E5010 Probe Cleanup and Adding autosuspend
Subsystems (3)
v9 cover E5010 Probe Cleanup and Adding autosuspend
Hello all,
I noticed a report from the intel bot[0] which pointed out a valid issue with the driver never calling e5010_init_device unless CONFIG_PM was enabled.
While looking at probe though, I noticed that the clk_enable was never properly called. That has now been fixed with a proper remove path as well.
Lastly, I cleaned up the goto jumps in probe to fix an issue where the m2m_dev wasn't being freed.
The rest of the series is setting up the driver for properly adding in support for autosuspend. Last patch adds in autosuspend support and moves e5010_init_device into probe for cases where PM isn't enabled.
Changes since v8:
- Patch 2: deleted new line I added in e5010_remove
- Patch 3 and 4: no code changed just commit message updates
Changes since v7:
- patch 3 subject wasn't formatted correctly after I updated it so wasn't properly showing up
Changes since v6:
- Patch 2 changed order of calls during remove cleanup. video unregister now happens before pm disable
- patch 3 was updated to account for the case where a reset was called and the device was never properly reinitialized
- patch 3 has moved setting the last_run_context to NULL as well to always update QP values when device is turned back on/reinitialized
- removed the r-b by Kendall, Devarsh, and Jai since I have made a slight code change requested by Devarsh
- patch 4 remove the last_run_context in runtime resume hook that it had before because I moved it to patch 3
- patch 4 has updates to commit message to accurately reflect why init_device was added to probe
- patch 4 has updates where disable autosuspend now comes before disable pm
Changes since v5:
- Fix patch 1 not properly removing all references to the fail_after_video_register_device label
Changes since v4:
- Change authorship of patch 1 since Myeonghun sent a similar patch first
- addressed Devarsh's comments in Patch 1 to follow similar naming structure as rest of probe function
- Moved the clk_enable call after pm_runtime_enable here so I can check success of that call. If runtime pm is enabled then don't enable the clock here
- Patch 3 no changes
- Patch 4 I added the macro requested by Devarsh instead of using magic number
- failure to power on now goes through the existing no_ready_buf_err path, so buffers are marked as error and the job is finished
Changes since v3:
- patches 1 and 2 are new in v4
- patch 3 remains unchanged
- patch 4:
- updated goto to match changes in patch 1
- update PM calls in probe
Best,
Brandon
[0]: https://lore.kernel.org/oe-kbuild-all/[email protected]/
Brandon Brnich (3):
media: imagination: e5010: Fix clk never enabled without CONFIG_PM
media: imagination: e5010: Update e5010_init_device Call Sites to
Ensure Proper Device State
media: imagination: e5010: Enable autosuspend for runtime PM
Myeonghun Pak (1):
media: imagination: e5010: release m2m device on probe failure
.../platform/imagination/e5010-jpeg-enc.c | 91 +++++++++++-------- .../platform/imagination/e5010-jpeg-enc.h | 1 + 2 files changed, 56 insertions(+), 36 deletions(-)
Signature · 1 line
2.43.0
PATCH v9 1/4 media: imagination: e5010: release m2m device on probe failure
From: Myeonghun Pak <[email protected]>
The probe path allocates the mem2mem device with v4l2_m2m_init(), but several later failures jump directly to the v4l2_device_unregister() unwind path. Since the driver core does not call .remove() after a failed probe, those paths leak the m2m device.
Route post-v4l2_m2m_init() failures through the existing m2m release step before unregistering the v4l2 device.
--- drivers/media/platform/imagination/e5010-jpeg-enc.c | 12 ++++++------ 1 file changed, 6 insertions(+), 6 deletions(-)
drivers/media/platform/imagination/e5010-jpeg-enc.c +6 -6
@@ -1072,14 +1072,14 @@ static int e5010_probe(struct platform_device *pdev)if (IS_ERR (e5010->core_base)) { ret =PTR_ERR (e5010->core_base);dev_err_probe (dev, ret,"Missing 'core' resources area\n" ); - goto fail_after_v4l2_register; + goto fail_after_v4l2_m2m_init; } e5010->mmu_base =devm_platform_ioremap_resource_byname (pdev,"mmu" );if (IS_ERR (e5010->mmu_base)) { ret =PTR_ERR (e5010->mmu_base);dev_err_probe (dev, ret,"Missing 'mmu' resources area\n" ); - goto fail_after_v4l2_register; + goto fail_after_v4l2_m2m_init; } e5010->last_context_run =NULL ;
@@ -1089,14 +1089,14 @@ static int e5010_probe(struct platform_device *pdev) E5010_MODULE_NAME, e5010);if (ret) {dev_err_probe (dev, ret,"failed to register IRQ %d\n" , irq); - goto fail_after_v4l2_register; + goto fail_after_v4l2_m2m_init; } e5010->clk =devm_clk_get (dev,NULL );if (IS_ERR (e5010->clk)) { ret =PTR_ERR (e5010->clk);dev_err_probe (dev, ret,"failed to get clock\n" ); - goto fail_after_v4l2_register; + goto fail_after_v4l2_m2m_init; }pm_runtime_enable (dev);
@@ -1104,7 +1104,7 @@ static int e5010_probe(struct platform_device *pdev) ret =video_register_device (e5010->vdev,VFL_TYPE_VIDEO ,0 );if (ret) {dev_err_probe (dev, ret,"failed to register video device\n" ); - goto fail_after_video_register_device; + goto fail_after_v4l2_m2m_init; } v4l2_info(&e5010->v4l2_dev, "Device registered as /dev/video%d\n",
@@ -1112,7 +1112,7 @@ static int e5010_probe(struct platform_device *pdev)return 0 ; -fail_after_video_register_device: +fail_after_v4l2_m2m_init:v4l2_m2m_release (e5010->m2m_dev); fail_after_v4l2_register:v4l2_device_unregister (&e5010->v4l2_dev);
Signature · 1 line
2.43.0
PATCH v9 2/4 media: imagination: e5010: Fix clk never enabled without CONFIG_PM
In reply to Brandon Brnichclk_prepare_enable is not called anywhere unless CONFIG_PM is set. This means without CONFIG_PM, the JPEG Encoder will not function properly.
Add this call during the probe sequence to properly configure the device. If video_registration fails after clk is enabled, clk would be left on. Make a label to cover this case.
Update e5010_remove to ensure that clock is getting disabled in the case where e5010 was not suspended when remove gets called.
--- .../platform/imagination/e5010-jpeg-enc.c | 19 +++++++++++++++++-- 1 file changed, 17 insertions(+), 2 deletions(-)
drivers/media/platform/imagination/e5010-jpeg-enc.c +17 -2
@@ -1101,10 +1101,18 @@ static int e5010_probe(struct platform_device *pdev)pm_runtime_enable (dev); +if (!pm_runtime_enabled (dev)) { + ret =clk_prepare_enable (e5010->clk); +if (ret) { +dev_err_probe (dev, ret,"failed to enable clock\n" ); + goto fail_after_v4l2_m2m_init; + } + } + ret =video_register_device (e5010->vdev,VFL_TYPE_VIDEO ,0 );if (ret) {dev_err_probe (dev, ret,"failed to register video device\n" ); - goto fail_after_v4l2_m2m_init; + goto fail_after_clock_enable; } v4l2_info(&e5010->v4l2_dev, "Device registered as /dev/video%d\n",
@@ -1112,6 +1120,11 @@ static int e5010_probe(struct platform_device *pdev)return 0 ; +fail_after_clock_enable: +if (pm_runtime_enabled (dev)) +pm_runtime_disable (dev); +else +clk_disable_unprepare (e5010->clk); fail_after_v4l2_m2m_init:v4l2_m2m_release (e5010->m2m_dev); fail_after_v4l2_register:
@@ -1125,10 +1138,12 @@ static void e5010_remove(struct platform_device *pdev) {struct e5010_dev *e5010 =platform_get_drvdata (pdev); -pm_runtime_disable (e5010->dev);video_unregister_device (e5010->vdev);v4l2_m2m_release (e5010->m2m_dev);v4l2_device_unregister (&e5010->v4l2_dev); +pm_runtime_disable (e5010->dev); +if (!pm_runtime_status_suspended (&pdev->dev)) +clk_disable_unprepare (e5010->clk); }static void e5010_vb2_buffers_return (struct vb2_queue *q,enum vb2_buffer_state state)
Signature · 1 line
2.43.0
PATCH v9 3/4 media: imagination: e5010: Update e5010_init_device Call Sites to Ensure Proper Device State
In reply to Brandon BrnichThe PM framework skips the runtime resume hook if the device was never on in the first place. This results in a crash in the system resume hook when a call to e5010_init_device() attempts a write to a powered down device.
Move e5010_init_device() to the resume hook to ensure register write only occurs when device is properly powered on.
e5010_reset() resets the whole IP, which reverts all registers, including the interrupt enables, input source and quantization tables, back to their reset defaults. Call e5010_init_device() after the reset so the next device_run() starts from a properly programmed device. e5010->last_context_run in e5010_init_device() should be set to NULL as well to guarantee the next device_run() reloads the user-configured QP tables instead of encoding with the default ones.
--- .../platform/imagination/e5010-jpeg-enc.c | 19 ++++++++++++------- 1 file changed, 12 insertions(+), 7 deletions(-)
drivers/media/platform/imagination/e5010-jpeg-enc.c +12 -7
@@ -986,6 +986,8 @@ static int e5010_init_device(struct e5010_dev *e5010) {int ret =0 ; + e5010->last_context_run =NULL ; +/*TODO: Set MMU in bypass mode until support for the same is added in driver*/ e5010_hw_bypass_mmu (e5010->mmu_base,1 );
@@ -1487,7 +1489,9 @@ static void e5010_device_run(void *priv) device_busy_err:e5010_reset (e5010->dev, e5010->core_base, e5010->mmu_base); - + ret =e5010_init_device (e5010); +if (ret) +v4l2_err (&e5010->v4l2_dev,"failed to initialize device after reset\n" ); no_ready_buf_err:if (s_vb) {v4l2_m2m_src_buf_remove_by_buf (ctx->fh.m2m_ctx, s_vb);
@@ -1516,6 +1520,13 @@ static int e5010_runtime_resume(struct device *dev)return ret; } + ret =e5010_init_device (e5010); +if (ret) { +dev_err (dev,"Failed to re-enable e5010 device\n" ); +clk_disable_unprepare (e5010->clk); +return ret; + } +return 0 ; }
@@ -1548,12 +1559,6 @@ static int e5010_resume(struct device *dev)if (ret <0 )return ret; - ret =e5010_init_device (e5010); -if (ret) { -dev_err (dev,"Failed to re-enable e5010 device\n" ); -return ret; - } -v4l2_m2m_resume (e5010->m2m_dev);return ret;
Signature · 1 line
2.43.0
PATCH v9 4/4 media: imagination: e5010: Enable autosuspend for runtime PM
In reply to Brandon BrnichCurrent implementation only has one call to turn the device on - start_streaming(). This can waste power of system as the device can be in STREAM_ON state, but no buffers are actually being submitted. To address this, device_run() will obtain a PM reference and will release it after interrupt completion is handled utilizing the pm autosuspend framework so device can shutdown if no other job is scheduled within 100ms.
Since e5010_runtime_resume() now initializes the hardware on every power-up, drop the e5010_init_device() call from start_streaming(). In the case where PM is not configured, e5010_init_device() would no longer be called since the only path was in start_streaming(). Move this call to e5010_probe() if runtime PM is not enabled so that the device registers are properly set.
--- .../platform/imagination/e5010-jpeg-enc.c | 47 +++++++++---------- .../platform/imagination/e5010-jpeg-enc.h | 1 + 2 files changed, 24 insertions(+), 24 deletions(-)
drivers/media/platform/imagination/e5010-jpeg-enc.c +23 -24
@@ -976,6 +976,7 @@ static irqreturn_t e5010_irq(int irq, void *data)v4l2_m2m_job_finish (e5010->m2m_dev, ctx->fh.m2m_ctx);dprintk (e5010,3 ,"ctx: 0x%p Finish job\n" , ctx); +pm_runtime_put_autosuspend (e5010->dev); job_unlock:spin_unlock (&e5010->hw_lock);
@@ -1101,6 +1102,8 @@ static int e5010_probe(struct platform_device *pdev) goto fail_after_v4l2_m2m_init; } +pm_runtime_set_autosuspend_delay (dev,JPEG_AUTOSUSPEND_DELAY_MS ); +pm_runtime_use_autosuspend (dev);pm_runtime_enable (dev);if (!pm_runtime_enabled (dev)) {
@@ -1109,6 +1112,11 @@ static int e5010_probe(struct platform_device *pdev)dev_err_probe (dev, ret,"failed to enable clock\n" ); goto fail_after_v4l2_m2m_init; } + ret =e5010_init_device (e5010); +if (ret) { +dev_err_probe (dev, ret,"failed to init device\n" ); + goto fail_after_clock_enable; + } } ret =video_register_device (e5010->vdev,VFL_TYPE_VIDEO ,0 );
@@ -1123,10 +1131,12 @@ static int e5010_probe(struct platform_device *pdev)return 0 ; fail_after_clock_enable: -if (pm_runtime_enabled (dev)) +if (pm_runtime_enabled (dev)) { +pm_runtime_dont_use_autosuspend (dev);pm_runtime_disable (dev); -else + }else {clk_disable_unprepare (e5010->clk); + } fail_after_v4l2_m2m_init:v4l2_m2m_release (e5010->m2m_dev); fail_after_v4l2_register:
@@ -1143,6 +1153,7 @@ static void e5010_remove(struct platform_device *pdev)video_unregister_device (e5010->vdev);v4l2_m2m_release (e5010->m2m_dev);v4l2_device_unregister (&e5010->v4l2_dev); +pm_runtime_dont_use_autosuspend (e5010->dev);pm_runtime_disable (e5010->dev);if (!pm_runtime_status_suspended (&pdev->dev))clk_disable_unprepare (e5010->clk);
@@ -1306,31 +1317,13 @@ static int e5010_encoder_cmd(struct file *file, void *priv,static int e5010_start_streaming (struct vb2_queue *q,unsigned int count) {struct e5010_context *ctx =vb2_get_drv_priv (q); -int ret;struct e5010_q_data *queue =get_queue (ctx, q->type);v4l2_m2m_update_start_streaming_state (ctx->fh.m2m_ctx, q); queue->sequence =0 ; - ret =pm_runtime_resume_and_get (ctx->e5010->dev); -if (ret <0 ) { -v4l2_err (&ctx->e5010->v4l2_dev,"failed to power up jpeg\n" ); - goto fail; - } - - ret =e5010_init_device (ctx->e5010); -if (ret) { -v4l2_err (&ctx->e5010->v4l2_dev,"failed to Enable e5010 device\n" ); - goto fail; - } -return 0 ; - -fail: -e5010_vb2_buffers_return (q,VB2_BUF_STATE_QUEUED ); - -return ret; }static void e5010_stop_streaming (struct vb2_queue *q)
@@ -1346,8 +1339,6 @@ static void e5010_stop_streaming(struct vb2_queue *q) v4l2_m2m_has_stopped(ctx->fh.m2m_ctx)) {v4l2_event_queue_fh (&ctx->fh, &e5010_eos_event); } - -pm_runtime_put_sync (ctx->e5010->dev); }static void e5010_device_run (void *priv)
@@ -1360,7 +1351,6 @@ static void e5010_device_run(void *priv)unsigned long flags;int num_planes = ctx->out_queue.fmt->num_planes; -spin_lock_irqsave (&e5010->hw_lock, flags); s_vb =v4l2_m2m_next_src_buf (ctx->fh.m2m_ctx);WARN_ON (!s_vb); d_vb =v4l2_m2m_next_dst_buf (ctx->fh.m2m_ctx);
@@ -1368,6 +1358,14 @@ static void e5010_device_run(void *priv)if (!s_vb || !d_vb) goto no_ready_buf_err; + ret =pm_runtime_resume_and_get (e5010->dev); +if (ret <0 ) { +dev_err (e5010->dev,"Device failed to turn on\n" ); + goto no_ready_buf_err; + } + +spin_lock_irqsave (&e5010->hw_lock, flags); + s_vb->sequence = ctx->out_queue.sequence++; d_vb->sequence = ctx->cap_queue.sequence++;
@@ -1492,6 +1490,8 @@ static void e5010_device_run(void *priv) ret =e5010_init_device (e5010);if (ret)v4l2_err (&e5010->v4l2_dev,"failed to initialize device after reset\n" ); +spin_unlock_irqrestore (&e5010->hw_lock, flags); +pm_runtime_put_autosuspend (e5010->dev); no_ready_buf_err:if (s_vb) {v4l2_m2m_src_buf_remove_by_buf (ctx->fh.m2m_ctx, s_vb);
@@ -1505,7 +1505,6 @@ static void e5010_device_run(void *priv)v4l2_m2m_buf_done (d_vb,VB2_BUF_STATE_ERROR ); }v4l2_m2m_job_finish (e5010->m2m_dev, ctx->fh.m2m_ctx); -spin_unlock_irqrestore (&e5010->hw_lock, flags); }#ifdef CONFIG_PM
drivers/media/platform/imagination/e5010-jpeg-enc.h +1 -0
@@ -23,6 +23,7 @@#define DEFAULT_HEIGHT 480#define E5010_MODULE_NAME "e5010"#define JPEG_MAX_BYTES_PER_PIXEL 2 +#define JPEG_AUTOSUSPEND_DELAY_MS 100/* JPEG marker definitions */ #define START_OF_IMAGE 0xFFD8
Signature · 1 line
2.43.0
Thakkar, Devarsh
maintainerHi Nicolas,
Quoted 2 lines · Brandon
Hello all,
[...]
Quoted 10 lines
Brandon Brnich (3):
media: imagination: e5010: Fix clk never enabled withoutCONFIG_PM
media: imagination: e5010: Updatee5010_init_deviceCall Sites to
Ensure Proper Device State
media: imagination: e5010: Enable autosuspend for runtime PMMyeonghun Pak (1):
media: imagination: e5010: release m2m device on probe failure
I've reviewed this series in detail, and it looks good to me as the driver maintainer. All four patches carry my Reviewed-by.
When you have a chance, would you be able to pick up this series ? Patches 1-3 are fixes tagged for stable, and patch 4 adds autosuspend support.
Please let me know if anything is needed from my side.
Regards
Devarsh