Re: [PATCH][DEBUG] x86/refcount: split up refcount saturation handling

[Kees Cook]

On Sat, Sep 2, 2017 at 3:35 AM, Ingo Molnar <mingo@xxxxxxxxxx> wrote:
>
> * Kees Cook <keescook@xxxxxxxxxxxx> wrote:
>
>> In support of debugging the problems Mike Galbraith has seen with
>> x86-refcount vs gcc vs network refcounts...
>>
>> This minimizes the differences between unchecked-refcount and x86-refcount
>> by changing the refcount_dec() failure case to not saturate. The reporting
>> of negative values is reduced to pr_warn from WARN to avoid spamming dmesg
>> (which may impact race conditions). Ratelimiting is disabled just to be
>> sure no reports are being dropped.
>>
>> Signed-off-by: Kees Cook <keescook@xxxxxxxxxxxx>
>> ---
>>  arch/x86/mm/extable.c | 51 ++++++++++++++++++++++++++++++++++-----------------
>>  kernel/panic.c        |  2 +-
>>  2 files changed, 35 insertions(+), 18 deletions(-)
>
> If this patch is still useful then please include it in your next refcount series.
> Better debuggability is always welcome.

It's not: this is while it seemed like something else was happening.
The most important reporting improvement is included in the final fix
patch.

-Kees

-- 
Kees Cook
Pixel Security