[PATCH] media: platform: Fix a kernel warning on clk control
Eddie James
eajames at linux.ibm.com
Sat Mar 30 07:08:29 AEDT 2019
On 3/28/19 4:25 PM, Jae Hyun Yoo wrote:
> Video engine clock control functions in the Aspeed video engine driver are
> being called from multiple context without any protection so video clocks
> can be disabled twice and eventually it causes a kernel warning with stack
> dump printing out like below:
>
> [ 120.034729] WARNING: CPU: 0 PID: 1334 at drivers/clk/clk.c:684 clk_core_unprepare+0x13c/0x170
> [ 120.043252] eclk-gate already unprepared
> [ 120.047283] CPU: 0 PID: 1334 Comm: obmc-ikvm Tainted: G W 5.0.3-b94b74e8b52db91fe4e99e0bb481ec8bf2b5b47c #1
> [ 120.058417] Hardware name: Generic DT based system
> [ 120.063219] Backtrace:
> [ 120.065787] [<80107cdc>] (dump_backtrace) from [<80107f10>] (show_stack+0x20/0x24)
> [ 120.073371] r7:803a4ff0 r6:00000009 r5:00000000 r4:96197e1c
> [ 120.079152] [<80107ef0>] (show_stack) from [<8068f7d8>] (dump_stack+0x20/0x28)
> [ 120.086479] [<8068f7b8>] (dump_stack) from [<8011604c>] (__warn.part.3+0xb4/0xdc)
> [ 120.094068] [<80115f98>] (__warn.part.3) from [<801160e0>] (warn_slowpath_fmt+0x6c/0x90)
> [ 120.102164] r6:000002ac r5:8080c0b8 r4:80a07008
> [ 120.106893] [<80116078>] (warn_slowpath_fmt) from [<803a4ff0>] (clk_core_unprepare+0x13c/0x170)
> [ 120.115686] r3:8080cf8c r2:8080c17c
> [ 120.119276] r7:97d68e58 r6:9df23200 r5:9668c260 r4:96459260
> [ 120.125046] [<803a4eb4>] (clk_core_unprepare) from [<803a707c>] (clk_unprepare+0x34/0x3c)
> [ 120.133226] r5:9668c260 r4:96459260
> [ 120.136932] [<803a7048>] (clk_unprepare) from [<804f34bc>] (aspeed_video_off+0x44/0x48)
> [ 120.145031] r5:9668c260 r4:9668cbc0
> [ 120.148647] [<804f3478>] (aspeed_video_off) from [<804f3fd0>] (aspeed_video_release+0x94/0x118)
> [ 120.157435] r5:966a0cb8 r4:966a0800
> [ 120.161049] [<804f3f3c>] (aspeed_video_release) from [<804d2c58>] (v4l2_release+0xd4/0xe8)
> [ 120.169404] r7:97d68e58 r6:9d087810 r5:9df23200 r4:966a0b20
> [ 120.175168] [<804d2b84>] (v4l2_release) from [<80236224>] (__fput+0x98/0x1c4)
> [ 120.182316] r5:96698e78 r4:9df23200
> [ 120.185994] [<8023618c>] (__fput) from [<802363b8>] (____fput+0x18/0x1c)
> [ 120.192712] r9:80a0700c r8:801011e4 r7:00000000 r6:80a64bbc r5:961dd560 r4:961dd89c
> [ 120.200562] [<802363a0>] (____fput) from [<80131c08>] (task_work_run+0x7c/0xa4)
> [ 120.207994] [<80131b8c>] (task_work_run) from [<80106884>] (do_work_pending+0x4a8/0x578)
> [ 120.216163] r7:801011e4 r6:80a07008 r5:96197fb0 r4:ffffe000
> [ 120.221856] [<801063dc>] (do_work_pending) from [<8010106c>] (slow_work_pending+0xc/0x20)
> [ 120.230116] Exception stack(0x96197fb0 to 0x96197ff8)
> [ 120.235254] 7fa0: 00000000 76ccf094 00000000 00000000
> [ 120.243438] 7fc0: 00000008 00a11978 7eab3c30 00000006 00000000 00000000 475b0fa4 00000000
> [ 120.251692] 7fe0: 00000002 7eab3a40 00000000 47720e38 80000010 00000008
> [ 120.258396] r10:00000000 r9:96196000 r8:801011e4 r7:00000006 r6:7eab3c30 r5:00a11978
> [ 120.266291] r4:00000008
>
> To prevent this issue, this commit adds spinlock protection and clock
> status checking logic into the Aspeed video engine driver.
Thanks Jae. Do you have a reliable way to reproduce this? Haven't seen
it myself.
>
> Fixes: d2b4387f3bdf ("media: platform: Add Aspeed Video Engine driver")
> Signed-off-by: Jae Hyun Yoo <jae.hyun.yoo at linux.intel.com>
> Cc: Eddie James <eajames at linux.ibm.com>
> Cc: Mauro Carvalho Chehab <mchehab at kernel.org>
> Cc: Joel Stanley <joel at jms.id.au>
> Cc: Andrew Jeffery <andrew at aj.id.au>
> ---
> drivers/media/platform/aspeed-video.c | 46 ++++++++++++++++++++-------
> 1 file changed, 35 insertions(+), 11 deletions(-)
>
> diff --git a/drivers/media/platform/aspeed-video.c b/drivers/media/platform/aspeed-video.c
> index 692e08ef38c0..9663ba4281a8 100644
> --- a/drivers/media/platform/aspeed-video.c
> +++ b/drivers/media/platform/aspeed-video.c
> @@ -227,6 +227,7 @@ struct aspeed_video {
> struct list_head buffers;
> unsigned long flags;
> unsigned int sequence;
> + bool is_video_on;
This can probably be a flag, like VIDEO_CLOCKS_ON, not a new boolean.
>
> unsigned int max_compressed_size;
> struct aspeed_video_addr srcs[2];
> @@ -495,20 +496,28 @@ static void aspeed_video_reset(struct aspeed_video *video)
>
> static void aspeed_video_off(struct aspeed_video *video)
> {
> - aspeed_video_reset(video);
> + if (video->is_video_on) {
> + aspeed_video_reset(video);
I'm working on a patch to remove the use of the reset line too.
> +
> + /* Turn off the relevant clocks */
> + clk_disable_unprepare(video->vclk);
> + clk_disable_unprepare(video->eclk);
>
> - /* Turn off the relevant clocks */
> - clk_disable_unprepare(video->vclk);
> - clk_disable_unprepare(video->eclk);
> + video->is_video_on = false;
> + }
> }
>
> static void aspeed_video_on(struct aspeed_video *video)
> {
> - /* Turn on the relevant clocks */
> - clk_prepare_enable(video->eclk);
> - clk_prepare_enable(video->vclk);
> + if (!video->is_video_on) {
> + /* Turn on the relevant clocks */
> + clk_prepare_enable(video->eclk);
> + clk_prepare_enable(video->vclk);
> +
> + aspeed_video_reset(video);
>
> - aspeed_video_reset(video);
> + video->is_video_on = true;
> + }
> }
>
> static void aspeed_video_bufs_done(struct aspeed_video *video,
> @@ -526,12 +535,14 @@ static void aspeed_video_bufs_done(struct aspeed_video *video,
>
> static void aspeed_video_irq_res_change(struct aspeed_video *video)
> {
> + spin_lock(&video->lock);
Is there a reason you're locking extra code? Why not put the lock in the
aspeed_video_on/off functions?
> dev_dbg(video->dev, "Resolution changed; resetting\n");
>
> set_bit(VIDEO_RES_CHANGE, &video->flags);
> clear_bit(VIDEO_FRAME_INPRG, &video->flags);
>
> aspeed_video_off(video);
> + spin_unlock(&video->lock);
> aspeed_video_bufs_done(video, VB2_BUF_STATE_ERROR);
>
> schedule_delayed_work(&video->res_work, RESOLUTION_CHANGE_DELAY);
> @@ -951,9 +962,13 @@ static void aspeed_video_init_regs(struct aspeed_video *video)
>
> static void aspeed_video_start(struct aspeed_video *video)
> {
> + unsigned long flags;
> +
> + spin_lock_irqsave(&video->lock, flags);
> aspeed_video_on(video);
>
> aspeed_video_init_regs(video);
> + spin_unlock_irqrestore(&video->lock, flags);
>
> /* Resolution set to 640x480 if no signal found */
> aspeed_video_get_resolution(video);
> @@ -969,6 +984,9 @@ static void aspeed_video_start(struct aspeed_video *video)
>
> static void aspeed_video_stop(struct aspeed_video *video)
> {
> + unsigned long flags;
> +
> + spin_lock_irqsave(&video->lock, flags);
> set_bit(VIDEO_STOPPED, &video->flags);
> cancel_delayed_work_sync(&video->res_work);
>
> @@ -982,6 +1000,7 @@ static void aspeed_video_stop(struct aspeed_video *video)
>
> video->v4l2_input_status = V4L2_IN_ST_NO_SIGNAL;
> video->flags = 0;
> + spin_unlock_irqrestore(&video->lock, flags);
> }
>
> static int aspeed_video_querycap(struct file *file, void *fh,
> @@ -1319,16 +1338,21 @@ static void aspeed_video_resolution_work(struct work_struct *work)
> struct delayed_work *dwork = to_delayed_work(work);
> struct aspeed_video *video = container_of(dwork, struct aspeed_video,
> res_work);
> - u32 input_status = video->v4l2_input_status;
> + unsigned long flags;
> + u32 input_status;
>
> + spin_lock_irqsave(&video->lock, flags);
> + input_status = video->v4l2_input_status;
> aspeed_video_on(video);
>
> /* Exit early in case no clients remain */
> - if (test_bit(VIDEO_STOPPED, &video->flags))
> + if (test_bit(VIDEO_STOPPED, &video->flags)) {
> + spin_unlock_irqrestore(&video->lock, flags);
> goto done;
> + }
>
> aspeed_video_init_regs(video);
> -
> + spin_unlock_irqrestore(&video->lock, flags);
> aspeed_video_get_resolution(video);
>
> if (video->detected_timings.width != video->active_timings.width ||
More information about the Linux-aspeed
mailing list