On 27 May 2016 at 12:41, Hoan Tran <[email protected]> wrote:
> Hi Ashwin,

Hi,

>
> Yes, I saw kernel crash.
> As cpc_desc_ptr is not NULL, cppc_cpufreq_init() still can pass then
> crash during cppc_get_perf_caps() access CPPC shared memory.
>
> It's not only "PCC channel request fail" can create this issue but
> "acpi_get_psd() fail" also creates it
>
> Thanks
> Hoan


Just for future reference. :)

https://web.archive.org/web/20080722025748/http://www.zip.com.au/~akpm/linux/patches/stuff/top-posting.txt


>
> On Fri, May 27, 2016 at 9:10 AM, Ashwin Chaugule
> <[email protected]> wrote:
>> On 25 May 2016 at 15:09, Hoan Tran <[email protected]> wrote:
>>> When CPPC fails to request PCC channel, the CPC data is freed
>>> and cpc_desc_ptr points to the invalid data. This change prevents
>>> this issue by moving cpc_desc_ptr assignment after PCC channel
>>> request.
>>>
>>> Signed-off-by: Hoan Tran <[email protected]>
>>> ---
>>>  drivers/acpi/cppc_acpi.c | 6 +++---
>>>  1 file changed, 3 insertions(+), 3 deletions(-)
>>>
>>> diff --git a/drivers/acpi/cppc_acpi.c b/drivers/acpi/cppc_acpi.c
>>> index 8adac69..85fd8f7 100644
>>> --- a/drivers/acpi/cppc_acpi.c
>>> +++ b/drivers/acpi/cppc_acpi.c
>>> @@ -595,9 +595,6 @@ int acpi_cppc_processor_probe(struct acpi_processor *pr)
>>>         /* Store CPU Logical ID */
>>>         cpc_ptr->cpu_id = pr->id;
>>>
>>> -       /* Plug it into this CPUs CPC descriptor. */
>>> -       per_cpu(cpc_desc_ptr, pr->id) = cpc_ptr;
>>> -
>>>         /* Parse PSD data for this CPU */
>>>         ret = acpi_get_psd(cpc_ptr, handle);
>>>         if (ret)
>>> @@ -610,6 +607,9 @@ int acpi_cppc_processor_probe(struct acpi_processor *pr)
>>>                         goto out_free;
>>>         }
>>>
>>> +       /* Plug PSD data into this CPUs CPC descriptor. */
>>> +       per_cpu(cpc_desc_ptr, pr->id) = cpc_ptr;
>>> +
>>
>> Are you seeing a real problem without this change? I'm missing where
>> this pointer is dereferenced if the PCC channel request fails.
>>

So, after freeing the cpc_ptr, we need to NULL the per-cpu pointer as
well. Alternately, not assign it until everything passes and rely on
the static declaration, which is the path you've taken here.

Acked-by: Ashwin Chaugule <[email protected]>

Reply via email to