2021-11-22 10:17:48

by Tiezhu Yang

[permalink] [raw]
Subject: [PATCH] MIPS: Fix using smp_processor_id() in preemptible in show_cpuinfo()

There exists the following issue under DEBUG_PREEMPT:

BUG: using smp_processor_id() in preemptible [00000000] code: systemd/1
caller is show_cpuinfo+0x460/0xea0
...
Call Trace:
[<ffffffff8020f0dc>] show_stack+0x94/0x128
[<ffffffff80e6cab4>] dump_stack_lvl+0x94/0xd8
[<ffffffff80e74c5c>] check_preemption_disabled+0x104/0x110
[<ffffffff802209c8>] show_cpuinfo+0x460/0xea0
[<ffffffff80539d54>] seq_read_iter+0xfc/0x4f8
[<ffffffff804fcc10>] new_sync_read+0x110/0x1b8
[<ffffffff804ff57c>] vfs_read+0x1b4/0x1d0
[<ffffffff804ffb18>] ksys_read+0xd0/0x110
[<ffffffff8021c090>] syscall_common+0x34/0x58

We can see the following call trace:
show_cpuinfo()
cpu_has_fpu
current_cpu_data
smp_processor_id()

$ addr2line -f -e vmlinux 0xffffffff802209c8
show_cpuinfo
arch/mips/kernel/proc.c:188

$ head -188 arch/mips/kernel/proc.c | tail -1
if (cpu_has_fpu)

arch/mips/include/asm/cpu-features.h
# define cpu_has_fpu (current_cpu_data.options & MIPS_CPU_FPU)

arch/mips/include/asm/cpu-info.h
#define current_cpu_data cpu_data[smp_processor_id()]

Based on the above analysis, fix the issue by disabling preemption
around cpu_has_fpu in show_cpuinfo().

Signed-off-by: Tiezhu Yang <[email protected]>
---
arch/mips/kernel/proc.c | 2 ++
1 file changed, 2 insertions(+)

diff --git a/arch/mips/kernel/proc.c b/arch/mips/kernel/proc.c
index 376a6e2..c6c2661 100644
--- a/arch/mips/kernel/proc.c
+++ b/arch/mips/kernel/proc.c
@@ -185,8 +185,10 @@ static int show_cpuinfo(struct seq_file *m, void *v)
seq_puts(m, " tx39_cache");
if (cpu_has_octeon_cache)
seq_puts(m, " octeon_cache");
+ preempt_disable();
if (cpu_has_fpu)
seq_puts(m, " fpu");
+ preempt_enable();
if (cpu_has_32fpr)
seq_puts(m, " 32fpr");
if (cpu_has_cache_cdex_p)
--
2.1.0



2021-11-25 09:37:32

by Thomas Bogendoerfer

[permalink] [raw]
Subject: Re: [PATCH] MIPS: Fix using smp_processor_id() in preemptible in show_cpuinfo()

On Mon, Nov 22, 2021 at 06:17:37PM +0800, Tiezhu Yang wrote:
> There exists the following issue under DEBUG_PREEMPT:
>
> BUG: using smp_processor_id() in preemptible [00000000] code: systemd/1
> caller is show_cpuinfo+0x460/0xea0
> ...
> Call Trace:
> [<ffffffff8020f0dc>] show_stack+0x94/0x128
> [<ffffffff80e6cab4>] dump_stack_lvl+0x94/0xd8
> [<ffffffff80e74c5c>] check_preemption_disabled+0x104/0x110
> [<ffffffff802209c8>] show_cpuinfo+0x460/0xea0
> [<ffffffff80539d54>] seq_read_iter+0xfc/0x4f8
> [<ffffffff804fcc10>] new_sync_read+0x110/0x1b8
> [<ffffffff804ff57c>] vfs_read+0x1b4/0x1d0
> [<ffffffff804ffb18>] ksys_read+0xd0/0x110
> [<ffffffff8021c090>] syscall_common+0x34/0x58
>
> We can see the following call trace:
> show_cpuinfo()
> cpu_has_fpu
> current_cpu_data
> smp_processor_id()
>
> $ addr2line -f -e vmlinux 0xffffffff802209c8
> show_cpuinfo
> arch/mips/kernel/proc.c:188
>
> $ head -188 arch/mips/kernel/proc.c | tail -1
> if (cpu_has_fpu)
>
> arch/mips/include/asm/cpu-features.h
> # define cpu_has_fpu (current_cpu_data.options & MIPS_CPU_FPU)
>
> arch/mips/include/asm/cpu-info.h
> #define current_cpu_data cpu_data[smp_processor_id()]
>
> Based on the above analysis, fix the issue by disabling preemption
> around cpu_has_fpu in show_cpuinfo().
>
> Signed-off-by: Tiezhu Yang <[email protected]>

missing

Fixes: 626bfa037299 ("MIPS: kernel: proc: add CPU option reporting")

> ---
> arch/mips/kernel/proc.c | 2 ++
> 1 file changed, 2 insertions(+)
>
> diff --git a/arch/mips/kernel/proc.c b/arch/mips/kernel/proc.c
> index 376a6e2..c6c2661 100644
> --- a/arch/mips/kernel/proc.c
> +++ b/arch/mips/kernel/proc.c
> @@ -185,8 +185,10 @@ static int show_cpuinfo(struct seq_file *m, void *v)
> seq_puts(m, " tx39_cache");
> if (cpu_has_octeon_cache)
> seq_puts(m, " octeon_cache");
> + preempt_disable();
> if (cpu_has_fpu)
> seq_puts(m, " fpu");
> + preempt_enable();

what about using raw_cpu_has_fpu() instead ?

Thomas.

--
Crap can work. Given enough thrust pigs will fly, but it's not necessarily a
good idea. [ RFC1925, 2.3 ]

2021-11-25 11:34:55

by Tiezhu Yang

[permalink] [raw]
Subject: Re: [PATCH] MIPS: Fix using smp_processor_id() in preemptible in show_cpuinfo()

On 11/25/2021 05:34 PM, Thomas Bogendoerfer wrote:
> On Mon, Nov 22, 2021 at 06:17:37PM +0800, Tiezhu Yang wrote:
>> There exists the following issue under DEBUG_PREEMPT:
>>
>> BUG: using smp_processor_id() in preemptible [00000000] code: systemd/1
>> caller is show_cpuinfo+0x460/0xea0
>> ...
>> Call Trace:
>> [<ffffffff8020f0dc>] show_stack+0x94/0x128
>> [<ffffffff80e6cab4>] dump_stack_lvl+0x94/0xd8
>> [<ffffffff80e74c5c>] check_preemption_disabled+0x104/0x110
>> [<ffffffff802209c8>] show_cpuinfo+0x460/0xea0
>> [<ffffffff80539d54>] seq_read_iter+0xfc/0x4f8
>> [<ffffffff804fcc10>] new_sync_read+0x110/0x1b8
>> [<ffffffff804ff57c>] vfs_read+0x1b4/0x1d0
>> [<ffffffff804ffb18>] ksys_read+0xd0/0x110
>> [<ffffffff8021c090>] syscall_common+0x34/0x58
>>
>> We can see the following call trace:
>> show_cpuinfo()
>> cpu_has_fpu
>> current_cpu_data
>> smp_processor_id()
>>
>> $ addr2line -f -e vmlinux 0xffffffff802209c8
>> show_cpuinfo
>> arch/mips/kernel/proc.c:188
>>
>> $ head -188 arch/mips/kernel/proc.c | tail -1
>> if (cpu_has_fpu)
>>
>> arch/mips/include/asm/cpu-features.h
>> # define cpu_has_fpu (current_cpu_data.options & MIPS_CPU_FPU)
>>
>> arch/mips/include/asm/cpu-info.h
>> #define current_cpu_data cpu_data[smp_processor_id()]
>>
>> Based on the above analysis, fix the issue by disabling preemption
>> around cpu_has_fpu in show_cpuinfo().
>>
>> Signed-off-by: Tiezhu Yang <[email protected]>
>
> missing
>
> Fixes: 626bfa037299 ("MIPS: kernel: proc: add CPU option reporting")

Thank you, will add in v2.

>
>> ---
>> arch/mips/kernel/proc.c | 2 ++
>> 1 file changed, 2 insertions(+)
>>
>> diff --git a/arch/mips/kernel/proc.c b/arch/mips/kernel/proc.c
>> index 376a6e2..c6c2661 100644
>> --- a/arch/mips/kernel/proc.c
>> +++ b/arch/mips/kernel/proc.c
>> @@ -185,8 +185,10 @@ static int show_cpuinfo(struct seq_file *m, void *v)
>> seq_puts(m, " tx39_cache");
>> if (cpu_has_octeon_cache)
>> seq_puts(m, " octeon_cache");
>> + preempt_disable();
>> if (cpu_has_fpu)
>> seq_puts(m, " fpu");
>> + preempt_enable();
>
> what about using raw_cpu_has_fpu() instead ?

OK, no problem, using raw_cpu_has_fpu can also fix the issue and looks
better, I will send v2 soon.

Thanks,
Tiezhu

>
> Thomas.
>