diff mbox series

ACPI: acpi_pad: fix crash in exit_round_robin()

Message ID 20240825141352.25280-1-snishika@redhat.com
State New
Headers show
Series ACPI: acpi_pad: fix crash in exit_round_robin() | expand

Commit Message

Seiji Nishikawa Aug. 25, 2024, 2:13 p.m. UTC
The kernel occasionally crashes in cpumask_clear_cpu(), which is called
within exit_round_robin(), because when executing clear_bit(nr, addr) with
nr set to 0xffffffff, the address calculation may cause misalignment within
the memory, leading to access to an invalid memory address.

----------
BUG: unable to handle kernel paging request at ffffffffe0740618
        ...
CPU: 3 PID: 2919323 Comm: acpi_pad/14 Kdump: loaded Tainted: G           OE  X --------- -  - 4.18.0-425.19.2.el8_7.x86_64 #1
        ...
RIP: 0010:power_saving_thread+0x313/0x411 [acpi_pad]
Code: 89 cd 48 89 d3 eb d1 48 c7 c7 55 70 72 c0 e8 64 86 b0 e4 c6 05 0d a1 02 00 01 e9 bc fd ff ff 45 89 e4 42 8b 04 a5 20 82 72 c0 <f0> 48 0f b3 05 f4 9c 01 00 42 c7 04 a5 20 82 72 c0 ff ff ff ff 31
RSP: 0018:ff72a5d51fa77ec8 EFLAGS: 00010202
RAX: 00000000ffffffff RBX: ff462981e5d8cb80 RCX: 0000000000000000
RDX: 0000000000000000 RSI: 0000000000000246 RDI: 0000000000000246
RBP: ff46297556959d80 R08: 0000000000000382 R09: ff46297c8d0f38d8
R10: 0000000000000000 R11: 0000000000000001 R12: 000000000000000e
R13: 0000000000000000 R14: ffffffffffffffff R15: 000000000000000e
FS:  0000000000000000(0000) GS:ff46297a800c0000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: ffffffffe0740618 CR3: 0000007e20410004 CR4: 0000000000771ee0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
PKRU: 55555554
Call Trace:
 ? acpi_pad_add+0x120/0x120 [acpi_pad]
 kthread+0x10b/0x130
 ? set_kthread_struct+0x50/0x50
 ret_from_fork+0x1f/0x40
        ...
CR2: ffffffffe0740618

crash> dis -lr ffffffffc0726923
        ...
/usr/src/debug/kernel-4.18.0-425.19.2.el8_7/linux-4.18.0-425.19.2.el8_7.x86_64/./include/linux/cpumask.h: 114
0xffffffffc0726918 <power_saving_thread+776>:	mov    %r12d,%r12d
/usr/src/debug/kernel-4.18.0-425.19.2.el8_7/linux-4.18.0-425.19.2.el8_7.x86_64/./include/linux/cpumask.h: 325
0xffffffffc072691b <power_saving_thread+779>:	mov    -0x3f8d7de0(,%r12,4),%eax
/usr/src/debug/kernel-4.18.0-425.19.2.el8_7/linux-4.18.0-425.19.2.el8_7.x86_64/./arch/x86/include/asm/bitops.h: 80
0xffffffffc0726923 <power_saving_thread+787>:	lock btr %rax,0x19cf4(%rip)        # 0xffffffffc0740620 <pad_busy_cpus_bits>

crash> px tsk_in_cpu[14]
$66 = 0xffffffff

crash> px 0xffffffffc072692c+0x19cf4
$99 = 0xffffffffc0740620

crash> sym 0xffffffffc0740620
ffffffffc0740620 (b) pad_busy_cpus_bits [acpi_pad]

crash> px pad_busy_cpus_bits[0]
$42 = 0xfffc0
----------

To fix this, ensure that tsk_in_cpu[tsk_index] != -1 before calling
cpumask_clear_cpu() in exit_round_robin(), just as it is done in
round_robin_cpu().

Signed-off-by: Seiji Nishikawa <snishika@redhat.com>
---
 drivers/acpi/acpi_pad.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

Comments

Rafael J. Wysocki Aug. 29, 2024, 11:21 a.m. UTC | #1
On Sun, Aug 25, 2024 at 4:16 PM Seiji Nishikawa <snishika@redhat.com> wrote:
>
> The kernel occasionally crashes in cpumask_clear_cpu(), which is called
> within exit_round_robin(), because when executing clear_bit(nr, addr) with
> nr set to 0xffffffff, the address calculation may cause misalignment within
> the memory, leading to access to an invalid memory address.
>
> ----------
> BUG: unable to handle kernel paging request at ffffffffe0740618
>         ...
> CPU: 3 PID: 2919323 Comm: acpi_pad/14 Kdump: loaded Tainted: G           OE  X --------- -  - 4.18.0-425.19.2.el8_7.x86_64 #1
>         ...
> RIP: 0010:power_saving_thread+0x313/0x411 [acpi_pad]
> Code: 89 cd 48 89 d3 eb d1 48 c7 c7 55 70 72 c0 e8 64 86 b0 e4 c6 05 0d a1 02 00 01 e9 bc fd ff ff 45 89 e4 42 8b 04 a5 20 82 72 c0 <f0> 48 0f b3 05 f4 9c 01 00 42 c7 04 a5 20 82 72 c0 ff ff ff ff 31
> RSP: 0018:ff72a5d51fa77ec8 EFLAGS: 00010202
> RAX: 00000000ffffffff RBX: ff462981e5d8cb80 RCX: 0000000000000000
> RDX: 0000000000000000 RSI: 0000000000000246 RDI: 0000000000000246
> RBP: ff46297556959d80 R08: 0000000000000382 R09: ff46297c8d0f38d8
> R10: 0000000000000000 R11: 0000000000000001 R12: 000000000000000e
> R13: 0000000000000000 R14: ffffffffffffffff R15: 000000000000000e
> FS:  0000000000000000(0000) GS:ff46297a800c0000(0000) knlGS:0000000000000000
> CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> CR2: ffffffffe0740618 CR3: 0000007e20410004 CR4: 0000000000771ee0
> DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
> DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
> PKRU: 55555554
> Call Trace:
>  ? acpi_pad_add+0x120/0x120 [acpi_pad]
>  kthread+0x10b/0x130
>  ? set_kthread_struct+0x50/0x50
>  ret_from_fork+0x1f/0x40
>         ...
> CR2: ffffffffe0740618
>
> crash> dis -lr ffffffffc0726923
>         ...
> /usr/src/debug/kernel-4.18.0-425.19.2.el8_7/linux-4.18.0-425.19.2.el8_7.x86_64/./include/linux/cpumask.h: 114
> 0xffffffffc0726918 <power_saving_thread+776>:   mov    %r12d,%r12d
> /usr/src/debug/kernel-4.18.0-425.19.2.el8_7/linux-4.18.0-425.19.2.el8_7.x86_64/./include/linux/cpumask.h: 325
> 0xffffffffc072691b <power_saving_thread+779>:   mov    -0x3f8d7de0(,%r12,4),%eax
> /usr/src/debug/kernel-4.18.0-425.19.2.el8_7/linux-4.18.0-425.19.2.el8_7.x86_64/./arch/x86/include/asm/bitops.h: 80
> 0xffffffffc0726923 <power_saving_thread+787>:   lock btr %rax,0x19cf4(%rip)        # 0xffffffffc0740620 <pad_busy_cpus_bits>
>
> crash> px tsk_in_cpu[14]
> $66 = 0xffffffff
>
> crash> px 0xffffffffc072692c+0x19cf4
> $99 = 0xffffffffc0740620
>
> crash> sym 0xffffffffc0740620
> ffffffffc0740620 (b) pad_busy_cpus_bits [acpi_pad]
>
> crash> px pad_busy_cpus_bits[0]
> $42 = 0xfffc0
> ----------
>
> To fix this, ensure that tsk_in_cpu[tsk_index] != -1 before calling
> cpumask_clear_cpu() in exit_round_robin(), just as it is done in
> round_robin_cpu().
>
> Signed-off-by: Seiji Nishikawa <snishika@redhat.com>
> ---
>  drivers/acpi/acpi_pad.c | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/acpi/acpi_pad.c b/drivers/acpi/acpi_pad.c
> index 350d3a892889..699fbc795cfe 100644
> --- a/drivers/acpi/acpi_pad.c
> +++ b/drivers/acpi/acpi_pad.c
> @@ -136,7 +136,8 @@ static void exit_round_robin(unsigned int tsk_index)
>  {
>         struct cpumask *pad_busy_cpus = to_cpumask(pad_busy_cpus_bits);
>
> -       cpumask_clear_cpu(tsk_in_cpu[tsk_index], pad_busy_cpus);
> +       if (tsk_in_cpu[tsk_index] != -1)
> +               cpumask_clear_cpu(tsk_in_cpu[tsk_index], pad_busy_cpus);
>         tsk_in_cpu[tsk_index] = -1;
>  }
>
> --

If you already put the check in there, it is generally better to avoid
updates to the same value or you artificially make a cache line hot,
so I've applied this version (modulo GMail-induced white space
breakage):

--- a/drivers/acpi/acpi_pad.c
+++ b/drivers/acpi/acpi_pad.c
@@ -136,8 +136,10 @@ static void exit_round_robin(unsigned int tsk_index)
 {
     struct cpumask *pad_busy_cpus = to_cpumask(pad_busy_cpus_bits);

-    cpumask_clear_cpu(tsk_in_cpu[tsk_index], pad_busy_cpus);
-    tsk_in_cpu[tsk_index] = -1;
+    if (tsk_in_cpu[tsk_index] != -1) {
+        cpumask_clear_cpu(tsk_in_cpu[tsk_index], pad_busy_cpus);
+        tsk_in_cpu[tsk_index] = -1;
+    }
 }

 static unsigned int idle_pct = 5; /* percentage */
diff mbox series

Patch

diff --git a/drivers/acpi/acpi_pad.c b/drivers/acpi/acpi_pad.c
index 350d3a892889..699fbc795cfe 100644
--- a/drivers/acpi/acpi_pad.c
+++ b/drivers/acpi/acpi_pad.c
@@ -136,7 +136,8 @@  static void exit_round_robin(unsigned int tsk_index)
 {
 	struct cpumask *pad_busy_cpus = to_cpumask(pad_busy_cpus_bits);
 
-	cpumask_clear_cpu(tsk_in_cpu[tsk_index], pad_busy_cpus);
+	if (tsk_in_cpu[tsk_index] != -1)
+		cpumask_clear_cpu(tsk_in_cpu[tsk_index], pad_busy_cpus);
 	tsk_in_cpu[tsk_index] = -1;
 }