Re: [PATCH v2] soc: qcom: llcc: Fix slice configuration values for SC8280XP

[Juerg Haefliger]

On Fri, 27 Jan 2023 16:47:24 +0200
Abel Vesa <abel.vesa@xxxxxxxxxx> wrote:

> These new values are now based on the latest LLCC SC table.
> 
> Fixes: ec69dfbdc426 ("soc: qcom: llcc: Add sc8180x and sc8280xp configurations")
> Signed-off-by: Abel Vesa <abel.vesa@xxxxxxxxxx>
> ---
> 
> The v1 is here:
> https://lore.kernel.org/all/20230126171636.2319496-1-abel.vesa@xxxxxxxxxx/
> 
> Changes since v1:
>  * dropped the LLCC_GPU and LLCC_WRCACHE max_cap changes
>  * took the new values from documentatio this time rather than
>    downstream kernel
> 
>  drivers/soc/qcom/llcc-qcom.c | 6 +++---
>  1 file changed, 3 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/soc/qcom/llcc-qcom.c b/drivers/soc/qcom/llcc-qcom.c
> index 23ce2f78c4ed..26efe12012a0 100644
> --- a/drivers/soc/qcom/llcc-qcom.c
> +++ b/drivers/soc/qcom/llcc-qcom.c
> @@ -191,9 +191,9 @@ static const struct llcc_slice_config sc8280xp_data[] = {
>  	{ LLCC_CVP,      28, 512,  3, 1, 0xfff, 0x0, 0, 0, 0, 1, 0, 0 },
>  	{ LLCC_APTCM,    30, 1024, 3, 1, 0x0,   0x1, 1, 0, 0, 1, 0, 0 },
>  	{ LLCC_WRCACHE,  31, 1024, 1, 1, 0xfff, 0x0, 0, 0, 0, 0, 1, 0 },
> -	{ LLCC_CVPFW,    32, 512,  1, 0, 0xfff, 0x0, 0, 0, 0, 1, 0, 0 },
> -	{ LLCC_CPUSS1,   33, 2048, 1, 1, 0xfff, 0x0, 0, 0, 0, 1, 0, 0 },
> -	{ LLCC_CPUHWT,   36, 512,  1, 1, 0xfff, 0x0, 0, 0, 0, 0, 1, 0 },
> +	{ LLCC_CVPFW,    17, 512,  1, 0, 0xfff, 0x0, 0, 0, 0, 1, 0, 0 },
> +	{ LLCC_CPUSS1,   3, 2048, 1, 1, 0xfff, 0x0, 0, 0, 0, 1, 0, 0 },
> +	{ LLCC_CPUHWT,   5, 512,  1, 1, 0xfff, 0x0, 0, 0, 0, 0, 1, 0 },
>  };

Without this patch UBSAN reports multiple of the following on a Lenovo x13s:

[    0.083079] ================================================================================
[    0.083085] UBSAN: shift-out-of-bounds in /build/linux-laptop-0y97Sr/linux-laptop-6.2.0/drivers/soc/qcom/llcc-qcom.c:772:45
[    0.083091] shift exponent 32 is too large for 32-bit type 'int'
[    0.083096] CPU: 5 PID: 1 Comm: swapper/0 Not tainted 6.2.0-1001-laptop #1-Ubuntu
[    0.083100] Hardware name: LENOVO 4810QL0100/4810QL0100, BIOS N3HET75C (1.47 ) 11/01/2022
[    0.083104] Call trace:
[    0.083107]  dump_backtrace+0x110/0x16c
[    0.083115]  show_stack+0x24/0x50
[    0.083119]  dump_stack_lvl+0x68/0x9c
[    0.083124]  dump_stack+0x1c/0x48
[    0.083126]  ubsan_epilogue+0x14/0x60
[    0.083132]  __ubsan_handle_shift_out_of_bounds+0xf0/0x1e0
[    0.083139]  _qcom_llcc_cfg_program+0x9c8/0xad4
[    0.083145]  qcom_llcc_probe+0x284/0x33c
[    0.083149]  platform_probe+0x74/0x110
[    0.083155]  really_probe+0xd8/0x430
[    0.083158]  __driver_probe_device+0x94/0x1e4
[    0.083161]  driver_probe_device+0x48/0x17c
[    0.083164]  __driver_attach+0x150/0x2cc
[    0.083168]  bus_for_each_dev+0x80/0x100
[    0.083173]  driver_attach+0x30/0x54
[    0.083176]  bus_add_driver+0x1d0/0x2a0
[    0.083179]  driver_register+0x84/0x184
[    0.083183]  __platform_driver_register+0x34/0x60
[    0.083187]  qcom_llcc_driver_init+0x2c/0x48
[    0.083193]  do_one_initcall+0x64/0x320
[    0.083196]  do_initcalls+0x188/0x1f8
[    0.083201]  kernel_init_freeable+0x170/0x1c0
[    0.083205]  kernel_init+0x3c/0x180
[    0.083209]  ret_from_fork+0x10/0x20
[    0.083213] ================================================================================

They're all gone with this patch. From that perspective:

Tested-by: Juerg Haefliger <juerg.haefliger@xxxxxxxxxxxxx>

...Juerg

  
>  static const struct llcc_slice_config sdm845_data[] =  {

Attachment: pgpy3NDgVnKsV.pgp
Description: OpenPGP digital signature