drm/msm: GPU fixes and features for 4.14
Subsystems (5)
cover drm/msm: GPU fixes and features for 4.14
Here is my stack of Adreno GPU goodness for 4.14. Starting off with a few fixes that could be appropriate for 4.13 if you deem them so and then moving into some new features:
- a5xx hardware fault detection - can detect a legitimate fault within microseconds and fires an interrupt so you should be able to recover and move on much faster than relying on the timer.
- submitqueues - submitqueues are the kernel side equivalent of a context. The user can set submitqueue specific flags and features (such as ringbuffer priority) and in the following patch we move to per-submitqueue fences.
- multiple ringbuffer preemption - provide for multiple ringbuffers, allow the user to select a ringbuffer to submit on and preempt between the different priorities. You've seen some of this code before but some of our newer features have cleaned things up significantly.
Jordan Crouse (17):
drm/msm: Remove some potentially blocked register ranges drm/msm: Allow hardware clock gating to be toggled drm/msm: Turn off hardware clock gating before reading A5XX registers drm/msm: Remove uneeded platform dev members
drm/msm: args->fence should be args->flags
drm/msm: Remove __user from __u64 data types
drm/msm: Add A5XX hardware fault detection
drm/msm: Add per-instance submit queues
drm/msm: Implement per-submitqueue fences
drm/msm: Attach the GPU MMU when it is created
drm/msm: Add a helper function for in-kernel buffer allocations drm/msm: Move memptrs to msm_gpu
drm/msm: Support multiple ringbuffers
drm/msm: Add a parameter query for the number of ringbuffers drm/msm: Shadow current pointer in the ring until command is complete drm/msm: Make the value of RB_CNTL (almost) generic drm/msm: Implement preemption for A5XX targets
drivers/gpu/drm/msm/Makefile | 4 +- drivers/gpu/drm/msm/adreno/a3xx_gpu.c | 12 +- drivers/gpu/drm/msm/adreno/a3xx_gpu.h | 1 - drivers/gpu/drm/msm/adreno/a4xx_gpu.c | 12 +- drivers/gpu/drm/msm/adreno/a4xx_gpu.h | 1 - drivers/gpu/drm/msm/adreno/a5xx_gpu.c | 366 +++++++++++++++++++++--------- drivers/gpu/drm/msm/adreno/a5xx_gpu.h | 105 ++++++++- drivers/gpu/drm/msm/adreno/a5xx_power.c | 20 +- drivers/gpu/drm/msm/adreno/a5xx_preempt.c | 294 ++++++++++++++++++++++++ drivers/gpu/drm/msm/adreno/adreno_gpu.c | 224 ++++++++---------- drivers/gpu/drm/msm/adreno/adreno_gpu.h | 43 ++-- drivers/gpu/drm/msm/msm_drv.c | 62 ++++- drivers/gpu/drm/msm/msm_drv.h | 28 ++- drivers/gpu/drm/msm/msm_fbdev.c | 35 ++- drivers/gpu/drm/msm/msm_fence.c | 2 +- drivers/gpu/drm/msm/msm_fence.h | 2 +- drivers/gpu/drm/msm/msm_gem.c | 46 ++++ drivers/gpu/drm/msm/msm_gem.h | 5 +- drivers/gpu/drm/msm/msm_gem_submit.c | 27 ++- drivers/gpu/drm/msm/msm_gpu.c | 259 +++++++++++++++------ drivers/gpu/drm/msm/msm_gpu.h | 54 ++++- drivers/gpu/drm/msm/msm_ringbuffer.c | 35 +-- drivers/gpu/drm/msm/msm_ringbuffer.h | 32 ++- drivers/gpu/drm/msm/msm_submitqueue.c | 160 +++++++++++++ include/uapi/drm/msm_drm.h | 30 ++- 25 files changed, 1422 insertions(+), 437 deletions(-) create mode 100644 drivers/gpu/drm/msm/adreno/a5xx_preempt.c create mode 100644 drivers/gpu/drm/msm/msm_submitqueue.c
Signature · 1 line
1.9.1
A message that is not in the archive (sent off-list or not yet received): 1501173766-24856-1-git-send-email-jcrouse-sgV2jX0FEOL9JmXXK+q4OQ@public.gmane.org
PATCH 1/17 drm/msm: Remove some potentially blocked register ranges
The 0xf400 and 0xf800 ranges are in the RBBM_SECVID block which may be protected from CPU access. Skip dumping them since they are minimally useful for debugging and they aren't worth a system hang.
--- drivers/gpu/drm/msm/adreno/a5xx_gpu.c | 49 +++++++++++++++++------------------ 1 file changed, 24 insertions(+), 25 deletions(-)
drivers/gpu/drm/msm/adreno/a5xx_gpu.c +24 -25
@@ -920,31 +920,30 @@ static irqreturn_t a5xx_irq(struct msm_gpu *gpu)0x0000 ,0x0002 ,0x0004 ,0x0020 ,0x0022 ,0x0026 ,0x0029 ,0x002B ,0x002E ,0x0035 ,0x0038 ,0x0042 ,0x0044 ,0x0044 ,0x0047 ,0x0095 ,0x0097 ,0x00BB ,0x03A0 ,0x0464 ,0x0469 ,0x046F ,0x04D2 ,0x04D3 , -0x04E0 ,0x0533 ,0x0540 ,0x0555 ,0xF400 ,0xF400 ,0xF800 ,0xF807 , -0x0800 ,0x081A ,0x081F ,0x0841 ,0x0860 ,0x0860 ,0x0880 ,0x08A0 , -0x0B00 ,0x0B12 ,0x0B15 ,0x0B28 ,0x0B78 ,0x0B7F ,0x0BB0 ,0x0BBD , -0x0BC0 ,0x0BC6 ,0x0BD0 ,0x0C53 ,0x0C60 ,0x0C61 ,0x0C80 ,0x0C82 , -0x0C84 ,0x0C85 ,0x0C90 ,0x0C98 ,0x0CA0 ,0x0CA0 ,0x0CB0 ,0x0CB2 , -0x2180 ,0x2185 ,0x2580 ,0x2585 ,0x0CC1 ,0x0CC1 ,0x0CC4 ,0x0CC7 , -0x0CCC ,0x0CCC ,0x0CD0 ,0x0CD8 ,0x0CE0 ,0x0CE5 ,0x0CE8 ,0x0CE8 , -0x0CEC ,0x0CF1 ,0x0CFB ,0x0D0E ,0x2100 ,0x211E ,0x2140 ,0x2145 , -0x2500 ,0x251E ,0x2540 ,0x2545 ,0x0D10 ,0x0D17 ,0x0D20 ,0x0D23 , -0x0D30 ,0x0D30 ,0x20C0 ,0x20C0 ,0x24C0 ,0x24C0 ,0x0E40 ,0x0E43 , -0x0E4A ,0x0E4A ,0x0E50 ,0x0E57 ,0x0E60 ,0x0E7C ,0x0E80 ,0x0E8E , -0x0E90 ,0x0E96 ,0x0EA0 ,0x0EA8 ,0x0EB0 ,0x0EB2 ,0xE140 ,0xE147 , -0xE150 ,0xE187 ,0xE1A0 ,0xE1A9 ,0xE1B0 ,0xE1B6 ,0xE1C0 ,0xE1C7 , -0xE1D0 ,0xE1D1 ,0xE200 ,0xE201 ,0xE210 ,0xE21C ,0xE240 ,0xE268 , -0xE000 ,0xE006 ,0xE010 ,0xE09A ,0xE0A0 ,0xE0A4 ,0xE0AA ,0xE0EB , -0xE100 ,0xE105 ,0xE380 ,0xE38F ,0xE3B0 ,0xE3B0 ,0xE400 ,0xE405 , -0xE408 ,0xE4E9 ,0xE4F0 ,0xE4F0 ,0xE280 ,0xE280 ,0xE282 ,0xE2A3 , -0xE2A5 ,0xE2C2 ,0xE940 ,0xE947 ,0xE950 ,0xE987 ,0xE9A0 ,0xE9A9 , -0xE9B0 ,0xE9B6 ,0xE9C0 ,0xE9C7 ,0xE9D0 ,0xE9D1 ,0xEA00 ,0xEA01 , -0xEA10 ,0xEA1C ,0xEA40 ,0xEA68 ,0xE800 ,0xE806 ,0xE810 ,0xE89A , -0xE8A0 ,0xE8A4 ,0xE8AA ,0xE8EB ,0xE900 ,0xE905 ,0xEB80 ,0xEB8F , -0xEBB0 ,0xEBB0 ,0xEC00 ,0xEC05 ,0xEC08 ,0xECE9 ,0xECF0 ,0xECF0 , -0xEA80 ,0xEA80 ,0xEA82 ,0xEAA3 ,0xEAA5 ,0xEAC2 ,0xA800 ,0xA8FF , -0xAC60 ,0xAC60 ,0xB000 ,0xB97F ,0xB9A0 ,0xB9BF , - ~0 +0x04E0 ,0x0533 ,0x0540 ,0x0555 ,0x0800 ,0x081A ,0x081F ,0x0841 , +0x0860 ,0x0860 ,0x0880 ,0x08A0 ,0x0B00 ,0x0B12 ,0x0B15 ,0x0B28 , +0x0B78 ,0x0B7F ,0x0BB0 ,0x0BBD ,0x0BC0 ,0x0BC6 ,0x0BD0 ,0x0C53 , +0x0C60 ,0x0C61 ,0x0C80 ,0x0C82 ,0x0C84 ,0x0C85 ,0x0C90 ,0x0C98 , +0x0CA0 ,0x0CA0 ,0x0CB0 ,0x0CB2 ,0x2180 ,0x2185 ,0x2580 ,0x2585 , +0x0CC1 ,0x0CC1 ,0x0CC4 ,0x0CC7 ,0x0CCC ,0x0CCC ,0x0CD0 ,0x0CD8 , +0x0CE0 ,0x0CE5 ,0x0CE8 ,0x0CE8 ,0x0CEC ,0x0CF1 ,0x0CFB ,0x0D0E , +0x2100 ,0x211E ,0x2140 ,0x2145 ,0x2500 ,0x251E ,0x2540 ,0x2545 , +0x0D10 ,0x0D17 ,0x0D20 ,0x0D23 ,0x0D30 ,0x0D30 ,0x20C0 ,0x20C0 , +0x24C0 ,0x24C0 ,0x0E40 ,0x0E43 ,0x0E4A ,0x0E4A ,0x0E50 ,0x0E57 , +0x0E60 ,0x0E7C ,0x0E80 ,0x0E8E ,0x0E90 ,0x0E96 ,0x0EA0 ,0x0EA8 , +0x0EB0 ,0x0EB2 ,0xE140 ,0xE147 ,0xE150 ,0xE187 ,0xE1A0 ,0xE1A9 , +0xE1B0 ,0xE1B6 ,0xE1C0 ,0xE1C7 ,0xE1D0 ,0xE1D1 ,0xE200 ,0xE201 , +0xE210 ,0xE21C ,0xE240 ,0xE268 ,0xE000 ,0xE006 ,0xE010 ,0xE09A , +0xE0A0 ,0xE0A4 ,0xE0AA ,0xE0EB ,0xE100 ,0xE105 ,0xE380 ,0xE38F , +0xE3B0 ,0xE3B0 ,0xE400 ,0xE405 ,0xE408 ,0xE4E9 ,0xE4F0 ,0xE4F0 , +0xE280 ,0xE280 ,0xE282 ,0xE2A3 ,0xE2A5 ,0xE2C2 ,0xE940 ,0xE947 , +0xE950 ,0xE987 ,0xE9A0 ,0xE9A9 ,0xE9B0 ,0xE9B6 ,0xE9C0 ,0xE9C7 , +0xE9D0 ,0xE9D1 ,0xEA00 ,0xEA01 ,0xEA10 ,0xEA1C ,0xEA40 ,0xEA68 , +0xE800 ,0xE806 ,0xE810 ,0xE89A ,0xE8A0 ,0xE8A4 ,0xE8AA ,0xE8EB , +0xE900 ,0xE905 ,0xEB80 ,0xEB8F ,0xEBB0 ,0xEBB0 ,0xEC00 ,0xEC05 , +0xEC08 ,0xECE9 ,0xECF0 ,0xECF0 ,0xEA80 ,0xEA80 ,0xEA82 ,0xEAA3 , +0xEAA5 ,0xEAC2 ,0xA800 ,0xA8FF ,0xAC60 ,0xAC60 ,0xB000 ,0xB97F , +0xB9A0 ,0xB9BF , ~0 };static void a5xx_dump (struct msm_gpu *gpu)
Signature · 6 lines
1.9.1 _______________________________________________ Freedreno mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/freedreno
PATCH 3/17 drm/msm: Turn off hardware clock gating before reading A5XX registers
In reply to a message not in the archiveOn A5XX GPU hardware clock gating needs to be turned off before reading certain GPU registers via AHB. Turn off HWCG before calling adreno_show() to safely dump all the registers without a system hang.
--- drivers/gpu/drm/msm/adreno/a5xx_gpu.c | 7 +++++++ 1 file changed, 7 insertions(+)
drivers/gpu/drm/msm/adreno/a5xx_gpu.c +7 -0
@@ -995,7 +995,14 @@ static void a5xx_show(struct msm_gpu *gpu, struct seq_file *m) {seq_printf (m,"status: %08x\n" ,gpu_read (gpu,REG_A5XX_RBBM_STATUS )); + +/* +* Temporarily disable hardware clock gating before going into +* adreno_show to avoid issues while reading the registers +*/ +a5xx_set_hwcg (gpu, false);adreno_show (gpu, m); +a5xx_set_hwcg (gpu, true); }#endif
Signature · 6 lines
1.9.1 _______________________________________________ Freedreno mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/freedreno
PATCH 6/17 drm/msm: Remove __user from __u64 data types
In reply to a message not in the archive__user should be used to identify user pointers and not __u64 variables containing pointers.
--- include/uapi/drm/msm_drm.h | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-)
include/uapi/drm/msm_drm.h +3 -3
@@ -171,7 +171,7 @@ struct drm_msm_gem_submit_cmd {__u32 size;/* in, cmdstream size */ __u32 pad;__u32 nr_relocs;/* in, number of submit_reloc's */ -__u64 __user relocs;/* in, ptr to array of submit_reloc's */ +__u64 relocs;/* in, ptr to array of submit_reloc's */ };/* Each buffer referenced elsewhere in the cmdstream submit (ie. the
@@ -215,8 +215,8 @@ struct drm_msm_gem_submit {__u32 fence;/* out */ __u32 nr_bos;/* in, number of submit_bo's */ __u32 nr_cmds;/* in, number of submit_cmd's */ -__u64 __user bos;/* in, ptr to array of submit_bo's */ -__u64 __user cmds;/* in, ptr to array of submit_cmd's */ +__u64 bos;/* in, ptr to array of submit_bo's */ +__u64 cmds;/* in, ptr to array of submit_cmd's */ __s32 fence_fd;/* in/out fence fd (see MSM_SUBMIT_FENCE_FD_IN/OUT) */ };
Signature · 6 lines
1.9.1 _______________________________________________ Freedreno mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/freedreno
PATCH 11/17 drm/msm: Add a helper function for in-kernel buffer allocations
In reply to a message not in the archiveNearly all of the buffer allocations for kernel allocate an buffer object, virtual address and GPU iova at the same time. Make a helper function to handle the details.
--- drivers/gpu/drm/msm/adreno/a5xx_gpu.c | 22 +++------------- drivers/gpu/drm/msm/adreno/a5xx_power.c | 14 +++------- drivers/gpu/drm/msm/adreno/adreno_gpu.c | 26 +++++-------------- drivers/gpu/drm/msm/msm_drv.h | 6 +++++ drivers/gpu/drm/msm/msm_fbdev.c | 35 ++++++++++--------------- drivers/gpu/drm/msm/msm_gem.c | 46 +++++++++++++++++++++++++++++++++ drivers/gpu/drm/msm/msm_ringbuffer.c | 12 ++++----- 7 files changed, 86 insertions(+), 75 deletions(-)
drivers/gpu/drm/msm/adreno/a5xx_gpu.c +4 -18
@@ -269,28 +269,14 @@ static int a5xx_me_init(struct msm_gpu *gpu)static struct drm_gem_object *a5xx_ucode_load_bo (struct msm_gpu *gpu,const struct firmware *fw,u64 *iova) { -struct drm_device *drm = gpu->dev;struct drm_gem_object *bo;void *ptr; - bo =msm_gem_new_locked (drm, fw->size -4 ,MSM_BO_UNCACHED ); -if (IS_ERR (bo)) -return bo; + ptr =msm_gem_kernel_new_locked (gpu->dev, fw->size -4 , +MSM_BO_UNCACHED |MSM_BO_GPU_READONLY , gpu->aspace, &bo, iova); - ptr =msm_gem_get_vaddr (bo); -if (!ptr) { -drm_gem_object_unreference (bo); -return ERR_PTR (-ENOMEM ); - } - -if (iova) { -int ret =msm_gem_get_iova (bo, gpu->aspace, iova); - -if (ret) { -drm_gem_object_unreference (bo); -return ERR_PTR (ret); - } - } +if (IS_ERR (ptr)) +return ERR_CAST (ptr);memcpy (ptr, &fw->data[4 ], fw->size -4 );
drivers/gpu/drm/msm/adreno/a5xx_power.c +4 -10
@@ -294,16 +294,10 @@ void a5xx_gpmu_ucode_init(struct msm_gpu *gpu)*/ bosize = (cmds_size + (cmds_size /TYPE4_MAX_PAYLOAD ) +1 ) <<2 ; - a5xx_gpu->gpmu_bo =msm_gem_new_locked (drm, bosize,MSM_BO_UNCACHED ); -if (IS_ERR (a5xx_gpu->gpmu_bo)) - goto err; - -if (msm_gem_get_iova (a5xx_gpu->gpmu_bo, gpu->aspace, - &a5xx_gpu->gpmu_iova)) - goto err; - - ptr =msm_gem_get_vaddr (a5xx_gpu->gpmu_bo); -if (!ptr) + ptr =msm_gem_kernel_new_locked (drm, bosize, +MSM_BO_UNCACHED |MSM_BO_GPU_READONLY , gpu->aspace, + &a5xx_gpu->gpmu_bo, &a5xx_gpu->gpmu_iova); +if (IS_ERR (ptr)) goto err;while (cmds_size >0 ) {
drivers/gpu/drm/msm/adreno/adreno_gpu.c +7 -19
@@ -384,29 +384,17 @@ int adreno_gpu_init(struct drm_device *drm, struct platform_device *pdev,return ret; } - adreno_gpu->memptrs_bo =msm_gem_new (drm,sizeof (*adreno_gpu->memptrs), -MSM_BO_UNCACHED ); -if (IS_ERR (adreno_gpu->memptrs_bo)) { - ret =PTR_ERR (adreno_gpu->memptrs_bo); - adreno_gpu->memptrs_bo =NULL ; -dev_err (drm->dev,"could not allocate memptrs: %d\n" , ret); -return ret; - } + adreno_gpu->memptrs =msm_gem_kernel_new (drm, +sizeof (*adreno_gpu->memptrs),MSM_BO_UNCACHED , gpu->aspace, + &adreno_gpu->memptrs_bo, &adreno_gpu->memptrs_iova); - adreno_gpu->memptrs =msm_gem_get_vaddr (adreno_gpu->memptrs_bo);if (IS_ERR (adreno_gpu->memptrs)) { -dev_err (drm->dev,"could not vmap memptrs\n" ); -return -ENOMEM ; - } - - ret =msm_gem_get_iova (adreno_gpu->memptrs_bo, gpu->aspace, - &adreno_gpu->memptrs_iova); -if (ret) { -dev_err (drm->dev,"could not map memptrs: %d\n" , ret); -return ret; + ret =PTR_ERR (adreno_gpu->memptrs); + adreno_gpu->memptrs =NULL ; +dev_err (drm->dev,"could not allocate memptrs: %d\n" , ret); } -return 0 ; +return ret; }void adreno_gpu_cleanup (struct adreno_gpu *adreno_gpu)
drivers/gpu/drm/msm/msm_drv.h +6 -0
@@ -235,6 +235,12 @@ struct drm_gem_object *msm_gem_new(struct drm_device *dev, uint32_t size, uint32_t flags);struct drm_gem_object *msm_gem_new_locked (struct drm_device *dev,uint32_t size,uint32_t flags); +void *msm_gem_kernel_new (struct drm_device *dev,uint32_t size, +uint32_t flags,struct msm_gem_address_space *aspace, +struct drm_gem_object **bo,uint64_t *iova); +void *msm_gem_kernel_new_locked (struct drm_device *dev,uint32_t size, +uint32_t flags,struct msm_gem_address_space *aspace, +struct drm_gem_object **bo,uint64_t *iova);struct drm_gem_object *msm_gem_import (struct drm_device *dev,struct dma_buf *dmabuf,struct sg_table *sgt);
drivers/gpu/drm/msm/msm_fbdev.c +14 -21
@@ -78,6 +78,7 @@ static int msm_fbdev_create(struct drm_fb_helper *helper,struct drm_framebuffer *fb =NULL ;struct fb_info *fbi =NULL ;struct drm_mode_fb_cmd2 mode_cmd = {0 }; +void *ptr;uint64_t paddr;int ret, size;
@@ -97,11 +98,18 @@ static int msm_fbdev_create(struct drm_fb_helper *helper,/* allocate backing bo */ size = mode_cmd.pitches[0 ] * mode_cmd.height;DBG ("allocating %d bytes for fb %d" , size, dev->primary->index); - fbdev->bo =msm_gem_new (dev, size,MSM_BO_SCANOUT | -MSM_BO_WC |MSM_BO_STOLEN ); -if (IS_ERR (fbdev->bo)) { - ret =PTR_ERR (fbdev->bo); - fbdev->bo =NULL ; + +/* +* NOTE: if we can be guaranteed to be able to map buffer +* in panic (ie. lock-safe, etc) we could avoid pinning the +* buffer now: +*/ + ptr =msm_gem_kernel_new (dev, size, +MSM_BO_SCANOUT |MSM_BO_WC |MSM_BO_STOLEN , + priv->kms->aspace, &fbdev->bo, &paddr); + +if (IS_ERR (ptr)) { + ret =PTR_ERR (ptr);dev_err (dev->dev,"failed to allocate buffer object: %d\n" , ret); goto fail; }
@@ -119,17 +127,6 @@ static int msm_fbdev_create(struct drm_fb_helper *helper,mutex_lock (&dev->struct_mutex); -/* -* NOTE: if we can be guaranteed to be able to map buffer -* in panic (ie. lock-safe, etc) we could avoid pinning the -* buffer now: -*/ - ret =msm_gem_get_iova (fbdev->bo, priv->kms->aspace, &paddr); -if (ret) { -dev_err (dev->dev,"failed to get buffer obj iova: %d\n" , ret); - goto fail_unlock; - } - fbi =drm_fb_helper_alloc_fbi (helper);if (IS_ERR (fbi)) {dev_err (dev->dev,"failed to allocate fb info\n" );
@@ -153,11 +150,7 @@ static int msm_fbdev_create(struct drm_fb_helper *helper, dev->mode_config.fb_base = paddr; - fbi->screen_base =msm_gem_get_vaddr (fbdev->bo); -if (IS_ERR (fbi->screen_base)) { - ret =PTR_ERR (fbi->screen_base); - goto fail_unlock; - } + fbi->screen_base = ptr; fbi->screen_size = fbdev->bo->size; fbi->fix.smem_start = paddr; fbi->fix.smem_len = fbdev->bo->size;
drivers/gpu/drm/msm/msm_gem.c +46 -0
49 lines not on this page · raw message
drivers/gpu/drm/msm/msm_ringbuffer.c +5 -7
@@ -33,16 +33,14 @@ struct msm_ringbuffer *msm_ringbuffer_new(struct msm_gpu *gpu, int size) } ring->gpu = gpu; - ring->bo =msm_gem_new (gpu->dev, size,MSM_BO_WC ); -if (IS_ERR (ring->bo)) { - ret =PTR_ERR (ring->bo); - ring->bo =NULL ; - goto fail; - } - ring->start =msm_gem_get_vaddr (ring->bo); +/* Pass NULL for the iova pointer - we will map it later */ + ring->start =msm_gem_kernel_new (gpu->dev, size,MSM_BO_WC , + gpu->aspace, &ring->bo,NULL ); +if (IS_ERR (ring->start)) { ret =PTR_ERR (ring->start); + ring->start =0 ; goto fail; } ring->end = ring->start + (size /4 );
Signature · 6 lines
1.9.1 _______________________________________________ Freedreno mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/freedreno
PATCH 13/17 drm/msm: Support multiple ringbuffers
In reply to a message not in the archiveAdd the infrastructure to support the idea of multiple ringbuffers. Assign each ringbuffer an id and use that as an index for the various ring specific operations.
The biggest delta is to support legacy fences. Each fence gets its own sequence number but the legacy functions expect to use a unique integer. To handle this we return a unique identifier for each submission but map it to a specific ring/sequence under the covers. Newer users use a dma_fence pointer anyway so they don't care about the actual sequence ID or ring.
The actual mechanics for multiple ringbuffers are very target specific so this code just allows for the possibility but still only defines one ringbuffer for each target family.
--- drivers/gpu/drm/msm/adreno/a3xx_gpu.c | 9 +- drivers/gpu/drm/msm/adreno/a4xx_gpu.c | 9 +- drivers/gpu/drm/msm/adreno/a5xx_gpu.c | 45 +++++----- drivers/gpu/drm/msm/adreno/a5xx_gpu.h | 2 +- drivers/gpu/drm/msm/adreno/a5xx_power.c | 6 +- drivers/gpu/drm/msm/adreno/adreno_gpu.c | 133 +++++++++++++++++------------ drivers/gpu/drm/msm/adreno/adreno_gpu.h | 20 +++-- drivers/gpu/drm/msm/msm_drv.h | 2 + drivers/gpu/drm/msm/msm_gem.h | 3 +- drivers/gpu/drm/msm/msm_gem_submit.c | 5 +- drivers/gpu/drm/msm/msm_gpu.c | 146 +++++++++++++++++++++----------- drivers/gpu/drm/msm/msm_gpu.h | 39 ++++----- drivers/gpu/drm/msm/msm_ringbuffer.c | 27 +++--- drivers/gpu/drm/msm/msm_ringbuffer.h | 19 ++++- 14 files changed, 287 insertions(+), 178 deletions(-)
drivers/gpu/drm/msm/adreno/a3xx_gpu.c +5 -4
@@ -44,7 +44,7 @@static bool a3xx_me_init (struct msm_gpu *gpu) { -struct msm_ringbuffer *ring = gpu->rb; +struct msm_ringbuffer *ring = gpu->rb[0 ];OUT_PKT3 (ring,CP_ME_INIT ,17 );OUT_RING (ring,0x000003f7 );
@@ -65,7 +65,7 @@ static bool a3xx_me_init(struct msm_gpu *gpu)OUT_RING (ring,0x00000000 );OUT_RING (ring,0x00000000 ); - gpu->funcs->flush (gpu); + gpu->funcs->flush (gpu, ring);return a3xx_idle (gpu); }
@@ -339,7 +339,7 @@ static void a3xx_destroy(struct msm_gpu *gpu)static bool a3xx_idle (struct msm_gpu *gpu) {/* wait for ringbuffer to drain: */ -if (!adreno_idle (gpu)) +if (!adreno_idle (gpu, gpu->rb[0 ]))return false;/* then wait for GPU to finish: */
@@ -446,6 +446,7 @@ static void a3xx_dump(struct msm_gpu *gpu) .recover = a3xx_recover, .submit = adreno_submit, .flush = adreno_flush, + .active_ring = adreno_active_ring, .irq = a3xx_irq, .destroy = a3xx_destroy,#ifdef CONFIG_DEBUG_FS
@@ -491,7 +492,7 @@ struct msm_gpu *a3xx_gpu_init(struct drm_device *dev) adreno_gpu->registers = a3xx_registers; adreno_gpu->reg_offsets = a3xx_register_offsets; - ret =adreno_gpu_init (dev, pdev, adreno_gpu, &funcs); + ret =adreno_gpu_init (dev, pdev, adreno_gpu, &funcs,1 );if (ret) goto fail;
drivers/gpu/drm/msm/adreno/a4xx_gpu.c +5 -4
@@ -116,7 +116,7 @@ static void a4xx_enable_hwcg(struct msm_gpu *gpu)static bool a4xx_me_init (struct msm_gpu *gpu) { -struct msm_ringbuffer *ring = gpu->rb; +struct msm_ringbuffer *ring = gpu->rb[0 ];OUT_PKT3 (ring,CP_ME_INIT ,17 );OUT_RING (ring,0x000003f7 );
@@ -137,7 +137,7 @@ static bool a4xx_me_init(struct msm_gpu *gpu)OUT_RING (ring,0x00000000 );OUT_RING (ring,0x00000000 ); - gpu->funcs->flush (gpu); + gpu->funcs->flush (gpu, ring);return a4xx_idle (gpu); }
@@ -337,7 +337,7 @@ static void a4xx_destroy(struct msm_gpu *gpu)static bool a4xx_idle (struct msm_gpu *gpu) {/* wait for ringbuffer to drain: */ -if (!adreno_idle (gpu)) +if (!adreno_idle (gpu, gpu->rb[0 ]))return false;/* then wait for GPU to finish: */
@@ -534,6 +534,7 @@ static int a4xx_get_timestamp(struct msm_gpu *gpu, uint64_t *value) .recover = a4xx_recover, .submit = adreno_submit, .flush = adreno_flush, + .active_ring = adreno_active_ring, .irq = a4xx_irq, .destroy = a4xx_destroy,#ifdef CONFIG_DEBUG_FS
@@ -573,7 +574,7 @@ struct msm_gpu *a4xx_gpu_init(struct drm_device *dev) adreno_gpu->registers = a4xx_registers; adreno_gpu->reg_offsets = a4xx_register_offsets; - ret =adreno_gpu_init (dev, pdev, adreno_gpu, &funcs); + ret =adreno_gpu_init (dev, pdev, adreno_gpu, &funcs,1 );if (ret) goto fail;
drivers/gpu/drm/msm/adreno/a5xx_gpu.c +23 -22
110 lines not on this page · raw message
drivers/gpu/drm/msm/adreno/a5xx_gpu.h +1 -1
@@ -57,7 +57,7 @@ static inline int spin_usecs(struct msm_gpu *gpu, uint32_t usecs, return -ETIMEDOUT; } -bool a5xx_idle (struct msm_gpu *gpu); +bool a5xx_idle (struct msm_gpu *gpu,struct msm_ringbuffer *ring);void a5xx_set_hwcg (struct msm_gpu *gpu,bool state);#endif /* __A5XX_GPU_H__ */
drivers/gpu/drm/msm/adreno/a5xx_power.c +3 -3
@@ -173,7 +173,7 @@ static int a5xx_gpmu_init(struct msm_gpu *gpu) {struct adreno_gpu *adreno_gpu =to_adreno_gpu (gpu);struct a5xx_gpu *a5xx_gpu =to_a5xx_gpu (adreno_gpu); -struct msm_ringbuffer *ring = gpu->rb; +struct msm_ringbuffer *ring = gpu->rb[0 ];if (!a5xx_gpu->gpmu_dwords)return 0 ;
@@ -192,9 +192,9 @@ static int a5xx_gpmu_init(struct msm_gpu *gpu)OUT_PKT7 (ring,CP_SET_PROTECTED_MODE ,1 );OUT_RING (ring,1 ); - gpu->funcs->flush (gpu); + gpu->funcs->flush (gpu, ring); -if (!a5xx_idle (gpu)) { +if (!a5xx_idle (gpu, ring)) {DRM_ERROR ("%s: Unable to load GPMU firmware. GPMU will not be active\n" , gpu->name);return -EINVAL ;
drivers/gpu/drm/msm/adreno/adreno_gpu.c +81 -52
245 lines not on this page · raw message
drivers/gpu/drm/msm/adreno/adreno_gpu.h +11 -9
@@ -185,17 +185,19 @@ static inline int adreno_is_a530(struct adreno_gpu *gpu)void adreno_recover (struct msm_gpu *gpu);void adreno_submit (struct msm_gpu *gpu,struct msm_gem_submit *submit,struct msm_file_private *ctx); -void adreno_flush (struct msm_gpu *gpu); -bool adreno_idle (struct msm_gpu *gpu); +void adreno_flush (struct msm_gpu *gpu,struct msm_ringbuffer *ring); +bool adreno_idle (struct msm_gpu *gpu,struct msm_ringbuffer *ring);#ifdef CONFIG_DEBUG_FS void adreno_show (struct msm_gpu *gpu,struct seq_file *m);#endif void adreno_dump_info (struct msm_gpu *gpu);void adreno_dump (struct msm_gpu *gpu); -void adreno_wait_ring (struct msm_gpu *gpu,uint32_t ndwords); +void adreno_wait_ring (struct msm_ringbuffer *ring,uint32_t ndwords); +struct msm_ringbuffer *adreno_active_ring (struct msm_gpu *gpu);int adreno_gpu_init (struct drm_device *drm,struct platform_device *pdev, -struct adreno_gpu *gpu,const struct adreno_gpu_funcs *funcs); +struct adreno_gpu *gpu,const struct adreno_gpu_funcs *funcs, +int nr_rings);void adreno_gpu_cleanup (struct adreno_gpu *gpu);
@@ -204,7 +206,7 @@ int adreno_gpu_init(struct drm_device *drm, struct platform_device *pdev,static inline void OUT_PKT0 (struct msm_ringbuffer *ring,uint16_t regindx,uint16_t cnt) { -adreno_wait_ring (ring->gpu, cnt+1 ); +adreno_wait_ring (ring, cnt+1 );OUT_RING (ring,CP_TYPE0_PKT | ((cnt-1 ) <<16 ) | (regindx &0x7FFF )); }
@@ -212,14 +214,14 @@ int adreno_gpu_init(struct drm_device *drm, struct platform_device *pdev,static inline void OUT_PKT2 (struct msm_ringbuffer *ring) { -adreno_wait_ring (ring->gpu,1 ); +adreno_wait_ring (ring,1 );OUT_RING (ring,CP_TYPE2_PKT ); }static inline void OUT_PKT3 (struct msm_ringbuffer *ring,uint8_t opcode,uint16_t cnt) { -adreno_wait_ring (ring->gpu, cnt+1 ); +adreno_wait_ring (ring, cnt+1 );OUT_RING (ring,CP_TYPE3_PKT | ((cnt-1 ) <<16 ) | ((opcode &0xFF ) <<8 )); }
@@ -241,14 +243,14 @@ static inline u32 PM4_PARITY(u32 val)static inline void OUT_PKT4 (struct msm_ringbuffer *ring,uint16_t regindx,uint16_t cnt) { -adreno_wait_ring (ring->gpu, cnt +1 ); +adreno_wait_ring (ring, cnt +1 );OUT_RING (ring,PKT4 (regindx, cnt)); }static inline void OUT_PKT7 (struct msm_ringbuffer *ring,uint8_t opcode,uint16_t cnt) { -adreno_wait_ring (ring->gpu, cnt +1 ); +adreno_wait_ring (ring, cnt +1 );OUT_RING (ring,CP_TYPE7_PKT | (cnt <<0 ) | (PM4_PARITY (cnt) <<15 ) | ((opcode &0x7F ) <<16 ) | (PM4_PARITY (opcode) <<23 )); }
drivers/gpu/drm/msm/msm_drv.h +2 -0
@@ -76,6 +76,8 @@ struct msm_vblank_ctrl {spinlock_t lock; }; +#define MSM_GPU_MAX_RINGS 1 +struct msm_drm_private {struct drm_device *dev;
drivers/gpu/drm/msm/msm_gem.h +2 -1
@@ -138,7 +138,7 @@ enum msm_gem_lock {struct msm_gem_submit {struct drm_device *dev;struct msm_gpu *gpu; -struct list_head node;/* node in gpu submit_list */ +struct list_head node;/* node in ring submit list */ struct list_head bo_list;struct ww_acquire_ctx ticket;uint32_t seqno;/* Sequence number of the submit on the ring */
@@ -146,6 +146,7 @@ struct msm_gem_submit {struct msm_gpu_submitqueue *queue;struct pid *pid;/* submitting process */ bool valid;/* true if no cmdstream patching needed */ +struct msm_ringbuffer *ring;unsigned int nr_cmds;unsigned int nr_bos;struct {
drivers/gpu/drm/msm/msm_gem_submit.c +4 -1
@@ -399,7 +399,7 @@ int msm_ioctl_gem_submit(struct drm_device *dev, void *data,struct msm_gpu_submitqueue *queue;int out_fence_fd =-1 ;unsigned i; -int ret; +int ret, ring;if (!gpu)return -ENXIO ;
@@ -532,6 +532,9 @@ int msm_ioctl_gem_submit(struct drm_device *dev, void *data, submit->nr_cmds = i; + ring =clamp_t (uint32_t, queue->prio,0 , gpu->nr_rings -1 ); + submit->ring = gpu->rb[ring]; + submit->fence =msm_fence_alloc (queue->fctx);if (IS_ERR (submit->fence)) { ret =PTR_ERR (submit->fence);
drivers/gpu/drm/msm/msm_gpu.c +98 -48
258 lines not on this page · raw message
drivers/gpu/drm/msm/msm_gpu.h +18 -21
@@ -33,7 +33,7 @@ struct msm_gpu_config {const char *irqname;uint64_t va_start;uint64_t va_end; -unsigned int ringsz; +unsigned int nr_rings; };/* So far, with hardware that I've seen to date, we can have:
@@ -57,8 +57,9 @@ struct msm_gpu_funcs {int (*pm_resume)(struct msm_gpu *gpu);void (*submit)(struct msm_gpu *gpu,struct msm_gem_submit *submit,struct msm_file_private *ctx); -void (*flush)(struct msm_gpu *gpu); +void (*flush)(struct msm_gpu *gpu,struct msm_ringbuffer *ring);irqreturn_t (*irq)(struct msm_gpu *irq); +struct msm_ringbuffer *(*active_ring)(struct msm_gpu *gpu);void (*recover)(struct msm_gpu *gpu);void (*destroy)(struct msm_gpu *gpu);#ifdef CONFIG_DEBUG_FS
@@ -67,14 +68,6 @@ struct msm_gpu_funcs {#endif }; -#define rbmemptr (gpu, member) \ - ((gpu)->memptrs_iova + offsetof(struct msm_rbmemptrs, member)) - -struct msm_rbmemptrs { -volatile uint32_t rptr; -volatile uint32_t fence; -}; -struct msm_gpu {const char *name;struct drm_device *dev;
@@ -93,9 +86,8 @@ struct msm_gpu {const struct msm_gpu_perfcntr *perfcntrs;uint32_t num_perfcntrs; -/* ringbuffer: */ -struct msm_ringbuffer *rb; -uint64_t rb_iova; +struct msm_ringbuffer *rb[MSM_GPU_MAX_RINGS ]; +int nr_rings;/* list of GEM active objects: */ struct list_head active_list;
@@ -133,21 +125,26 @@ struct msm_gpu {#define DRM_MSM_HANGCHECK_PERIOD 500/* in ms */ #define DRM_MSM_HANGCHECK_JIFFIES msecs_to_jiffies(DRM_MSM_HANGCHECK_PERIOD)struct timer_list hangcheck_timer; -uint32_t hangcheck_fence;struct work_struct recover_work; -struct list_head submit_list; - -struct msm_rbmemptrs *memptrs;struct drm_gem_object *memptrs_bo; -uint64_t memptrs_iova; - - }; +/* It turns out that all targets use the same ringbuffer size */ +#define MSM_GPU_RINGBUFFER_SZ SZ_32K +static inline bool msm_gpu_active (struct msm_gpu *gpu) { -return gpu->seqno > gpu->memptrs->fence; +int i; + +for (i =0 ; i < gpu->nr_rings; i++) { +struct msm_ringbuffer *ring = gpu->rb[i]; + +if (ring->seqno > ring->memptrs->fence) +return true; + } + +return false; }/* Perf-Counters:
drivers/gpu/drm/msm/msm_ringbuffer.c +17 -10
@@ -18,13 +18,14 @@#include "msm_ringbuffer.h" #include "msm_gpu.h" -struct msm_ringbuffer *msm_ringbuffer_new (struct msm_gpu *gpu,int size) +struct msm_ringbuffer *msm_ringbuffer_new (struct msm_gpu *gpu,int id, +void *memptrs,uint64_t memptrs_iova) {struct msm_ringbuffer *ring;int ret; -if (WARN_ON (!is_power_of_2 (size))) -return ERR_PTR (-EINVAL ); +/* We assume everwhere that MSM_GPU_RINGBUFFER_SZ is a power of 2 */ +BUILD_BUG_ON (!is_power_of_2 (MSM_GPU_RINGBUFFER_SZ )); ring =kzalloc (sizeof (*ring),GFP_KERNEL );if (!ring) {
@@ -33,32 +34,38 @@ struct msm_ringbuffer *msm_ringbuffer_new(struct msm_gpu *gpu, int size) } ring->gpu = gpu; - + ring->id = id;/* Pass NULL for the iova pointer - we will map it later */ - ring->start =msm_gem_kernel_new (gpu->dev, size,MSM_BO_WC , - gpu->aspace, &ring->bo,NULL ); + ring->start =msm_gem_kernel_new (gpu->dev,MSM_GPU_RINGBUFFER_SZ , +MSM_BO_WC , gpu->aspace, &ring->bo,NULL );if (IS_ERR (ring->start)) { ret =PTR_ERR (ring->start); ring->start =0 ; goto fail; } - ring->end = ring->start + (size /4 ); + ring->end = ring->start + (MSM_GPU_RINGBUFFER_SZ >>2 ); ring->cur = ring->start; - ring->size = size; + ring->memptrs = memptrs; + ring->memptrs_iova = memptrs_iova; + +INIT_LIST_HEAD (&ring->submits);return ring; fail: -if (ring) -msm_ringbuffer_destroy (ring); +msm_ringbuffer_destroy (ring);return ERR_PTR (ret); }void msm_ringbuffer_destroy (struct msm_ringbuffer *ring) { +if (IS_ERR_OR_NULL (ring)) +return ; +if (ring->bo) { +msm_gem_put_iova (ring->bo, ring->gpu->aspace);msm_gem_put_vaddr (ring->bo);drm_gem_object_unreference_unlocked (ring->bo); }
drivers/gpu/drm/msm/msm_ringbuffer.h +17 -2
@@ -20,14 +20,29 @@#include "msm_drv.h" +#define rbmemptr (ring, member) \ + ((ring)->memptrs_iova + offsetof(struct msm_rbmemptrs, member)) + +struct msm_rbmemptrs { +volatile uint32_t rptr; +volatile uint32_t fence; +}; +struct msm_ringbuffer {struct msm_gpu *gpu; -int size; +int id;struct drm_gem_object *bo;uint32_t *start, *end, *cur; +struct list_head submits; +uint64_t iova; +uint32_t seqno; +uint32_t hangcheck_fence; +struct msm_rbmemptrs *memptrs; +uint64_t memptrs_iova; }; -struct msm_ringbuffer *msm_ringbuffer_new (struct msm_gpu *gpu,int size); +struct msm_ringbuffer *msm_ringbuffer_new (struct msm_gpu *gpu,int id, +void *memptrs,uint64_t memptrs_iova);void msm_ringbuffer_destroy (struct msm_ringbuffer *ring);/* ringbuffer helpers (the parts that are same for a3xx/a2xx/z180..) */
Signature · 6 lines
1.9.1 _______________________________________________ Freedreno mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/freedreno
kbuild test robot
build botRe: [PATCH 13/17] drm/msm: Support multiple ringbuffers
url: https://github.com/0day-ci/linux/commits/Jordan-Crouse/drm-msm-GPU-fixes-and-features-for-4-14/20170729-063809 config: arm-multi_v7_defconfig (attached as .config) compiler…
Hi Jordan,
[auto build test WARNING on v4.13-rc2]
[also build test WARNING on next-20170728]
[if your patch is applied to the wrong git tree, please drop us a note to help improve the system]
url: https://github.com/0day-ci/linux/commits/Jordan-Crouse/drm-msm-GPU-fixes-and-features-for-4-14/20170729-063809
config: arm-multi_v7_defconfig (attached as .config)
compiler: arm-linux-gnueabi-gcc (Debian 6.1.1-9) 6.1.1 20160705
reproduce:
wget https://raw.githubusercontent.com/01org/lkp-tests/master/sbin/make.cross -O ~/bin/make.cross chmod +x ~/bin/make.cross
# save the attached .config to linux build tree
make.cross ARCH=arm
All warnings (new ones prefixed by >>):
In file included from drivers/gpu/drm/msm/msm_drv.h:37:0,
from drivers/gpu/drm/msm/msm_gpu.h:24,
from drivers/gpu/drm/msm/msm_gpu.c:18:
drivers/gpu/drm/msm/msm_gpu.c: In function 'msm_gpu_init':
>> drivers/gpu/drm/msm/msm_gpu.c:743:31: warning: format '%lu' expects argument of type 'long unsigned int', but argument 7 has type 'unsigned int' [-Wformat=]
DRM_DEV_INFO_ONCE(drm->dev, "Only creating %lu ringbuffers\n",
^
include/drm/drmP.h:208:60: note: in definition of macro 'DRM_DEV_INFO'
drm_dev_printk(dev, KERN_INFO, DRM_UT_NONE, __func__, "", fmt, \
^~~
>> drivers/gpu/drm/msm/msm_gpu.c:743:3: note: in expansion of macro 'DRM_DEV_INFO_ONCE'
DRM_DEV_INFO_ONCE(drm->dev, "Only creating %lu ringbuffers\n",
^~~~~~~~~~~~~~~~~vim +743 drivers/gpu/drm/msm/msm_gpu.c
648
649 int msm_gpu_init(struct drm_device *drm, struct platform_device *pdev,
650 struct msm_gpu *gpu, const struct msm_gpu_funcs *funcs,
651 const char *name, struct msm_gpu_config *config)
652 {
653 int i, ret, nr_rings = config->nr_rings;
654 void *memptrs;
655 uint64_t memptrs_iova;
656
657 if (WARN_ON(gpu->num_perfcntrs > ARRAY_SIZE(gpu->last_cntrs)))
658 gpu->num_perfcntrs = ARRAY_SIZE(gpu->last_cntrs);
659
660 gpu->dev = drm;
661 gpu->funcs = funcs;
662 gpu->name = name;
663
664 INIT_LIST_HEAD(&gpu->active_list);
665 INIT_WORK(&gpu->retire_work, retire_worker);
666 INIT_WORK(&gpu->recover_work, recover_worker);
667
668
669 setup_timer(&gpu->hangcheck_timer, hangcheck_handler,
670 (unsigned long)gpu);
671
672 spin_lock_init(&gpu->perf_lock);
673
674
675 /* Map registers: */
676 gpu->mmio = msm_ioremap(pdev, config->ioname, name);
677 if (IS_ERR(gpu->mmio)) {
678 ret = PTR_ERR(gpu->mmio);
679 goto fail;
680 }
681
682 /* Get Interrupt: */
683 gpu->irq = platform_get_irq_byname(pdev, config->irqname);
684 if (gpu->irq < 0) {
685 ret = gpu->irq;
686 dev_err(drm->dev, "failed to get irq: %d\n", ret);
687 goto fail;
688 }
689
690 ret = devm_request_irq(&pdev->dev, gpu->irq, irq_handler,
691 IRQF_TRIGGER_HIGH, gpu->name, gpu);
692 if (ret) {
693 dev_err(drm->dev, "failed to request IRQ%u: %d\n", gpu->irq, ret);
694 goto fail;
695 }
696
697 ret = get_clocks(pdev, gpu);
698 if (ret)
699 goto fail;
700
701 gpu->ebi1_clk = msm_clk_get(pdev, "bus");
702 DBG("ebi1_clk: %p", gpu->ebi1_clk);
703 if (IS_ERR(gpu->ebi1_clk))
704 gpu->ebi1_clk = NULL;
705
706 /* Acquire regulators: */
707 gpu->gpu_reg = devm_regulator_get(&pdev->dev, "vdd");
708 DBG("gpu_reg: %p", gpu->gpu_reg);
709 if (IS_ERR(gpu->gpu_reg))
710 gpu->gpu_reg = NULL;
711
712 gpu->gpu_cx = devm_regulator_get(&pdev->dev, "vddcx");
713 DBG("gpu_cx: %p", gpu->gpu_cx);
714 if (IS_ERR(gpu->gpu_cx))
715 gpu->gpu_cx = NULL;
716
717 gpu->pdev = pdev;
718 platform_set_drvdata(pdev, gpu);
719
720 bs_init(gpu);
721
722 gpu->aspace = msm_gpu_create_address_space(gpu, pdev,
723 config->va_start, config->va_end);
724
725 if (gpu->aspace == NULL)
726 dev_info(drm->dev, "%s: no IOMMU, fallback to VRAM carveout!\n", name);
727 else if (IS_ERR(gpu->aspace)) {
728 ret = PTR_ERR(gpu->aspace);
729 goto fail;
730 }
731
732 memptrs = msm_gem_kernel_new(drm, sizeof(*gpu->memptrs_bo),
733 MSM_BO_UNCACHED, gpu->aspace, &gpu->memptrs_bo,
734 &memptrs_iova);
735
736 if (IS_ERR(memptrs)) {
737 ret = PTR_ERR(memptrs);
738 dev_err(drm->dev, "could not allocate memptrs: %d\n", ret);
739 goto fail;
740 }
741
742 if (nr_rings > ARRAY_SIZE(gpu->rb)) {
> 743 DRM_DEV_INFO_ONCE(drm->dev, "Only creating %lu ringbuffers\n",
744 ARRAY_SIZE(gpu->rb));
745 nr_rings = ARRAY_SIZE(gpu->rb);
746 }
747
748 /* Create ringbuffer(s): */
749 for (i = 0; i < nr_rings; i++) {
750 gpu->rb[i] = msm_ringbuffer_new(gpu, i, memptrs, memptrs_iova);
751
752 if (IS_ERR(gpu->rb[i])) {
753 ret = PTR_ERR(gpu->rb[i]);
754 dev_err(drm->dev,
755 "could not create ringbuffer %d: %d\n", i, ret);
756 goto fail;
757 }
758
759 memptrs += sizeof(struct msm_rbmemptrs);
760 memptrs_iova += sizeof(struct msm_rbmemptrs);
761 }
762
763 gpu->nr_rings = nr_rings;
764
765 return 0;
766
767 fail:
768 for (i = 0; i < ARRAY_SIZE(gpu->rb); i++) {
769 msm_ringbuffer_destroy(gpu->rb[i]);
770 gpu->rb[i] = NULL;
771 }
772
773 if (gpu->memptrs_bo) {
774 msm_gem_put_vaddr(gpu->memptrs_bo);
775 msm_gem_put_iova(gpu->memptrs_bo, gpu->aspace);
776 drm_gem_object_unreference_unlocked(gpu->memptrs_bo);
777 }
778
779 if (gpu->aspace) {
780 gpu->aspace->mmu->funcs->detach(gpu->aspace->mmu,
781 NULL, 0);
782 msm_gem_address_space_put(gpu->aspace);
783 }
784
785 platform_set_drvdata(pdev, NULL);
786 return ret;
787 }
788
---
0-DAY kernel test infrastructure Open Source Technology Center
https://lists.01.org/pipermail/kbuild-all Intel Corporation
PATCH 16/17 drm/msm: Make the value of RB_CNTL (almost) generic
In reply to a message not in the archiveWe use a global ringbuffer size and block size for all targets and at least for 5XX preemption we need to know the value the RB_CNTL in several locations so it makes sense to calculate it once and use it everywhere.
The only monkey wrench is that we need to disable the RPTR shadow for A430 targets but that only needs to be done once and doesn't affect A5XX so we can or in the value at init time.
--- drivers/gpu/drm/msm/adreno/adreno_gpu.c | 12 +++++++----- drivers/gpu/drm/msm/msm_gpu.h | 5 +++++ 2 files changed, 12 insertions(+), 5 deletions(-)
drivers/gpu/drm/msm/adreno/adreno_gpu.c +7 -5
@@ -21,7 +21,6 @@#include "msm_gem.h" #include "msm_mmu.h" -#define RB_BLKSIZE 32int adreno_get_param (struct msm_gpu *gpu,uint32_t param,uint64_t *value) {
@@ -89,11 +88,14 @@ int adreno_hw_init(struct msm_gpu *gpu) ring->memptrs->rptr =0 ; } -/* Setup REG_CP_RB_CNTL: */ +/* +* Setup REG_CP_RB_CNTL. The same value is used across targets (with +* the excpetion of A430 that disables the RPTR shadow) - the cacluation +* for the ringbuffer size and block size is moved to msm_gpu.h for the +* pre-processor to deal with and the A430 variant is ORed in here +*/ adreno_gpu_write (adreno_gpu,REG_ADRENO_CP_RB_CNTL , -/* size is log2(quad-words): */ -AXXX_CP_RB_CNTL_BUFSZ (ilog2 (MSM_GPU_RINGBUFFER_SZ /8 )) | -AXXX_CP_RB_CNTL_BLKSZ (ilog2 (RB_BLKSIZE /8 )) | +MSM_GPU_RB_CNTL_DEFAULT | (adreno_is_a430 (adreno_gpu) ?AXXX_CP_RB_CNTL_NO_UPDATE :0 ));/* Setup ringbuffer address - use ringbuffer[0] for GPU init */
drivers/gpu/drm/msm/msm_gpu.h +5 -0
@@ -132,6 +132,11 @@ struct msm_gpu {/* It turns out that all targets use the same ringbuffer size */ #define MSM_GPU_RINGBUFFER_SZ SZ_32K +#define MSM_GPU_RINGBUFFER_BLKSIZE 32 + +#define MSM_GPU_RB_CNTL_DEFAULT \ + (AXXX_CP_RB_CNTL_BUFSZ(ilog2(MSM_GPU_RINGBUFFER_SZ / 8)) | \ + AXXX_CP_RB_CNTL_BLKSZ(ilog2(MSM_GPU_RINGBUFFER_BLKSIZE / 8)))static inline bool msm_gpu_active (struct msm_gpu *gpu) {
Signature · 6 lines
1.9.1 _______________________________________________ Freedreno mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/freedreno
PATCH 2/17 drm/msm: Allow hardware clock gating to be toggled
In reply to Jordan CrouseThere are some use cases wherein we need to turn off hardware clock gating before reading certain registers. Modify the A5XX HWCG function to allow user to enable or disable clock gating at will.
--- drivers/gpu/drm/msm/adreno/a5xx_gpu.c | 42 ++++++++--------------------------- drivers/gpu/drm/msm/adreno/a5xx_gpu.h | 1 + 2 files changed, 10 insertions(+), 33 deletions(-)
drivers/gpu/drm/msm/adreno/a5xx_gpu.c +9 -33
@@ -117,12 +117,10 @@ static void a5xx_submit(struct msm_gpu *gpu, struct msm_gem_submit *submit, gpu->funcs->flush (gpu); } -struct a5xx_hwcg { +static const struct {u32 offset;u32 value; -}; - -static const struct a5xx_hwcg a530_hwcg[] = { +} a5xx_hwcg[] = { {REG_A5XX_RBBM_CLOCK_CNTL_SP0 ,0x02222222 }, {REG_A5XX_RBBM_CLOCK_CNTL_SP1 ,0x02222222 }, {REG_A5XX_RBBM_CLOCK_CNTL_SP2 ,0x02222222 },
@@ -217,38 +215,16 @@ struct a5xx_hwcg { {REG_A5XX_RBBM_CLOCK_DELAY_VFD, 0x00002222} }; -static const struct { -int (*test)(struct adreno_gpu *gpu); -const struct a5xx_hwcg *regs; -unsigned int count; -} a5xx_hwcg_regs[] = { - { adreno_is_a530, a530_hwcg,ARRAY_SIZE (a530_hwcg), }, -}; - -static void _a5xx_enable_hwcg (struct msm_gpu *gpu, -const struct a5xx_hwcg *regs,unsigned int count) +void a5xx_set_hwcg (struct msm_gpu *gpu,bool state) {unsigned int i; -for (i =0 ; i < count; i++) -gpu_write (gpu, regs[i].offset, regs[i].value); - -gpu_write (gpu,REG_A5XX_RBBM_CLOCK_CNTL ,0xAAA8AA00 ); -gpu_write (gpu,REG_A5XX_RBBM_ISDB_CNT ,0x182 ); -} +for (i =0 ; i <ARRAY_SIZE (a5xx_hwcg); i++) +gpu_write (gpu, a5xx_hwcg[i].offset, + state ? a5xx_hwcg[i].value :0 ); -static void a5xx_enable_hwcg (struct msm_gpu *gpu) -{ -struct adreno_gpu *adreno_gpu =to_adreno_gpu (gpu); -unsigned int i; - -for (i =0 ; i <ARRAY_SIZE (a5xx_hwcg_regs); i++) { -if (a5xx_hwcg_regs[i].test (adreno_gpu)) { -_a5xx_enable_hwcg (gpu, a5xx_hwcg_regs[i].regs, - a5xx_hwcg_regs[i].count); -return ; - } - } +gpu_write (gpu,REG_A5XX_RBBM_CLOCK_CNTL , state ?0xAAA8AA00 :0 ); +gpu_write (gpu,REG_A5XX_RBBM_ISDB_CNT , state ?0x182 :0x180 ); }static int a5xx_me_init (struct msm_gpu *gpu)
@@ -545,7 +521,7 @@ static int a5xx_hw_init(struct msm_gpu *gpu)gpu_write (gpu,REG_A5XX_RBBM_AHB_CNTL1 ,0xA6FFFFFF );/* Enable HWCG */ -a5xx_enable_hwcg (gpu); +a5xx_set_hwcg (gpu, true);gpu_write (gpu,REG_A5XX_RBBM_AHB_CNTL2 ,0x0000003F );
drivers/gpu/drm/msm/adreno/a5xx_gpu.h +1 -0
@@ -59,5 +59,6 @@ static inline int spin_usecs(struct msm_gpu *gpu, uint32_t usecs, }bool a5xx_idle (struct msm_gpu *gpu); +void a5xx_set_hwcg (struct msm_gpu *gpu,bool state);#endif /* __A5XX_GPU_H__ */
Signature · 1 line
1.9.1
PATCH 4/17 drm/msm: Remove uneeded platform dev members
In reply to Jordan CrouseCommit eeb754746b14 ("drm/msm/gpu: use pm-runtime") adds a pointer for the GPU platform device to the msm_gpu struct so we can happily remove the same pointers from the individual GPU structs.
--- drivers/gpu/drm/msm/adreno/a3xx_gpu.c | 2 -- drivers/gpu/drm/msm/adreno/a3xx_gpu.h | 1 - drivers/gpu/drm/msm/adreno/a4xx_gpu.c | 2 -- drivers/gpu/drm/msm/adreno/a4xx_gpu.h | 1 - drivers/gpu/drm/msm/adreno/a5xx_gpu.c | 3 +-- drivers/gpu/drm/msm/adreno/a5xx_gpu.h | 1 - 6 files changed, 1 insertion(+), 9 deletions(-)
drivers/gpu/drm/msm/adreno/a3xx_gpu.c +0 -2
@@ -486,8 +486,6 @@ struct msm_gpu *a3xx_gpu_init(struct drm_device *dev) adreno_gpu = &a3xx_gpu->base; gpu = &adreno_gpu->base; - a3xx_gpu->pdev = pdev; - gpu->perfcntrs = perfcntrs; gpu->num_perfcntrs =ARRAY_SIZE (perfcntrs);
drivers/gpu/drm/msm/adreno/a3xx_gpu.h +0 -1
@@ -28,7 +28,6 @@struct a3xx_gpu {struct adreno_gpu base; -struct platform_device *pdev;/* if OCMEM is used for GMEM: */ uint32_t ocmem_base;
drivers/gpu/drm/msm/adreno/a4xx_gpu.c +0 -2
@@ -568,8 +568,6 @@ struct msm_gpu *a4xx_gpu_init(struct drm_device *dev) adreno_gpu = &a4xx_gpu->base; gpu = &adreno_gpu->base; - a4xx_gpu->pdev = pdev; - gpu->perfcntrs =NULL ; gpu->num_perfcntrs =0 ;
drivers/gpu/drm/msm/adreno/a4xx_gpu.h +0 -1
@@ -23,7 +23,6 @@struct a4xx_gpu {struct adreno_gpu base; -struct platform_device *pdev;/* if OCMEM is used for GMEM: */ uint32_t ocmem_base;
drivers/gpu/drm/msm/adreno/a5xx_gpu.c +1 -2
@@ -397,7 +397,7 @@ static int a5xx_zap_shader_init(struct msm_gpu *gpu)static bool loaded;struct adreno_gpu *adreno_gpu =to_adreno_gpu (gpu);struct a5xx_gpu *a5xx_gpu =to_a5xx_gpu (adreno_gpu); -struct platform_device *pdev = a5xx_gpu->pdev; +struct platform_device *pdev = gpu->pdev;int ret;/*
@@ -1046,7 +1046,6 @@ struct msm_gpu *a5xx_gpu_init(struct drm_device *dev) adreno_gpu = &a5xx_gpu->base; gpu = &adreno_gpu->base; - a5xx_gpu->pdev = pdev; adreno_gpu->registers = a5xx_registers; adreno_gpu->reg_offsets = a5xx_register_offsets;
drivers/gpu/drm/msm/adreno/a5xx_gpu.h +0 -1
@@ -23,7 +23,6 @@struct a5xx_gpu {struct adreno_gpu base; -struct platform_device *pdev;struct drm_gem_object *pm4_bo;uint64_t pm4_iova;
Signature · 1 line
1.9.1
PATCH 5/17 drm/msm: args->fence should be args->flags
In reply to Jordan CrouseFix a typo in msm_ioctl_gem_submit - check args->flags for the MSM_SUBMIT_NO_IMPLICIT flag instead of args->fence.
--- drivers/gpu/drm/msm/msm_gem_submit.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/gpu/drm/msm/msm_gem_submit.c +1 -1
@@ -451,7 +451,7 @@ int msm_ioctl_gem_submit(struct drm_device *dev, void *data,if (ret) goto out; -if (!(args->fence &MSM_SUBMIT_NO_IMPLICIT )) { +if (!(args->flags &MSM_SUBMIT_NO_IMPLICIT )) { ret =submit_fence_sync (submit);if (ret) goto out;
Signature · 1 line
1.9.1
PATCH 7/17 drm/msm: Add A5XX hardware fault detection
In reply to Jordan CrouseThe A5XX GPU has really good hardware fault detection that can detect a abnormal hardware condition and fire an interrupt in a matter of milliseconds which is a lot better than waiting for the hangcheck timer.
Enable the interrupt and log information before kicking off recovery.
--- drivers/gpu/drm/msm/adreno/a5xx_gpu.c | 26 ++++++++++++++++++++++++++ 1 file changed, 26 insertions(+)
drivers/gpu/drm/msm/adreno/a5xx_gpu.c +26 -0
@@ -438,6 +438,7 @@ static int a5xx_zap_shader_init(struct msm_gpu *gpu)A5XX_RBBM_INT_0_MASK_RBBM_ETS_MS_TIMEOUT | \A5XX_RBBM_INT_0_MASK_RBBM_ATB_ASYNC_OVERFLOW | \A5XX_RBBM_INT_0_MASK_CP_HW_ERROR | \ +A5XX_RBBM_INT_0_MASK_MISC_HANG_DETECT | \A5XX_RBBM_INT_0_MASK_CP_CACHE_FLUSH_TS | \A5XX_RBBM_INT_0_MASK_UCHE_OOB_ACCESS | \ A5XX_RBBM_INT_0_MASK_GPMU_VOLTAGE_DROOP)
@@ -843,6 +844,28 @@ static void a5xx_gpmu_err_irq(struct msm_gpu *gpu)dev_err_ratelimited (gpu->dev->dev,"GPMU | voltage droop\n" ); } +static void a5xx_fault_detect_irq (struct msm_gpu *gpu) +{ +struct drm_device *dev = gpu->dev; +struct msm_drm_private *priv = dev->dev_private; +struct msm_ringbuffer *ring = gpu->funcs->active_ring (gpu); + +dev_err (dev->dev,"gpu fault ring %d fence %x status %8.8X rb %4.4x/%4.4x ib1 %16.16llX/%4.4x ib2 %16.16llX/%4.4x\n" , + ring ? ring->id :-1 , ring ? ring->seqno :0 , +gpu_read (gpu,REG_A5XX_RBBM_STATUS ), +gpu_read (gpu,REG_A5XX_CP_RB_RPTR ), +gpu_read (gpu,REG_A5XX_CP_RB_WPTR ), +gpu_read64 (gpu,REG_A5XX_CP_IB1_BASE ,REG_A5XX_CP_IB1_BASE_HI ), +gpu_read (gpu,REG_A5XX_CP_IB1_BUFSZ ), +gpu_read64 (gpu,REG_A5XX_CP_IB2_BASE ,REG_A5XX_CP_IB2_BASE_HI ), +gpu_read (gpu,REG_A5XX_CP_IB2_BUFSZ )); + +/* Turn off the hangcheck timer to keep it from bothering us */ +del_timer (&gpu->hangcheck_timer); + +queue_work (priv->wq, &gpu->recover_work); +} +#define RBBM_ERROR_MASK \ (A5XX_RBBM_INT_0_MASK_RBBM_AHB_ERROR | \ A5XX_RBBM_INT_0_MASK_RBBM_TRANSFER_TIMEOUT | \
@@ -869,6 +892,9 @@ static irqreturn_t a5xx_irq(struct msm_gpu *gpu)if (status &A5XX_RBBM_INT_0_MASK_CP_HW_ERROR )a5xx_cp_err_irq (gpu); +if (status &A5XX_RBBM_INT_0_MASK_MISC_HANG_DETECT ) +a5xx_fault_detect_irq (gpu); +if (status &A5XX_RBBM_INT_0_MASK_UCHE_OOB_ACCESS )a5xx_uche_err_irq (gpu);
Signature · 1 line
1.9.1
PATCH 8/17 drm/msm: Add per-instance submit queues
In reply to Jordan CrouseCurrently the behavior of a command stream is provided by the user application during submission and the application is expected to internally maintain the settings for each 'context' or 'rendering queue' and specify the correct ones.
This works okay for simple cases but as applications become more complex we will want to set context specific flags and do various permission checks to allow certain contexts to enable additional privileges.
Add kernel-side submit queues to be analogous to 'contexts' or 'rendering queues' on the application side. Each file descriptor instance will maintain its own list of queues. Queues cannot be shared between file descriptors.
For backwards compatibility context id '0' is defined as a default context specifying no priority and no special flags. This is intended to be the usual configuration for 99% of applications so that a garden variety application can function correctly without creating a queue. Only those applications requiring the specific benefit of different queues need create one.
--- drivers/gpu/drm/msm/Makefile | 3 +- drivers/gpu/drm/msm/msm_drv.c | 51 +++++++++++-- drivers/gpu/drm/msm/msm_drv.h | 20 +++-- drivers/gpu/drm/msm/msm_gem.h | 1 + drivers/gpu/drm/msm/msm_gem_submit.c | 13 +++- drivers/gpu/drm/msm/msm_gpu.h | 15 ++++ drivers/gpu/drm/msm/msm_submitqueue.c | 139 ++++++++++++++++++++++++++++++++++ include/uapi/drm/msm_drm.h | 22 ++++++ 8 files changed, 249 insertions(+), 15 deletions(-) create mode 100644 drivers/gpu/drm/msm/msm_submitqueue.c
drivers/gpu/drm/msm/Makefile +2 -1
@@ -57,7 +57,8 @@ msm-y := \ msm_iommu.o \ msm_perf.o \ msm_rd.o \ - msm_ringbuffer.o + msm_ringbuffer.o \ + msm_submitqueue.o msm-$(CONFIG_DRM_FBDEV_EMULATION) += msm_fbdev.o msm-$(CONFIG_COMMON_CLK) += mdp/mdp4/mdp4_lvds_pll.o
drivers/gpu/drm/msm/msm_drv.c +44 -7
87 lines not on this page · raw message
drivers/gpu/drm/msm/msm_drv.h +15 -5
@@ -58,11 +58,9 @@#define NUM_DOMAINS 2/* one for KMS, then one per gpu core (?) */ struct msm_file_private { -/* currently we don't do anything useful with this.. but when -* per-context address spaces are supported we'd keep track of -* the context's page-tables here. -*/ -int dummy; +rwlock_t queuelock; +struct list_head submitqueues; +int queueid; };enum msm_mdp_plane_property {
@@ -315,6 +313,18 @@ void __iomem *msm_ioremap(struct platform_device *pdev, const char *name,void msm_writel (u32 data,void __iomem *addr);u32 msm_readl (const void __iomem *addr); +struct msm_gpu_submitqueue ; +int msm_submitqueue_init (struct msm_file_private *ctx); +struct msm_gpu_submitqueue *msm_submitqueue_get (struct msm_file_private *ctx, +u32 id); +int msm_submitqueue_create (struct msm_file_private *ctx,u32 prio, +u32 flags,u32 *id); +int msm_submitqueue_remove (struct msm_file_private *ctx,u32 id); +void msm_submitqueue_close (struct msm_file_private *ctx); + +void msm_submitqueue_destroy (struct kref *kref); + +#define DBG (fmt, ...) DRM_DEBUG_DRIVER(fmt"\n", ##__VA_ARGS__)#define VERB (fmt, ...) if (0) DRM_DEBUG_DRIVER(fmt"\n", ##__VA_ARGS__)
drivers/gpu/drm/msm/msm_gem.h +1 -0
@@ -142,6 +142,7 @@ struct msm_gem_submit {struct list_head bo_list;struct ww_acquire_ctx ticket;struct dma_fence *fence; +struct msm_gpu_submitqueue *queue;struct pid *pid;/* submitting process */ bool valid;/* true if no cmdstream patching needed */ unsigned int nr_cmds;
drivers/gpu/drm/msm/msm_gem_submit.c +11 -2
@@ -31,7 +31,8 @@#define BO_PINNED 0x2000static struct msm_gem_submit *submit_create (struct drm_device *dev, -struct msm_gpu *gpu,uint32_t nr_bos,uint32_t nr_cmds) +struct msm_gpu *gpu,struct msm_gpu_submitqueue *queue, +uint32_t nr_bos,uint32_t nr_cmds) {struct msm_gem_submit *submit; uint64_t sz = sizeof(*submit) + (nr_bos * sizeof(submit->bos[0])) +
@@ -49,6 +50,7 @@ static struct msm_gem_submit *submit_create(struct drm_device *dev, submit->fence =NULL ; submit->pid =get_pid (task_pid (current)); submit->cmd = (void *)&submit->bos[nr_bos]; + submit->queue = queue;/* initially, until copy_from_user() and bo lookup succeeds: */ submit->nr_bos =0 ;
@@ -66,6 +68,8 @@ void msm_gem_submit_free(struct msm_gem_submit *submit)dma_fence_put (submit->fence);list_del (&submit->node);put_pid (submit->pid); +msm_submitqueue_put (submit->queue); +kfree (submit); }
@@ -391,6 +395,7 @@ int msm_ioctl_gem_submit(struct drm_device *dev, void *data,struct msm_gpu *gpu = priv->gpu;struct dma_fence *in_fence =NULL ;struct sync_file *sync_file =NULL ; +struct msm_gpu_submitqueue *queue;int out_fence_fd =-1 ;unsigned i;int ret;
@@ -407,6 +412,10 @@ int msm_ioctl_gem_submit(struct drm_device *dev, void *data,if (MSM_PIPE_FLAGS (args->flags) & ~MSM_SUBMIT_FLAGS )return -EINVAL ; + queue =msm_submitqueue_get (ctx, args->queueid); +if (!queue) +return -ENOENT ; +if (args->flags &MSM_SUBMIT_FENCE_FD_IN ) { in_fence =sync_file_get_fence (args->fence_fd);
@@ -437,7 +446,7 @@ int msm_ioctl_gem_submit(struct drm_device *dev, void *data, } priv->struct_mutex_task = current; - submit =submit_create (dev, gpu, args->nr_bos, args->nr_cmds); + submit =submit_create (dev, gpu, queue, args->nr_bos, args->nr_cmds);if (!submit) { ret = -ENOMEM ; goto out_unlock;
drivers/gpu/drm/msm/msm_gpu.h +15 -0
@@ -150,6 +150,15 @@ struct msm_gpu_perfcntr {const char *name; }; +struct msm_gpu_submitqueue { +int id; +u32 flags; +u32 prio; +int faults; +struct list_head node; +struct kref ref; +}; +static inline void gpu_write (struct msm_gpu *gpu,u32 reg,u32 data) {msm_writel (data, gpu->mmio + (reg <<2 ));
@@ -223,4 +232,10 @@ int msm_gpu_init(struct drm_device *drm, struct platform_device *pdev,void __init adreno_register (void ); void __exit adreno_unregister(void); +static inline void msm_submitqueue_put (struct msm_gpu_submitqueue *queue) +{ +if (queue) +kref_put (&queue->ref, msm_submitqueue_destroy); +} +#endif /* __MSM_GPU_H__ */
drivers/gpu/drm/msm/msm_submitqueue.c new file +139 -0
139 lines not on this page · raw message
include/uapi/drm/msm_drm.h +22 -0
@@ -218,6 +218,7 @@ struct drm_msm_gem_submit {__u64 bos;/* in, ptr to array of submit_bo's */ __u64 cmds;/* in, ptr to array of submit_cmd's */ __s32 fence_fd;/* in/out fence fd (see MSM_SUBMIT_FENCE_FD_IN/OUT) */ +__u32 queueid;/* in, submitqueue id */ };/* The normal way to synchronize with the GPU is just to CPU_PREP on
@@ -254,6 +255,20 @@ struct drm_msm_gem_madvise {__u32 retained;/* out, whether backing store still exists */ }; +/* +* Draw queues allow the user to set specific submission parameter. Command +* submissions specify a specific submitqueue to use. ID 0 is reserved for +* backwards compatibility as a "default" submitqueue +*/ + +#define MSM_SUBMITQUEUE_FLAGS (0) + +struct drm_msm_submitqueue { +__u32 flags;/* in, MSM_SUBMITQUEUE_x */ +__u32 prio;/* in, Priority level */ +__u32 id;/* out, identifier */ +}; +#define DRM_MSM_GET_PARAM 0x00/* placeholder: #define DRM_MSM_SET_PARAM 0x01
@@ -265,6 +280,11 @@ struct drm_msm_gem_madvise {#define DRM_MSM_GEM_SUBMIT 0x06#define DRM_MSM_WAIT_FENCE 0x07#define DRM_MSM_GEM_MADVISE 0x08 +/* placeholder: +#define DRM_MSM_GEM_SVM_NEW 0x09 +*/ +#define DRM_MSM_SUBMITQUEUE_NEW 0x0A +#define DRM_MSM_SUBMITQUEUE_CLOSE 0x0B#define DRM_IOCTL_MSM_GET_PARAM DRM_IOWR(DRM_COMMAND_BASE + DRM_MSM_GET_PARAM, struct drm_msm_param)#define DRM_IOCTL_MSM_GEM_NEW DRM_IOWR(DRM_COMMAND_BASE + DRM_MSM_GEM_NEW, struct drm_msm_gem_new)
@@ -274,6 +294,8 @@ struct drm_msm_gem_madvise {#define DRM_IOCTL_MSM_GEM_SUBMIT DRM_IOWR(DRM_COMMAND_BASE + DRM_MSM_GEM_SUBMIT, struct drm_msm_gem_submit)#define DRM_IOCTL_MSM_WAIT_FENCE DRM_IOW (DRM_COMMAND_BASE + DRM_MSM_WAIT_FENCE, struct drm_msm_wait_fence)#define DRM_IOCTL_MSM_GEM_MADVISE DRM_IOWR(DRM_COMMAND_BASE + DRM_MSM_GEM_MADVISE, struct drm_msm_gem_madvise) +#define DRM_IOCTL_MSM_SUBMITQUEUE_NEW DRM_IOWR(DRM_COMMAND_BASE + DRM_MSM_SUBMITQUEUE_NEW, struct drm_msm_submitqueue) +#define DRM_IOCTL_MSM_SUBMITQUEUE_CLOSE DRM_IOW (DRM_COMMAND_BASE + DRM_MSM_SUBMITQUEUE_CLOSE, struct drm_msm_submitqueue)#if defined(__cplusplus) }
Signature · 1 line
1.9.1
PATCH 9/17 drm/msm: Implement per-submitqueue fences
In reply to Jordan CrouseCurrently we use a single fence context for the entire GPU. That works until we start dealing with multiple ringbuffers that will all need their own sequence number.
So in order for the user to continue using the fence functions we will need some way to uniquely identify a submission that gets mapped into the appropriate ringbuffer and sequence ID.
The best way to do this is to move to per-submitqueue fences. This allows each user context to have their own monotonically increasing identifier and we can easily map back and forth.
To implement this, create a fence context for each submitqueue and assign user fences from that context. Each submission will have be assigned a submitqueue fence and a unique sequence ID for the target ringbuffer. When the submission is retired on the ring the fence will get signaled.
struct drm_msm_wait_fence in the UABI needs to be extended to include the queue ID for the fence being queried. Legacy applications that don't use submitqueues will use queue 0 by default.
--- drivers/gpu/drm/msm/adreno/a5xx_gpu.c | 4 ++-- drivers/gpu/drm/msm/adreno/adreno_gpu.c | 10 ++++----- drivers/gpu/drm/msm/msm_drv.c | 19 ++++++++++++----- drivers/gpu/drm/msm/msm_drv.h | 6 +++--- drivers/gpu/drm/msm/msm_fence.c | 2 +- drivers/gpu/drm/msm/msm_fence.h | 2 +- drivers/gpu/drm/msm/msm_gem.h | 1 + drivers/gpu/drm/msm/msm_gem_submit.c | 7 ++++--- drivers/gpu/drm/msm/msm_gpu.c | 36 +++++++++++++++++++-------------- drivers/gpu/drm/msm/msm_gpu.h | 8 +++++--- drivers/gpu/drm/msm/msm_submitqueue.c | 33 ++++++++++++++++++++++++------ include/uapi/drm/msm_drm.h | 1 + 12 files changed, 85 insertions(+), 44 deletions(-)
drivers/gpu/drm/msm/adreno/a5xx_gpu.c +2 -2
@@ -106,13 +106,13 @@ static void a5xx_submit(struct msm_gpu *gpu, struct msm_gem_submit *submit, }OUT_PKT4 (ring,REG_A5XX_CP_SCRATCH_REG (2 ),1 ); -OUT_RING (ring, submit->fence->seqno); +OUT_RING (ring, submit->seqno);OUT_PKT7 (ring,CP_EVENT_WRITE ,4 );OUT_RING (ring,CACHE_FLUSH_TS | (1 <<31 ));OUT_RING (ring,lower_32_bits (rbmemptr (adreno_gpu, fence)));OUT_RING (ring,upper_32_bits (rbmemptr (adreno_gpu, fence))); -OUT_RING (ring, submit->fence->seqno); +OUT_RING (ring, submit->seqno); gpu->funcs->flush (gpu); }
drivers/gpu/drm/msm/adreno/adreno_gpu.c +5 -5
@@ -75,7 +75,7 @@ int adreno_hw_init(struct msm_gpu *gpu) gpu->rb->cur = gpu->rb->start;/* reset completed fence seqno: */ - adreno_gpu->memptrs->fence = gpu->fctx->completed_fence; + adreno_gpu->memptrs->fence = gpu->seqno; adreno_gpu->memptrs->rptr =0 ;/* Setup REG_CP_RB_CNTL: */
@@ -165,7 +165,7 @@ void adreno_submit(struct msm_gpu *gpu, struct msm_gem_submit *submit, }OUT_PKT0 (ring,REG_AXXX_CP_SCRATCH_REG2 ,1 ); -OUT_RING (ring, submit->fence->seqno); +OUT_RING (ring, submit->seqno);if (adreno_is_a3xx (adreno_gpu) ||adreno_is_a4xx (adreno_gpu)) {/* Flush HLSQ lazy updates to make sure there is nothing
@@ -182,7 +182,7 @@ void adreno_submit(struct msm_gpu *gpu, struct msm_gem_submit *submit,OUT_PKT3 (ring,CP_EVENT_WRITE ,3 );OUT_RING (ring,CACHE_FLUSH_TS );OUT_RING (ring,rbmemptr (adreno_gpu, fence)); -OUT_RING (ring, submit->fence->seqno); +OUT_RING (ring, submit->seqno);/* we could maybe be clever and only CP_COND_EXEC the interrupt: */ OUT_PKT3 (ring,CP_INTERRUPT ,1 );
@@ -255,7 +255,7 @@ void adreno_show(struct msm_gpu *gpu, struct seq_file *m) adreno_gpu->rev.patchid);seq_printf (m,"fence: %d/%d\n" , adreno_gpu->memptrs->fence, - gpu->fctx->last_fence); + gpu->seqno);seq_printf (m,"rptr: %d\n" ,get_rptr (adreno_gpu));seq_printf (m,"rb wptr: %d\n" ,get_wptr (gpu->rb));
@@ -290,7 +290,7 @@ void adreno_dump_info(struct msm_gpu *gpu) adreno_gpu->rev.patchid);printk ("fence: %d/%d\n" , adreno_gpu->memptrs->fence, - gpu->fctx->last_fence); + gpu->seqno);printk ("rptr: %d\n" ,get_rptr (adreno_gpu));printk ("rb wptr: %d\n" ,get_wptr (gpu->rb)); }
drivers/gpu/drm/msm/msm_drv.c +14 -5
@@ -510,7 +510,7 @@ static void load_gpu(struct drm_device *dev)mutex_unlock (&init_lock); } -static int context_init (struct drm_file *file) +static int context_init (struct drm_device *dev,struct drm_file *file) {struct msm_file_private *ctx;
@@ -518,7 +518,7 @@ static int context_init(struct drm_file *file)if (!ctx)return -ENOMEM ; -msm_submitqueue_init (ctx); +msm_submitqueue_init (dev, ctx); file->driver_priv = ctx;
@@ -532,7 +532,7 @@ static int msm_open(struct drm_device *dev, struct drm_file *file)*/ load_gpu (dev); -return context_init (file); +return context_init (dev, file); }static void context_close (struct msm_file_private *ctx)
@@ -746,6 +746,8 @@ static int msm_ioctl_wait_fence(struct drm_device *dev, void *data,struct msm_drm_private *priv = dev->dev_private;struct drm_msm_wait_fence *args = data;ktime_t timeout =to_ktime (args->timeout); +struct msm_gpu_submitqueue *queue; +int ret;if (args->pad) {DRM_ERROR ("invalid pad: %08x\n" , args->pad);
@@ -755,7 +757,14 @@ static int msm_ioctl_wait_fence(struct drm_device *dev, void *data,if (!priv->gpu)return 0 ; -return msm_wait_fence (priv->gpu->fctx, args->fence, &timeout, true); + queue =msm_submitqueue_get (file->driver_priv, args->queueid); +if (!queue) +return -ENOENT ; + + ret =msm_wait_fence (queue->fctx, args->fence, &timeout, true); + +msm_submitqueue_put (queue); +return ret; } static int msm_ioctl_gem_madvise(struct drm_device *dev, void *data,
@@ -805,7 +814,7 @@ static int msm_ioctl_submitqueue_new(struct drm_device *dev, void *data,if (args->flags & ~MSM_SUBMITQUEUE_FLAGS )return -EINVAL ; -return msm_submitqueue_create (file->driver_priv, args->prio, +return msm_submitqueue_create (dev, file->driver_priv, args->prio, args->flags, &args->id); }
drivers/gpu/drm/msm/msm_drv.h +3 -3
@@ -314,11 +314,11 @@ void __iomem *msm_ioremap(struct platform_device *pdev, const char *name,u32 msm_readl (const void __iomem *addr);struct msm_gpu_submitqueue ; -int msm_submitqueue_init (struct msm_file_private *ctx); +int msm_submitqueue_init (struct drm_device *drm,struct msm_file_private *ctx);struct msm_gpu_submitqueue *msm_submitqueue_get (struct msm_file_private *ctx,u32 id); -int msm_submitqueue_create (struct msm_file_private *ctx,u32 prio, -u32 flags,u32 *id); +int msm_submitqueue_create (struct drm_device *drm,struct msm_file_private *ctx, +u32 prio,u32 flags,u32 *id);int msm_submitqueue_remove (struct msm_file_private *ctx,u32 id);void msm_submitqueue_close (struct msm_file_private *ctx);
drivers/gpu/drm/msm/msm_fence.c +1 -1
@@ -31,7 +31,7 @@ struct msm_fence_context *return ERR_PTR (-ENOMEM ); fctx->dev = dev; - fctx->name = name; +strncpy (fctx->name, name,sizeof (fctx->name)); fctx->context =dma_fence_context_alloc (1 );init_waitqueue_head (&fctx->event);spin_lock_init (&fctx->spinlock);
drivers/gpu/drm/msm/msm_fence.h +1 -1
@@ -22,7 +22,7 @@struct msm_fence_context {struct drm_device *dev; -const char *name; +char name[32 ];unsigned context;/* last_fence == completed_fence --> no pending work */ uint32_t last_fence;/* last assigned fence */
drivers/gpu/drm/msm/msm_gem.h +1 -0
@@ -141,6 +141,7 @@ struct msm_gem_submit {struct list_head node;/* node in gpu submit_list */ struct list_head bo_list;struct ww_acquire_ctx ticket; +uint32_t seqno;/* Sequence number of the submit on the ring */ struct dma_fence *fence;struct msm_gpu_submitqueue *queue;struct pid *pid;/* submitting process */
drivers/gpu/drm/msm/msm_gem_submit.c +4 -3
@@ -233,7 +233,8 @@ static int submit_fence_sync(struct msm_gem_submit *submit)struct msm_gem_object *msm_obj = submit->bos[i].obj;bool write = submit->bos[i].flags &MSM_SUBMIT_BO_WRITE ; - ret =msm_gem_sync_object (&msm_obj->base, submit->gpu->fctx, write); + ret =msm_gem_sync_object (&msm_obj->base, submit->queue->fctx, + write);if (ret)break ; }
@@ -426,7 +427,7 @@ int msm_ioctl_gem_submit(struct drm_device *dev, void *data,* Wait if the fence is from a foreign context, or if the fence * array contains any fence from a foreign context. */ -if (!dma_fence_match_context (in_fence, gpu->fctx->context)) { +if (!dma_fence_match_context (in_fence, queue->fctx->context)) { ret =dma_fence_wait (in_fence, true);if (ret)return ret;
@@ -531,7 +532,7 @@ int msm_ioctl_gem_submit(struct drm_device *dev, void *data, submit->nr_cmds = i; - submit->fence =msm_fence_alloc (gpu->fctx); + submit->fence =msm_fence_alloc (queue->fctx);if (IS_ERR (submit->fence)) { ret =PTR_ERR (submit->fence); submit->fence =NULL ;
drivers/gpu/drm/msm/msm_gpu.c +21 -15
@@ -221,6 +221,19 @@ int msm_gpu_hw_init(struct msm_gpu *gpu)* Hangcheck detection for locked gpu: */ +static void update_fences (struct msm_gpu *gpu,uint32_t fence) +{ +struct msm_gem_submit *submit; + +list_for_each_entry (submit, &gpu->submit_list, node) { +if (submit->seqno > fence) +break ; + +msm_update_fence (submit->queue->fctx, + submit->fence->seqno); + } +} +static void retire_submits (struct msm_gpu *gpu);static void recover_worker (struct work_struct *work)
@@ -230,13 +243,13 @@ static void recover_worker(struct work_struct *work)struct msm_gem_submit *submit;uint32_t fence = gpu->funcs->last_fence (gpu); -msm_update_fence (gpu->fctx, fence +1 ); +update_fences (gpu, fence +1 );mutex_lock (&dev->struct_mutex);dev_err (dev->dev,"%s: hangcheck recover!\n" , gpu->name);list_for_each_entry (submit, &gpu->submit_list, node) { -if (submit->fence->seqno == (fence +1 )) { +if (submit->seqno == (fence +1 )) {struct task_struct *task;rcu_read_lock ();
@@ -286,7 +299,7 @@ static void hangcheck_handler(unsigned long data)if (fence != gpu->hangcheck_fence) {/* some progress has been made.. ya! */ gpu->hangcheck_fence = fence; - }else if (fence < gpu->fctx->last_fence) { + }else if (fence < gpu->seqno) {/* no progress and not done.. hung! */ gpu->hangcheck_fence = fence; dev_err(dev->dev, "%s: hangcheck detected gpu lockup!\n",
@@ -294,12 +307,12 @@ static void hangcheck_handler(unsigned long data)dev_err (dev->dev,"%s: completed fence: %u\n" , gpu->name, fence);dev_err (dev->dev,"%s: submitted fence: %u\n" , - gpu->name, gpu->fctx->last_fence); + gpu->name, gpu->seqno);queue_work (priv->wq, &gpu->recover_work); }/* if still more pending work, reset the hangcheck timer: */ -if (gpu->fctx->last_fence > gpu->hangcheck_fence) +if (gpu->seqno > gpu->hangcheck_fence)hangcheck_timer_reset (gpu);/* workaround for missing irq: */
@@ -451,7 +464,7 @@ static void retire_worker(struct work_struct *work)struct drm_device *dev = gpu->dev;uint32_t fence = gpu->funcs->last_fence (gpu); -msm_update_fence (gpu->fctx, fence); +update_fences (gpu, fence);mutex_lock (&dev->struct_mutex);retire_submits (gpu);
@@ -480,6 +493,8 @@ void msm_gpu_submit(struct msm_gpu *gpu, struct msm_gem_submit *submit,msm_gpu_hw_init (gpu); + submit->seqno = ++gpu->seqno; +list_add_tail (&submit->node, &gpu->submit_list);msm_rd_dump_submit (submit);
@@ -575,12 +590,6 @@ int msm_gpu_init(struct drm_device *drm, struct platform_device *pdev, gpu->dev = drm; gpu->funcs = funcs; gpu->name = name; - gpu->fctx =msm_fence_context_alloc (drm, name); -if (IS_ERR (gpu->fctx)) { - ret =PTR_ERR (gpu->fctx); - gpu->fctx =NULL ; - goto fail; - }INIT_LIST_HEAD (&gpu->active_list);INIT_WORK (&gpu->retire_work, retire_worker);
@@ -693,7 +702,4 @@ void msm_gpu_cleanup(struct msm_gpu *gpu)msm_gem_put_iova (gpu->rb->bo, gpu->aspace);msm_ringbuffer_destroy (gpu->rb); } - -if (gpu->fctx) -msm_fence_context_free (gpu->fctx); }
drivers/gpu/drm/msm/msm_gpu.h +5 -3
@@ -93,8 +93,8 @@ struct msm_gpu {/* list of GEM active objects: */ struct list_head active_list; -/* fencing: */ -struct msm_fence_context *fctx; +/* The sequence number for ring submissions */ +uint32_t seqno;/* does gpu need hw_init? */ bool needs_hw_init;
@@ -134,7 +134,7 @@ struct msm_gpu {static inline bool msm_gpu_active (struct msm_gpu *gpu) { -return gpu->fctx->last_fence > gpu->funcs->last_fence (gpu); +return gpu->seqno > gpu->funcs->last_fence (gpu); }/* Perf-Counters:
@@ -157,6 +157,8 @@ struct msm_gpu_submitqueue {int faults;struct list_head node;struct kref ref; + +struct msm_fence_context *fctx; };static inline void gpu_write (struct msm_gpu *gpu,u32 reg,u32 data)
drivers/gpu/drm/msm/msm_submitqueue.c +27 -6
@@ -19,6 +19,7 @@ void msm_submitqueue_destroy(struct kref *kref)struct msm_gpu_submitqueue *queue =container_of (kref, struct msm_gpu_submitqueue, ref); +msm_fence_context_free (queue->fctx);kfree (queue); }
@@ -60,16 +61,17 @@ void msm_submitqueue_close(struct msm_file_private *ctx)msm_submitqueue_put (entry); } -int msm_submitqueue_create (struct msm_file_private *ctx,u32 prio,u32 flags, -u32 *id) +int msm_submitqueue_create (struct drm_device *drm,struct msm_file_private *ctx, +u32 prio,u32 flags,u32 *id) {struct msm_gpu_submitqueue *queue; +char name[32 ]; +int ret =0 ;if (!ctx)return -ENODEV ; queue =kzalloc (sizeof (*queue),GFP_KERNEL ); -if (!queue)return -ENOMEM ;
@@ -86,12 +88,31 @@ int msm_submitqueue_create(struct msm_file_private *ctx, u32 prio, u32 flags,list_add_tail (&queue->node, &ctx->submitqueues); +/* +* Get another reference to the queue before releasing the lock +* so it does't get destroyed while we are still setting it up +*/ +kref_get (&queue->ref); +write_unlock (&ctx->queuelock); -return 0 ; +/* FIXME: What should the name be? */ +sprintf (name,"gpu-queue-%d" , queue->id); + +/* Allocate a fence domain for the queue */ + + queue->fctx =msm_fence_context_alloc (drm, name); +if (IS_ERR (queue->fctx)) { + ret =PTR_ERR (queue->fctx); +msm_submitqueue_put (queue); + } + +msm_submitqueue_put (queue); + +return ret; } -int msm_submitqueue_init (struct msm_file_private *ctx) +int msm_submitqueue_init (struct drm_device *drm,struct msm_file_private *ctx) {if (!ctx)return 0 ;
@@ -104,7 +125,7 @@ int msm_submitqueue_init(struct msm_file_private *ctx)* Add the "default" submitqueue with id 0 * "medium" priority (3) and no flags */ -return msm_submitqueue_create (ctx,3 ,0 ,NULL ); +return msm_submitqueue_create (drm, ctx,3 ,0 ,NULL ); }int msm_submitqueue_remove (struct msm_file_private *ctx,u32 id)
include/uapi/drm/msm_drm.h +1 -0
@@ -232,6 +232,7 @@ struct drm_msm_wait_fence {__u32 fence;/* in */ __u32 pad;struct drm_msm_timespec timeout;/* in */ +__u32 queueid;/* in, submitqueue id that owns the fence */ };/* madvise provides a way to tell the kernel in case a buffers contents
Signature · 1 line
1.9.1
PATCH 10/17 drm/msm: Attach the GPU MMU when it is created
In reply to Jordan CrouseCurrently the GPU MMU is attached in the adreno_gpu code but as more and more of the GPU initialization moves to the generic GPU path we have a need to map and use GPU memory earlier and earlier. There isn't any reason to defer attaching the MMU until later so attach it right after the address space is created so it can be used immediately.
--- drivers/gpu/drm/msm/adreno/adreno_gpu.c | 27 ++--------- drivers/gpu/drm/msm/msm_gpu.c | 83 ++++++++++++++++++++++----------- 2 files changed, 61 insertions(+), 49 deletions(-)
drivers/gpu/drm/msm/adreno/adreno_gpu.c +4 -23
@@ -330,11 +330,6 @@ void adreno_wait_ring(struct msm_gpu *gpu, uint32_t ndwords)DRM_ERROR ("%s: timeout waiting for ringbuffer space\n" , gpu->name); } -static const char *iommu_ports[] = { -"gfx3d_user" ,"gfx3d_priv" , -"gfx3d1_user" ,"gfx3d1_priv" , -}; -int adreno_gpu_init (struct drm_device *drm,struct platform_device *pdev,struct adreno_gpu *adreno_gpu,const struct adreno_gpu_funcs *funcs) {
@@ -366,15 +361,15 @@ int adreno_gpu_init(struct drm_device *drm, struct platform_device *pdev, adreno_gpu_config.ringsz =RB_SIZE ; +pm_runtime_set_autosuspend_delay (&pdev->dev,DRM_MSM_INACTIVE_PERIOD ); +pm_runtime_use_autosuspend (&pdev->dev); +pm_runtime_enable (&pdev->dev); + ret =msm_gpu_init (drm, pdev, &adreno_gpu->base, &funcs->base, adreno_gpu->info->name, &adreno_gpu_config);if (ret)return ret; -pm_runtime_set_autosuspend_delay (&pdev->dev,DRM_MSM_INACTIVE_PERIOD ); -pm_runtime_use_autosuspend (&pdev->dev); -pm_runtime_enable (&pdev->dev); - ret =request_firmware (&adreno_gpu->pm4, adreno_gpu->info->pm4fw, drm->dev);if (ret) { dev_err(drm->dev, "failed to load %s PM4 firmware: %d\n",
@@ -389,14 +384,6 @@ int adreno_gpu_init(struct drm_device *drm, struct platform_device *pdev,return ret; } -if (gpu->aspace && gpu->aspace->mmu) { -struct msm_mmu *mmu = gpu->aspace->mmu; - ret = mmu->funcs->attach (mmu, iommu_ports, -ARRAY_SIZE (iommu_ports)); -if (ret) -return ret; - } - adreno_gpu->memptrs_bo =msm_gem_new (drm,sizeof (*adreno_gpu->memptrs),MSM_BO_UNCACHED );if (IS_ERR (adreno_gpu->memptrs_bo)) {
@@ -439,10 +426,4 @@ void adreno_gpu_cleanup(struct adreno_gpu *adreno_gpu)release_firmware (adreno_gpu->pfp);msm_gpu_cleanup (gpu); - -if (gpu->aspace) { - gpu->aspace->mmu->funcs->detach (gpu->aspace->mmu, - iommu_ports,ARRAY_SIZE (iommu_ports)); -msm_gem_address_space_put (gpu->aspace); - } }
drivers/gpu/drm/msm/msm_gpu.c +57 -26
114 lines not on this page · raw message
Signature · 1 line
1.9.1
PATCH 12/17 drm/msm: Move memptrs to msm_gpu
In reply to Jordan CrouseWhen we move to multiple ringbuffers we're going to store the data in the memptrs on a per-ring basis. In order to prepare for that move the current memptrs from the adreno namespace into msm_gpu. This is way cleaner and immediately lets us kill off some sub functions so there is much less cost later when we do move to per-ring structs.
--- drivers/gpu/drm/msm/adreno/a3xx_gpu.c | 1 - drivers/gpu/drm/msm/adreno/a4xx_gpu.c | 1 - drivers/gpu/drm/msm/adreno/a5xx_gpu.c | 6 ++-- drivers/gpu/drm/msm/adreno/adreno_gpu.c | 52 ++++++++------------------------- drivers/gpu/drm/msm/adreno/adreno_gpu.h | 16 ---------- drivers/gpu/drm/msm/msm_gpu.c | 35 ++++++++++++++++++++-- drivers/gpu/drm/msm/msm_gpu.h | 17 +++++++++-- 7 files changed, 61 insertions(+), 67 deletions(-)
drivers/gpu/drm/msm/adreno/a3xx_gpu.c +0 -1
@@ -444,7 +444,6 @@ static void a3xx_dump(struct msm_gpu *gpu) .pm_suspend = msm_gpu_pm_suspend, .pm_resume = msm_gpu_pm_resume, .recover = a3xx_recover, - .last_fence = adreno_last_fence, .submit = adreno_submit, .flush = adreno_flush, .irq = a3xx_irq,
drivers/gpu/drm/msm/adreno/a4xx_gpu.c +0 -1
@@ -532,7 +532,6 @@ static int a4xx_get_timestamp(struct msm_gpu *gpu, uint64_t *value) .pm_suspend = a4xx_pm_suspend, .pm_resume = a4xx_pm_resume, .recover = a4xx_recover, - .last_fence = adreno_last_fence, .submit = adreno_submit, .flush = adreno_flush, .irq = a4xx_irq,
drivers/gpu/drm/msm/adreno/a5xx_gpu.c +2 -4
@@ -83,7 +83,6 @@ static int zap_shader_load_mdt(struct device *dev, const char *fwname)static void a5xx_submit (struct msm_gpu *gpu,struct msm_gem_submit *submit,struct msm_file_private *ctx) { -struct adreno_gpu *adreno_gpu =to_adreno_gpu (gpu);struct msm_drm_private *priv = gpu->dev->dev_private;struct msm_ringbuffer *ring = gpu->rb;unsigned int i, ibs =0 ;
@@ -110,8 +109,8 @@ static void a5xx_submit(struct msm_gpu *gpu, struct msm_gem_submit *submit,OUT_PKT7 (ring,CP_EVENT_WRITE ,4 );OUT_RING (ring,CACHE_FLUSH_TS | (1 <<31 )); -OUT_RING (ring,lower_32_bits (rbmemptr (adreno_gpu, fence))); -OUT_RING (ring,upper_32_bits (rbmemptr (adreno_gpu, fence))); +OUT_RING (ring,lower_32_bits (rbmemptr (gpu, fence))); +OUT_RING (ring,upper_32_bits (rbmemptr (gpu, fence)));OUT_RING (ring, submit->seqno); gpu->funcs->flush (gpu);
@@ -1025,7 +1024,6 @@ static void a5xx_show(struct msm_gpu *gpu, struct seq_file *m) .pm_suspend = a5xx_pm_suspend, .pm_resume = a5xx_pm_resume, .recover = a5xx_recover, - .last_fence = adreno_last_fence, .submit = a5xx_submit, .flush = adreno_flush, .irq = a5xx_irq,
drivers/gpu/drm/msm/adreno/adreno_gpu.c +12 -40
106 lines not on this page · raw message
drivers/gpu/drm/msm/adreno/adreno_gpu.h +0 -16
@@ -82,14 +82,6 @@ struct adreno_info {const struct adreno_info *adreno_info (struct adreno_rev rev); -#define rbmemptr (adreno_gpu, member) \ - ((adreno_gpu)->memptrs_iova + offsetof(struct adreno_rbmemptrs, member)) - -struct adreno_rbmemptrs { -volatile uint32_t rptr; -volatile uint32_t fence; -}; -struct adreno_gpu {struct msm_gpu base;struct adreno_rev rev;
@@ -104,13 +96,6 @@ struct adreno_gpu {/* firmware: */ const struct firmware *pm4, *pfp; -/* ringbuffer rptr/wptr: */ -// TODO should this be in msm_ringbuffer? I think it would be -// different for z180.. -struct adreno_rbmemptrs *memptrs; -struct drm_gem_object *memptrs_bo; -uint64_t memptrs_iova; -/* * Register offsets are different between some GPUs. * GPU specific offsets will be exported by GPU specific
@@ -197,7 +182,6 @@ static inline int adreno_is_a530(struct adreno_gpu *gpu)int adreno_get_param (struct msm_gpu *gpu,uint32_t param,uint64_t *value);int adreno_hw_init (struct msm_gpu *gpu); -uint32_t adreno_last_fence (struct msm_gpu *gpu);void adreno_recover (struct msm_gpu *gpu);void adreno_submit (struct msm_gpu *gpu,struct msm_gem_submit *submit,struct msm_file_private *ctx);
drivers/gpu/drm/msm/msm_gpu.c +32 -3
@@ -241,7 +241,7 @@ static void recover_worker(struct work_struct *work) struct msm_gpu *gpu = container_of(work, struct msm_gpu, recover_work);struct drm_device *dev = gpu->dev;struct msm_gem_submit *submit; -uint32_t fence = gpu->funcs->last_fence (gpu); +uint32_t fence = gpu->memptrs->fence;update_fences (gpu, fence +1 );
@@ -294,7 +294,7 @@ static void hangcheck_handler(unsigned long data)struct msm_gpu *gpu = (struct msm_gpu *)data;struct drm_device *dev = gpu->dev;struct msm_drm_private *priv = dev->dev_private; -uint32_t fence = gpu->funcs->last_fence (gpu); +uint32_t fence = gpu->memptrs->fence;if (fence != gpu->hangcheck_fence) {/* some progress has been made.. ya! */
@@ -462,7 +462,7 @@ static void retire_worker(struct work_struct *work) { struct msm_gpu *gpu = container_of(work, struct msm_gpu, retire_work);struct drm_device *dev = gpu->dev; -uint32_t fence = gpu->funcs->last_fence (gpu); +uint32_t fence = gpu->memptrs->fence;update_fences (gpu, fence);
@@ -698,6 +698,17 @@ int msm_gpu_init(struct drm_device *drm, struct platform_device *pdev, goto fail; } + gpu->memptrs =msm_gem_kernel_new (drm,sizeof (*gpu->memptrs_bo), +MSM_BO_UNCACHED , gpu->aspace, &gpu->memptrs_bo, + &gpu->memptrs_iova); + +if (IS_ERR (gpu->memptrs)) { + ret =PTR_ERR (gpu->memptrs); + gpu->memptrs =NULL ; +dev_err (drm->dev,"could not allocate memptrs: %d\n" , ret); + goto fail; + } +/* Create ringbuffer: */ gpu->rb =msm_ringbuffer_new (gpu, config->ringsz);if (IS_ERR (gpu->rb)) {
@@ -710,6 +721,18 @@ int msm_gpu_init(struct drm_device *drm, struct platform_device *pdev,return 0 ; fail: +if (gpu->memptrs_bo) { +msm_gem_put_vaddr (gpu->memptrs_bo); +msm_gem_put_iova (gpu->memptrs_bo, gpu->aspace); +drm_gem_object_unreference_unlocked (gpu->memptrs_bo); + } + +if (gpu->aspace) { + gpu->aspace->mmu->funcs->detach (gpu->aspace->mmu, +NULL ,0 ); +msm_gem_address_space_put (gpu->aspace); + } +platform_set_drvdata (pdev,NULL );return ret; }
@@ -728,6 +751,12 @@ void msm_gpu_cleanup(struct msm_gpu *gpu)msm_ringbuffer_destroy (gpu->rb); } +if (gpu->memptrs_bo) { +msm_gem_put_vaddr (gpu->memptrs_bo); +msm_gem_put_iova (gpu->memptrs_bo, gpu->aspace); +drm_gem_object_unreference_unlocked (gpu->memptrs_bo); + } +if (gpu->aspace) { gpu->aspace->mmu->funcs->detach (gpu->aspace->mmu,NULL ,0 );
drivers/gpu/drm/msm/msm_gpu.h +15 -2
@@ -59,7 +59,6 @@ struct msm_gpu_funcs { struct msm_file_private *ctx);void (*flush)(struct msm_gpu *gpu);irqreturn_t (*irq)(struct msm_gpu *irq); -uint32_t (*last_fence)(struct msm_gpu *gpu);void (*recover)(struct msm_gpu *gpu);void (*destroy)(struct msm_gpu *gpu);#ifdef CONFIG_DEBUG_FS
@@ -68,6 +67,14 @@ struct msm_gpu_funcs {#endif }; +#define rbmemptr (gpu, member) \ + ((gpu)->memptrs_iova + offsetof(struct msm_rbmemptrs, member)) + +struct msm_rbmemptrs { +volatile uint32_t rptr; +volatile uint32_t fence; +}; +struct msm_gpu {const char *name;struct drm_device *dev;
@@ -130,11 +137,17 @@ struct msm_gpu {struct work_struct recover_work;struct list_head submit_list; + +struct msm_rbmemptrs *memptrs; +struct drm_gem_object *memptrs_bo; +uint64_t memptrs_iova; + + };static inline bool msm_gpu_active (struct msm_gpu *gpu) { -return gpu->seqno > gpu->funcs->last_fence (gpu); +return gpu->seqno > gpu->memptrs->fence; }/* Perf-Counters:
Signature · 1 line
1.9.1
PATCH 14/17 drm/msm: Add a parameter query for the number of ringbuffers
In reply to Jordan CrouseIn order to manage ringbuffer priority to its fullest userspace should know how many ringbuffers it has to work with. Add a parameter to return the number of active rings.
--- drivers/gpu/drm/msm/adreno/adreno_gpu.c | 3 +++ include/uapi/drm/msm_drm.h | 1 + 2 files changed, 4 insertions(+)
drivers/gpu/drm/msm/adreno/adreno_gpu.c +3 -0
@@ -50,6 +50,9 @@ int adreno_get_param(struct msm_gpu *gpu, uint32_t param, uint64_t *value)if (adreno_gpu->funcs->get_timestamp)return adreno_gpu->funcs->get_timestamp (gpu, value);return -EINVAL ; +case MSM_PARAM_NR_RINGS : + *value = gpu->nr_rings; +return 0 ;default :DBG ("%s: invalid param: %u" , gpu->name, param);return -EINVAL ;
include/uapi/drm/msm_drm.h +1 -0
@@ -73,6 +73,7 @@ struct drm_msm_timespec {#define MSM_PARAM_MAX_FREQ 0x04#define MSM_PARAM_TIMESTAMP 0x05#define MSM_PARAM_GMEM_BASE 0x06 +#define MSM_PARAM_NR_RINGS 0x07struct drm_msm_param {__u32 pipe;/* in, MSM_PIPE_x */
Signature · 1 line
1.9.1
PATCH 15/17 drm/msm: Shadow current pointer in the ring until command is complete
In reply to Jordan CrouseAdd a shadow pointer to track the current command being written into the ring. Don't commit it as 'cur' until the command is submitted. Because 'cur' is used to construct the software copy of the wptr this ensures that somebody peeking in on the ring doesn't assume that a command is inflight while it is being written. This isn't a huge deal with a single ring (though technically the hangcheck could assume the system is prematurely busy when it isn't) but it will be rather important for preemption where the decision to preempt is based on a non-empty ringbuffer. Without a shadow an aggressive preemption scheme could assume that the ringbuffer is non empty and switch to it before the CPU is done writing the command and boom.
Even though preemption won't be supported for all targets because of the way the code is organized it is simpler to make this generic for all targets. The extra load for non-preemption targets should be minimal.
--- drivers/gpu/drm/msm/adreno/adreno_gpu.c | 9 +++++++-- drivers/gpu/drm/msm/msm_ringbuffer.c | 1 + drivers/gpu/drm/msm/msm_ringbuffer.h | 12 ++++++++---- 3 files changed, 16 insertions(+), 6 deletions(-)
drivers/gpu/drm/msm/adreno/adreno_gpu.c +7 -2
@@ -82,6 +82,7 @@ int adreno_hw_init(struct msm_gpu *gpu) } ring->cur = ring->start; + ring->next = ring->start;/* reset completed fence seqno: */ ring->memptrs->fence = ring->seqno;
@@ -226,12 +227,15 @@ void adreno_flush(struct msm_gpu *gpu, struct msm_ringbuffer *ring)struct adreno_gpu *adreno_gpu =to_adreno_gpu (gpu);uint32_t wptr; +/* Copy the shadow to the actual register */ + ring->cur = ring->next; +/* * Mask wptr value that we calculate to fit in the HW range. This is * to account for the possibility that the last command fit exactly into * the ringbuffer and rb->next hasn't wrapped to zero yet */ - wptr =get_wptr (ring) % (MSM_GPU_RINGBUFFER_SZ >>2 ); + wptr = (ring->cur - ring->start) % (MSM_GPU_RINGBUFFER_SZ >>2 );/* ensure writes to ringbuffer have hit system memory: */ mb ();
@@ -343,7 +347,8 @@ static uint32_t ring_freewords(struct msm_ringbuffer *ring) {struct adreno_gpu *adreno_gpu =to_adreno_gpu (ring->gpu);uint32_t size =MSM_GPU_RINGBUFFER_SZ >>2 ; -uint32_t wptr =get_wptr (ring); +/* Use ring->next to calculate free size */ +uint32_t wptr = ring->next - ring->start;uint32_t rptr =get_rptr (adreno_gpu, ring);return (rptr + (size -1 ) - wptr) % size; }
drivers/gpu/drm/msm/msm_ringbuffer.c +1 -0
@@ -45,6 +45,7 @@ struct msm_ringbuffer *msm_ringbuffer_new(struct msm_gpu *gpu, int id, goto fail; } ring->end = ring->start + (MSM_GPU_RINGBUFFER_SZ >>2 ); + ring->next = ring->start; ring->cur = ring->start; ring->memptrs = memptrs;
drivers/gpu/drm/msm/msm_ringbuffer.h +8 -4
@@ -32,7 +32,7 @@ struct msm_ringbuffer {struct msm_gpu *gpu;int id;struct drm_gem_object *bo; -uint32_t *start, *end, *cur; +uint32_t *start, *end, *cur, *next;struct list_head submits;uint64_t iova;uint32_t seqno;
@@ -50,9 +50,13 @@ struct msm_ringbuffer *msm_ringbuffer_new(struct msm_gpu *gpu, int id,static inline void OUT_RING (struct msm_ringbuffer *ring,uint32_t data) { -if (ring->cur == ring->end) - ring->cur = ring->start; - *(ring->cur++) = data; +/* +* ring->next points to the current command being written - it won't be +* committed as ring->cur until the flush +*/ +if (ring->next == ring->end) + ring->next = ring->start; + *(ring->next++) = data; }#endif /* __MSM_RINGBUFFER_H__ */
Signature · 1 line
1.9.1
PATCH 17/17 drm/msm: Implement preemption for A5XX targets
In reply to Jordan CrouseImplement preemption for A5XX targets - this allows multiple ringbuffers for different priorities with automatic preemption of a lower priority ringbuffer if a higher one is ready.
--- drivers/gpu/drm/msm/Makefile | 1 + drivers/gpu/drm/msm/adreno/a5xx_gpu.c | 176 +++++++++++++++++- drivers/gpu/drm/msm/adreno/a5xx_gpu.h | 101 +++++++++- drivers/gpu/drm/msm/adreno/a5xx_preempt.c | 294 ++++++++++++++++++++++++++++++ drivers/gpu/drm/msm/adreno/adreno_gpu.c | 14 +- drivers/gpu/drm/msm/adreno/adreno_gpu.h | 7 +- drivers/gpu/drm/msm/msm_drv.h | 2 +- drivers/gpu/drm/msm/msm_gpu.c | 5 +- drivers/gpu/drm/msm/msm_ringbuffer.c | 1 + drivers/gpu/drm/msm/msm_ringbuffer.h | 1 + drivers/gpu/drm/msm/msm_submitqueue.c | 4 +- 11 files changed, 584 insertions(+), 22 deletions(-) create mode 100644 drivers/gpu/drm/msm/adreno/a5xx_preempt.c
drivers/gpu/drm/msm/Makefile +1 -0
@@ -8,6 +8,7 @@ msm-y := \ adreno/a4xx_gpu.o \ adreno/a5xx_gpu.o \ adreno/a5xx_power.o \ + adreno/a5xx_preempt.o \ hdmi/hdmi.o \ hdmi/hdmi_audio.o \ hdmi/hdmi_bridge.o \
drivers/gpu/drm/msm/adreno/a5xx_gpu.c +171 -5
270 lines not on this page · raw message
drivers/gpu/drm/msm/adreno/a5xx_gpu.h +100 -1
118 lines not on this page · raw message
drivers/gpu/drm/msm/adreno/a5xx_preempt.c new file +294 -0
294 lines not on this page · raw message
drivers/gpu/drm/msm/adreno/adreno_gpu.c +5 -9
@@ -111,11 +111,6 @@ int adreno_hw_init(struct msm_gpu *gpu)return 0 ; } -static uint32_t get_wptr (struct msm_ringbuffer *ring) -{ -return ring->cur - ring->start; -} -/* Use this helper to read rptr, since a430 doesn't update rptr in memory */ static uint32_t get_rptr (struct adreno_gpu *adreno_gpu,struct msm_ringbuffer *ring)
@@ -170,7 +165,7 @@ void adreno_submit(struct msm_gpu *gpu, struct msm_gem_submit *submit,case MSM_SUBMIT_CMD_BUF :OUT_PKT3 (ring,adreno_is_a430 (adreno_gpu) ?CP_INDIRECT_BUFFER_PFE :CP_INDIRECT_BUFFER_PFD ,2 ); -OUT_RING (ring, submit->cmd[i].iova); +OUT_RING (ring,lower_32_bits (submit->cmd[i].iova));OUT_RING (ring, submit->cmd[i].size);OUT_PKT2 (ring);break ;
@@ -237,7 +232,7 @@ void adreno_flush(struct msm_gpu *gpu, struct msm_ringbuffer *ring)* to account for the possibility that the last command fit exactly into * the ringbuffer and rb->next hasn't wrapped to zero yet */ - wptr = (ring->cur - ring->start) % (MSM_GPU_RINGBUFFER_SZ >>2 ); + wptr =get_wptr (ring);/* ensure writes to ringbuffer have hit system memory: */ mb ();
@@ -255,8 +250,9 @@ bool adreno_idle(struct msm_gpu *gpu, struct msm_ringbuffer *ring)return true;/* TODO maybe we need to reset GPU here to recover from hang? */ -DRM_ERROR ("%s: timeout waiting to drain ringbuffer %d!\n" , gpu->name, - ring->id); +DRM_ERROR ("%s: timeout waiting to drain ringbuffer %d rptr/wptr = %X/%X\n" , + gpu->name, ring->id,get_rptr (adreno_gpu, ring), wptr); +return false; }
drivers/gpu/drm/msm/adreno/adreno_gpu.h +6 -1
@@ -2,7 +2,7 @@* Copyright (C) 2013 Red Hat * Author: Rob Clark <[email protected]> * -* Copyright (c) 2014 The Linux Foundation. All rights reserved. +* Copyright (c) 2014,2017 The Linux Foundation. All rights reserved. * * This program is free software; you can redistribute it and/or modify it * under the terms of the GNU General Public License version 2 as published by
@@ -309,6 +309,11 @@ static inline void adreno_gpu_write64(struct adreno_gpu *gpu,adreno_gpu_write (gpu, hi,upper_32_bits (data)); } +static inline uint32_t get_wptr (struct msm_ringbuffer *ring) +{ +return (ring->cur - ring->start) % (MSM_GPU_RINGBUFFER_SZ >>2 ); +} +/* * Given a register and a count, return a value to program into * REG_CP_PROTECT_REG(n) - this will block both reads and writes for _len
drivers/gpu/drm/msm/msm_drv.h +1 -1
@@ -76,7 +76,7 @@ struct msm_vblank_ctrl {spinlock_t lock; }; -#define MSM_GPU_MAX_RINGS 1 +#define MSM_GPU_MAX_RINGS 4struct msm_drm_private {
drivers/gpu/drm/msm/msm_gpu.c +2 -3
@@ -295,8 +295,7 @@ static void recover_worker(struct work_struct *work)* Replay all remaining submits starting with highest priority * ring */ - -for (i = gpu->nr_rings -1 ; i >=0 ; i--) { +for (i =0 ; i < gpu->nr_rings; i++) {struct msm_ringbuffer *ring = gpu->rb[i];list_for_each_entry (submit, &ring->submits, node)
@@ -476,7 +475,7 @@ static void retire_submits(struct msm_gpu *gpu)WARN_ON (!mutex_is_locked (&dev->struct_mutex));/* Retire the commits starting with highest priority */ -for (i = gpu->nr_rings -1 ; i >=0 ; i--) { +for (i =0 ; i < gpu->nr_rings; i++) {struct msm_ringbuffer *ring = gpu->rb[i];list_for_each_entry_safe (submit, tmp, &ring->submits, node) {
drivers/gpu/drm/msm/msm_ringbuffer.c +1 -0
@@ -52,6 +52,7 @@ struct msm_ringbuffer *msm_ringbuffer_new(struct msm_gpu *gpu, int id, ring->memptrs_iova = memptrs_iova;INIT_LIST_HEAD (&ring->submits); +spin_lock_init (&ring->lock);return ring;
drivers/gpu/drm/msm/msm_ringbuffer.h +1 -0
@@ -39,6 +39,7 @@ struct msm_ringbuffer {uint32_t hangcheck_fence;struct msm_rbmemptrs *memptrs;uint64_t memptrs_iova; +spinlock_t lock; }; struct msm_ringbuffer *msm_ringbuffer_new(struct msm_gpu *gpu, int id,
drivers/gpu/drm/msm/msm_submitqueue.c +2 -2
@@ -123,9 +123,9 @@ int msm_submitqueue_init(struct drm_device *drm, struct msm_file_private *ctx)/* * Add the "default" submitqueue with id 0 -* "medium" priority (3) and no flags +* "low" priority (2) and no flags */ -return msm_submitqueue_create (drm, ctx,3 ,0 ,NULL ); +return msm_submitqueue_create (drm, ctx,2 ,0 ,NULL ); }int msm_submitqueue_remove (struct msm_file_private *ctx,u32 id)
Signature · 1 line
1.9.1