drm/v3d: Delay the scheduler timeout if we're still making progress.
Subsystems (4)
PATCH 1/4 drm/v3d: Delay the scheduler timeout if we're still making progress.
GTF-GLES2.gtf.GL.acos.acos_float_vert_xvary submits jobs that take 4 seconds at maximum resolution, but we still want to reset quickly if a job is really hung. Sample the CL's current address and the return address (since we call into tile lists repeatedly) and if either has changed then assume we've made progress.
--- drivers/gpu/drm/v3d/v3d_drv.h | 2 ++ drivers/gpu/drm/v3d/v3d_regs.h | 1 + drivers/gpu/drm/v3d/v3d_sched.c | 18 ++++++++++++++++++ 3 files changed, 21 insertions(+)
drivers/gpu/drm/v3d/v3d_drv.h +2 -0
@@ -189,6 +189,8 @@ struct v3d_job {/* GPU virtual addresses of the start/end of the CL job. */ u32 start, end; + +u32 timedout_ctca, timedout_ctra; };struct v3d_exec_info {
drivers/gpu/drm/v3d/v3d_regs.h +1 -0
@@ -222,6 +222,7 @@#define V3D_CLE_CTNCA (n) (V3D_CLE_CT0CA + 4 * n)#define V3D_CLE_CT0RA 0x00118#define V3D_CLE_CT1RA 0x0011c +#define V3D_CLE_CTNRA (n) (V3D_CLE_CT0RA + 4 * n)#define V3D_CLE_CT0LC 0x00120#define V3D_CLE_CT1LC 0x00124#define V3D_CLE_CT0PC 0x00128
drivers/gpu/drm/v3d/v3d_sched.c +18 -0
@@ -153,7 +153,25 @@ v3d_job_timedout(struct drm_sched_job *sched_job)struct v3d_job *job =to_v3d_job (sched_job);struct v3d_exec_info *exec = job->exec;struct v3d_dev *v3d = exec->v3d; +enum v3d_queue job_q = job == &exec->bin ?V3D_BIN :V3D_RENDER ;enum v3d_queue q; +u32 ctca =V3D_CORE_READ (0 ,V3D_CLE_CTNCA (job_q)); +u32 ctra =V3D_CORE_READ (0 ,V3D_CLE_CTNRA (job_q)); + +/* If the current address or return address have changed, then +* the GPU has probably made progress and we should delay the +* reset. This could fail if the GPU got in an infinite loop +* in the CL, but that is pretty unlikely outside of an i-g-t +* testcase. +*/ +if (job->timedout_ctca != ctca || job->timedout_ctra != ctra) { + job->timedout_ctca = ctca; + job->timedout_ctra = ctra; + +schedule_delayed_work (&job->base.work_tdr, + job->base.sched->timeout); +return ; + }mutex_lock (&v3d->reset_lock);
Signature · 6 lines
2.18.0 _______________________________________________ dri-devel mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/dri-devel
PATCH 2/4 drm/v3d: Remove unnecessary dma_fence_ops.
The dma-fence core as of commit 418cc6ca0607 ("dma-fence: Make ->wait callback optional") provides appropriate defaults for these methods.
--- drivers/gpu/drm/v3d/v3d_fence.c | 12 ------------ 1 file changed, 12 deletions(-)
drivers/gpu/drm/v3d/v3d_fence.c +0 -12
@@ -35,19 +35,7 @@ static const char *v3d_fence_get_timeline_name(struct dma_fence *fence)return "v3d-render" ; } -static bool v3d_fence_enable_signaling (struct dma_fence *fence) -{ -return true; -} -const struct dma_fence_ops v3d_fence_ops = { .get_driver_name = v3d_fence_get_driver_name, .get_timeline_name = v3d_fence_get_timeline_name, - .enable_signaling = v3d_fence_enable_signaling, -/* Each of our fences gets signaled as complete by the IRQ -* handler, so we rely on the core's tracking of signaling. -*/ - .signaled =NULL , - .wait = dma_fence_default_wait, - .release = dma_fence_free, };
Signature · 6 lines
2.18.0 _______________________________________________ dri-devel mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/dri-devel
PATCH 3/4 drm/v3d: Add missing v3d documentation structure.
In reply to Eric AnholtThis was a failure of "git add" on my part -- we already referenced the doc from drivers.rst.
--- Documentation/gpu/v3d.rst | 28 ++++++++++++++++++++++++++++ 1 file changed, 28 insertions(+) create mode 100644 Documentation/gpu/v3d.rst
Documentation/gpu/v3d.rst new file +28 -0
@@ -0,0 +1,28 @@ +===================================== + drm/v3d Broadcom V3D Graphics Driver +===================================== + +.. kernel-doc:: drivers/gpu/drm/v3d/v3d_drv.c + :doc: Broadcom V3D Graphics Driver + +GPU buffer object (BO) management +--------------------------------- + +.. kernel-doc:: drivers/gpu/drm/v3d/v3d_bo.c + :doc: V3D GEM BO management support + +Address space management +=========================================== +.. kernel-doc:: drivers/gpu/drm/v3d/v3d_mmu.c + :doc: Broadcom V3D MMU + +GPU Scheduling +=========================================== +.. kernel-doc:: drivers/gpu/drm/v3d/v3d_sched.c + :doc: Broadcom V3D scheduling + +Interrupts +-------------- + +.. kernel-doc:: drivers/gpu/drm/v3d/v3d_irq.c + :doc: Interrupt management for the V3D engine
Signature · 6 lines
2.18.0 _______________________________________________ dri-devel mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/dri-devel
PATCH 4/4 drm/v3d: Fix a grammar nit in the scheduler docs.
In reply to Eric Anholt--- drivers/gpu/drm/v3d/v3d_sched.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/gpu/drm/v3d/v3d_sched.c +2 -2
@@ -14,8 +14,8 @@* to the HW only when it has completed the last one, instead of * filling up the CT[01]Q FIFOs with jobs. Similarly, we use * v3d_job_dependency() to manage the dependency between bin and -* render, instead of having the clients submit jobs with using the -* HW's semaphores to interlock between them. +* render, instead of having the clients submit jobs using the HW's +* semaphores to interlock between them. */ #include <linux/kthread.h>
Signature · 6 lines
2.18.0 _______________________________________________ dri-devel mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/dri-devel
Quoted 8 lines · Eric Anholt
GTF-GLES2.gtf.GL.acos.acos_float_vert_xvary submits jobs that take 4 seconds at maximum resolution, but we still want to reset quickly if a job is really hung. Sample the CL's current address and the return address (since we call into tile lists repeatedly) and if either has changed then assume we've made progress.
Signed-off-by Eric Anholt <[email protected]>Cc Lucas Stach <[email protected]>
Series is:
Quoted 64 lines
--- drivers/gpu/drm/v3d/v3d_drv.h | 2 ++ drivers/gpu/drm/v3d/v3d_regs.h | 1 + drivers/gpu/drm/v3d/v3d_sched.c | 18 ++++++++++++++++++ 3 files changed, 21 insertions(+)drivers/gpu/drm/v3d/v3d_drv.h +2
-0@@ -189,6 +189,8 @@ struct v3d_job {/* GPU virtual addresses of the start/end of the CL job. */ u32 start, end; + +u32 timedout_ctca, timedout_ctra; };struct v3d_exec_info {drivers/gpu/drm/v3d/v3d_regs.h +1
-0@@ -222,6 +222,7 @@#define V3D_CLE_CTNCA (n) (V3D_CLE_CT0CA + 4 * n)#define V3D_CLE_CT0RA 0x00118#define V3D_CLE_CT1RA 0x0011c +#define V3D_CLE_CTNRA (n) (V3D_CLE_CT0RA + 4 * n)#define V3D_CLE_CT0LC 0x00120#define V3D_CLE_CT1LC 0x00124#define V3D_CLE_CT0PC 0x00128drivers/gpu/drm/v3d/v3d_sched.c +18
-1@@ -153,7 +153,25 @@ v3d_job_timedout(struct drm_sched_job *sched_job)struct v3d_job *job =to_v3d_job (sched_job);struct v3d_exec_info *exec = job->exec;struct v3d_dev *v3d = exec->v3d; +enum v3d_queue job_q = job == &exec->bin ?V3D_BIN :V3D_RENDER ;enum v3d_queue q; +u32 ctca =V3D_CORE_READ (0 ,V3D_CLE_CTNCA (job_q)); +u32 ctra =V3D_CORE_READ (0 ,V3D_CLE_CTNRA (job_q)); + +/* If the current address or return address have changed, then +* the GPU has probably made progress and we should delay the +* reset. This could fail if the GPU got in an infinite loop +* in the CL, but that is pretty unlikely outside of an i-g-t +* testcase. +*/ +if (job->timedout_ctca != ctca || job->timedout_ctra != ctra) { + job->timedout_ctca = ctca; + job->timedout_ctra = ctra; + +schedule_delayed_work (&job->base.work_tdr, + job->base.sched->timeout); +return ; + }mutex_lock (&v3d->reset_lock); --2.18.0
Quoted 4 lines
_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
Quoted 48 lines · Eric Anholt
GTF-GLES2.gtf.GL.acos.acos_float_vert_xvary submits jobs that take 4 seconds at maximum resolution, but we still want to reset quickly if a job is really hung. Sample the CL's current address and the return address (since we call into tile lists repeatedly) and if either has changed then assume we've made progress.
Signed-off-by Eric Anholt <[email protected]>Cc Lucas Stach <[email protected]>--- drivers/gpu/drm/v3d/v3d_drv.h | 2 ++ drivers/gpu/drm/v3d/v3d_regs.h | 1 + drivers/gpu/drm/v3d/v3d_sched.c | 18 ++++++++++++++++++ 3 files changed, 21 insertions(+)drivers/gpu/drm/v3d/v3d_drv.h +2
-0@@ -189,6 +189,8 @@ struct v3d_job {/* GPU virtual addresses of the start/end of the CL job. */ u32 start, end; + +u32 timedout_ctca, timedout_ctra; };struct v3d_exec_info {drivers/gpu/drm/v3d/v3d_regs.h +1
-0@@ -222,6 +222,7 @@#define V3D_CLE_CTNCA (n) (V3D_CLE_CT0CA + 4 * n)#define V3D_CLE_CT0RA 0x00118#define V3D_CLE_CT1RA 0x0011c +#define V3D_CLE_CTNRA (n) (V3D_CLE_CT0RA + 4 * n)#define V3D_CLE_CT0LC 0x00120#define V3D_CLE_CT1LC 0x00124#define V3D_CLE_CT0PC 0x00128drivers/gpu/drm/v3d/v3d_sched.c +1
-0@@ -153,7 +153,25 @@ v3d_job_timedout(struct drm_sched_job *sched_job)struct v3d_job *job =to_v3d_job (sched_job);struct v3d_exec_info *exec = job->exec;struct v3d_dev *v3d = exec->v3d; +enum v3d_queue job_q = job == &exec->bin ?V3D_BIN :V3D_RENDER ;
drivers/gpu/drm/v3d/v3d_sched.c v3d_job_timedout(struct drm_sched_job *sched_job) · line 166
Quoted 10 lines
drivers/gpu/drm/v3d/v3d_sched.c +9
-0enum v3d_queue q; +u32 ctca =V3D_CORE_READ (0 ,V3D_CLE_CTNCA (job_q)); +u32 ctra =V3D_CORE_READ (0 ,V3D_CLE_CTNRA (job_q)); + +/* If the current address or return address have changed, then +* the GPU has probably made progress and we should delay the +* reset. This could fail if the GPU got in an infinite loop +* in the CL, but that is pretty unlikely outside of an i-g-t +* testcase. +*/
In i915 we have a credit system that allows only a limited amount of postponing the timeout. That would block the fairly easy DOS. Otoh I have no idea how robust v3d is against DOS attacks at the ioctl level and whether that's worth it.
From your description I'm assuming that a loop in the shaders won't trick this, so ARB_robustness/WebGL should still be happy.
Anyway, Ack on the entire pile.
-Daniel
Quoted 18 lines
drivers/gpu/drm/v3d/v3d_sched.c +8
-0+if (job->timedout_ctca != ctca || job->timedout_ctra != ctra) { + job->timedout_ctca = ctca; + job->timedout_ctra = ctra; + +schedule_delayed_work (&job->base.work_tdr, + job->base.sched->timeout); +return ; + }mutex_lock (&v3d->reset_lock);Signature · 6 lines
2.18.0 _______________________________________________ dri-devel mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/dri-devel
Signature · 7 lines
Daniel Vetter Software Engineer, Intel Corporation http://blog.ffwll.ch _______________________________________________ dri-devel mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/dri-devel
Am Dienstag, den 03.07.2018, 10:05 -0700 schrieb Eric Anholt:
Quoted 5 lines
GTF-GLES2.gtf.GL.acos.acos_float_vert_xvary submits jobs that take 4 seconds at maximum resolution, but we still want to reset quickly if a job is really hung. Sample the CL's current address and the return address (since we call into tile lists repeatedly) and if either has changed then assume we've made progress.
So this means you are doubling your timeout? AFAICS for the first time you hit the timeout handler the cached ctca and ctra values will probably always differ from the current values. Maybe this warrants a mention in the commit message, as it's changing the behavior of the scheduler timeout.
Also how easy is it for userspace to construct such an infinite loop in the CL? Thinking about a rogue client DoSing the GPU while exploiting this check in the timeout handler to stay under the radar...
Regards,
Lucas
Quoted 56 lines
Signed-off-by Eric Anholt <[email protected]>Quoted 1 line
Cc Lucas Stach <[email protected]>---
drivers/gpu/drm/v3d/v3d_drv.h | 2 ++ drivers/gpu/drm/v3d/v3d_regs.h | 1 +drivers/gpu/drm/v3d/v3d_sched.c| 18 ++++++++++++++++++3 files changed, 21 insertions(+)drivers/gpu/drm/v3d/v3d_drv.h +0
-0@@ -189,6 +189,8 @@ struct v3d_job {Quoted 2 lines
/* GPU virtual addresses of the start/end of the CL job. */ u32 start, end;
+
drivers/gpu/drm/v3d/v3d_drv.h struct v3d_job { · line 193
Quoted 1 line
drivers/gpu/drm/v3d/v3d_drv.h +1
-0+u32 timedout_ctca, timedout_ctra;};
structv3d_exec_info{drivers/gpu/drm/v3d/v3d_regs.h +1
-0@@ -222,6 +222,7 @@ #define V3D_CLE_CTNCA(n) (V3D_CLE_CT0CA + 4 * n) #define V3D_CLE_CT0RA 0x00118 #define V3D_CLE_CT1RA 0x0011c +#define V3D_CLE_CTNRA (n) (V3D_CLE_CT0RA + 4 * n) #define V3D_CLE_CT0LC 0x00120 #define V3D_CLE_CT1LC 0x00124 #define V3D_CLE_CT0PC 0x00128drivers/gpu/drm/v3d/v3d_sched.c +1
-0@@ -153,7 +153,25 @@ v3d_job_timedout(struct drm_sched_job *sched_job) > struct v3d_job *job = to_v3d_job(sched_job); > struct v3d_exec_info *exec = job->exec; > struct v3d_dev *v3d = exec->v3d; > + enum v3d_queue job_q = job == &exec->bin ? V3D_BIN : V3D_RENDER; > enum v3d_queue q; > + u32 ctca = V3D_CORE_READ(0, V3D_CLE_CTNCA(job_q)); > + u32 ctra = V3D_CORE_READ(0, V3D_CLE_CTNRA(job_q)); +drivers/gpu/drm/v3d/v3d_sched.c v3d_job_timedout(struct drm_sched_job *sched_job) · line 169
Quoted 9 lines
drivers/gpu/drm/v3d/v3d_sched.c +9
-0+/* If the current address or return address have changed, then +* the GPU has probably made progress and we should delay the +* reset. This could fail if the GPU got in an infinite loop +* in the CL, but that is pretty unlikely outside of an i-g-t +* testcase. +*/ +if (job->timedout_ctca != ctca || job->timedout_ctra != ctra) { + job->timedout_ctca = ctca; + job->timedout_ctra = ctra;+
Quoted 7 lines
drivers/gpu/drm/v3d/v3d_sched.c v3d_job_timedout(struct drm_sched_job *sched_job) · line 174
Quoted 4 lines
drivers/gpu/drm/v3d/v3d_sched.c +4
-0+schedule_delayed_work (&job->base.work_tdr, + job->base.sched->timeout); +return ; + }Quoted 1 line
mutex_lock(&v3d->reset_lock);
_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
Quoted 7 lines · Lucas Stach
Am Dienstag, den 03.07.2018, 10:05 -0700 schrieb Eric Anholt:
Quoted 5 lines
GTF-GLES2.gtf.GL.acos.acos_float_vert_xvary submits jobs that take 4 seconds at maximum resolution, but we still want to reset quickly if a job is really hung. Sample the CL's current address and the return address (since we call into tile lists repeatedly) and if either has changed then assume we've made progress.
Quoted 5 lines
So this means you are doubling your timeout? AFAICS for the first time you hit the timeout handler the cached ctca and ctra values will probably always differ from the current values. Maybe this warrants a mention in the commit message, as it's changing the behavior of the scheduler timeout.
I supposes that doubles the minimum timeout, but I don't think there's any principled choice behind that value.
Quoted 3 lines
Also how easy is it for userspace to construct such an infinite loop in the CL? Thinking about a rogue client DoSing the GPU while exploiting this check in the timeout handler to stay under the radar...
You'd need to have a big enough CL that you don't sample the same location twice in a row, but otherwise it's trivial and equivalent to a V3D33 igt case I wrote. I don't think we as the kernel particularly cares to protect from that case, though -- it's mainly "does a broken WebGL shader take down your desktop?" which we will still be protecting from. If you wanted to protect from a general userspace attacker, you could have a maximum 1 minute timeout or something, but I'm not sure your life is actually much better when you let an arbitrary number of clients submit many jobs to round-robin through each of which has a long timeout like that.
_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
Am Dienstag, den 03.07.2018, 10:05 -0700 schrieb Eric Anholt:
Quoted 8 lines
GTF-GLES2.gtf.GL.acos.acos_float_vert_xvary submits jobs that take 4 seconds at maximum resolution, but we still want to reset quickly if a job is really hung. Sample the CL's current address and the return address (since we call into tile lists repeatedly) and if either has changed then assume we've made progress.
Quoted 1 line
Signed-off-by Eric Anholt <[email protected]>Cc Lucas Stach <[email protected]>
Quoted 54 lines
---
drivers/gpu/drm/v3d/v3d_drv.h | 2 ++ drivers/gpu/drm/v3d/v3d_regs.h | 1 +drivers/gpu/drm/v3d/v3d_sched.c| 18 ++++++++++++++++++3 files changed, 21 insertions(+)drivers/gpu/drm/v3d/v3d_drv.h +0
-0@@ -189,6 +189,8 @@ struct v3d_job {Quoted 2 lines
/* GPU virtual addresses of the start/end of the CL job. */ u32 start, end;
+
drivers/gpu/drm/v3d/v3d_drv.h struct v3d_job { · line 193
Quoted 1 line
drivers/gpu/drm/v3d/v3d_drv.h +1
-0+u32 timedout_ctca, timedout_ctra;};
structv3d_exec_info{drivers/gpu/drm/v3d/v3d_regs.h +1
-0@@ -222,6 +222,7 @@ #define V3D_CLE_CTNCA(n) (V3D_CLE_CT0CA + 4 * n) #define V3D_CLE_CT0RA 0x00118 #define V3D_CLE_CT1RA 0x0011c +#define V3D_CLE_CTNRA (n) (V3D_CLE_CT0RA + 4 * n) #define V3D_CLE_CT0LC 0x00120 #define V3D_CLE_CT1LC 0x00124 #define V3D_CLE_CT0PC 0x00128drivers/gpu/drm/v3d/v3d_sched.c +1
-0@@ -153,7 +153,25 @@ v3d_job_timedout(struct drm_sched_job *sched_job) > struct v3d_job *job = to_v3d_job(sched_job); > struct v3d_exec_info *exec = job->exec; > struct v3d_dev *v3d = exec->v3d; > + enum v3d_queue job_q = job == &exec->bin ? V3D_BIN : V3D_RENDER; > enum v3d_queue q; > + u32 ctca = V3D_CORE_READ(0, V3D_CLE_CTNCA(job_q)); > + u32 ctra = V3D_CORE_READ(0, V3D_CLE_CTNRA(job_q)); +drivers/gpu/drm/v3d/v3d_sched.c v3d_job_timedout(struct drm_sched_job *sched_job) · line 169
Quoted 9 lines
drivers/gpu/drm/v3d/v3d_sched.c +9
-0+/* If the current address or return address have changed, then +* the GPU has probably made progress and we should delay the +* reset. This could fail if the GPU got in an infinite loop +* in the CL, but that is pretty unlikely outside of an i-g-t +* testcase. +*/ +if (job->timedout_ctca != ctca || job->timedout_ctra != ctra) { + job->timedout_ctca = ctca; + job->timedout_ctra = ctra;+
Quoted 7 lines
drivers/gpu/drm/v3d/v3d_sched.c v3d_job_timedout(struct drm_sched_job *sched_job) · line 174
Quoted 4 lines
drivers/gpu/drm/v3d/v3d_sched.c +4
-0+schedule_delayed_work (&job->base.work_tdr, + job->base.sched->timeout); +return ; + }Quoted 1 line
mutex_lock(&v3d->reset_lock);
_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel