Re: [PATCH] firmware: cs_dsp: avoid large local variables

[Richard Fitzgerald <rf@opensource.cirrus.com>]

On 16/12/2024 8:33 am, Arnd Bergmann wrote:
> From: Arnd Bergmann <arnd@arndb.de>
> 
> Having 1280 bytes of local variables on the stack exceeds the limit
> on 32-bit architectures:
> 
> drivers/firmware/cirrus/test/cs_dsp_test_bin.c: In function 'bin_patch_mixed_packed_unpacked_random':
> drivers/firmware/cirrus/test/cs_dsp_test_bin.c:2097:1: error: the frame size of 1784 bytes is larger than 1024 bytes [-Werror=frame-larger-than=]
> 
> Use dynamic allocation for the largest two here.
> 
> Fixes: dd0b6b1f29b9 ("firmware: cs_dsp: Add KUnit testing of bin file download")
> Signed-off-by: Arnd Bergmann <arnd@arndb.de>
> ---
>   .../firmware/cirrus/test/cs_dsp_test_bin.c    | 36 +++++++++++--------
>   1 file changed, 22 insertions(+), 14 deletions(-)
> 
> diff --git a/drivers/firmware/cirrus/test/cs_dsp_test_bin.c b/drivers/firmware/cirrus/test/cs_dsp_test_bin.c
> index 689190453307..f1955813919c 100644
> --- a/drivers/firmware/cirrus/test/cs_dsp_test_bin.c
> +++ b/drivers/firmware/cirrus/test/cs_dsp_test_bin.c
> @@ -1978,8 +1978,10 @@ static void bin_patch_mixed_packed_unpacked_random(struct kunit *test)
>   		 4, 51, 76, 72, 16,  6, 39, 62, 15, 41, 28, 73, 53, 40, 45, 54,
>   		14, 55, 46, 66, 64, 59, 23,  9, 67, 47, 19, 71, 35, 18, 42,  1,
>   	};
> -	u32 packed_payload[80][3];
> -	u32 unpacked_payload[80];
> +	struct {
> +		u32 packed[80][3];
> +		u32 unpacked[80];
> +	} *payload;
>   	u32 readback[3];
>   	unsigned int alg_base_words, patch_pos_words;
>   	unsigned int alg_base_in_packed_regs, patch_pos_in_packed_regs;
> @@ -1988,8 +1990,12 @@ static void bin_patch_mixed_packed_unpacked_random(struct kunit *test)
>   	struct firmware *fw;
>   	int i;
>   
> -	get_random_bytes(packed_payload, sizeof(packed_payload));
> -	get_random_bytes(unpacked_payload, sizeof(unpacked_payload));
> +	payload = kmalloc(sizeof(*payload), GFP_KERNEL);

Should be kunit_kmalloc() so the kunit framework frees it automatically
if any test case fails and the test terminates early.

Apart from that it looks ok to me.