Re: [PATCH] media: chips-media: wave5: fix panic on decoding DECODED_IDX_FLAG_SKIP

[Mattijs Korpershoek <mkorpershoek@baylibre.com>]

Hi Nicolas,

Thank you for your review.

On Fri, Dec 01, 2023 at 15:18, Nicolas Dufresne <nicolas@ndufresne.ca> wrote:

> Hi Mattijs,
>
> Le mercredi 29 novembre 2023 à 11:37 +0100, Mattijs Korpershoek a écrit :
>> The display frame region information received from the vpu also
>> contains the frame display index: info->index_frame_display.
>> 
>> This index, being a s32, can be negative when a skip option is passed.
>> In that case, its value is DECODED_IDX_FLAG_SKIP (-2).
>> 
>> When disp_idx == -2, the following exception occurs:
>> 
>> [ 1530.782246][ T1900] Hardware name: Texas Instruments AM62P5 SK (DT)
>> [ 1530.788501][ T1900] pstate: a0400005 (NzCv daif +PAN -UAO -TCO -DIT -SSBS BTYPE=--)
>> [ 1530.796144][ T1900] pc : wave5_vpu_dec_get_output_info+0x300/0x308 [wave5]
>> [ 1530.803060][ T1900] lr : wave5_vpu_dec_get_output_info+0x80/0x308 [wave5]
>> [ 1530.809873][ T1900] sp : ffffffc00b85bc00
>> [ 1530.813872][ T1900] x29: ffffffc00b85bc00 x28: 0000000000000000 x27: 0000000000000001
>> [ 1530.821695][ T1900] x26: 00000000fffffffd x25: 00000000ffffffff x24: ffffff8812820000
>> [ 1530.829516][ T1900] x23: ffffff88199f7840 x22: ffffff8873f5e000 x21: ffffffc00b85bc58
>> [ 1530.837336][ T1900] x20: 0000000000000000 x19: ffffff88199f7920 x18: ffffffc00a899030
>> [ 1530.845156][ T1900] x17: 00000000529c6ef0 x16: 00000000529c6ef0 x15: 0000000000198487
>> [ 1530.852975][ T1900] x14: ffffffc009f2b650 x13: 0000000000058016 x12: 0000000005000000
>> [ 1530.860795][ T1900] x11: 0000000000000000 x10: 0000000000000000 x9 : 0000000000000000
>> [ 1530.868615][ T1900] x8 : 0000000000000000 x7 : 0000000000000000 x6 : 0000000000004086
>> [ 1530.876434][ T1900] x5 : 0000000000000001 x4 : ffffffc001454b94 x3 : ffffffc001454d94
>> [ 1530.884256][ T1900] x2 : ffffffc00b8201d0 x1 : 0000000000000020 x0 : 0000000000000000
>> [ 1530.892087][ T1900] Call trace:
>> [ 1530.895225][ T1900]  wave5_vpu_dec_get_output_info+0x300/0x308 [wave5]
>> [ 1530.901788][ T1900]  wave5_vpu_dec_finish_decode+0x6c/0x3dc [wave5]
>> [ 1530.908081][ T1900]  wave5_vpu_irq_thread+0x140/0x168 [wave5]
>> [ 1530.913856][ T1900]  irq_thread_fn+0x44/0xa4
>> [ 1530.918154][ T1900]  irq_thread+0x15c/0x288
>> [ 1530.922330][ T1900]  kthread+0x104/0x1d4
>> [ 1530.926247][ T1900]  ret_from_fork+0x10/0x20
>> [ 1530.930520][ T1900] Code: 2a1f03ea 2a1f03eb 35ffef2c 17ffff74 (d42aa240)
>> [ 1530.937296][ T1900] ---[ end trace 0000000000000000 ]---
>> [ 1530.942596][ T1900] Kernel panic - not syncing: BRK handler: Fatal exception
>> [ 1530.949629][ T1900] SMP: stopping secondary CPUs
>> [ 1530.954244][ T1900] Kernel Offset: disabled
>> [ 1530.958415][ T1900] CPU features: 0x00,00000000,00800184,0000421b
>> [ 1530.964496][ T1900] Memory Limit: none
>> 
>> Move the disp_info assignment after testing that the index is positive
>> to avoid the exception.
>> 
>> Fixes: 45d1a2b93277 ("media: chips-media: wave5: Add vpuapi layer")
>> Signed-off-by: Mattijs Korpershoek <mkorpershoek@baylibre.com>
>> ---
>>  drivers/media/platform/chips-media/wave5/wave5-vpuapi.c | 2 +-
>>  1 file changed, 1 insertion(+), 1 deletion(-)
>> 
>> diff --git a/drivers/media/platform/chips-media/wave5/wave5-vpuapi.c b/drivers/media/platform/chips-media/wave5/wave5-vpuapi.c
>> index 86b3993722db..1a3efb638dde 100644
>> --- a/drivers/media/platform/chips-media/wave5/wave5-vpuapi.c
>> +++ b/drivers/media/platform/chips-media/wave5/wave5-vpuapi.c
>> @@ -508,8 +508,8 @@ int wave5_vpu_dec_get_output_info(struct vpu_instance *inst, struct dec_output_i
>>  	info->rc_decoded = rect_info;
>>  
>>  	disp_idx = info->index_frame_display;
>> -	disp_info = &p_dec_info->dec_out_info[disp_idx];
>>  	if (info->index_frame_display >= 0 && info->index_frame_display < WAVE5_MAX_FBS) {
>> +		disp_info = &p_dec_info->dec_out_info[disp_idx];
>
> I could not reproduce the crash, it probably depends on the compiler version and
> compiler options. This negative index should normally generate a pointer, even
> if that points to bad location. I'd like to understand how that lead to a crash.
> If you have further information on this, I really like to get to the bottom on
> these subjects.

I can reproduce it consistently with my setup:
https://pastebin.com/0Mdfw1EC

Here is a copy of the .config when my kernel is build:
https://pastebin.com/RHxLV8cf

I build the kernel via KLeaf (Android's way of building kernels) so I'm
not sure how I can provide you with the exact compile command.

I do know that LTO is used, if that gives any hint.

If prefer chatting about this, I'm also reachable on irc (oftc.net, nickname:
mkorpershoek)

Let me know if other information is needed to get to the bottom on this.

>
> Meanwhile, this fix is obviously correct, we should not do that unless we have a
> valid index.
>
> Reviewed-by: Nicolas Dufresne <nicolas.dufresne@collabora.com>
>
>>  		if (info->index_frame_display != info->index_frame_decoded) {
>>  			/*
>>  			 * when index_frame_decoded < 0, and index_frame_display >= 0
>> 
>> ---
>> base-commit: a00b3f296eac3d43328615c3113e1a74143fc67a
>> change-id: 20231129-wave5-panic-82ea2d30042f
>> 
>> Best regards,