Thread (20 messages) 20 messages, 9 authors, 2015-01-26

[next-20150119]regression (mm)?

From: Tyler Baker <hidden>
Date: 2015-01-26 22:38:28
Also in: linux-mm, linux-next, linux-omap

On 26 January 2015 at 04:00, Kirill A. Shutemov [off-list ref] wrote:
On Fri, Jan 23, 2015 at 10:37:46PM -0600, Nishanth Menon wrote:
quoted
On 03:13-20150124, Kirill A. Shutemov wrote:
quoted
quoted
quoted
quoted
On 09:39-20150123, Tyler Baker wrote:
[...]
quoted
quoted
quoted
quoted
quoted
I just reviewed the boot logs for next-20150123 and there still seems
to be a related issue. I've been boot testing
multi_v7_defconfig+CONFIG_ARM_LPAE=y kernel configurations which still
seem broken.
[...]
quoted
Okay, proof of concept patch is below. It's going to break every other
architecture with FIRST_USER_ADDRESS != 0, but I think it's cleaner way to
go.
Testing on my end:

just ran through this set (+ logs similar to Tyler's from my side):

next-20150123 (multi_v7_defconfig == !LPAE)
 1:    BeagleBoard-X15(am57xx-evm): BOOT: PASS: http://paste.ubuntu.org.cn/2219449
 2:                     dra72x-evm: BOOT: PASS: http://paste.ubuntu.org.cn/2219450
 3:                     dra7xx-evm: BOOT: PASS: http://paste.ubuntu.org.cn/2219451
 4:                      omap5-evm: BOOT: PASS: http://paste.ubuntu.org.cn/2219452
TOTAL = 4 boards, Booted Boards = 4, No Boot boards = 0

next-20150123-LPAE-Logging enabled[1] (multi_v7_defconfig +LPAE)
 1:    BeagleBoard-X15(am57xx-evm): BOOT: FAIL: http://paste.ubuntu.org.cn/2220938
 2:                     dra72x-evm: BOOT: FAIL: http://paste.ubuntu.org.cn/2220943
 3:                     dra7xx-evm: BOOT: FAIL: http://paste.ubuntu.org.cn/2220947
 4:                      omap5-evm: BOOT: FAIL: http://paste.ubuntu.org.cn/2220955
TOTAL = 4 boards, Booted Boards = 0, No Boot boards = 4

next-20150123-LPAE-new-patch [2] (multi_v7_defconfig + LPAE)
 1:    BeagleBoard-X15(am57xx-evm): BOOT: PASS: http://paste.ubuntu.org.cn/2221047
 2:                     dra72x-evm: BOOT: PASS: http://paste.ubuntu.org.cn/2221065
 3:                     dra7xx-evm: BOOT: PASS: http://paste.ubuntu.org.cn/2221069
 4:                      omap5-evm: BOOT: PASS: http://paste.ubuntu.org.cn/2221070
TOTAL = 4 boards, Booted Boards = 4, No Boot boards = 0

next-20150123-new-patch[2] (multi_v7_defconfig == !LPAE)
 1:                     am335x-evm: BOOT: PASS: http://paste.ubuntu.org.cn/2221277
 2:                      am335x-sk: BOOT: PASS: http://paste.ubuntu.org.cn/2221278
 3:                      am437x-sk: BOOT: FAIL: http://paste.ubuntu.org.cn/2221279 (unrelated)
 4:                    am43xx-epos: BOOT: PASS: http://paste.ubuntu.org.cn/2221280
 5:                   am43xx-gpevm: BOOT: PASS: http://paste.ubuntu.org.cn/2221281
 6:    BeagleBoard-X15(am57xx-evm): BOOT: PASS: http://paste.ubuntu.org.cn/2221282
 7:                 BeagleBoard-XM: BOOT: FAIL: http://paste.ubuntu.org.cn/2221283 (unrelated)
 8:            beagleboard-vanilla: BOOT: PASS: http://paste.ubuntu.org.cn/2221284
 9:               beaglebone-black: BOOT: PASS: http://paste.ubuntu.org.cn/2221285
10:                     beaglebone: BOOT: FAIL: http://paste.ubuntu.org.cn/2221286 (unrelated)
11:                     dra72x-evm: BOOT: PASS: http://paste.ubuntu.org.cn/2221287
12:                     dra7xx-evm: BOOT: PASS: http://paste.ubuntu.org.cn/2221288
13:                      omap5-evm: BOOT: PASS: http://paste.ubuntu.org.cn/2221289
14:                  pandaboard-es: BOOT: PASS: http://paste.ubuntu.org.cn/2221290
15:             pandaboard-vanilla: BOOT: PASS: http://paste.ubuntu.org.cn/2221291
16:                        sdp4430: BOOT: PASS: http://paste.ubuntu.org.cn/2221292
TOTAL = 16 boards, Booted Boards = 13, No Boot boards = 3

next-20150123-new-patch[2] (omap2plus_defconfig)
 1:                     am335x-evm: BOOT: PASS: http://paste.ubuntu.org.cn/2221653
 2:                      am335x-sk: BOOT: PASS: http://paste.ubuntu.org.cn/2221654
 3:                      am437x-sk: BOOT: PASS: http://paste.ubuntu.org.cn/2221656
 4:                    am43xx-epos: BOOT: PASS: http://paste.ubuntu.org.cn/2221659
 5:                   am43xx-gpevm: BOOT: PASS: http://paste.ubuntu.org.cn/2221660
 6:    BeagleBoard-X15(am57xx-evm): BOOT: PASS: http://paste.ubuntu.org.cn/2221661
 7:                 BeagleBoard-XM: BOOT: PASS: http://paste.ubuntu.org.cn/2221670
 8:            beagleboard-vanilla: BOOT: PASS: http://paste.ubuntu.org.cn/2221676
 9:               beaglebone-black: BOOT: PASS: http://paste.ubuntu.org.cn/2221683
10:                     beaglebone: BOOT: PASS: http://paste.ubuntu.org.cn/2221690
11:                     dra72x-evm: BOOT: PASS: http://paste.ubuntu.org.cn/2221692
12:                     dra7xx-evm: BOOT: PASS: http://paste.ubuntu.org.cn/2221695
13:                      omap5-evm: BOOT: PASS: http://paste.ubuntu.org.cn/2221700
14:                  pandaboard-es: BOOT: PASS: http://paste.ubuntu.org.cn/2221704
15:             pandaboard-vanilla: BOOT: PASS: http://paste.ubuntu.org.cn/2221707
16:                        sdp4430: BOOT: PASS: http://paste.ubuntu.org.cn/2221713
TOTAL = 16 boards, Booted Boards = 16, No Boot boards = 0
Okay thanks. Here's proper patch.

From 8f9845ab8d972164b700ff3e3ce53484cceb942b Mon Sep 17 00:00:00 2001
From: "Kirill A. Shutemov" <redacted>
Date: Mon, 26 Jan 2015 12:07:54 +0200
Subject: [PATCH 1/2] mm: fix false-positive warning on exit due mm_nr_pmds(mm)

The problem is that we check nr_ptes/nr_pmds in exit_mmap() which happens
*before* pgd_free(). And if an arch does pte/pmd allocation in pgd_alloc()
and frees them in pgd_free() we see offset in counters by the time of the
checks.

We tried to workaround this by offsetting expected counter value
according to FIRST_USER_ADDRESS for both nr_pte and nr_pmd in
exit_mmap(). But it doesn't work in some cases:

1. ARM with LPAE enabled also has non-zero USER_PGTABLES_CEILING, but
   upper addresses occupied with huge pmd entries, so the trick with
   offsetting expected counter value will get really ugly: we will have
   to apply it nr_pmds, but not nr_ptes.

2. Metag has non-zero FIRST_USER_ADDRESS, but doesn't do allocation
   pte/pmd page tables allocation in pgd_alloc(), just setup a pgd entry
   which is allocated at boot and shared accross all processes.

The proposal is to move the check to check_mm() which happens *after*
pgd_free() and do proper accounting during pgd_alloc() and pgd_free()
which would bring counters to zero if nothing leaked.

Signed-off-by: Kirill A. Shutemov <redacted>
I've tested this patch on top of linux-next [1] on a various array of
arm, arm64 and x86 hardware. I can confirm the issue with
CONFIG_ARM_LPAE=y has been resolved with no additional regressions
detected. The results can be found here [2].

Feel free to add:

Tested-by: Tyler Baker <redacted>
quoted hunk ↗ jump to hunk
Reported-by: Tyler Baker <redacted>
Tested-by: Nishanth Menon <nm@ti.com>
Cc: Russell King <redacted>
Cc: James Hogan <redacted>
Cc: Guan Xuetao <redacted>
---
 arch/arm/mm/pgd.c       | 4 ++++
 arch/unicore32/mm/pgd.c | 3 +++
 kernel/fork.c           | 8 ++++++++
 mm/mmap.c               | 5 -----
 4 files changed, 15 insertions(+), 5 deletions(-)
diff --git a/arch/arm/mm/pgd.c b/arch/arm/mm/pgd.c
index 249379535be2..a3681f11dd9f 100644
--- a/arch/arm/mm/pgd.c
+++ b/arch/arm/mm/pgd.c
@@ -97,6 +97,7 @@ pgd_t *pgd_alloc(struct mm_struct *mm)

 no_pte:
        pmd_free(mm, new_pmd);
+       mm_dec_nr_pmds(mm);
 no_pmd:
        pud_free(mm, new_pud);
 no_pud:
@@ -130,9 +131,11 @@ void pgd_free(struct mm_struct *mm, pgd_t *pgd_base)
        pte = pmd_pgtable(*pmd);
        pmd_clear(pmd);
        pte_free(mm, pte);
+       atomic_long_dec(&mm->nr_ptes);
 no_pmd:
        pud_clear(pud);
        pmd_free(mm, pmd);
+       mm_dec_nr_pmds(mm);
 no_pud:
        pgd_clear(pgd);
        pud_free(mm, pud);
@@ -152,6 +155,7 @@ no_pgd:
                pmd = pmd_offset(pud, 0);
                pud_clear(pud);
                pmd_free(mm, pmd);
+               mm_dec_nr_pmds(mm);
                pgd_clear(pgd);
                pud_free(mm, pud);
        }
diff --git a/arch/unicore32/mm/pgd.c b/arch/unicore32/mm/pgd.c
index 08b8d4295e70..1bc00d0305d4 100644
--- a/arch/unicore32/mm/pgd.c
+++ b/arch/unicore32/mm/pgd.c
@@ -69,6 +69,7 @@ pgd_t *get_pgd_slow(struct mm_struct *mm)

 no_pte:
        pmd_free(mm, new_pmd);
+       mm_dec_nr_pmds(mm);
 no_pmd:
        free_pages((unsigned long)new_pgd, 0);
 no_pgd:
@@ -96,7 +97,9 @@ void free_pgd_slow(struct mm_struct *mm, pgd_t *pgd)
        pte = pmd_pgtable(*pmd);
        pmd_clear(pmd);
        pte_free(mm, pte);
+       atomic_long_dec(&mm->nr_ptes);
        pmd_free(mm, pmd);
+       mm_dec_nr_pmds(mm)
 free:
        free_pages((unsigned long) pgd, 0);
 }
diff --git a/kernel/fork.c b/kernel/fork.c
index c99098c52641..76d6f292274c 100644
--- a/kernel/fork.c
+++ b/kernel/fork.c
@@ -606,6 +606,14 @@ static void check_mm(struct mm_struct *mm)
                        printk(KERN_ALERT "BUG: Bad rss-counter state "
                                          "mm:%p idx:%d val:%ld\n", mm, i, x);
        }
+
+       if (atomic_long_read(&mm->nr_ptes))
+               pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld",
+                               atomic_long_read(&mm->nr_ptes));
+       if (mm_nr_pmds(mm))
+               pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld",
+                               mm_nr_pmds(mm));
+
 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
        VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
 #endif
diff --git a/mm/mmap.c b/mm/mmap.c
index 6a7d36d133fb..c5f44682c0d1 100644
--- a/mm/mmap.c
+++ b/mm/mmap.c
@@ -2851,11 +2851,6 @@ void exit_mmap(struct mm_struct *mm)
                vma = remove_vma(vma);
        }
        vm_unacct_memory(nr_accounted);
-
-       WARN_ON(atomic_long_read(&mm->nr_ptes) >
-                       round_up(FIRST_USER_ADDRESS, PMD_SIZE) >> PMD_SHIFT);
-       WARN_ON(mm_nr_pmds(mm) >
-                       round_up(FIRST_USER_ADDRESS, PUD_SIZE) >> PUD_SHIFT);
 }

 /* Insert vm structure into process list sorted by address
--
 Kirill A. Shutemov
[1] https://git.linaro.org/people/tyler.baker/linux-next.git/shortlog/refs/heads/next-testing
[2] http://kernelci.org/boot/all/job/tbaker/kernel/v3.19-rc5-5174-g384ba8a33c70/

Thanks,

Tyler
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help