On Thu, Jun 15, 2023, at 10:46, Benjamin Gaignard wrote: > Le 15/06/2023 à 09:35, kernel test robot a écrit : >> >> If you fix the issue in a separate patch/commit (i.e. not just a new version of >> the same patch/commit), kindly add following tags >> | Reported-by: kernel test robot <lkp@xxxxxxxxx> >> | Closes: https://lore.kernel.org/oe-kbuild-all/202306151506.goHEegOd-lkp@xxxxxxxxx/ >> >> All warnings (new ones prefixed by >>): >> >> drivers/media/platform/verisilicon/rockchip_vpu981_hw_av1_dec.c: In function 'rockchip_vpu981_av1_dec_set_segmentation': >>>> drivers/media/platform/verisilicon/rockchip_vpu981_hw_av1_dec.c:1022:1: warning: the frame size of 1112 bytes is larger than 1024 bytes [-Wframe-larger-than=] >> 1022 | } >> | ^ > > Arnd have alreaday send a patch to solve this gcc issue: > https://lore.kernel.org/lkml/20230601151832.3632525-1-arnd@xxxxxxxxxx/ I just ran into the above issue with my workaround applied, so I think this is caused by something else. I can see one 256 byte variable that is a little too large to fit on the stack, but that is not the main issue: u32 segval[V4L2_AV1_MAX_SEGMENTS][V4L2_AV1_SEG_LVL_MAX] ; The actual problem appears to be the construction of temporary arrays on the stack like #define AV1_DEC_REG(b, s, m) \ ((const struct hantro_reg) { \ .base = AV1_SWREG(b), \ .shift = s, \ .mask = m, \ }) #define av1_dec_out_ec_bypass AV1_DEC_REG(3, 8, 0x1) hantro_reg_write(vpu, &av1_dec_out_ec_bypass, 1); Each one of these adds 12 bytes to the stack, and there are a lot of them here. If KASAN_STACK is active, they cannot overlap, and there is an addition redzone around each one, but the case I saw does not use KASAN or UBSAN. I've tried a few things like removing the debug printf in each register access, but that did not help. Maybe you have some other idea for how to simplify this code. Arnd
