drm/i915: Silence build error with UBSAN

Submitted by Jani Nikula on Oct. 16, 2018, 9:59 a.m.

Details

Message ID 87sh16ky9d.fsf@intel.com
State New
Headers show
Series "drm/i915: Silence build error with UBSAN" ( rev: 2 ) in Intel GFX

Not browsing as part of any series.

Commit Message

Jani Nikula Oct. 16, 2018, 9:59 a.m.
On Mon, 15 Oct 2018, Stephen Boyd <swboyd@chromium.org> wrote:
> When I enable UBSAN and compile this driver with clang I get the
> following build error:
>
> drivers/gpu/drm/i915/intel_engine_cs.o: In function `intel_engine_init_execlist':
> drivers/gpu/drm/i915/intel_engine_cs.c:411: undefined reference to `__compiletime_assert_411'
>
> from what I can figure out, the compiler can't optimize
> execlists_num_ports() sufficiently enough at compile time to figure out
> that the 'execlists->port_mask = 1' assignment one line above the
> BUILD_BUG_ON_NOT_POWER_OF_2 check will make execlists_num_ports() return
> 2. Most likely that's because UBSAN is going to check the load inside
> execlists_num_ports() and that check isn't omitted so the optimizer
> can't optimize away the whole function.

See [1] for a better explanation.

[1] http://mid.mail-archive.com/20181009171401.14980-1-natechancellor@gmail.com

> So let's just change this check to cause a build error when the maximum
> number of ports isn't a power of two. It looks like this is similar to
> what's being checked here so this might work well enough.

That's not the same thing. I guess I'd go with the below instead,
similar to the check on the next line. I guess both of the checks could
be static on gcc.

BR,
Jani.



---


>
> Cc: Masahiro Yamada <yamada.masahiro@socionext.com>
> Cc: Michal Marek <michal.lkml@markovi.net>
> Cc: Andrey Ryabinin <aryabinin@virtuozzo.com>
> Signed-off-by: Stephen Boyd <swboyd@chromium.org>
> ---
>  drivers/gpu/drm/i915/intel_engine_cs.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/drivers/gpu/drm/i915/intel_engine_cs.c b/drivers/gpu/drm/i915/intel_engine_cs.c
> index 217ed3ee1cab..bdf75628ed83 100644
> --- a/drivers/gpu/drm/i915/intel_engine_cs.c
> +++ b/drivers/gpu/drm/i915/intel_engine_cs.c
> @@ -463,7 +463,7 @@ static void intel_engine_init_execlist(struct intel_engine_cs *engine)
>  	struct intel_engine_execlists * const execlists = &engine->execlists;
>  
>  	execlists->port_mask = 1;
> -	BUILD_BUG_ON_NOT_POWER_OF_2(execlists_num_ports(execlists));
> +	BUILD_BUG_ON_NOT_POWER_OF_2(EXECLIST_MAX_PORTS);
>  	GEM_BUG_ON(execlists_num_ports(execlists) > EXECLIST_MAX_PORTS);
>  
>  	execlists->queue_priority = INT_MIN;

Patch hide | download patch | download mbox

diff --git a/drivers/gpu/drm/i915/intel_engine_cs.c b/drivers/gpu/drm/i915/intel_engine_cs.c
index f27dbe26bcc1..897d5a557d88 100644
--- a/drivers/gpu/drm/i915/intel_engine_cs.c
+++ b/drivers/gpu/drm/i915/intel_engine_cs.c
@@ -461,12 +461,14 @@  static void intel_engine_init_batch_pool(struct intel_engine_cs *engine)
 	i915_gem_batch_pool_init(&engine->batch_pool, engine);
 }
 
+#define IS_POWER_OF_2(n) ((n) != 0 && ((n) & ((n) - 1)) == 0)
+
 static void intel_engine_init_execlist(struct intel_engine_cs *engine)
 {
 	struct intel_engine_execlists * const execlists = &engine->execlists;
 
 	execlists->port_mask = 1;
-	BUILD_BUG_ON_NOT_POWER_OF_2(execlists_num_ports(execlists));
+	GEM_BUG_ON(!IS_POWER_OF_2(execlists_num_ports(execlists)));
 	GEM_BUG_ON(execlists_num_ports(execlists) > EXECLIST_MAX_PORTS);
 
 	execlists->queue_priority = INT_MIN;

Comments

Quoting Jani Nikula (2018-10-16 10:59:42)
> On Mon, 15 Oct 2018, Stephen Boyd <swboyd@chromium.org> wrote:
> > When I enable UBSAN and compile this driver with clang I get the
> > following build error:
> >
> > drivers/gpu/drm/i915/intel_engine_cs.o: In function `intel_engine_init_execlist':
> > drivers/gpu/drm/i915/intel_engine_cs.c:411: undefined reference to `__compiletime_assert_411'
> >
> > from what I can figure out, the compiler can't optimize
> > execlists_num_ports() sufficiently enough at compile time to figure out
> > that the 'execlists->port_mask = 1' assignment one line above the
> > BUILD_BUG_ON_NOT_POWER_OF_2 check will make execlists_num_ports() return
> > 2. Most likely that's because UBSAN is going to check the load inside
> > execlists_num_ports() and that check isn't omitted so the optimizer
> > can't optimize away the whole function.
> 
> See [1] for a better explanation.
> 
> [1] http://mid.mail-archive.com/20181009171401.14980-1-natechancellor@gmail.com
> 
> > So let's just change this check to cause a build error when the maximum
> > number of ports isn't a power of two. It looks like this is similar to
> > what's being checked here so this might work well enough.
> 
> That's not the same thing. I guess I'd go with the below instead,
> similar to the check on the next line. I guess both of the checks could
> be static on gcc.
> 
> BR,
> Jani.
> 
> 
> diff --git a/drivers/gpu/drm/i915/intel_engine_cs.c b/drivers/gpu/drm/i915/intel_engine_cs.c
> index f27dbe26bcc1..897d5a557d88 100644
> --- a/drivers/gpu/drm/i915/intel_engine_cs.c
> +++ b/drivers/gpu/drm/i915/intel_engine_cs.c
> @@ -461,12 +461,14 @@ static void intel_engine_init_batch_pool(struct intel_engine_cs *engine)
>         i915_gem_batch_pool_init(&engine->batch_pool, engine);
>  }
>  
> +#define IS_POWER_OF_2(n) ((n) != 0 && ((n) & ((n) - 1)) == 0)
> +
>  static void intel_engine_init_execlist(struct intel_engine_cs *engine)
>  {
>         struct intel_engine_execlists * const execlists = &engine->execlists;
>  
>         execlists->port_mask = 1;
> -       BUILD_BUG_ON_NOT_POWER_OF_2(execlists_num_ports(execlists));
> +       GEM_BUG_ON(!IS_POWER_OF_2(execlists_num_ports(execlists)));

That should be happy with is_power_of_2() from log2.h
-Chris
On Tue, 16 Oct 2018, Chris Wilson <chris@chris-wilson.co.uk> wrote:
> Quoting Jani Nikula (2018-10-16 10:59:42)
>> diff --git a/drivers/gpu/drm/i915/intel_engine_cs.c b/drivers/gpu/drm/i915/intel_engine_cs.c
>> index f27dbe26bcc1..897d5a557d88 100644
>> --- a/drivers/gpu/drm/i915/intel_engine_cs.c
>> +++ b/drivers/gpu/drm/i915/intel_engine_cs.c
>> @@ -461,12 +461,14 @@ static void intel_engine_init_batch_pool(struct intel_engine_cs *engine)
>>         i915_gem_batch_pool_init(&engine->batch_pool, engine);
>>  }
>>  
>> +#define IS_POWER_OF_2(n) ((n) != 0 && ((n) & ((n) - 1)) == 0)
>> +
>>  static void intel_engine_init_execlist(struct intel_engine_cs *engine)
>>  {
>>         struct intel_engine_execlists * const execlists = &engine->execlists;
>>  
>>         execlists->port_mask = 1;
>> -       BUILD_BUG_ON_NOT_POWER_OF_2(execlists_num_ports(execlists));
>> +       GEM_BUG_ON(!IS_POWER_OF_2(execlists_num_ports(execlists)));
>
> That should be happy with is_power_of_2() from log2.h

D'oh. Thanks. I was sure there was one, I looked for it, but my git
greps were all upper case. :/

I'll spin a new version, along with the other clang build fix.

BR,
Jani.