[Intel-gfx] [PATCH] drm/i915: Refine mmio debug flow to avoid bad unlock balance detected.

Andi Shyti andi.shyti at linux.intel.com
Tue Jul 11 10:31:16 UTC 2023


Hi Lee,

I'm really sorry for the late reply, I had some holidays in
between.

On Tue, Jul 04, 2023 at 04:07:27PM +0800, Lee Shawn C wrote:
> Perform reboot stresss test on a kernel build with lockdebug flag enabled.
> Bad unlock balanace detected would happened sometimes. Below is the
> problematic scenario. If params.mmio_debug value was changed at step #4.
> And it would trigger this issue. Modify code flow that decide to
> enable/disable mmio debug before unclaimed_reg_debug() can avoid
> this symptom.
> 
>     1. GEN6_READ_HEADER is called with params.mmio_debug = 0
>     2. unclaimed_reg_debug(before = true) is called
>     3. unclaimed_reg_debug return without taking a lock
>        because params.mmio_debug == 0
>     4. other thread modifies params.mmio_debug to 1
>     5. GEN6_READ_FOOTER is called with params.mmio_debug != 0
>     6. unclaimed_reg_debug tries to assert non-taken lock (first WARN)
>     7. unclaimed_reg_debug tries to release non-taken lock (second WARN)
> 
> Closes:https://gitlab.freedesktop.org/drm/intel/-/issues/8749
> Signed-off-by: Lee Shawn C <shawn.c.lee at intel.com>
> Cc: Uma Shankar <uma.shankar at intel.com>
> Cc: Matt Roper <matthew.d.roper at intel.com>
> Cc: Andi Shyti <andi.shyti at linux.intel.com>
> ---
>  drivers/gpu/drm/i915/intel_uncore.c | 10 +++++-----
>  1 file changed, 5 insertions(+), 5 deletions(-)
> 
> diff --git a/drivers/gpu/drm/i915/intel_uncore.c b/drivers/gpu/drm/i915/intel_uncore.c
> index 796ebfe6c550..9d665978cc43 100644
> --- a/drivers/gpu/drm/i915/intel_uncore.c
> +++ b/drivers/gpu/drm/i915/intel_uncore.c
> @@ -1931,9 +1931,6 @@ unclaimed_reg_debug(struct intel_uncore *uncore,
>  		    const bool read,
>  		    const bool before)
>  {
> -	if (likely(!uncore->i915->params.mmio_debug) || !uncore->debug)
> -		return;
> -

this is a very good catch! I'm fine with the change from my side:

Reviewed-by: Andi Shyti <andi.shyti at linux.intel.com> 

But I'd like to hear from someone else, as well, Matt, Daniele?

Thank you,
Andi


More information about the Intel-gfx mailing list