[PATCH] media: verisilicon: Additional fix for the crash when opening the driver

[Michael Tretter]

On Tue, 23 May 2023 18:36:09 +0200, Benjamin Gaignard wrote:
> 
> Le 23/05/2023 à 18:25, Benjamin Gaignard a écrit :
> > This fixes the following issue observed on Odroid-M1 board:
> > 
> >   Unable to handle kernel NULL pointer dereference at virtual address 0000000000000008
> >   Mem abort info:
> >   ...
> >   Modules linked in: crct10dif_ce hantro_vpu snd_soc_simple_card snd_soc_simple_card_utils v4l2_vp9 v4l2_h264 rockchip_saradc v4l2_mem2mem videobuf2_dma_contig videobuf2_memops rtc_rk808 videobuf2_v4l2 industrialio_triggered_buffer rockchip_thermal dwmac_rk stmmac_platform stmmac videodev kfifo_buf display_connector videobuf2_common pcs_xpcs mc rockchipdrm analogix_dp dw_mipi_dsi dw_hdmi drm_display_helper panfrost drm_shmem_helper gpu_sched ip_tables x_tables ipv6
> >   CPU: 3 PID: 176 Comm: v4l_id Not tainted 6.3.0-rc7-next-20230420 #13481
> >   Hardware name: Hardkernel ODROID-M1 (DT)
> >   pstate: 60400009 (nZCv daif +PAN -UAO -TCO -DIT -SSBS BTYPE=--)
> >   pc : hantro_try_fmt+0xa0/0x278 [hantro_vpu]
> >   lr : hantro_try_fmt+0x94/0x278 [hantro_vpu]
> >   ...
> >   Call trace:
> >    hantro_try_fmt+0xa0/0x278 [hantro_vpu]
> >    hantro_set_fmt_out+0x3c/0x298 [hantro_vpu]
> >    hantro_reset_raw_fmt+0x98/0x128 [hantro_vpu]
> >    hantro_set_fmt_cap+0x240/0x254 [hantro_vpu]
> >    hantro_reset_encoded_fmt+0x94/0xcc [hantro_vpu]
> >    hantro_reset_fmts+0x18/0x38 [hantro_vpu]
> >    hantro_open+0xd4/0x20c [hantro_vpu]
> >    v4l2_open+0x80/0x120 [videodev]
> >    chrdev_open+0xc0/0x22c
> >    do_dentry_open+0x13c/0x48c
> >    vfs_open+0x2c/0x38
> >    path_openat+0x550/0x934
> >    do_filp_open+0x80/0x12c
> >    do_sys_openat2+0xb4/0x168
> >    __arm64_sys_openat+0x64/0xac
> >    invoke_syscall+0x48/0x114
> >    el0_svc_common+0x100/0x120
> >    do_el0_svc+0x3c/0xa8
> >    el0_svc+0x40/0xa8
> >    el0t_64_sync_handler+0xb8/0xbc
> >    el0t_64_sync+0x190/0x194
> >   Code: 97fc8a7f f940aa80 52864a61 72a686c1 (b9400800)
> >   ---[ end trace 0000000000000000 ]---
> > 
> > Fixes: db6f68b51e5c ("media: verisilicon: Do not set context src/dst formats in reset functions")

This patch partially reverts the previous commit. I wonder whether the reason
for resetting the context format only if the targeted queue is not busy still
stands.

> > 
> > Signed-off-by: Benjamin Gaignard <benjamin.gaignard at collabora.com>

Tested-by: Michael Tretter <m.tretter at pengutronix.de>

> > ---
> 
> Diederick, Marek, Michael,
> I have tested this patch on my boards and I see no regressions on
> decoder part and no more crash when probing the encoder.
> Could you test it on your side to confirm it is ok ?
> 
> Thorsten, I try/test regzbot commands, please tell me if it is correct.
> 
> #regzbot ^introduced db6f68b51e5c
> #regzbot title media: verisilicon: null pointer dereference in try_fmt
> #regzbot ignore-activity
> 
> 
> >   drivers/media/platform/verisilicon/hantro_v4l2.c | 6 ++++--
> >   1 file changed, 4 insertions(+), 2 deletions(-)
> > 
> > diff --git a/drivers/media/platform/verisilicon/hantro_v4l2.c b/drivers/media/platform/verisilicon/hantro_v4l2.c
> > index 835518534e3b..61cfaaf4e927 100644
> > --- a/drivers/media/platform/verisilicon/hantro_v4l2.c
> > +++ b/drivers/media/platform/verisilicon/hantro_v4l2.c
> > @@ -397,10 +397,12 @@ hantro_reset_raw_fmt(struct hantro_ctx *ctx, int bit_depth)
> >   	if (!raw_vpu_fmt)
> >   		return -EINVAL;
> > -	if (ctx->is_encoder)
> > +	if (ctx->is_encoder) {
> >   		encoded_fmt = &ctx->dst_fmt;
> > -	else
> > +		ctx->vpu_src_fmt = raw_vpu_fmt;
> > +	} else {
> >   		encoded_fmt = &ctx->src_fmt;
> > +	}
> >   	hantro_reset_fmt(&raw_fmt, raw_vpu_fmt);
> >   	raw_fmt.width = encoded_fmt->width;
>