[PATCH] drm/nouveau/fb: mark ramfuc_reg() noinline

[PATCH] drm/nouveau/fb: mark ramfuc_reg() noinline

[Paul Bolle]

Building ramnve0.o triggers a GCC warning on 32 bits x86:
    drivers/gpu/drm/nouveau/core/subdev/fb/ramnve0.c: In function 'nve0_ram_ctor':
    drivers/gpu/drm/nouveau/core/subdev/fb/ramnve0.c:1253:1: warning: the frame size of 1496 bytes is larger than 1024 bytes [-Wframe-larger-than=]

This warning is caused by ramfuc_reg(), which is inlined 74 times in
nve0_ram_ctor(). Mark it noinline to silence this warning.

Signed-off-by: Paul Bolle <pebolle@tiscali.nl>
---
Compile tested (on 32 bits x86) only. I've no Nvidia cards at hand, so I
can't really test it.

This assumes this function - a constructor, apparently - isn't called
often, so the overhead calling of 74 functions is acceptable. (The same
goes for the similar functions in [...]/ramnva3.c and in
[...]/ramnvc0.c, though these call ramfuc_reg() not quite as often.)
Perhaps there are other downsides to not inlining this function too. So
proper testing will probably be needed.

 drivers/gpu/drm/nouveau/core/subdev/fb/ramfuc.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/nouveau/core/subdev/fb/ramfuc.h b/drivers/gpu/drm/nouveau/core/subdev/fb/ramfuc.h
index 0f57fcf..04e3849 100644
--- a/drivers/gpu/drm/nouveau/core/subdev/fb/ramfuc.h
+++ b/drivers/gpu/drm/nouveau/core/subdev/fb/ramfuc.h
@@ -26,7 +26,7 @@ ramfuc_reg2(u32 addr1, u32 addr2)
 	};
 }
 
-static inline struct ramfuc_reg
+static noinline struct ramfuc_reg
 ramfuc_reg(u32 addr)
 {
 	return ramfuc_reg2(addr, addr);
-- 
1.8.4.2

Re: [PATCH] drm/nouveau/fb: mark ramfuc_reg() noinline

[Paul Bolle]

Paul Bolle schreef op vr 10-01-2014 om 11:37 [+0100]:
> Building ramnve0.o triggers a GCC warning on 32 bits x86:
>     drivers/gpu/drm/nouveau/core/subdev/fb/ramnve0.c: In function 'nve0_ram_ctor':
>     drivers/gpu/drm/nouveau/core/subdev/fb/ramnve0.c:1253:1: warning: the frame size of 1496 bytes is larger than 1024 bytes [-Wframe-larger-than=]
> 
> This warning is caused by ramfuc_reg(), which is inlined 74 times in
> nve0_ram_ctor(). Mark it noinline to silence this warning.
> 
> Signed-off-by: Paul Bolle <pebolle@tiscali.nl>
> ---
> Compile tested (on 32 bits x86) only. I've no Nvidia cards at hand, so I
> can't really test it.
> 
> This assumes this function - a constructor, apparently - isn't called

Here "this function" refers to nve0_ram_ctor().

> often, so the overhead calling of 74 functions is acceptable. (The same
> goes for the similar functions in [...]/ramnva3.c and in
> [...]/ramnvc0.c, though these call ramfuc_reg() not quite as often.)
> Perhaps there are other downsides to not inlining this function too. So

And here it refers to ramfuc_reg(). So my writing was a bit confusing.

> proper testing will probably be needed.

Building v3.14-rc5 on 32 bit x86 still triggers this warning. Has anyone
tried to review or test this patch?


Paul Bolle

>  drivers/gpu/drm/nouveau/core/subdev/fb/ramfuc.h | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/drivers/gpu/drm/nouveau/core/subdev/fb/ramfuc.h b/drivers/gpu/drm/nouveau/core/subdev/fb/ramfuc.h
> index 0f57fcf..04e3849 100644
> --- a/drivers/gpu/drm/nouveau/core/subdev/fb/ramfuc.h
> +++ b/drivers/gpu/drm/nouveau/core/subdev/fb/ramfuc.h
> @@ -26,7 +26,7 @@ ramfuc_reg2(u32 addr1, u32 addr2)
>  	};
>  }
>  
> -static inline struct ramfuc_reg
> +static noinline struct ramfuc_reg
>  ramfuc_reg(u32 addr)
>  {
>  	return ramfuc_reg2(addr, addr);