From: Yukimasa Sugizaki ysugi@idein.jp
Hi,
The current V3D scheduler has two issues where CSD jobs are resubmitted regardless of the previous timed-out flag, and where the timer is not restarted for timed-out CL/CSD jobs (which we wish to continue running). The second one is due to the DRM scheduler API change and fixed in a similar way to [1]. A kernel command-line option to set the default timeout value is also added.
I tested this patchset with Piglit and our CSD programs in [2]. Because it is hard to get the current upstream kernel to work on BCM2711, I used the kernel from rpi-5.8.y tree [3]. There still are problems where some Piglit tests get longer time to finish running (3610 minutes to 3650 minutes in total), and some ones result in the invalid memory read errors with unknown reasons:
[17086.230959] v3d fec00000.v3d: MMU error from client CLE (4) at 0xac1000, pte invalid [17086.238722] v3d fec00000.v3d: MMU error from client CLE (4) at 0x1b61000, pte invalid [18643.303188] v3d fec00000.v3d: MMU error from client L2T (0) at 0x15bff00, pte invalid [18655.933748] v3d fec00000.v3d: MMU error from client L2T (0) at 0x15bff00, pte invalid
However, most of the CL/CSD programs are now working happily without kernel warnings and errors.
Regards, Sugizaki
[1] https://patchwork.kernel.org/patch/11732895/ [2] https://github.com/Idein/py-videocore6 [3] https://github.com/raspberrypi/linux/tree/rpi-5.8.y
Yukimasa Sugizaki (3): drm/v3d: Don't resubmit guilty CSD jobs drm/v3d: Correctly restart the timer when progress is made drm/v3d: Add job timeout module param
drivers/gpu/drm/v3d/v3d_sched.c | 62 +++++++++++++++++++++++++++++++++-------- 1 file changed, 51 insertions(+), 11 deletions(-)
-- 2.7.4
From: Yukimasa Sugizaki ysugi@idein.jp
The previous code misses a check for the timeout error set by drm_sched_resubmit_jobs(), which results in an infinite GPU reset loop if once a timeout occurs:
[ 178.799106] v3d fec00000.v3d: [drm:v3d_reset [v3d]] *ERROR* Resetting GPU for hang. [ 178.807836] v3d fec00000.v3d: [drm:v3d_reset [v3d]] *ERROR* V3D_ERR_STAT: 0x00001000 [ 179.839132] v3d fec00000.v3d: [drm:v3d_reset [v3d]] *ERROR* Resetting GPU for hang. [ 179.847865] v3d fec00000.v3d: [drm:v3d_reset [v3d]] *ERROR* V3D_ERR_STAT: 0x00001000 [ 180.879146] v3d fec00000.v3d: [drm:v3d_reset [v3d]] *ERROR* Resetting GPU for hang. [ 180.887925] v3d fec00000.v3d: [drm:v3d_reset [v3d]] *ERROR* V3D_ERR_STAT: 0x00001000 [ 181.919188] v3d fec00000.v3d: [drm:v3d_reset [v3d]] *ERROR* Resetting GPU for hang. [ 181.928002] v3d fec00000.v3d: [drm:v3d_reset [v3d]] *ERROR* V3D_ERR_STAT: 0x00001000 ...
This commit adds the check for timeout as in v3d_{bin,render}_job_run():
[ 66.408962] v3d fec00000.v3d: [drm:v3d_reset [v3d]] *ERROR* Resetting GPU for hang. [ 66.417734] v3d fec00000.v3d: [drm:v3d_reset [v3d]] *ERROR* V3D_ERR_STAT: 0x00001000 [ 66.428296] [drm] Skipping CSD job resubmission due to previous error (-125)
, where -125 is -ECANCELED, though users currently have no way other than inspecting the dmesg to check if the timeout has occurred.
Signed-off-by: Yukimasa Sugizaki ysugi@idein.jp --- drivers/gpu/drm/v3d/v3d_sched.c | 11 +++++++++++ 1 file changed, 11 insertions(+)
diff --git a/drivers/gpu/drm/v3d/v3d_sched.c b/drivers/gpu/drm/v3d/v3d_sched.c index 0747614a78f0..001216f22017 100644 --- a/drivers/gpu/drm/v3d/v3d_sched.c +++ b/drivers/gpu/drm/v3d/v3d_sched.c @@ -226,6 +226,17 @@ v3d_csd_job_run(struct drm_sched_job *sched_job) struct dma_fence *fence; int i;
+ /* This error is set to -ECANCELED by drm_sched_resubmit_jobs() if this + * job timed out more than sched_job->sched->hang_limit times. + */ + int error = sched_job->s_fence->finished.error; + + if (unlikely(error < 0)) { + DRM_WARN("Skipping CSD job resubmission due to previous error (%d)\n", + error); + return ERR_PTR(error); + } + v3d->csd_job = job;
v3d_invalidate_caches(v3d); -- 2.7.4
I've tested the patch and confirmed that applies correctly over drm-next.
I've also confirmed that the timeout happens with the described test case by the developer.
https://github.com/raspberrypi/linux/pull/3816#issuecomment-682251862
Considering this is my first review of a patch in v3d kernel side I think this patch is fine.
Reviewed-by: Jose Maria Casanova Crespo jmcasanova@igalia.com
On 3/9/20 18:48, Yukimasa Sugizaki wrote:
From: Yukimasa Sugizaki ysugi@idein.jp
The previous code misses a check for the timeout error set by drm_sched_resubmit_jobs(), which results in an infinite GPU reset loop if once a timeout occurs:
[ 178.799106] v3d fec00000.v3d: [drm:v3d_reset [v3d]] *ERROR* Resetting GPU for hang. [ 178.807836] v3d fec00000.v3d: [drm:v3d_reset [v3d]] *ERROR* V3D_ERR_STAT: 0x00001000 [ 179.839132] v3d fec00000.v3d: [drm:v3d_reset [v3d]] *ERROR* Resetting GPU for hang. [ 179.847865] v3d fec00000.v3d: [drm:v3d_reset [v3d]] *ERROR* V3D_ERR_STAT: 0x00001000 [ 180.879146] v3d fec00000.v3d: [drm:v3d_reset [v3d]] *ERROR* Resetting GPU for hang. [ 180.887925] v3d fec00000.v3d: [drm:v3d_reset [v3d]] *ERROR* V3D_ERR_STAT: 0x00001000 [ 181.919188] v3d fec00000.v3d: [drm:v3d_reset [v3d]] *ERROR* Resetting GPU for hang. [ 181.928002] v3d fec00000.v3d: [drm:v3d_reset [v3d]] *ERROR* V3D_ERR_STAT: 0x00001000 ...
This commit adds the check for timeout as in v3d_{bin,render}_job_run():
[ 66.408962] v3d fec00000.v3d: [drm:v3d_reset [v3d]] *ERROR* Resetting GPU for hang. [ 66.417734] v3d fec00000.v3d: [drm:v3d_reset [v3d]] *ERROR* V3D_ERR_STAT: 0x00001000 [ 66.428296] [drm] Skipping CSD job resubmission due to previous error (-125)
, where -125 is -ECANCELED, though users currently have no way other than inspecting the dmesg to check if the timeout has occurred.
Signed-off-by: Yukimasa Sugizaki ysugi@idein.jp
drivers/gpu/drm/v3d/v3d_sched.c | 11 +++++++++++ 1 file changed, 11 insertions(+)
diff --git a/drivers/gpu/drm/v3d/v3d_sched.c b/drivers/gpu/drm/v3d/v3d_sched.c index 0747614a78f0..001216f22017 100644 --- a/drivers/gpu/drm/v3d/v3d_sched.c +++ b/drivers/gpu/drm/v3d/v3d_sched.c @@ -226,6 +226,17 @@ v3d_csd_job_run(struct drm_sched_job *sched_job) struct dma_fence *fence; int i;
/* This error is set to -ECANCELED by drm_sched_resubmit_jobs() if this
* job timed out more than sched_job->sched->hang_limit times.*/int error = sched_job->s_fence->finished.error;
if (unlikely(error < 0)) {
DRM_WARN("Skipping CSD job resubmission due to previous error (%d)\n",error);return ERR_PTR(error);}
v3d->csd_job = job;
v3d_invalidate_caches(v3d);
-- 2.7.4
dri-devel mailing list dri-devel@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/dri-devel
From: Yukimasa Sugizaki ysugi@idein.jp
The V3D scheduler wants a timed-out job to continue running if it made progress. However, the current DRM scheduler removes the timed-out job from ring_mirror_list and thus the timer is not restarted automatically, resulting in an infinite timeout. We need stop and restart the DRM scheduler to rearm the timer.
Fixes: 135517d3565b ("drm/scheduler: Avoid accessing freed bad job.") Signed-off-by: Yukimasa Sugizaki ysugi@idein.jp --- drivers/gpu/drm/v3d/v3d_sched.c | 27 +++++++++++++++++++++++++++ 1 file changed, 27 insertions(+)
diff --git a/drivers/gpu/drm/v3d/v3d_sched.c b/drivers/gpu/drm/v3d/v3d_sched.c index 001216f22017..feef0c749e68 100644 --- a/drivers/gpu/drm/v3d/v3d_sched.c +++ b/drivers/gpu/drm/v3d/v3d_sched.c @@ -312,9 +312,24 @@ v3d_cl_job_timedout(struct drm_sched_job *sched_job, enum v3d_queue q, u32 ctca = V3D_CORE_READ(0, V3D_CLE_CTNCA(q)); u32 ctra = V3D_CORE_READ(0, V3D_CLE_CTNRA(q));
+ /* If we've made progress, skip reset and let the timer get + * rearmed. + */ if (*timedout_ctca != ctca || *timedout_ctra != ctra) { *timedout_ctca = ctca; *timedout_ctra = ctra; + + /* Because the timed-out job has been removed from + * ring_mirror_list in drm_sched_job_timedout(), we need to + * stop and restart the scheduler to rearm the timer. + * Holding the reset_lock is necessary for concurrent + * v3d_gpu_reset_for_timeout(). + */ + mutex_lock(&v3d->reset_lock); + drm_sched_stop(sched_job->sched, sched_job); + drm_sched_start(sched_job->sched, sched_job); + mutex_unlock(&v3d->reset_lock); + return; }
@@ -359,6 +374,18 @@ v3d_csd_job_timedout(struct drm_sched_job *sched_job) */ if (job->timedout_batches != batches) { job->timedout_batches = batches; + + /* Because the timed-out job has been removed from + * ring_mirror_list in drm_sched_job_timedout(), we need to + * stop and restart the scheduler to rearm the timer. + * Holding the reset_lock is necessary for concurrent + * v3d_gpu_reset_for_timeout(). + */ + mutex_lock(&v3d->reset_lock); + drm_sched_stop(sched_job->sched, sched_job); + drm_sched_start(sched_job->sched, sched_job); + mutex_unlock(&v3d->reset_lock); + return; }
-- 2.7.4
From: Yukimasa Sugizaki ysugi@idein.jp
The default timeout is 500 ms which is too short for some workloads including Piglit. Adding this parameter will help users to run heavier tasks.
Signed-off-by: Yukimasa Sugizaki ysugi@idein.jp --- drivers/gpu/drm/v3d/v3d_sched.c | 24 +++++++++++++----------- 1 file changed, 13 insertions(+), 11 deletions(-)
diff --git a/drivers/gpu/drm/v3d/v3d_sched.c b/drivers/gpu/drm/v3d/v3d_sched.c index feef0c749e68..983efb018560 100644 --- a/drivers/gpu/drm/v3d/v3d_sched.c +++ b/drivers/gpu/drm/v3d/v3d_sched.c @@ -19,11 +19,17 @@ */
#include <linux/kthread.h> +#include <linux/moduleparam.h>
#include "v3d_drv.h" #include "v3d_regs.h" #include "v3d_trace.h"
+static uint timeout = 500; +module_param(timeout, uint, 0444); +MODULE_PARM_DESC(timeout, + "Timeout for a job in ms (0 means infinity and default is 500 ms)"); + static struct v3d_job * to_v3d_job(struct drm_sched_job *sched_job) { @@ -432,13 +438,13 @@ v3d_sched_init(struct v3d_dev *v3d) { int hw_jobs_limit = 1; int job_hang_limit = 0; - int hang_limit_ms = 500; + long timeout_jiffies = timeout ? + msecs_to_jiffies(timeout) : MAX_SCHEDULE_TIMEOUT; int ret;
ret = drm_sched_init(&v3d->queue[V3D_BIN].sched, &v3d_bin_sched_ops, - hw_jobs_limit, job_hang_limit, - msecs_to_jiffies(hang_limit_ms), + hw_jobs_limit, job_hang_limit, timeout_jiffies, "v3d_bin"); if (ret) { dev_err(v3d->drm.dev, "Failed to create bin scheduler: %d.", ret); @@ -447,8 +453,7 @@ v3d_sched_init(struct v3d_dev *v3d)
ret = drm_sched_init(&v3d->queue[V3D_RENDER].sched, &v3d_render_sched_ops, - hw_jobs_limit, job_hang_limit, - msecs_to_jiffies(hang_limit_ms), + hw_jobs_limit, job_hang_limit, timeout_jiffies, "v3d_render"); if (ret) { dev_err(v3d->drm.dev, "Failed to create render scheduler: %d.", @@ -459,8 +464,7 @@ v3d_sched_init(struct v3d_dev *v3d)
ret = drm_sched_init(&v3d->queue[V3D_TFU].sched, &v3d_tfu_sched_ops, - hw_jobs_limit, job_hang_limit, - msecs_to_jiffies(hang_limit_ms), + hw_jobs_limit, job_hang_limit, timeout_jiffies, "v3d_tfu"); if (ret) { dev_err(v3d->drm.dev, "Failed to create TFU scheduler: %d.", @@ -472,8 +476,7 @@ v3d_sched_init(struct v3d_dev *v3d) if (v3d_has_csd(v3d)) { ret = drm_sched_init(&v3d->queue[V3D_CSD].sched, &v3d_csd_sched_ops, - hw_jobs_limit, job_hang_limit, - msecs_to_jiffies(hang_limit_ms), + hw_jobs_limit, job_hang_limit, timeout_jiffies, "v3d_csd"); if (ret) { dev_err(v3d->drm.dev, "Failed to create CSD scheduler: %d.", @@ -484,8 +487,7 @@ v3d_sched_init(struct v3d_dev *v3d)
ret = drm_sched_init(&v3d->queue[V3D_CACHE_CLEAN].sched, &v3d_cache_clean_sched_ops, - hw_jobs_limit, job_hang_limit, - msecs_to_jiffies(hang_limit_ms), + hw_jobs_limit, job_hang_limit, timeout_jiffies, "v3d_cache_clean"); if (ret) { dev_err(v3d->drm.dev, "Failed to create CACHE_CLEAN scheduler: %d.", -- 2.7.4
On 3/9/20 18:48, Yukimasa Sugizaki wrote:
From: Yukimasa Sugizaki ysugi@idein.jp
The default timeout is 500 ms which is too short for some workloads including Piglit. Adding this parameter will help users to run heavier tasks.
Signed-off-by: Yukimasa Sugizaki ysugi@idein.jp
drivers/gpu/drm/v3d/v3d_sched.c | 24 +++++++++++++----------- 1 file changed, 13 insertions(+), 11 deletions(-)
diff --git a/drivers/gpu/drm/v3d/v3d_sched.c b/drivers/gpu/drm/v3d/v3d_sched.c index feef0c749e68..983efb018560 100644 --- a/drivers/gpu/drm/v3d/v3d_sched.c +++ b/drivers/gpu/drm/v3d/v3d_sched.c @@ -19,11 +19,17 @@ */
#include <linux/kthread.h> +#include <linux/moduleparam.h>
#include "v3d_drv.h" #include "v3d_regs.h" #include "v3d_trace.h"
+static uint timeout = 500; +module_param(timeout, uint, 0444); +MODULE_PARM_DESC(timeout,
- "Timeout for a job in ms (0 means infinity and default is 500 ms)");
I think that parameter definition could be included at v3d_drv.c as other drivers do. Then we would have all future parameters together. In that case we would need also to include the extern definition at v3d_drv.h for the driver variable.
The param name could be more descriptive like "sched_timeout_ms" in the lima driver.
I wonder if it would make sense to have different timeouts parameters for different type of jobs we have. At least this series come from the need additional time to complete compute jobs. This is what amdgpu does, but we probably don't need something such complex.
I think it would also make sense to increase our default value for the compute jobs.
What do you think?
In any case I think that having this general scheduling timeout parameter as other drivers do is a good idea.
Regards,
Chema Casanova
static struct v3d_job * to_v3d_job(struct drm_sched_job *sched_job) { @@ -432,13 +438,13 @@ v3d_sched_init(struct v3d_dev *v3d) { int hw_jobs_limit = 1; int job_hang_limit = 0;
- int hang_limit_ms = 500;
long timeout_jiffies = timeout ?
msecs_to_jiffies(timeout) : MAX_SCHEDULE_TIMEOUT;int ret;
ret = drm_sched_init(&v3d->queue[V3D_BIN].sched, &v3d_bin_sched_ops,
hw_jobs_limit, job_hang_limit,msecs_to_jiffies(hang_limit_ms),
hw_jobs_limit, job_hang_limit, timeout_jiffies, "v3d_bin");@@ -447,8 +453,7 @@ v3d_sched_init(struct v3d_dev *v3d)
ret = drm_sched_init(&v3d->queue[V3D_RENDER].sched, &v3d_render_sched_ops,
hw_jobs_limit, job_hang_limit,msecs_to_jiffies(hang_limit_ms),
hw_jobs_limit, job_hang_limit, timeout_jiffies, "v3d_render");@@ -459,8 +464,7 @@ v3d_sched_init(struct v3d_dev *v3d)
ret = drm_sched_init(&v3d->queue[V3D_TFU].sched, &v3d_tfu_sched_ops,
hw_jobs_limit, job_hang_limit,msecs_to_jiffies(hang_limit_ms),
hw_jobs_limit, job_hang_limit, timeout_jiffies, "v3d_tfu");@@ -472,8 +476,7 @@ v3d_sched_init(struct v3d_dev *v3d) if (v3d_has_csd(v3d)) { ret = drm_sched_init(&v3d->queue[V3D_CSD].sched, &v3d_csd_sched_ops,
hw_jobs_limit, job_hang_limit,msecs_to_jiffies(hang_limit_ms),
hw_jobs_limit, job_hang_limit, timeout_jiffies, "v3d_csd");@@ -484,8 +487,7 @@ v3d_sched_init(struct v3d_dev *v3d)
ret = drm_sched_init(&v3d->queue[V3D_CACHE_CLEAN].sched, &v3d_cache_clean_sched_ops,
hw_jobs_limit, job_hang_limit,msecs_to_jiffies(hang_limit_ms),
hw_jobs_limit, job_hang_limit, timeout_jiffies, "v3d_cache_clean");-- 2.7.4
dri-devel mailing list dri-devel@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/dri-devel
On Thu, Feb 4, 2021 at 10:09 AM Chema Casanova jmcasanova@igalia.com wrote:
On 3/9/20 18:48, Yukimasa Sugizaki wrote:
From: Yukimasa Sugizaki ysugi@idein.jp
The default timeout is 500 ms which is too short for some workloads including Piglit. Adding this parameter will help users to run heavier tasks.
Signed-off-by: Yukimasa Sugizaki ysugi@idein.jp
drivers/gpu/drm/v3d/v3d_sched.c | 24 +++++++++++++----------- 1 file changed, 13 insertions(+), 11 deletions(-)
diff --git a/drivers/gpu/drm/v3d/v3d_sched.c b/drivers/gpu/drm/v3d/v3d_sched.c index feef0c749e68..983efb018560 100644 --- a/drivers/gpu/drm/v3d/v3d_sched.c +++ b/drivers/gpu/drm/v3d/v3d_sched.c @@ -19,11 +19,17 @@ */
#include <linux/kthread.h> +#include <linux/moduleparam.h>
#include "v3d_drv.h" #include "v3d_regs.h" #include "v3d_trace.h"
+static uint timeout = 500; +module_param(timeout, uint, 0444); +MODULE_PARM_DESC(timeout,
"Timeout for a job in ms (0 means infinity and default is 500 ms)");I think that parameter definition could be included at v3d_drv.c as other drivers do. Then we would have all future parameters together. In that case we would need also to include the extern definition at v3d_drv.h for the driver variable.
The param name could be more descriptive like "sched_timeout_ms" in the lima driver.
I wonder if it would make sense to have different timeouts parameters for different type of jobs we have. At least this series come from the need additional time to complete compute jobs. This is what amdgpu does, but we probably don't need something such complex.
I think it would also make sense to increase our default value for the compute jobs.
What do you think?
In any case I think that having this general scheduling timeout parameter as other drivers do is a good idea.
Having a param for being able to test if the job completes eventually is a good idea, but if tests are triggering timeouts, then you probably need to investigate what's going on in the driver -- it's not like you want .5second unpreemptible jobs to be easy to produce.
Also, for CS, I wonder if we have another reg besides CSD_CURRENT_CFG4 we could be looking at for "we're making progress on the job". Half a second with no discernible progress sounds like a bug.
On 05/02/2021, Eric Anholt eric@anholt.net wrote:
On Thu, Feb 4, 2021 at 10:09 AM Chema Casanova jmcasanova@igalia.com wrote:
On 3/9/20 18:48, Yukimasa Sugizaki wrote:
From: Yukimasa Sugizaki ysugi@idein.jp
The default timeout is 500 ms which is too short for some workloads including Piglit. Adding this parameter will help users to run heavier tasks.
Signed-off-by: Yukimasa Sugizaki ysugi@idein.jp
drivers/gpu/drm/v3d/v3d_sched.c | 24 +++++++++++++----------- 1 file changed, 13 insertions(+), 11 deletions(-)
diff --git a/drivers/gpu/drm/v3d/v3d_sched.c b/drivers/gpu/drm/v3d/v3d_sched.c index feef0c749e68..983efb018560 100644 --- a/drivers/gpu/drm/v3d/v3d_sched.c +++ b/drivers/gpu/drm/v3d/v3d_sched.c @@ -19,11 +19,17 @@ */
#include <linux/kthread.h> +#include <linux/moduleparam.h>
#include "v3d_drv.h" #include "v3d_regs.h" #include "v3d_trace.h"
+static uint timeout = 500; +module_param(timeout, uint, 0444); +MODULE_PARM_DESC(timeout,
"Timeout for a job in ms (0 means infinity and default is 500ms)");
I think that parameter definition could be included at v3d_drv.c as other drivers do. Then we would have all future parameters together. In that case we would need also to include the extern definition at v3d_drv.h for the driver variable.
The param name could be more descriptive like "sched_timeout_ms" in the lima driver.
I wonder if it would make sense to have different timeouts parameters for different type of jobs we have. At least this series come from the need additional time to complete compute jobs. This is what amdgpu does, but we probably don't need something such complex.
I think it would also make sense to increase our default value for the compute jobs.
What do you think?
In any case I think that having this general scheduling timeout parameter as other drivers do is a good idea.
I agree with your observations. I'm going to add bin_timeout_ms, render_timeout_ms, tfu_timeout_ms, v3d_timeout_ms, and cache_clean_timeout_ms parameters to v3d_drv.c instead of the sole timeout parameter. Though not sophisticated, this should be enough.
Having a param for being able to test if the job completes eventually is a good idea, but if tests are triggering timeouts, then you probably need to investigate what's going on in the driver -- it's not like you want .5second unpreemptible jobs to be easy to produce.
Also, for CS, I wonder if we have another reg besides CSD_CURRENT_CFG4 we could be looking at for "we're making progress on the job". Half a second with no discernible progress sounds like a bug.
If binning/render/TFU/cache_clean jobs keep running over 0.5 seconds, then it should be a bug because they normally complete processing within the period. However, a CSD job can do anything. For example, SaschaWillems's ray tracing example requires about 0.7 seconds to compose a frame with compute shader with Igalia's Vulkan driver. Another example is our matrix computation library for QPU: https://github.com/Idein/qmkl6 It requires about 1.1 seconds to multiply two 2048x2048 matrices. Because in general we do not know what is the expected behavior of a QPU program and what is not, we also cannot detect a progress the program is making. This is why we want to have the timeout parameters to be able to be modified.
Regards, Y. Sugizaki.
On 5/2/21 13:28, Yukimasa Sugizaki wrote:
On 05/02/2021, Eric Anholt eric@anholt.net wrote:
On Thu, Feb 4, 2021 at 10:09 AM Chema Casanova jmcasanova@igalia.com wrote:
On 3/9/20 18:48, Yukimasa Sugizaki wrote:
From: Yukimasa Sugizaki ysugi@idein.jp
The default timeout is 500 ms which is too short for some workloads including Piglit. Adding this parameter will help users to run heavier tasks.
Signed-off-by: Yukimasa Sugizaki ysugi@idein.jp
drivers/gpu/drm/v3d/v3d_sched.c | 24 +++++++++++++----------- 1 file changed, 13 insertions(+), 11 deletions(-)
diff --git a/drivers/gpu/drm/v3d/v3d_sched.c b/drivers/gpu/drm/v3d/v3d_sched.c index feef0c749e68..983efb018560 100644 --- a/drivers/gpu/drm/v3d/v3d_sched.c +++ b/drivers/gpu/drm/v3d/v3d_sched.c @@ -19,11 +19,17 @@ */
#include <linux/kthread.h> +#include <linux/moduleparam.h>
#include "v3d_drv.h" #include "v3d_regs.h" #include "v3d_trace.h"
+static uint timeout = 500; +module_param(timeout, uint, 0444); +MODULE_PARM_DESC(timeout,
"Timeout for a job in ms (0 means infinity and default is 500ms)");
I think that parameter definition could be included at v3d_drv.c as other drivers do. Then we would have all future parameters together. In that case we would need also to include the extern definition at v3d_drv.h for the driver variable.
The param name could be more descriptive like "sched_timeout_ms" in the lima driver.
I wonder if it would make sense to have different timeouts parameters for different type of jobs we have. At least this series come from the need additional time to complete compute jobs. This is what amdgpu does, but we probably don't need something such complex.
I think it would also make sense to increase our default value for the compute jobs.
What do you think?
In any case I think that having this general scheduling timeout parameter as other drivers do is a good idea.
I agree with your observations. I'm going to add bin_timeout_ms, render_timeout_ms, tfu_timeout_ms, v3d_timeout_ms, and cache_clean_timeout_ms parameters to v3d_drv.c instead of the sole timeout parameter. Though not sophisticated, this should be enough.
Having a param for being able to test if the job completes eventually is a good idea, but if tests are triggering timeouts, then you probably need to investigate what's going on in the driver -- it's not like you want .5second unpreemptible jobs to be easy to produce.
Also, for CS, I wonder if we have another reg besides CSD_CURRENT_CFG4 we could be looking at for "we're making progress on the job". Half a second with no discernible progress sounds like a bug.
If binning/render/TFU/cache_clean jobs keep running over 0.5 seconds, then it should be a bug because they normally complete processing within the period. However, a CSD job can do anything. For example, SaschaWillems's ray tracing example requires about 0.7 seconds to compose a frame with compute shader with Igalia's Vulkan driver.
I've been checking the values of the registers with SaschaWillems computeraytracing on job submission, timeout evaluation and job destroy. The driver detects progress. CSD_CURRENT_CFG4 values are decreased as expected based in the number of pending batches. Next debug info starts with 262144 - 1, and we can see it ends with 0xfffffffff. So, I think that using CSD_CURRENT_CFG4 is working as expected. We could see also that CSD_CURRENT_ID0 and CSD_CURRENT_ID1 could also be used to track progress as we see how values are changing for X, Y and Z component of current workgroup ID. But comparisons on CSD_CURRENT_CFG4 are simpler.
10661.766782] [drm] CSD launched: timeout_batches (0x00000000) | cfg4 (0x0003fff9) | id0 (0x00000080) | id1 (0x00000000) | status (0x00000056) [10662.272345] [drm] CSD timeout check: timeout_batches (0x00000000) -> cfg4 (0x0002dfaf) | id0 (0x00050000) | id1 (0x00000024) | status (0x00000056) [10662.792324] [drm] CSD timeout check: timeout_batches (0x0002dfaf) -> cfg4 (0x00018c1f) | id0 (0x003e0000) | id1 (0x0000004e) | status (0x00000056) [10663.302350] [drm] CSD timeout check: timeout_batches (0x00018c1f) -> cfg4 (0x0000724f) | id0 (0x005b0000) | id1 (0x00000071) | status (0x00000056) [10663.487139] [drm] CSD free: timeout_batches (0x0000724f) -> cfg4 (0xffffffff) | id0 (0x00000000) | id1 (0x00010000) | status (0x00000060)
Another example is our matrix computation library for QPU: https://github.com/Idein/qmkl6 It requires about 1.1 seconds to multiply two 2048x2048 matrices. Because in general we do not know what is the expected behavior of a QPU program and what is not, we also cannot detect a progress the program is making. This is why we want to have the timeout parameters to be able to be modified.
In case of qmkl6, all jobs are launched with CSD_CURRENT_CFG4 is initialized as 7 (Number of batches minus 1) so when the job starts, all the batches are on execution. So, the HW cannot report any progress, and CFG4 reports that there are no pending batches queued with ~0.
This is the trace of v3d_submit_csd_ioctl calls showing CFG4.
sscal-qmkl6-26222 [002] .... 11978.399780: v3d_submit_csd_ioctl: dev=1, CFG4 0x00000007, CFG5 0x00560000, CFG6 0x00280000 sscal-qmkl6-26222 [002] .... 11978.518960: v3d_submit_csd_ioctl: dev=1, CFG4 0x00000007, CFG5 0x00560000, CFG6 0x00280000
Here is a example of the process finishing before reaching the timeout:
[10339.374376] [drm] CSD launched: timeout_batches (0x00000000) | cfg4 (0x00000001) | id0 (0x00080800) | id1 (0x00000000) | status (0x00000b34) [10339.470097] [drm] CSD free: timeout_batches (0x00000000) -> cfg4 (0xffffffff) | id0 (0x00080800) | id1 (0x00000000) | status (0x00000b40)
Here is the example of a job that reaches the timeout and RESET:
[10339.980849] [drm] CSD launched: timeout_batches (0x00000000) | cfg4 (0x00000000) | id0 (0x00080800) | id1 (0x00000000) | status (0x00000b44) [10340.497731] [drm] CSD timeout: timeout_batches (0x00000000) -> cfg4 (0xffffffff) | id0 (0x00080800) | id1 (0x00000000) | status (0x00000b44) [10341.007815] [drm] CSD timeout RESET: timeout_batches (0xffffffff) | cfg4 (0xffffffff) | id0 (0x00080800) | id1 (0x00000000) | status (0x00000b44) [10341.008143] v3d fec00000.v3d: [drm:v3d_reset [v3d]] *ERROR* Resetting GPU for hang. [10341.008241] v3d fec00000.v3d: [drm:v3d_reset [v3d]] *ERROR* V3D_ERR_STAT: 0x00001000 [10341.008311] [drm] Skipping CSD job resubmission due to previous error (-125) [10341.008375] [drm] CSD free: timeout_batches (0xffffffff) -> cfg4 (0x00000000) | id0 (0x00000000) | id1 (0x00000000) | status (0x00000000)
In the general case we need at least to reach the timeout twice to reset the GPU.
Jobs start with 0 on timeout_batches, then the value is updated to 0xffffffff and at next timeout evaluation there is no progress as all batches have already been submitted. Then we reset the GPU.
I don't know if the algorithms in qmkl6 or py-videocore6 cannot be launched using more batches. But if this is not the case adding the timeout module parameter seems to be our first solution. It would probably be enough adding a general timeout "timeout_ms" for debugging any kind of job and a specific "csd_timeout_ms" that only applies to CSD jobs and could be used by systems that need to execute this kind of jobs. When the CSD timeout is reached we could probably notify with a DRM_WARN suggesting increasing the csd_timeout.
Kind regards,
Chema Casanova
On 11/02/2021, Chema Casanova jmcasanova@igalia.com wrote:
On 5/2/21 13:28, Yukimasa Sugizaki wrote:
On 05/02/2021, Eric Anholt eric@anholt.net wrote:
On Thu, Feb 4, 2021 at 10:09 AM Chema Casanova jmcasanova@igalia.com wrote:
On 3/9/20 18:48, Yukimasa Sugizaki wrote:
From: Yukimasa Sugizaki ysugi@idein.jp
The default timeout is 500 ms which is too short for some workloads including Piglit. Adding this parameter will help users to run heavier tasks.
Signed-off-by: Yukimasa Sugizaki ysugi@idein.jp
drivers/gpu/drm/v3d/v3d_sched.c | 24 +++++++++++++----------- 1 file changed, 13 insertions(+), 11 deletions(-)
diff --git a/drivers/gpu/drm/v3d/v3d_sched.c b/drivers/gpu/drm/v3d/v3d_sched.c index feef0c749e68..983efb018560 100644 --- a/drivers/gpu/drm/v3d/v3d_sched.c +++ b/drivers/gpu/drm/v3d/v3d_sched.c @@ -19,11 +19,17 @@ */
#include <linux/kthread.h> +#include <linux/moduleparam.h>
#include "v3d_drv.h" #include "v3d_regs.h" #include "v3d_trace.h"
+static uint timeout = 500; +module_param(timeout, uint, 0444); +MODULE_PARM_DESC(timeout,
"Timeout for a job in ms (0 means infinity and default is 500ms)");
I think that parameter definition could be included at v3d_drv.c as other drivers do. Then we would have all future parameters together. In that case we would need also to include the extern definition at v3d_drv.h for the driver variable.
The param name could be more descriptive like "sched_timeout_ms" in the lima driver.
I wonder if it would make sense to have different timeouts parameters for different type of jobs we have. At least this series come from the need additional time to complete compute jobs. This is what amdgpu does, but we probably don't need something such complex.
I think it would also make sense to increase our default value for the compute jobs.
What do you think?
In any case I think that having this general scheduling timeout parameter as other drivers do is a good idea.
I agree with your observations. I'm going to add bin_timeout_ms, render_timeout_ms, tfu_timeout_ms, v3d_timeout_ms, and cache_clean_timeout_ms parameters to v3d_drv.c instead of the sole timeout parameter. Though not sophisticated, this should be enough.
Having a param for being able to test if the job completes eventually is a good idea, but if tests are triggering timeouts, then you probably need to investigate what's going on in the driver -- it's not like you want .5second unpreemptible jobs to be easy to produce.
Also, for CS, I wonder if we have another reg besides CSD_CURRENT_CFG4 we could be looking at for "we're making progress on the job". Half a second with no discernible progress sounds like a bug.
If binning/render/TFU/cache_clean jobs keep running over 0.5 seconds, then it should be a bug because they normally complete processing within the period. However, a CSD job can do anything. For example, SaschaWillems's ray tracing example requires about 0.7 seconds to compose a frame with compute shader with Igalia's Vulkan driver.
I've been checking the values of the registers with SaschaWillems computeraytracing on job submission, timeout evaluation and job destroy. The driver detects progress. CSD_CURRENT_CFG4 values are decreased as expected based in the number of pending batches. Next debug info starts with 262144 - 1, and we can see it ends with 0xfffffffff. So, I think that using CSD_CURRENT_CFG4 is working as expected. We could see also that CSD_CURRENT_ID0 and CSD_CURRENT_ID1 could also be used to track progress as we see how values are changing for X, Y and Z component of current workgroup ID. But comparisons on CSD_CURRENT_CFG4 are simpler.
10661.766782] [drm] CSD launched: timeout_batches (0x00000000) | cfg4 (0x0003fff9) | id0 (0x00000080) | id1 (0x00000000) | status (0x00000056) [10662.272345] [drm] CSD timeout check: timeout_batches (0x00000000) -> cfg4 (0x0002dfaf) | id0 (0x00050000) | id1 (0x00000024) | status (0x00000056) [10662.792324] [drm] CSD timeout check: timeout_batches (0x0002dfaf) -> cfg4 (0x00018c1f) | id0 (0x003e0000) | id1 (0x0000004e) | status (0x00000056) [10663.302350] [drm] CSD timeout check: timeout_batches (0x00018c1f) -> cfg4 (0x0000724f) | id0 (0x005b0000) | id1 (0x00000071) | status (0x00000056) [10663.487139] [drm] CSD free: timeout_batches (0x0000724f) -> cfg4 (0xffffffff) | id0 (0x00000000) | id1 (0x00010000) | status (0x00000060)
Another example is our matrix computation library for QPU: https://github.com/Idein/qmkl6 It requires about 1.1 seconds to multiply two 2048x2048 matrices. Because in general we do not know what is the expected behavior of a QPU program and what is not, we also cannot detect a progress the program is making. This is why we want to have the timeout parameters to be able to be modified.
In case of qmkl6, all jobs are launched with CSD_CURRENT_CFG4 is initialized as 7 (Number of batches minus 1) so when the job starts, all the batches are on execution. So, the HW cannot report any progress, and CFG4 reports that there are no pending batches queued with ~0.
This is the trace of v3d_submit_csd_ioctl calls showing CFG4.
sscal-qmkl6-26222 [002] .... 11978.399780: v3d_submit_csd_ioctl:dev=1, CFG4 0x00000007, CFG5 0x00560000, CFG6 0x00280000 sscal-qmkl6-26222 [002] .... 11978.518960: v3d_submit_csd_ioctl: dev=1, CFG4 0x00000007, CFG5 0x00560000, CFG6 0x00280000
Here is a example of the process finishing before reaching the timeout:
[10339.374376] [drm] CSD launched: timeout_batches (0x00000000) | cfg4 (0x00000001) | id0 (0x00080800) | id1 (0x00000000) | status (0x00000b34) [10339.470097] [drm] CSD free: timeout_batches (0x00000000) -> cfg4 (0xffffffff) | id0 (0x00080800) | id1 (0x00000000) | status (0x00000b40)
Here is the example of a job that reaches the timeout and RESET:
[10339.980849] [drm] CSD launched: timeout_batches (0x00000000) | cfg4 (0x00000000) | id0 (0x00080800) | id1 (0x00000000) | status (0x00000b44) [10340.497731] [drm] CSD timeout: timeout_batches (0x00000000) -> cfg4 (0xffffffff) | id0 (0x00080800) | id1 (0x00000000) | status (0x00000b44) [10341.007815] [drm] CSD timeout RESET: timeout_batches (0xffffffff) | cfg4 (0xffffffff) | id0 (0x00080800) | id1 (0x00000000) | status (0x00000b44) [10341.008143] v3d fec00000.v3d: [drm:v3d_reset [v3d]] *ERROR* Resetting GPU for hang. [10341.008241] v3d fec00000.v3d: [drm:v3d_reset [v3d]] *ERROR* V3D_ERR_STAT: 0x00001000 [10341.008311] [drm] Skipping CSD job resubmission due to previous error (-125) [10341.008375] [drm] CSD free: timeout_batches (0xffffffff) -> cfg4 (0x00000000) | id0 (0x00000000) | id1 (0x00000000) | status (0x00000000)
In the general case we need at least to reach the timeout twice to reset the GPU.
Jobs start with 0 on timeout_batches, then the value is updated to 0xffffffff and at next timeout evaluation there is no progress as all batches have already been submitted. Then we reset the GPU.
I don't know if the algorithms in qmkl6 or py-videocore6 cannot be launched using more batches. But if this is not the case adding the timeout module parameter seems to be our first solution. It would probably be enough adding a general timeout "timeout_ms" for debugging any kind of job and a specific "csd_timeout_ms" that only applies to CSD jobs and could be used by systems that need to execute this kind of jobs. When the CSD timeout is reached we could probably notify with a DRM_WARN suggesting increasing the csd_timeout.
Kind regards,
Chema Casanova
Yes, our CSD codes perform computation with just eight batches (one batch per QPU) for two reasons. One is for efficiency and the other is that we cannot always divide the problem to run within 500 milliseconds. So we too think it makes sense to add the timeout parameters.
Thank you for the suggestion on the parameters. Though I've completed implementing them, we are facing an issue where very occasionally v3d_irq() is called by a finished CSD job during the reset by v3d_gpu_reset_for_timeout(). In this situation, v3d_irq() reads V3D_CTL_INT_STS as 0xdeadbeef (no response), and the remaining operations cause a kernel panic. I suspect this can be fixed by holding v3d->reset_lock in v3d_irq(), but we've not completed testing it yet. I appreciate your patience.
Regards, Y. Sugizaki.
dri-devel@lists.freedesktop.org
-
Chema Casanova -
Eric Anholt -
Yukimasa Sugizaki