Re: [PATCH] acpi: osl: Replace GFP_ATOMIC with GFP_KERNEL in acpi_os_execute

[Rafael J. Wysocki]

On Thu, Feb 8, 2018 at 2:41 PM, Jia-Ju Bai <baijiaju1990@xxxxxxxxx> wrote:
>
>
> On 2018/2/8 18:20, Rafael J. Wysocki wrote:
>>
>> On Thursday, February 8, 2018 11:13:10 AM CET Chris Wilson wrote:
>>>
>>> Quoting Rafael J. Wysocki (2018-02-08 09:51:41)
>>>>
>>>> On Thursday, January 25, 2018 11:13:41 AM CET Jia-Ju Bai wrote:
>>>>>
>>>>> After checking all possible call chains to acpi_os_execute here,
>>>>> my tool finds that acpi_os_execute is never called in atomic context.
>>>>> And acpi_os_execute calls acpi_debugger_create_thread
>>>>> which calls mutex_lock,
>>>>> thus it proves again that acpi_os_execute can
>>>>> call functions which may sleep.
>>>>> Thus GFP_ATOMIC is not necessary, and it can be replaced with
>>>>> GFP_KERNEL.
>>>>>
>>>>> This is found by a static analysis tool named DCNS written by myself.
>>>>>
>>>>> Signed-off-by: Jia-Ju Bai <baijiaju1990@xxxxxxxxx>
>>>>> ---
>>>>>   drivers/acpi/osl.c |    2 +-
>>>>>   1 file changed, 1 insertion(+), 1 deletion(-)
>>>>>
>>>>> diff --git a/drivers/acpi/osl.c b/drivers/acpi/osl.c
>>>>> index 3bb46cb..8ee605e 100644
>>>>> --- a/drivers/acpi/osl.c
>>>>> +++ b/drivers/acpi/osl.c
>>>>> @@ -1066,7 +1066,7 @@ acpi_status acpi_os_execute(acpi_execute_type
>>>>> type,
>>>>>         * having a static work_struct.
>>>>>         */
>>>>>   -     dpc = kzalloc(sizeof(struct acpi_os_dpc), GFP_ATOMIC);
>>>>> +     dpc = kzalloc(sizeof(struct acpi_os_dpc), GFP_KERNEL);
>>>>>        if (!dpc)
>>>>>                return AE_NO_MEMORY;
>>>>>
>>>>
>>>> Applied, thanks!
>>>
>>> Hmm, not this patch per se, but
>>>
>>> https://intel-gfx-ci.01.org/tree/next/next-20180207/fi-bxt-dsi/dmesg0.log
>>> [  111.378236] BUG: sleeping function called from invalid context at
>>> mm/slab.h:420
>>> [  111.378259] in_atomic(): 1, irqs_disabled(): 1, pid: 1701, name:
>>> gem_exec_flush
>>> [  111.378275] INFO: lockdep is turned off.
>>> [  111.378277] irq event stamp: 0
>>> [  111.378280] hardirqs last  enabled at (0): [<          (null)>]
>>> (null)
>>> [  111.378286] hardirqs last disabled at (0): [<00000000a01fa473>]
>>> copy_process.part.7+0x2f1/0x1db0
>>> [  111.378290] softirqs last  enabled at (0): [<00000000a01fa473>]
>>> copy_process.part.7+0x2f1/0x1db0
>>> [  111.378292] softirqs last disabled at (0): [<          (null)>]
>>> (null)
>>> [  111.378293] Preemption disabled at:
>>> [  111.378298] [<ffffffffa18f14f6>] __mutex_lock+0x56/0x9b0
>>> [  111.378311] CPU: 1 PID: 1701 Comm: gem_exec_flush Tainted: G     U  W
>>> 4.15.0-next-20180207-g5d1c98967100-next-20180207 #1
>>> [  111.378313] Hardware name: Intel Corp. Broxton P/Apollolake RVP1A,
>>> BIOS APLKRVPA.X64.0150.B11.1608081044 08/08/2016
>>> [  111.378314] Call Trace:
>>> [  111.378318]  <IRQ>
>>> [  111.378323]  dump_stack+0x5f/0x86
>>> [  111.378328]  ___might_sleep+0x1d9/0x240
>>> [  111.378334]  ? acpi_os_execute+0x2d/0x130
>>> [  111.378338]  kmem_cache_alloc_trace+0x1ae/0x2b0
>>> [  111.378344]  ? acpi_ev_asynch_enable_gpe+0x2d/0x2d
>>> [  111.378347]  acpi_os_execute+0x2d/0x130
>>> [  111.378351]  acpi_ev_gpe_dispatch+0xd7/0x120
>>> [  111.378355]  acpi_ev_gpe_detect+0x156/0x195
>>> [  111.378362]  acpi_ev_sci_xrupt_handler+0x16/0x28
>>> [  111.378365]  acpi_irq+0xd/0x30
>>> [  111.378369]  __handle_irq_event_percpu+0x3c/0x340
>>> [  111.378374]  handle_irq_event_percpu+0x1b/0x50
>>> [  111.378378]  handle_irq_event+0x2f/0x50
>>> [  111.378381]  handle_fasteoi_irq+0x93/0x150
>>> [  111.378386]  handle_irq+0x11/0x20
>>> [  111.378390]  do_IRQ+0x5e/0x120
>>> [  111.378395]  common_interrupt+0xbb/0xbb
>>> [  111.378397]  </IRQ>
>>>
>>> does tell us that acpi_os_execute() is called in irq context.
>>
>> Well, right, thanks!
>>
>> I overlooked this instance, so dropping the patch.
>
>
> Sorry for my false positive.

No worries.

> My tool missed that acpi_ev_sci_xrupt_handler() is an interrupt handler, sorry.

That just means that the tool is not perfect, which is nothing unusual. :-)

I only would suggest double checking its findings before sending out
patches next time.