Re: [PATCH] drm: a6xx: avoid excessive stack usage

["Arnd Bergmann" <arnd@arndb.de>]

On Mon, Oct 21, 2024, at 09:25, Akhil P Oommen wrote:
> On Sat, Oct 19, 2024 at 04:14:13PM +0300, Dmitry Baryshkov wrote:
>> On Sat, Oct 19, 2024 at 03:01:46PM +0530, Akhil P Oommen wrote:
>> > On Fri, Oct 18, 2024 at 03:11:38PM +0000, Arnd Bergmann wrote:
>> > > From: Arnd Bergmann <arnd@arndb.de>
>> > > 
>> > > Clang-19 and above sometimes end up with multiple copies of the large
>> > > a6xx_hfi_msg_bw_table structure on the stack. The problem is that
>> > > a6xx_hfi_send_bw_table() calls a number of device specific functions to
>> > > fill the structure, but these create another copy of the structure on
>> > > the stack which gets copied to the first.
>> > > 
>> > > If the functions get inlined, that busts the warning limit:
>> > > 
>> > > drivers/gpu/drm/msm/adreno/a6xx_hfi.c:631:12: error: stack frame size (1032) exceeds limit (1024) in 'a6xx_hfi_send_bw_table' [-Werror,-Wframe-larger-than]
>> > 
>> > Why does this warning says that the limit is 1024? 1024 bytes is too small, isn't it?
>> 
>> Kernel stacks are expected to be space limited, so 1024 is a logical
>> limit for a single function.
>
> Thanks for the clarification. I think it is better to move this table to
> struct a6xx_gmu which is required anyway when we implement dynamic generation
> of bw table. Also, we can skip initializing it in subsequent gpu wake ups.
>
> Arnd, do you think that would be sufficient? I can send that patch if you
> want help.

Yes, that should work. I actually tried first to turn the model
specific data into static const structures but that ended up 
not working because some of them have a couple of dynamically
computed values. I think that would have been even nicer.

      Arnd