise(MADV_SOFT_OFFLINE), and we see -EBUSY when hugepage migration succeeded and error containment failed: commit 6bc9b56433b76e40d11099338d27fbc5cd2935ca Author: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com> Date: Thu Aug 23 17:00:38 2018 -0700 mm: fix race on soft-offlining free huge pages , so we don't have to consider this EBUSY as error, but a good report for application. Your change meets the change. Feel free to add my ack: Acked-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com> Thanks, - Naoya On Fri, Jun 07, 2019 at 05:52:13PM +0800, Li Wang wrote: > The test#2 is going to simulate the race condition, where move_pages() > and soft offline are called on a single hugetlb page concurrently. But, > it return EBUSY and report FAIL in soft-offline a moving hugepage as a > result sometimes. > > The root cause seems a call to page_huge_active return false, then the > soft offline action will failed to isolate hugepage with EBUSY return as > below call trace: > > In Parent: > madvise(..., MADV_SOFT_OFFLINE) > ... > soft_offline_page > soft_offline_in_use_page > soft_offline_huge_page > isolate_huge_page > page_huge_active --> return false at here > > In Child: > move_pages() > ... > do_move_pages > do_move_pages_to_node > add_page_for_migration > isolate_huge_page --> it has already isolated the hugepage > > In this patch, I simply regard the returned EBUSY as a normal situation and > mask it in error handler. Because move_pages is calling add_page_for_migration > to isolate hugepage before do migration, so that's very possible to hit the > collision and return EBUSY on the same page. > > Error log: > ---------- > move_pages12.c:235: INFO: Free RAM 8386256 kB > move_pages12.c:253: INFO: Increasing 2048kB hugepages pool on node 0 to 4 > move_pages12.c:263: INFO: Increasing 2048kB hugepages pool on node 1 to 6 > move_pages12.c:179: INFO: Allocating and freeing 4 hugepages on node 0 > move_pages12.c:179: INFO: Allocating and freeing 4 hugepages on node 1 > move_pages12.c:169: PASS: Bug not reproduced > move_pages12.c:81: FAIL: madvise failed: SUCCESS > move_pages12.c:81: FAIL: madvise failed: SUCCESS > move_pages12.c:143: BROK: mmap((nil),4194304,3,262178,-1,0) failed: ENOMEM > move_pages12.c:114: FAIL: move_pages failed: EINVAL > > Dmesg: > ------ > [165435.492170] soft offline: 0x61c00 hugepage failed to isolate > [165435.590252] soft offline: 0x61c00 hugepage failed to isolate > [165435.725493] soft offline: 0x61400 hugepage failed to isolate > > Other two fixes in this patch: > * use TERRNO(but not TTERRNO) to catch madvise(..., MADV_SOFT_OFFLINE) errno > * go out test when hugepage allocating failed with ENOMEM > > Signed-off-by: Li Wang <liwang@redhat.com> > Cc: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com> > Cc: Xiao Yang <yangx.jy@cn.fujitsu.com> > Cc: Yang Xu <xuyang2018.jy@cn.fujitsu.com> > --- > .../kernel/syscalls/move_pages/move_pages12.c | 33 ++++++++++++++----- > 1 file changed, 24 insertions(+), 9 deletions(-) > > diff --git a/testcases/kernel/syscalls/move_pages/move_pages12.c b/testcases/kernel/syscalls/move_pages/move_pages12.c > index 964b712fb..c446396dc 100644 > --- a/testcases/kernel/syscalls/move_pages/move_pages12.c > +++ b/testcases/kernel/syscalls/move_pages/move_pages12.c > @@ -77,8 +77,8 @@ static void *addr; > static int do_soft_offline(int tpgs) > { > if (madvise(addr, tpgs * hpsz, MADV_SOFT_OFFLINE) == -1) { > - if (errno != EINVAL) > - tst_res(TFAIL | TTERRNO, "madvise failed"); > + if (errno != EINVAL && errno != EBUSY) > + tst_res(TFAIL | TERRNO, "madvise failed"); > return errno; > } > return 0; > @@ -121,7 +121,8 @@ static void do_child(int tpgs) > > static void do_test(unsigned int n) > { > - int i; > + int i, ret; > + void *ptr; > pid_t cpid = -1; > int status; > unsigned int twenty_percent = (tst_timeout_remaining() / 5); > @@ -136,24 +137,37 @@ static void do_test(unsigned int n) > do_child(tcases[n].tpages); > > for (i = 0; i < LOOPS; i++) { > - void *ptr; > + ptr = mmap(NULL, tcases[n].tpages * hpsz, > + PROT_READ | PROT_WRITE, > + MAP_PRIVATE | MAP_ANONYMOUS | MAP_HUGETLB, -1, 0); > + if (ptr == MAP_FAILED) { > + if (errno == ENOMEM) { > + tst_res(TCONF, > + "Cannot allocate hugepage, memory too fragmented?"); > + goto out; > + } > + > + tst_brk(TBROK | TERRNO, "Cannot allocate hugepage"); > + } > > - ptr = SAFE_MMAP(NULL, tcases[n].tpages * hpsz, > - PROT_READ | PROT_WRITE, > - MAP_PRIVATE | MAP_ANONYMOUS | MAP_HUGETLB, -1, 0); > if (ptr != addr) > tst_brk(TBROK, "Failed to mmap at desired addr"); > > memset(addr, 0, tcases[n].tpages * hpsz); > > if (tcases[n].offline) { > - if (do_soft_offline(tcases[n].tpages) == EINVAL) { > + ret = do_soft_offline(tcases[n].tpages); > + > + if (ret == EINVAL) { > SAFE_KILL(cpid, SIGKILL); > SAFE_WAITPID(cpid, &status, 0); > SAFE_MUNMAP(addr, tcases[n].tpages * hpsz); > tst_res(TCONF, > "madvise() didn't support MADV_SOFT_OFFLINE"); > return; > + } else if (ret == EBUSY) { > + SAFE_MUNMAP(addr, tcases[n].tpages * hpsz); > + goto out; > } > } > > @@ -163,9 +177,10 @@ static void do_test(unsigned int n) > break; > } > > +out: > SAFE_KILL(cpid, SIGKILL); > SAFE_WAITPID(cpid, &status, 0); > - if (!WIFEXITED(status)) > + if (!WIFEXITED(status) && ptr != MAP_FAILED) > tst_res(TPASS, "Bug not reproduced"); > } > > -- > 2.20.1 > >
> Hi Li Wang, > > Thank you for maintaining the testcase. > > Recently (since 4.19) we have a semantics change on the return value of > madvise(MADV_SOFT_OFFLINE), and we see -EBUSY when hugepage migration > succeeded and error containment failed: > > commit 6bc9b56433b76e40d11099338d27fbc5cd2935ca > Author: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com> > Date: Thu Aug 23 17:00:38 2018 -0700 > > mm: fix race on soft-offlining free huge pages > > , so we don't have to consider this EBUSY as error, but a good report > for application. Your change meets the change. > > Feel free to add my ack: > > Acked-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com> > > Thanks, > - Naoya > > On Fri, Jun 07, 2019 at 05:52:13PM +0800, Li Wang wrote: >> The test#2 is going to simulate the race condition, where move_pages() >> and soft offline are called on a single hugetlb page concurrently. But, >> it return EBUSY and report FAIL in soft-offline a moving hugepage as a >> result sometimes. >> >> The root cause seems a call to page_huge_active return false, then the >> soft offline action will failed to isolate hugepage with EBUSY return as >> below call trace: >> >> In Parent: >> madvise(..., MADV_SOFT_OFFLINE) >> ... >> soft_offline_page >> soft_offline_in_use_page >> soft_offline_huge_page >> isolate_huge_page >> page_huge_active --> return false at here >> >> In Child: >> move_pages() >> ... >> do_move_pages >> do_move_pages_to_node >> add_page_for_migration >> isolate_huge_page --> it has already isolated the hugepage >> >> In this patch, I simply regard the returned EBUSY as a normal situation and >> mask it in error handler. Because move_pages is calling add_page_for_migration >> to isolate hugepage before do migration, so that's very possible to hit the >> collision and return EBUSY on the same page. >> >> Error log: >> ---------- >> move_pages12.c:235: INFO: Free RAM 8386256 kB >> move_pages12.c:253: INFO: Increasing 2048kB hugepages pool on node 0 to 4 >> move_pages12.c:263: INFO: Increasing 2048kB hugepages pool on node 1 to 6 >> move_pages12.c:179: INFO: Allocating and freeing 4 hugepages on node 0 >> move_pages12.c:179: INFO: Allocating and freeing 4 hugepages on node 1 >> move_pages12.c:169: PASS: Bug not reproduced >> move_pages12.c:81: FAIL: madvise failed: SUCCESS >> move_pages12.c:81: FAIL: madvise failed: SUCCESS >> move_pages12.c:143: BROK: mmap((nil),4194304,3,262178,-1,0) failed: ENOMEM >> move_pages12.c:114: FAIL: move_pages failed: EINVAL >> >> Dmesg: >> ------ >> [165435.492170] soft offline: 0x61c00 hugepage failed to isolate >> [165435.590252] soft offline: 0x61c00 hugepage failed to isolate >> [165435.725493] soft offline: 0x61400 hugepage failed to isolate >> >> Other two fixes in this patch: >> * use TERRNO(but not TTERRNO) to catch madvise(..., MADV_SOFT_OFFLINE) errno >> * go out test when hugepage allocating failed with ENOMEM Hi Li Your patch can handle EBUSY errno correctly for soft offline. But move page may be killed by SIGBUS because of MCE when we soft offline concurrently. That leads to move_page failed with ESRCH. Also, move page may fails with ENOMEM . Do you notice it ? I think ESRCH error can represent the soft offline bug not reproduce because it don't trigger a crash. What do you think about it? err_log: tst_test.c:1096: INFO: Timeout per run is 0h 05m 00s move_pages12.c:236: INFO: Free RAM 119568 kB move_pages12.c:254: INFO: Increasing 2048kB hugepages pool on node 0 to 83 move_pages12.c:264: INFO: Increasing 2048kB hugepages pool on node 1 to 94 move_pages12.c:180: INFO: Allocating and freeing 4 hugepages on node 0 move_pages12.c:180: INFO: Allocating and freeing 4 hugepages on node 1 move_pages12.c:170: PASS: Bug not reproduced tst_test.c:1141: BROK: Test killed by SIGBUS! Summary: passed 1 failed 0 skipped 0 warnings 0 move_pages12.c:114: FAIL: move_pages failed: ESRCH dmesg [ 9868.180669] MCE: Killing move_pages12:29616 due to hardware memory corruption fault at 2aaaaac00018 [ 9990.049875] Soft offlining page 50e00 at 2aaaaac00000 [ 9990.052218] Soft offlining page 50c00 at 2aaaaae00000 [ 9990.060395] Soft offlining page 51000 at 2aaaaac00000 Kind Regards, Yang Xu >> Signed-off-by: Li Wang <liwang@redhat.com> >> Cc: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com> >> Cc: Xiao Yang <yangx.jy@cn.fujitsu.com> >> Cc: Yang Xu <xuyang2018.jy@cn.fujitsu.com> >> --- >> .../kernel/syscalls/move_pages/move_pages12.c | 33 ++++++++++++++----- >> 1 file changed, 24 insertions(+), 9 deletions(-) >> >> diff --git a/testcases/kernel/syscalls/move_pages/move_pages12.c b/testcases/kernel/syscalls/move_pages/move_pages12.c >> index 964b712fb..c446396dc 100644 >> --- a/testcases/kernel/syscalls/move_pages/move_pages12.c >> +++ b/testcases/kernel/syscalls/move_pages/move_pages12.c >> @@ -77,8 +77,8 @@ static void *addr; >> static int do_soft_offline(int tpgs) >> { >> if (madvise(addr, tpgs * hpsz, MADV_SOFT_OFFLINE) == -1) { >> - if (errno != EINVAL) >> - tst_res(TFAIL | TTERRNO, "madvise failed"); >> + if (errno != EINVAL && errno != EBUSY) >> + tst_res(TFAIL | TERRNO, "madvise failed"); >> return errno; >> } >> return 0; >> @@ -121,7 +121,8 @@ static void do_child(int tpgs) >> >> static void do_test(unsigned int n) >> { >> - int i; >> + int i, ret; >> + void *ptr; >> pid_t cpid = -1; >> int status; >> unsigned int twenty_percent = (tst_timeout_remaining() / 5); >> @@ -136,24 +137,37 @@ static void do_test(unsigned int n) >> do_child(tcases[n].tpages); >> >> for (i = 0; i < LOOPS; i++) { >> - void *ptr; >> + ptr = mmap(NULL, tcases[n].tpages * hpsz, >> + PROT_READ | PROT_WRITE, >> + MAP_PRIVATE | MAP_ANONYMOUS | MAP_HUGETLB, -1, 0); >> + if (ptr == MAP_FAILED) { >> + if (errno == ENOMEM) { >> + tst_res(TCONF, >> + "Cannot allocate hugepage, memory too fragmented?"); >> + goto out; >> + } >> + >> + tst_brk(TBROK | TERRNO, "Cannot allocate hugepage"); >> + } >> >> - ptr = SAFE_MMAP(NULL, tcases[n].tpages * hpsz, >> - PROT_READ | PROT_WRITE, >> - MAP_PRIVATE | MAP_ANONYMOUS | MAP_HUGETLB, -1, 0); >> if (ptr != addr) >> tst_brk(TBROK, "Failed to mmap at desired addr"); >> >> memset(addr, 0, tcases[n].tpages * hpsz); >> >> if (tcases[n].offline) { >> - if (do_soft_offline(tcases[n].tpages) == EINVAL) { >> + ret = do_soft_offline(tcases[n].tpages); >> + >> + if (ret == EINVAL) { >> SAFE_KILL(cpid, SIGKILL); >> SAFE_WAITPID(cpid, &status, 0); >> SAFE_MUNMAP(addr, tcases[n].tpages * hpsz); >> tst_res(TCONF, >> "madvise() didn't support MADV_SOFT_OFFLINE"); >> return; >> + } else if (ret == EBUSY) { >> + SAFE_MUNMAP(addr, tcases[n].tpages * hpsz); >> + goto out; >> } >> } >> >> @@ -163,9 +177,10 @@ static void do_test(unsigned int n) >> break; >> } >> >> +out: >> SAFE_KILL(cpid, SIGKILL); >> SAFE_WAITPID(cpid, &status, 0); >> - if (!WIFEXITED(status)) >> + if (!WIFEXITED(status) && ptr != MAP_FAILED) >> tst_res(TPASS, "Bug not reproduced"); >> } >> >> -- >> 2.20.1 >> >> > > . >
Hi Xu Yang, On Fri, Jun 21, 2019 at 1:58 PM Yang Xu <xuyang2018.jy@cn.fujitsu.com> wrote: > > > Hi Li Wang, > > > > Thank you for maintaining the testcase. > > > > Recently (since 4.19) we have a semantics change on the return value of > > madvise(MADV_SOFT_OFFLINE), and we see -EBUSY when hugepage migration > > succeeded and error containment failed: > > > > commit 6bc9b56433b76e40d11099338d27fbc5cd2935ca > > Author: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com> > > Date: Thu Aug 23 17:00:38 2018 -0700 > > > > mm: fix race on soft-offlining free huge pages > > > > , so we don't have to consider this EBUSY as error, but a good report > > for application. Your change meets the change. > > > > Feel free to add my ack: > > > > Acked-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com> > > > > Thanks, > > - Naoya > > > > On Fri, Jun 07, 2019 at 05:52:13PM +0800, Li Wang wrote: > >> The test#2 is going to simulate the race condition, where move_pages() > >> and soft offline are called on a single hugetlb page concurrently. But, > >> it return EBUSY and report FAIL in soft-offline a moving hugepage as a > >> result sometimes. > >> > >> The root cause seems a call to page_huge_active return false, then the > >> soft offline action will failed to isolate hugepage with EBUSY return as > >> below call trace: > >> > >> In Parent: > >> madvise(..., MADV_SOFT_OFFLINE) > >> ... > >> soft_offline_page > >> soft_offline_in_use_page > >> soft_offline_huge_page > >> isolate_huge_page > >> page_huge_active --> return false at here > >> > >> In Child: > >> move_pages() > >> ... > >> do_move_pages > >> do_move_pages_to_node > >> add_page_for_migration > >> isolate_huge_page --> it has already isolated the hugepage > >> > >> In this patch, I simply regard the returned EBUSY as a normal situation > and > >> mask it in error handler. Because move_pages is calling > add_page_for_migration > >> to isolate hugepage before do migration, so that's very possible to hit > the > >> collision and return EBUSY on the same page. > >> > >> Error log: > >> ---------- > >> move_pages12.c:235: INFO: Free RAM 8386256 kB > >> move_pages12.c:253: INFO: Increasing 2048kB hugepages pool on node 0 to > 4 > >> move_pages12.c:263: INFO: Increasing 2048kB hugepages pool on node 1 to > 6 > >> move_pages12.c:179: INFO: Allocating and freeing 4 hugepages on node 0 > >> move_pages12.c:179: INFO: Allocating and freeing 4 hugepages on node 1 > >> move_pages12.c:169: PASS: Bug not reproduced > >> move_pages12.c:81: FAIL: madvise failed: SUCCESS > >> move_pages12.c:81: FAIL: madvise failed: SUCCESS > >> move_pages12.c:143: BROK: mmap((nil),4194304,3,262178,-1,0) failed: > ENOMEM > >> move_pages12.c:114: FAIL: move_pages failed: EINVAL > >> > >> Dmesg: > >> ------ > >> [165435.492170] soft offline: 0x61c00 hugepage failed to isolate > >> [165435.590252] soft offline: 0x61c00 hugepage failed to isolate > >> [165435.725493] soft offline: 0x61400 hugepage failed to isolate > >> > >> Other two fixes in this patch: > >> * use TERRNO(but not TTERRNO) to catch madvise(..., MADV_SOFT_OFFLINE) > errno > >> * go out test when hugepage allocating failed with ENOMEM > Hi Li > > Your patch can handle EBUSY errno correctly for soft offline. > But move page may be killed by SIGBUS because of MCE when we soft > offline concurrently. > That leads to move_page failed with ESRCH. Also, move page may fails > with ENOMEM . > Do you notice it ? > I didn't get this failure, it seems not related to this patch. Two questions: 1. which kernel version do you test? 2. can you reproduce this without my patch? > > I think ESRCH error can represent the soft offline bug not reproduce > because it don't trigger a crash. > What do you think about it? > Maybe, but it needs to check details on your kernel. > > err_log: > tst_test.c:1096: INFO: Timeout per run is 0h 05m 00s > move_pages12.c:236: INFO: Free RAM 119568 kB > move_pages12.c:254: INFO: Increasing 2048kB hugepages pool on node 0 to 83 > move_pages12.c:264: INFO: Increasing 2048kB hugepages pool on node 1 to 94 > move_pages12.c:180: INFO: Allocating and freeing 4 hugepages on node 0 > move_pages12.c:180: INFO: Allocating and freeing 4 hugepages on node 1 > move_pages12.c:170: PASS: Bug not reproduced > tst_test.c:1141: BROK: Test killed by SIGBUS! > > Summary: > passed 1 > failed 0 > skipped 0 > warnings 0 > > move_pages12.c:114: FAIL: move_pages failed: ESRCH > > dmesg > [ 9868.180669] MCE: Killing move_pages12:29616 due to hardware memory > corruption fault at 2aaaaac00018 > [ 9990.049875] Soft offlining page 50e00 at 2aaaaac00000 > [ 9990.052218] Soft offlining page 50c00 at 2aaaaae00000 > [ 9990.060395] Soft offlining page 51000 at 2aaaaac00000 > > Kind Regards, > Yang Xu > > >> Signed-off-by: Li Wang <liwang@redhat.com> > >> Cc: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com> > >> Cc: Xiao Yang <yangx.jy@cn.fujitsu.com> > >> Cc: Yang Xu <xuyang2018.jy@cn.fujitsu.com> > >> --- > >> .../kernel/syscalls/move_pages/move_pages12.c | 33 ++++++++++++++----- > >> 1 file changed, 24 insertions(+), 9 deletions(-) > >> > >> diff --git a/testcases/kernel/syscalls/move_pages/move_pages12.c > b/testcases/kernel/syscalls/move_pages/move_pages12.c > >> index 964b712fb..c446396dc 100644 > >> --- a/testcases/kernel/syscalls/move_pages/move_pages12.c > >> +++ b/testcases/kernel/syscalls/move_pages/move_pages12.c > >> @@ -77,8 +77,8 @@ static void *addr; > >> static int do_soft_offline(int tpgs) > >> { > >> if (madvise(addr, tpgs * hpsz, MADV_SOFT_OFFLINE) == -1) { > >> - if (errno != EINVAL) > >> - tst_res(TFAIL | TTERRNO, "madvise failed"); > >> + if (errno != EINVAL && errno != EBUSY) > >> + tst_res(TFAIL | TERRNO, "madvise failed"); > >> return errno; > >> } > >> return 0; > >> @@ -121,7 +121,8 @@ static void do_child(int tpgs) > >> > >> static void do_test(unsigned int n) > >> { > >> - int i; > >> + int i, ret; > >> + void *ptr; > >> pid_t cpid = -1; > >> int status; > >> unsigned int twenty_percent = (tst_timeout_remaining() / 5); > >> @@ -136,24 +137,37 @@ static void do_test(unsigned int n) > >> do_child(tcases[n].tpages); > >> > >> for (i = 0; i < LOOPS; i++) { > >> - void *ptr; > >> + ptr = mmap(NULL, tcases[n].tpages * hpsz, > >> + PROT_READ | PROT_WRITE, > >> + MAP_PRIVATE | MAP_ANONYMOUS | MAP_HUGETLB, > -1, 0); > >> + if (ptr == MAP_FAILED) { > >> + if (errno == ENOMEM) { > >> + tst_res(TCONF, > >> + "Cannot allocate hugepage, memory > too fragmented?"); > >> + goto out; > >> + } > >> + > >> + tst_brk(TBROK | TERRNO, "Cannot allocate > hugepage"); > >> + } > >> > >> - ptr = SAFE_MMAP(NULL, tcases[n].tpages * hpsz, > >> - PROT_READ | PROT_WRITE, > >> - MAP_PRIVATE | MAP_ANONYMOUS | MAP_HUGETLB, -1, 0); > >> if (ptr != addr) > >> tst_brk(TBROK, "Failed to mmap at desired addr"); > >> > >> memset(addr, 0, tcases[n].tpages * hpsz); > >> > >> if (tcases[n].offline) { > >> - if (do_soft_offline(tcases[n].tpages) == EINVAL) { > >> + ret = do_soft_offline(tcases[n].tpages); > >> + > >> + if (ret == EINVAL) { > >> SAFE_KILL(cpid, SIGKILL); > >> SAFE_WAITPID(cpid, &status, 0); > >> SAFE_MUNMAP(addr, tcases[n].tpages * hpsz); > >> tst_res(TCONF, > >> "madvise() didn't support > MADV_SOFT_OFFLINE"); > >> return; > >> + } else if (ret == EBUSY) { > >> + SAFE_MUNMAP(addr, tcases[n].tpages * hpsz); > >> + goto out; > >> } > >> } > >> > >> @@ -163,9 +177,10 @@ static void do_test(unsigned int n) > >> break; > >> } > >> > >> +out: > >> SAFE_KILL(cpid, SIGKILL); > >> SAFE_WAITPID(cpid, &status, 0); > >> - if (!WIFEXITED(status)) > >> + if (!WIFEXITED(status) && ptr != MAP_FAILED) > >> tst_res(TPASS, "Bug not reproduced"); > >> } > >> > >> -- > >> 2.20.1 > >> > >> > > > > . > > > > > >
on 2019/06/24 10:43, Li Wang wrote: > Hi Xu Yang, > > On Fri, Jun 21, 2019 at 1:58 PM Yang Xu <xuyang2018.jy@cn.fujitsu.com > <mailto:xuyang2018.jy@cn.fujitsu.com>> wrote: > > > > Hi Li Wang, > > > > Thank you for maintaining the testcase. > > > > Recently (since 4.19) we have a semantics change on the return > value of > > madvise(MADV_SOFT_OFFLINE), and we see -EBUSY when hugepage > migration > > succeeded and error containment failed: > > > > commit 6bc9b56433b76e40d11099338d27fbc5cd2935ca > > Author: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com > <mailto:n-horiguchi@ah.jp.nec.com>> > > Date: Thu Aug 23 17:00:38 2018 -0700 > > > > mm: fix race on soft-offlining free huge pages > > > > , so we don't have to consider this EBUSY as error, but a good > report > > for application. Your change meets the change. > > > > Feel free to add my ack: > > > > Acked-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com > <mailto:n-horiguchi@ah.jp.nec.com>> > > > > Thanks, > > - Naoya > > > > On Fri, Jun 07, 2019 at 05:52:13PM +0800, Li Wang wrote: > >> The test#2 is going to simulate the race condition, where > move_pages() > >> and soft offline are called on a single hugetlb page > concurrently. But, > >> it return EBUSY and report FAIL in soft-offline a moving > hugepage as a > >> result sometimes. > >> > >> The root cause seems a call to page_huge_active return false, > then the > >> soft offline action will failed to isolate hugepage with EBUSY > return as > >> below call trace: > >> > >> In Parent: > >> madvise(..., MADV_SOFT_OFFLINE) > >> ... > >> soft_offline_page > >> soft_offline_in_use_page > >> soft_offline_huge_page > >> isolate_huge_page > >> page_huge_active --> return false at here > >> > >> In Child: > >> move_pages() > >> ... > >> do_move_pages > >> do_move_pages_to_node > >> add_page_for_migration > >> isolate_huge_page --> it has already isolated the > hugepage > >> > >> In this patch, I simply regard the returned EBUSY as a normal > situation and > >> mask it in error handler. Because move_pages is calling > add_page_for_migration > >> to isolate hugepage before do migration, so that's very > possible to hit the > >> collision and return EBUSY on the same page. > >> > >> Error log: > >> ---------- > >> move_pages12.c:235: INFO: Free RAM 8386256 kB > >> move_pages12.c:253: INFO: Increasing 2048kB hugepages pool on > node 0 to 4 > >> move_pages12.c:263: INFO: Increasing 2048kB hugepages pool on > node 1 to 6 > >> move_pages12.c:179: INFO: Allocating and freeing 4 hugepages on > node 0 > >> move_pages12.c:179: INFO: Allocating and freeing 4 hugepages on > node 1 > >> move_pages12.c:169: PASS: Bug not reproduced > >> move_pages12.c:81: FAIL: madvise failed: SUCCESS > >> move_pages12.c:81: FAIL: madvise failed: SUCCESS > >> move_pages12.c:143: BROK: mmap((nil),4194304,3,262178,-1,0) > failed: ENOMEM > >> move_pages12.c:114: FAIL: move_pages failed: EINVAL > >> > >> Dmesg: > >> ------ > >> [165435.492170] soft offline: 0x61c00 hugepage failed to isolate > >> [165435.590252] soft offline: 0x61c00 hugepage failed to isolate > >> [165435.725493] soft offline: 0x61400 hugepage failed to isolate > >> > >> Other two fixes in this patch: > >> * use TERRNO(but not TTERRNO) to catch madvise(..., > MADV_SOFT_OFFLINE) errno > >> * go out test when hugepage allocating failed with ENOMEM > Hi Li > > Your patch can handle EBUSY errno correctly for soft offline. > But move page may be killed by SIGBUS because of MCE when we > soft offline concurrently. > That leads to move_page failed with ESRCH. Also, move page may > fails with ENOMEM . > Do you notice it ? > > > I didn't get this failure, it seems not related to this patch. Two > questions: > > 1. which kernel version do you test? > 2. can you reproduce this without my patch? Hi Li I test it on 3.10.0-957.el7.x86_64 kvm(my machine was not support numa and i enable it on kvm. as below: <cpu mode='custom' match='exact' check='full'> <model fallback='forbid'>Penryn</model> <feature policy='require' name='x2apic'/> <feature policy='require' name='hypervisor'/> <numa> <cell id='0' cpus='0' memory='1048576' unit='KiB'/> <cell id='1' cpus='1' memory='1048576' unit='KiB'/> </numa> </cpu> Does it only exist on kvm and doesn't exist on physical machine? I don't have physical machine that supports numa. And the fix patch has been merged since 3.10.0-957.el7.x86_64 . Yes, I can reproduce this without your patch because MCE kills child process and move_page gets ESRCH error. > > > I think ESRCH error can represent the soft offline bug not > reproduce because it don't trigger a crash. > What do you think about it? > > > Maybe, but it needs to check details on your kernel. > > > err_log: > tst_test.c:1096: INFO: Timeout per run is 0h 05m 00s > move_pages12.c:236: INFO: Free RAM 119568 kB > move_pages12.c:254: INFO: Increasing 2048kB hugepages pool on node > 0 to 83 > move_pages12.c:264: INFO: Increasing 2048kB hugepages pool on node > 1 to 94 > move_pages12.c:180: INFO: Allocating and freeing 4 hugepages on node 0 > move_pages12.c:180: INFO: Allocating and freeing 4 hugepages on node 1 > move_pages12.c:170: PASS: Bug not reproduced > tst_test.c:1141: BROK: Test killed by SIGBUS! > > Summary: > passed 1 > failed 0 > skipped 0 > warnings 0 > > move_pages12.c:114: FAIL: move_pages failed: ESRCH > > dmesg > [ 9868.180669] MCE: Killing move_pages12:29616 due to hardware > memory corruption fault at 2aaaaac00018 > [ 9990.049875] Soft offlining page 50e00 at 2aaaaac00000 > [ 9990.052218] Soft offlining page 50c00 at 2aaaaae00000 > [ 9990.060395] Soft offlining page 51000 at 2aaaaac00000 > > Kind Regards, > Yang Xu > > >> Signed-off-by: Li Wang <liwang@redhat.com > <mailto:liwang@redhat.com>> > >> Cc: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com > <mailto:n-horiguchi@ah.jp.nec.com>> > >> Cc: Xiao Yang <yangx.jy@cn.fujitsu.com > <mailto:yangx.jy@cn.fujitsu.com>> > >> Cc: Yang Xu <xuyang2018.jy@cn.fujitsu.com > <mailto:xuyang2018.jy@cn.fujitsu.com>> > >> --- > >> .../kernel/syscalls/move_pages/move_pages12.c | 33 > ++++++++++++++----- > >> 1 file changed, 24 insertions(+), 9 deletions(-) > >> > >> diff --git > a/testcases/kernel/syscalls/move_pages/move_pages12.c > b/testcases/kernel/syscalls/move_pages/move_pages12.c > >> index 964b712fb..c446396dc 100644 > >> --- a/testcases/kernel/syscalls/move_pages/move_pages12.c > >> +++ b/testcases/kernel/syscalls/move_pages/move_pages12.c > >> @@ -77,8 +77,8 @@ static void *addr; > >> static int do_soft_offline(int tpgs) > >> { > >> if (madvise(addr, tpgs * hpsz, MADV_SOFT_OFFLINE) == -1) { > >> - if (errno != EINVAL) > >> - tst_res(TFAIL | TTERRNO, "madvise failed"); > >> + if (errno != EINVAL && errno != EBUSY) > >> + tst_res(TFAIL | TERRNO, "madvise failed"); > >> return errno; > >> } > >> return 0; > >> @@ -121,7 +121,8 @@ static void do_child(int tpgs) > >> > >> static void do_test(unsigned int n) > >> { > >> - int i; > >> + int i, ret; > >> + void *ptr; > >> pid_t cpid = -1; > >> int status; > >> unsigned int twenty_percent = (tst_timeout_remaining() / 5); > >> @@ -136,24 +137,37 @@ static void do_test(unsigned int n) > >> do_child(tcases[n].tpages); > >> > >> for (i = 0; i < LOOPS; i++) { > >> - void *ptr; > >> + ptr = mmap(NULL, tcases[n].tpages * hpsz, > >> + PROT_READ | PROT_WRITE, > >> + MAP_PRIVATE | MAP_ANONYMOUS | > MAP_HUGETLB, -1, 0); > >> + if (ptr == MAP_FAILED) { > >> + if (errno == ENOMEM) { > >> + tst_res(TCONF, > >> + "Cannot allocate hugepage, > memory too fragmented?"); > >> + goto out; > >> + } > >> + > >> + tst_brk(TBROK | TERRNO, "Cannot allocate > hugepage"); > >> + } > >> > >> - ptr = SAFE_MMAP(NULL, tcases[n].tpages * hpsz, > >> - PROT_READ | PROT_WRITE, > >> - MAP_PRIVATE | MAP_ANONYMOUS | MAP_HUGETLB, > -1, 0); > >> if (ptr != addr) > >> tst_brk(TBROK, "Failed to mmap at desired > addr"); > >> > >> memset(addr, 0, tcases[n].tpages * hpsz); > >> > >> if (tcases[n].offline) { > >> - if (do_soft_offline(tcases[n].tpages) == > EINVAL) { > >> + ret = do_soft_offline(tcases[n].tpages); > >> + > >> + if (ret == EINVAL) { > >> SAFE_KILL(cpid, SIGKILL); > >> SAFE_WAITPID(cpid, &status, 0); > >> SAFE_MUNMAP(addr, tcases[n].tpages > * hpsz); > >> tst_res(TCONF, > >> "madvise() didn't support > MADV_SOFT_OFFLINE"); > >> return; > >> + } else if (ret == EBUSY) { > >> + SAFE_MUNMAP(addr, tcases[n].tpages > * hpsz); > >> + goto out; > >> } > >> } > >> > >> @@ -163,9 +177,10 @@ static void do_test(unsigned int n) > >> break; > >> } > >> > >> +out: > >> SAFE_KILL(cpid, SIGKILL); > >> SAFE_WAITPID(cpid, &status, 0); > >> - if (!WIFEXITED(status)) > >> + if (!WIFEXITED(status) && ptr != MAP_FAILED) > >> tst_res(TPASS, "Bug not reproduced"); > >> } > >> > >> -- > >> 2.20.1 > >> > >> > > > > . > > > > > > > > -- > Regards, > Li Wang
Hi! > + if (ret == EINVAL) { > SAFE_KILL(cpid, SIGKILL); > SAFE_WAITPID(cpid, &status, 0); > SAFE_MUNMAP(addr, tcases[n].tpages * hpsz); > tst_res(TCONF, > "madvise() didn't support MADV_SOFT_OFFLINE"); > return; > + } else if (ret == EBUSY) { > + SAFE_MUNMAP(addr, tcases[n].tpages * hpsz); > + goto out; Shouldn't we continue with the test here rather than exit? I guess that there is no harm in doing a few more iterations if we manage to hit EBUSY, or is there a good reason to exit the test here? Otherwise the patch looks good.
Hi Xu, On Thu, Jun 27, 2019 at 10:50 AM Yang Xu <xuyang2018.jy@cn.fujitsu.com> wrote: > ... >> Hi Li >> >> Your patch can handle EBUSY errno correctly for soft offline. >> But move page may be killed by SIGBUS because of MCE when we soft >> offline concurrently. >> That leads to move_page failed with ESRCH. Also, move page may fails >> with ENOMEM . >> Do you notice it ? >> > > I didn't get this failure, it seems not related to this patch. Two > questions: > > 1. which kernel version do you test? > 2. can you reproduce this without my patch? > > Hi Li > > I test it on 3.10.0-957.el7.x86_64 kvm(my machine was not support numa > and i enable it on kvm. as below: > <cpu mode='custom' match='exact' check='full'> > <model fallback='forbid'>Penryn</model> > <feature policy='require' name='x2apic'/> > <feature policy='require' name='hypervisor'/> > <numa> > <cell id='0' cpus='0' memory='1048576' unit='KiB'/> > <cell id='1' cpus='1' memory='1048576' unit='KiB'/> > </numa> > </cpu> > > Does it only exist on kvm and doesn't exist on physical machine? I don't > have physical machine that supports numa. > I can reproduce your problem on bare metal too, it seems like you hit the bug as the commit 6bc9b56433b (mm: fix race on soft-offlining free huge pages) described, which Naoya pointed out before: See: + /* + * We set PG_hwpoison only when the migration source hugepage + * was successfully dissolved, because otherwise hwpoisoned + * hugepage remains on free hugepage list, then userspace will + * find it as SIGBUS by allocation failure. That's not expected + * in soft-offlining. + */ + ret = dissolve_free_huge_page(page); + if (!ret) { + if (set_hwpoison_free_buddy_page(page)) + num_poisoned_pages_inc(); + } And, this bz still exists in the latest rhel7 kernel, I will open a bug to RHEL7 product.
On Wed, Jul 3, 2019 at 9:10 PM Cyril Hrubis <chrubis@suse.cz> wrote: > Hi! > > + if (ret == EINVAL) { > > SAFE_KILL(cpid, SIGKILL); > > SAFE_WAITPID(cpid, &status, 0); > > SAFE_MUNMAP(addr, tcases[n].tpages * hpsz); > > tst_res(TCONF, > > "madvise() didn't support > MADV_SOFT_OFFLINE"); > > return; > > + } else if (ret == EBUSY) { > > + SAFE_MUNMAP(addr, tcases[n].tpages * hpsz); > > + goto out; > > Shouldn't we continue with the test here rather than exit? > > I guess that there is no harm in doing a few more iterations if we > manage to hit EBUSY, or is there a good reason to exit the test here? > Yes, we can do more iterations then, but it probably makes no sense. The reason I guess is that, if we get an EBUSY on the hugepage offline, that means the page is already being isolated by move_pages() in the child at that moment and we can't really release it. So in the next iteration, the mmap() will be failed with ENOMEM(since we only have 1 huge page in /proc/.../nr_hugepages). To confirm that, I change the code to continue after get EBUSY, but it couldn't: # ./move_pages12 tst_test.c:1100: INFO: Timeout per run is 0h 05m 00s move_pages12.c:251: INFO: Free RAM 30860672 kB move_pages12.c:269: INFO: Increasing 2048kB hugepages pool on node 0 to 4 move_pages12.c:279: INFO: Increasing 2048kB hugepages pool on node 1 to 5 move_pages12.c:195: INFO: Allocating and freeing 4 hugepages on node 0 move_pages12.c:195: INFO: Allocating and freeing 4 hugepages on node 1 move_pages12.c:185: PASS: Bug not reproduced move_pages12.c:146: CONF: Cannot allocate hugepage, memory too fragmented? > > Otherwise the patch looks good. > Thanks for review.
> iteration, the mmap() will be failed with ENOMEM(since we only have 1 huge > page in /proc/.../nr_hugepages). > Sentence correction: It is not "only have 1 huge page in nr_hugepages", I mixed this test with another case, sorry about that. But the justification is the same, we don't have enough memory for the parent does mmap(..., MAP_HUGETLB) in a new loop.
Hi! > > iteration, the mmap() will be failed with ENOMEM(since we only have 1 huge > > page in /proc/.../nr_hugepages). > > > > Sentence correction: > It is not "only have 1 huge page in nr_hugepages", I mixed this test > with another case, sorry about that. > > But the justification is the same, we don't have enough memory for the > parent does mmap(..., MAP_HUGETLB) in a new loop. I guess I get it now, if we attempt to continue after EBUSY we unmap() the memory but that unmap() will happen asynchronously because the migration is in progress and we hit ENOMEM just in the next iteration of the loop. Should we then attempt to retry the mmap() on ENOMEM as well, ideally with exponential backoff? Unfortunately we cannot reuse the TST_RETRY_FUNC() as it is because it exits the test with TBROK on failure, we need a function that actually returns the last function return value on timeout.
On Wed, Jul 10, 2019 at 9:00 PM Cyril Hrubis <chrubis@suse.cz> wrote: > Hi! > > > iteration, the mmap() will be failed with ENOMEM(since we only have 1 > huge > > > page in /proc/.../nr_hugepages). > > > > > > > Sentence correction: > > It is not "only have 1 huge page in nr_hugepages", I mixed this test > > with another case, sorry about that. > > > > But the justification is the same, we don't have enough memory for the > > parent does mmap(..., MAP_HUGETLB) in a new loop. > > I guess I get it now, if we attempt to continue after EBUSY we unmap() > the memory but that unmap() will happen asynchronously because the > migration is in progress and we hit ENOMEM just in the next iteration of > the loop. > > Should we then attempt to retry the mmap() on ENOMEM as well, ideally > with exponential backoff? > Not very sure if that worth to do. > > Unfortunately we cannot reuse the TST_RETRY_FUNC() as it is because it > exits the test with TBROK on failure, we need a function that actually > returns the last function return value on timeout. > Yes, we could define a new TST_WAIT_FUNC() to return mmap() returned value on timeout, but it seems hard to give an expected return(ERET) value for that function, in this case, we could define the ERET as addr since we know it, but for most situations, we can't make sure what is the address being returned. Once the returned address is not equal ERET, then it will retry the mmap() and do not unmmap() the previous memory. That will be terrible.
Hi Cyril, Would you mind if I apply this patch? or do you have other thoughts besides retry mmap on ENOMEM?
Hi! > Would you mind if I apply this patch? or do you have other thoughts > besides retry mmap on ENOMEM? My only concern is that we may exit the test too soon if we do not attempt to retry.
diff --git a/testcases/kernel/syscalls/move_pages/move_pages12.c b/testcases/kernel/syscalls/move_pages/move_pages12.c index 964b712fb..c446396dc 100644 --- a/testcases/kernel/syscalls/move_pages/move_pages12.c +++ b/testcases/kernel/syscalls/move_pages/move_pages12.c @@ -77,8 +77,8 @@ static void *addr; static int do_soft_offline(int tpgs) { if (madvise(addr, tpgs * hpsz, MADV_SOFT_OFFLINE) == -1) { - if (errno != EINVAL) - tst_res(TFAIL | TTERRNO, "madvise failed"); + if (errno != EINVAL && errno != EBUSY) + tst_res(TFAIL | TERRNO, "madvise failed"); return errno; } return 0; @@ -121,7 +121,8 @@ static void do_child(int tpgs) static void do_test(unsigned int n) { - int i; + int i, ret; + void *ptr; pid_t cpid = -1; int status; unsigned int twenty_percent = (tst_timeout_remaining() / 5); @@ -136,24 +137,37 @@ static void do_test(unsigned int n) do_child(tcases[n].tpages); for (i = 0; i < LOOPS; i++) { - void *ptr; + ptr = mmap(NULL, tcases[n].tpages * hpsz, + PROT_READ | PROT_WRITE, + MAP_PRIVATE | MAP_ANONYMOUS | MAP_HUGETLB, -1, 0); + if (ptr == MAP_FAILED) { + if (errno == ENOMEM) { + tst_res(TCONF, + "Cannot allocate hugepage, memory too fragmented?"); + goto out; + } + + tst_brk(TBROK | TERRNO, "Cannot allocate hugepage"); + } - ptr = SAFE_MMAP(NULL, tcases[n].tpages * hpsz, - PROT_READ | PROT_WRITE, - MAP_PRIVATE | MAP_ANONYMOUS | MAP_HUGETLB, -1, 0); if (ptr != addr) tst_brk(TBROK, "Failed to mmap at desired addr"); memset(addr, 0, tcases[n].tpages * hpsz); if (tcases[n].offline) { - if (do_soft_offline(tcases[n].tpages) == EINVAL) { + ret = do_soft_offline(tcases[n].tpages); + + if (ret == EINVAL) { SAFE_KILL(cpid, SIGKILL); SAFE_WAITPID(cpid, &status, 0); SAFE_MUNMAP(addr, tcases[n].tpages * hpsz); tst_res(TCONF, "madvise() didn't support MADV_SOFT_OFFLINE"); return; + } else if (ret == EBUSY) { + SAFE_MUNMAP(addr, tcases[n].tpages * hpsz); + goto out; } } @@ -163,9 +177,10 @@ static void do_test(unsigned int n) break; } +out: SAFE_KILL(cpid, SIGKILL); SAFE_WAITPID(cpid, &status, 0); - if (!WIFEXITED(status)) + if (!WIFEXITED(status) && ptr != MAP_FAILED) tst_res(TPASS, "Bug not reproduced"); }
The test#2 is going to simulate the race condition, where move_pages() and soft offline are called on a single hugetlb page concurrently. But, it return EBUSY and report FAIL in soft-offline a moving hugepage as a result sometimes. The root cause seems a call to page_huge_active return false, then the soft offline action will failed to isolate hugepage with EBUSY return as below call trace: In Parent: madvise(..., MADV_SOFT_OFFLINE) ... soft_offline_page soft_offline_in_use_page soft_offline_huge_page isolate_huge_page page_huge_active --> return false at here In Child: move_pages() ... do_move_pages do_move_pages_to_node add_page_for_migration isolate_huge_page --> it has already isolated the hugepage In this patch, I simply regard the returned EBUSY as a normal situation and mask it in error handler. Because move_pages is calling add_page_for_migration to isolate hugepage before do migration, so that's very possible to hit the collision and return EBUSY on the same page. Error log: ---------- move_pages12.c:235: INFO: Free RAM 8386256 kB move_pages12.c:253: INFO: Increasing 2048kB hugepages pool on node 0 to 4 move_pages12.c:263: INFO: Increasing 2048kB hugepages pool on node 1 to 6 move_pages12.c:179: INFO: Allocating and freeing 4 hugepages on node 0 move_pages12.c:179: INFO: Allocating and freeing 4 hugepages on node 1 move_pages12.c:169: PASS: Bug not reproduced move_pages12.c:81: FAIL: madvise failed: SUCCESS move_pages12.c:81: FAIL: madvise failed: SUCCESS move_pages12.c:143: BROK: mmap((nil),4194304,3,262178,-1,0) failed: ENOMEM move_pages12.c:114: FAIL: move_pages failed: EINVAL Dmesg: ------ [165435.492170] soft offline: 0x61c00 hugepage failed to isolate [165435.590252] soft offline: 0x61c00 hugepage failed to isolate [165435.725493] soft offline: 0x61400 hugepage failed to isolate Other two fixes in this patch: * use TERRNO(but not TTERRNO) to catch madvise(..., MADV_SOFT_OFFLINE) errno * go out test when hugepage allocating failed with ENOMEM Signed-off-by: Li Wang <liwang@redhat.com> Cc: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com> Cc: Xiao Yang <yangx.jy@cn.fujitsu.com> Cc: Yang Xu <xuyang2018.jy@cn.fujitsu.com> --- .../kernel/syscalls/move_pages/move_pages12.c | 33 ++++++++++++++----- 1 file changed, 24 insertions(+), 9 deletions(-)