[PATCH 00/10] drm/msm: some stuff I'm working on for 4.8

[PATCH 00/10] drm/msm: some stuff I'm working on for 4.8

[Rob Clark]

Also working on fence-fd support for submit ioctl, but that is
depending on some other patches from Gustavo, and not so much
actually tested yet, so unlikely to be 4.8 material.  But I'll
send an RFC at least in near future.

Main interesting thing here is, I think, shrinker.  Currently
it is limited to purging madvise'd buffers (userspace bo cache)
and vmaps.  Maybe someday I'll setup a swap partition and take
this further, but the more important things to solve in the
short term are keeping the userspace bo-cache from causing
problems on devices w/ less memory (<= 1G) and keeping vmap's
from being a problem on 32b systems.

Rob Clark (10):
  drm/msm: use mutex_lock_interruptible for submit ioctl
  drm/msm: add madvise ioctl
  drm/msm: add put_iova() helper
  drm/msm: shrinker support
  drm/msm: change gem->vmap() to get/put
  drm/msm: wire up vmap shrinker
  drm/msm: deal with arbitrary # of cmd buffers
  drm/msm: bump kernel api version
  drm/msm/rd: split out snapshot_buf helper
  drm/msm/rd: add module param to dump all bo's

 drivers/gpu/drm/msm/Makefile            |   1 +
 drivers/gpu/drm/msm/adreno/adreno_gpu.c |   6 +-
 drivers/gpu/drm/msm/dsi/dsi_host.c      |   5 +-
 drivers/gpu/drm/msm/msm_drv.c           |  59 ++++++++++-
 drivers/gpu/drm/msm/msm_drv.h           |  17 +++-
 drivers/gpu/drm/msm/msm_fbdev.c         |   3 +-
 drivers/gpu/drm/msm/msm_gem.c           | 131 ++++++++++++++++++++++---
 drivers/gpu/drm/msm/msm_gem.h           |  23 ++++-
 drivers/gpu/drm/msm/msm_gem_prime.c     |   4 +-
 drivers/gpu/drm/msm/msm_gem_shrinker.c  | 168 ++++++++++++++++++++++++++++++++
 drivers/gpu/drm/msm/msm_gem_submit.c    |  26 +++--
 drivers/gpu/drm/msm/msm_rd.c            |  66 ++++++++++---
 drivers/gpu/drm/msm/msm_ringbuffer.c    |   6 +-
 include/uapi/drm/msm_drm.h              |  25 ++++-
 14 files changed, 485 insertions(+), 55 deletions(-)
 create mode 100644 drivers/gpu/drm/msm/msm_gem_shrinker.c

-- 
2.5.5

[PATCH 01/10] drm/msm: use mutex_lock_interruptible for submit ioctl

[Rob Clark]

Be kinder to things that do lots of signal handling (ie. Xorg)

Signed-off-by: Rob Clark <robdclark@gmail.com>
---
 drivers/gpu/drm/msm/msm_gem_submit.c | 13 +++++++++----
 1 file changed, 9 insertions(+), 4 deletions(-)

diff --git a/drivers/gpu/drm/msm/msm_gem_submit.c b/drivers/gpu/drm/msm/msm_gem_submit.c
index eb4bb8b..ea5f709 100644
--- a/drivers/gpu/drm/msm/msm_gem_submit.c
+++ b/drivers/gpu/drm/msm/msm_gem_submit.c
@@ -372,11 +372,15 @@ int msm_ioctl_gem_submit(struct drm_device *dev, void *data,
 	if (args->nr_cmds > MAX_CMDS)
 		return -EINVAL;
 
-	submit = submit_create(dev, gpu, args->nr_bos);
-	if (!submit)
-		return -ENOMEM;
+	ret = mutex_lock_interruptible(&dev->struct_mutex);
+	if (ret)
+		return ret;
 
-	mutex_lock(&dev->struct_mutex);
+	submit = submit_create(dev, gpu, args->nr_bos);
+	if (!submit) {
+		ret = -ENOMEM;
+		goto out_unlock;
+	}
 
 	ret = submit_lookup_objects(submit, args, file);
 	if (ret)
@@ -462,6 +466,7 @@ out:
 	submit_cleanup(submit);
 	if (ret)
 		msm_gem_submit_free(submit);
+out_unlock:
 	mutex_unlock(&dev->struct_mutex);
 	return ret;
 }
-- 
2.5.5

[PATCH 02/10] drm/msm: add madvise ioctl

[Rob Clark]

Doesn't do anything too interesting until we wire up shrinker.  Pretty
much lifted from i915.

Signed-off-by: Rob Clark <robdclark@gmail.com>
---
 drivers/gpu/drm/msm/msm_drv.c | 39 +++++++++++++++++++++++++++++++++++++++
 drivers/gpu/drm/msm/msm_drv.h |  1 +
 drivers/gpu/drm/msm/msm_gem.c | 35 +++++++++++++++++++++++++++++++++--
 drivers/gpu/drm/msm/msm_gem.h |  5 +++++
 include/uapi/drm/msm_drm.h    | 25 ++++++++++++++++++++++++-
 5 files changed, 102 insertions(+), 3 deletions(-)

diff --git a/drivers/gpu/drm/msm/msm_drv.c b/drivers/gpu/drm/msm/msm_drv.c
index 9c65409..ddf6878 100644
--- a/drivers/gpu/drm/msm/msm_drv.c
+++ b/drivers/gpu/drm/msm/msm_drv.c
@@ -690,6 +690,44 @@ static int msm_ioctl_wait_fence(struct drm_device *dev, void *data,
 	return msm_wait_fence(priv->gpu->fctx, args->fence, &timeout, true);
 }
 
+static int msm_ioctl_gem_madvise(struct drm_device *dev, void *data,
+		struct drm_file *file)
+{
+	struct drm_msm_gem_madvise *args = data;
+	struct drm_gem_object *obj;
+	int ret;
+
+	switch (args->madv) {
+	case MSM_MADV_DONTNEED:
+	case MSM_MADV_WILLNEED:
+		break;
+	default:
+		return -EINVAL;
+	}
+
+	ret = mutex_lock_interruptible(&dev->struct_mutex);
+	if (ret)
+		return ret;
+
+	obj = drm_gem_object_lookup(file, args->handle);
+	if (!obj) {
+		ret = -ENOENT;
+		goto unlock;
+	}
+
+	ret = msm_gem_madvise(obj, args->madv);
+	if (ret >= 0) {
+		args->retained = ret;
+		ret = 0;
+	}
+
+	drm_gem_object_unreference(obj);
+
+unlock:
+	mutex_unlock(&dev->struct_mutex);
+	return ret;
+}
+
 static const struct drm_ioctl_desc msm_ioctls[] = {
 	DRM_IOCTL_DEF_DRV(MSM_GET_PARAM,    msm_ioctl_get_param,    DRM_AUTH|DRM_RENDER_ALLOW),
 	DRM_IOCTL_DEF_DRV(MSM_GEM_NEW,      msm_ioctl_gem_new,      DRM_AUTH|DRM_RENDER_ALLOW),
@@ -698,6 +736,7 @@ static const struct drm_ioctl_desc msm_ioctls[] = {
 	DRM_IOCTL_DEF_DRV(MSM_GEM_CPU_FINI, msm_ioctl_gem_cpu_fini, DRM_AUTH|DRM_RENDER_ALLOW),
 	DRM_IOCTL_DEF_DRV(MSM_GEM_SUBMIT,   msm_ioctl_gem_submit,   DRM_AUTH|DRM_RENDER_ALLOW),
 	DRM_IOCTL_DEF_DRV(MSM_WAIT_FENCE,   msm_ioctl_wait_fence,   DRM_AUTH|DRM_RENDER_ALLOW),
+	DRM_IOCTL_DEF_DRV(MSM_GEM_MADVISE,  msm_ioctl_gem_madvise,  DRM_AUTH|DRM_RENDER_ALLOW),
 };
 
 static const struct vm_operations_struct vm_ops = {
diff --git a/drivers/gpu/drm/msm/msm_drv.h b/drivers/gpu/drm/msm/msm_drv.h
index 5b2963f..f757ade 100644
--- a/drivers/gpu/drm/msm/msm_drv.h
+++ b/drivers/gpu/drm/msm/msm_drv.h
@@ -191,6 +191,7 @@ int msm_gem_prime_pin(struct drm_gem_object *obj);
 void msm_gem_prime_unpin(struct drm_gem_object *obj);
 void *msm_gem_vaddr_locked(struct drm_gem_object *obj);
 void *msm_gem_vaddr(struct drm_gem_object *obj);
+int msm_gem_madvise(struct drm_gem_object *obj, unsigned madv);
 int msm_gem_sync_object(struct drm_gem_object *obj,
 		struct msm_fence_context *fctx, bool exclusive);
 void msm_gem_move_to_active(struct drm_gem_object *obj,
diff --git a/drivers/gpu/drm/msm/msm_gem.c b/drivers/gpu/drm/msm/msm_gem.c
index 69836f5..c40db08 100644
--- a/drivers/gpu/drm/msm/msm_gem.c
+++ b/drivers/gpu/drm/msm/msm_gem.c
@@ -413,6 +413,21 @@ void *msm_gem_vaddr(struct drm_gem_object *obj)
 	return ret;
 }
 
+/* Update madvise status, returns true if not purged, else
+ * false or -errno.
+ */
+int msm_gem_madvise(struct drm_gem_object *obj, unsigned madv)
+{
+	struct msm_gem_object *msm_obj = to_msm_bo(obj);
+
+	WARN_ON(!mutex_is_locked(&obj->dev->struct_mutex));
+
+	if (msm_obj->madv != __MSM_MADV_PURGED)
+		msm_obj->madv = madv;
+
+	return (msm_obj->madv != __MSM_MADV_PURGED);
+}
+
 /* must be called before _move_to_active().. */
 int msm_gem_sync_object(struct drm_gem_object *obj,
 		struct msm_fence_context *fctx, bool exclusive)
@@ -464,6 +479,7 @@ void msm_gem_move_to_active(struct drm_gem_object *obj,
 		struct msm_gpu *gpu, bool exclusive, struct fence *fence)
 {
 	struct msm_gem_object *msm_obj = to_msm_bo(obj);
+	WARN_ON(msm_obj->madv != MSM_MADV_WILLNEED);
 	msm_obj->gpu = gpu;
 	if (exclusive)
 		reservation_object_add_excl_fence(msm_obj->resv, fence);
@@ -532,13 +548,27 @@ void msm_gem_describe(struct drm_gem_object *obj, struct seq_file *m)
 	struct reservation_object_list *fobj;
 	struct fence *fence;
 	uint64_t off = drm_vma_node_start(&obj->vma_node);
+	const char *madv;
 
 	WARN_ON(!mutex_is_locked(&obj->dev->struct_mutex));
 
-	seq_printf(m, "%08x: %c %2d (%2d) %08llx %p %zu\n",
+	switch (msm_obj->madv) {
+	case __MSM_MADV_PURGED:
+		madv = " purged";
+		break;
+	case MSM_MADV_DONTNEED:
+		madv = " purgeable";
+		break;
+	case MSM_MADV_WILLNEED:
+	default:
+		madv = "";
+		break;
+	}
+
+	seq_printf(m, "%08x: %c %2d (%2d) %08llx %p %zu%s\n",
 			msm_obj->flags, is_active(msm_obj) ? 'A' : 'I',
 			obj->name, obj->refcount.refcount.counter,
-			off, msm_obj->vaddr, obj->size);
+			off, msm_obj->vaddr, obj->size, madv);
 
 	rcu_read_lock();
 	fobj = rcu_dereference(robj->fence);
@@ -688,6 +718,7 @@ static int msm_gem_new_impl(struct drm_device *dev,
 		msm_obj->vram_node = (void *)&msm_obj[1];
 
 	msm_obj->flags = flags;
+	msm_obj->madv = MSM_MADV_WILLNEED;
 
 	if (resv) {
 		msm_obj->resv = resv;
diff --git a/drivers/gpu/drm/msm/msm_gem.h b/drivers/gpu/drm/msm/msm_gem.h
index 9facd4b..fa8e1f1 100644
--- a/drivers/gpu/drm/msm/msm_gem.h
+++ b/drivers/gpu/drm/msm/msm_gem.h
@@ -29,6 +29,11 @@ struct msm_gem_object {
 
 	uint32_t flags;
 
+	/**
+	 * Advice: are the backing pages purgeable?
+	 */
+	uint8_t madv;
+
 	/* And object is either:
 	 *  inactive - on priv->inactive_list
 	 *  active   - on one one of the gpu's active_list..  well, at
diff --git a/include/uapi/drm/msm_drm.h b/include/uapi/drm/msm_drm.h
index bf19d2c..49f778d 100644
--- a/include/uapi/drm/msm_drm.h
+++ b/include/uapi/drm/msm_drm.h
@@ -201,6 +201,27 @@ struct drm_msm_wait_fence {
 	struct drm_msm_timespec timeout;   /* in */
 };
 
+/* madvise provides a way to tell the kernel in case a buffers contents
+ * can be discarded under memory pressure, which is useful for userspace
+ * bo cache where we want to optimistically hold on to buffer allocate
+ * and potential mmap, but allow the pages to be discarded under memory
+ * pressure.
+ *
+ * Typical usage would involve madvise(DONTNEED) when buffer enters BO
+ * cache, and madvise(WILLNEED) if trying to recycle buffer from BO cache.
+ * In the WILLNEED case, 'retained' indicates to userspace whether the
+ * backing pages still exist.
+ */
+#define MSM_MADV_WILLNEED 0       /* backing pages are needed, status returned in 'retained' */
+#define MSM_MADV_DONTNEED 1       /* backing pages not needed */
+#define __MSM_MADV_PURGED 2       /* internal state */
+
+struct drm_msm_gem_madvise {
+	__u32 handle;         /* in, GEM handle */
+	__u32 madv;           /* in, MSM_MADV_x */
+	__u32 retained;       /* out, whether backing store still exists */
+};
+
 #define DRM_MSM_GET_PARAM              0x00
 /* placeholder:
 #define DRM_MSM_SET_PARAM              0x01
@@ -211,7 +232,8 @@ struct drm_msm_wait_fence {
 #define DRM_MSM_GEM_CPU_FINI           0x05
 #define DRM_MSM_GEM_SUBMIT             0x06
 #define DRM_MSM_WAIT_FENCE             0x07
-#define DRM_MSM_NUM_IOCTLS             0x08
+#define DRM_MSM_GEM_MADVISE            0x08
+#define DRM_MSM_NUM_IOCTLS             0x09
 
 #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)
@@ -220,6 +242,7 @@ struct drm_msm_wait_fence {
 #define DRM_IOCTL_MSM_GEM_CPU_FINI     DRM_IOW (DRM_COMMAND_BASE + DRM_MSM_GEM_CPU_FINI, struct drm_msm_gem_cpu_fini)
 #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)
 
 #if defined(__cplusplus)
 }
-- 
2.5.5

[PATCH 03/10] drm/msm: add put_iova() helper

[Rob Clark]

We'll need this too for shrinker/purging.

Signed-off-by: Rob Clark <robdclark@gmail.com>
---
 drivers/gpu/drm/msm/msm_gem.c | 30 +++++++++++++++++++++---------
 1 file changed, 21 insertions(+), 9 deletions(-)

diff --git a/drivers/gpu/drm/msm/msm_gem.c b/drivers/gpu/drm/msm/msm_gem.c
index c40db08..2636c27 100644
--- a/drivers/gpu/drm/msm/msm_gem.c
+++ b/drivers/gpu/drm/msm/msm_gem.c
@@ -276,6 +276,26 @@ uint64_t msm_gem_mmap_offset(struct drm_gem_object *obj)
 	return offset;
 }
 
+static void
+put_iova(struct drm_gem_object *obj)
+{
+	struct drm_device *dev = obj->dev;
+	struct msm_drm_private *priv = obj->dev->dev_private;
+	struct msm_gem_object *msm_obj = to_msm_bo(obj);
+	int id;
+
+	WARN_ON(!mutex_is_locked(&dev->struct_mutex));
+
+	for (id = 0; id < ARRAY_SIZE(msm_obj->domain); id++) {
+		struct msm_mmu *mmu = priv->mmus[id];
+		if (mmu && msm_obj->domain[id].iova) {
+			uint32_t offset = msm_obj->domain[id].iova;
+			mmu->funcs->unmap(mmu, offset, msm_obj->sgt, obj->size);
+			msm_obj->domain[id].iova = 0;
+		}
+	}
+}
+
 /* should be called under struct_mutex.. although it can be called
  * from atomic context without struct_mutex to acquire an extra
  * iova ref if you know one is already held.
@@ -608,9 +628,7 @@ void msm_gem_describe_objects(struct list_head *list, struct seq_file *m)
 void msm_gem_free_object(struct drm_gem_object *obj)
 {
 	struct drm_device *dev = obj->dev;
-	struct msm_drm_private *priv = obj->dev->dev_private;
 	struct msm_gem_object *msm_obj = to_msm_bo(obj);
-	int id;
 
 	WARN_ON(!mutex_is_locked(&dev->struct_mutex));
 
@@ -619,13 +637,7 @@ void msm_gem_free_object(struct drm_gem_object *obj)
 
 	list_del(&msm_obj->mm_list);
 
-	for (id = 0; id < ARRAY_SIZE(msm_obj->domain); id++) {
-		struct msm_mmu *mmu = priv->mmus[id];
-		if (mmu && msm_obj->domain[id].iova) {
-			uint32_t offset = msm_obj->domain[id].iova;
-			mmu->funcs->unmap(mmu, offset, msm_obj->sgt, obj->size);
-		}
-	}
+	put_iova(obj);
 
 	if (obj->import_attach) {
 		if (msm_obj->vaddr)
-- 
2.5.5

[PATCH 04/10] drm/msm: shrinker support

[Rob Clark]

For a first step, only purge obj->madv==DONTNEED objects.  We could be
more agressive and next try unpinning inactive objects..  but that is
only useful if you have swap.

Signed-off-by: Rob Clark <robdclark@gmail.com>
---
 drivers/gpu/drm/msm/Makefile           |   1 +
 drivers/gpu/drm/msm/msm_drv.c          |   5 ++
 drivers/gpu/drm/msm/msm_drv.h          |   8 +++
 drivers/gpu/drm/msm/msm_gem.c          |  32 +++++++++
 drivers/gpu/drm/msm/msm_gem.h          |   6 ++
 drivers/gpu/drm/msm/msm_gem_shrinker.c | 128 +++++++++++++++++++++++++++++++++
 6 files changed, 180 insertions(+)
 create mode 100644 drivers/gpu/drm/msm/msm_gem_shrinker.c

diff --git a/drivers/gpu/drm/msm/Makefile b/drivers/gpu/drm/msm/Makefile
index 60cb026..2bb2c4b 100644
--- a/drivers/gpu/drm/msm/Makefile
+++ b/drivers/gpu/drm/msm/Makefile
@@ -45,6 +45,7 @@ msm-y := \
 	msm_fence.o \
 	msm_gem.o \
 	msm_gem_prime.o \
+	msm_gem_shrinker.o \
 	msm_gem_submit.o \
 	msm_gpu.o \
 	msm_iommu.o \
diff --git a/drivers/gpu/drm/msm/msm_drv.c b/drivers/gpu/drm/msm/msm_drv.c
index ddf6878..3e15a50 100644
--- a/drivers/gpu/drm/msm/msm_drv.c
+++ b/drivers/gpu/drm/msm/msm_drv.c
@@ -195,6 +195,8 @@ static int msm_drm_uninit(struct device *dev)
 		kfree(vbl_ev);
 	}
 
+	msm_gem_shrinker_cleanup(ddev);
+
 	drm_kms_helper_poll_fini(ddev);
 
 	drm_connector_unregister_all(ddev);
@@ -352,6 +354,7 @@ static int msm_drm_init(struct device *dev, struct drm_driver *drv)
 	}
 
 	ddev->dev_private = priv;
+	priv->dev = ddev;
 
 	priv->wq = alloc_ordered_workqueue("msm", 0);
 	priv->atomic_wq = alloc_ordered_workqueue("msm:atomic", 0);
@@ -376,6 +379,8 @@ static int msm_drm_init(struct device *dev, struct drm_driver *drv)
 	if (ret)
 		goto fail;
 
+	msm_gem_shrinker_init(ddev);
+
 	switch (get_mdp_ver(pdev)) {
 	case 4:
 		kms = mdp4_kms_init(ddev);
diff --git a/drivers/gpu/drm/msm/msm_drv.h b/drivers/gpu/drm/msm/msm_drv.h
index f757ade..bcf33cf 100644
--- a/drivers/gpu/drm/msm/msm_drv.h
+++ b/drivers/gpu/drm/msm/msm_drv.h
@@ -77,6 +77,8 @@ struct msm_vblank_ctrl {
 
 struct msm_drm_private {
 
+	struct drm_device *dev;
+
 	struct msm_kms *kms;
 
 	/* subordinate devices, if present: */
@@ -147,6 +149,8 @@ struct msm_drm_private {
 		struct drm_mm mm;
 	} vram;
 
+	struct shrinker shrinker;
+
 	struct msm_vblank_ctrl vblank_ctrl;
 };
 
@@ -165,6 +169,9 @@ void msm_gem_submit_free(struct msm_gem_submit *submit);
 int msm_ioctl_gem_submit(struct drm_device *dev, void *data,
 		struct drm_file *file);
 
+void msm_gem_shrinker_init(struct drm_device *dev);
+void msm_gem_shrinker_cleanup(struct drm_device *dev);
+
 int msm_gem_mmap_obj(struct drm_gem_object *obj,
 			struct vm_area_struct *vma);
 int msm_gem_mmap(struct file *filp, struct vm_area_struct *vma);
@@ -192,6 +199,7 @@ void msm_gem_prime_unpin(struct drm_gem_object *obj);
 void *msm_gem_vaddr_locked(struct drm_gem_object *obj);
 void *msm_gem_vaddr(struct drm_gem_object *obj);
 int msm_gem_madvise(struct drm_gem_object *obj, unsigned madv);
+void msm_gem_purge(struct drm_gem_object *obj);
 int msm_gem_sync_object(struct drm_gem_object *obj,
 		struct msm_fence_context *fctx, bool exclusive);
 void msm_gem_move_to_active(struct drm_gem_object *obj,
diff --git a/drivers/gpu/drm/msm/msm_gem.c b/drivers/gpu/drm/msm/msm_gem.c
index 2636c27..444d0b5 100644
--- a/drivers/gpu/drm/msm/msm_gem.c
+++ b/drivers/gpu/drm/msm/msm_gem.c
@@ -448,6 +448,38 @@ int msm_gem_madvise(struct drm_gem_object *obj, unsigned madv)
 	return (msm_obj->madv != __MSM_MADV_PURGED);
 }
 
+void msm_gem_purge(struct drm_gem_object *obj)
+{
+	struct drm_device *dev = obj->dev;
+	struct msm_gem_object *msm_obj = to_msm_bo(obj);
+
+	WARN_ON(!mutex_is_locked(&dev->struct_mutex));
+	WARN_ON(!is_purgeable(msm_obj));
+	WARN_ON(obj->import_attach);
+
+	put_iova(obj);
+
+	vunmap(msm_obj->vaddr);
+	msm_obj->vaddr = NULL;
+
+	put_pages(obj);
+
+	msm_obj->madv = __MSM_MADV_PURGED;
+
+	drm_vma_node_unmap(&obj->vma_node, dev->anon_inode->i_mapping);
+	drm_gem_free_mmap_offset(obj);
+
+	/* Our goal here is to return as much of the memory as
+	 * is possible back to the system as we are called from OOM.
+	 * To do this we must instruct the shmfs to drop all of its
+	 * backing pages, *now*.
+	 */
+	shmem_truncate_range(file_inode(obj->filp), 0, (loff_t)-1);
+
+	invalidate_mapping_pages(file_inode(obj->filp)->i_mapping,
+			0, (loff_t)-1);
+}
+
 /* must be called before _move_to_active().. */
 int msm_gem_sync_object(struct drm_gem_object *obj,
 		struct msm_fence_context *fctx, bool exclusive)
diff --git a/drivers/gpu/drm/msm/msm_gem.h b/drivers/gpu/drm/msm/msm_gem.h
index fa8e1f1..631dab5 100644
--- a/drivers/gpu/drm/msm/msm_gem.h
+++ b/drivers/gpu/drm/msm/msm_gem.h
@@ -77,6 +77,12 @@ static inline bool is_active(struct msm_gem_object *msm_obj)
 	return msm_obj->gpu != NULL;
 }
 
+static inline bool is_purgeable(struct msm_gem_object *msm_obj)
+{
+	return (msm_obj->madv == MSM_MADV_DONTNEED) && msm_obj->sgt &&
+			!msm_obj->base.dma_buf && !msm_obj->base.import_attach;
+}
+
 #define MAX_CMDS 4
 
 /* Created per submit-ioctl, to track bo's and cmdstream bufs, etc,
diff --git a/drivers/gpu/drm/msm/msm_gem_shrinker.c b/drivers/gpu/drm/msm/msm_gem_shrinker.c
new file mode 100644
index 0000000..2bbd6d0
--- /dev/null
+++ b/drivers/gpu/drm/msm/msm_gem_shrinker.c
@@ -0,0 +1,128 @@
+/*
+ * Copyright (C) 2016 Red Hat
+ * Author: Rob Clark <robdclark@gmail.com>
+ *
+ * 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
+ * the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
+ * more details.
+ *
+ * You should have received a copy of the GNU General Public License along with
+ * this program.  If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#include "msm_drv.h"
+#include "msm_gem.h"
+
+static bool mutex_is_locked_by(struct mutex *mutex, struct task_struct *task)
+{
+	if (!mutex_is_locked(mutex))
+		return false;
+
+#if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_MUTEXES)
+	return mutex->owner == task;
+#else
+	/* Since UP may be pre-empted, we cannot assume that we own the lock */
+	return false;
+#endif
+}
+
+static bool msm_gem_shrinker_lock(struct drm_device *dev, bool *unlock)
+{
+	if (!mutex_trylock(&dev->struct_mutex)) {
+		if (!mutex_is_locked_by(&dev->struct_mutex, current))
+			return false;
+		*unlock = false;
+	} else {
+		*unlock = true;
+	}
+
+	return true;
+}
+
+
+static unsigned long
+msm_gem_shrinker_count(struct shrinker *shrinker, struct shrink_control *sc)
+{
+	struct msm_drm_private *priv =
+		container_of(shrinker, struct msm_drm_private, shrinker);
+	struct drm_device *dev = priv->dev;
+	struct msm_gem_object *msm_obj;
+	unsigned long count = 0;
+	bool unlock;
+
+	if (!msm_gem_shrinker_lock(dev, &unlock))
+		return 0;
+
+	list_for_each_entry(msm_obj, &priv->inactive_list, mm_list) {
+		if (is_purgeable(msm_obj))
+			count += msm_obj->base.size >> PAGE_SHIFT;
+	}
+
+	if (unlock)
+		mutex_unlock(&dev->struct_mutex);
+
+	return count;
+}
+
+static unsigned long
+msm_gem_shrinker_scan(struct shrinker *shrinker, struct shrink_control *sc)
+{
+	struct msm_drm_private *priv =
+		container_of(shrinker, struct msm_drm_private, shrinker);
+	struct drm_device *dev = priv->dev;
+	struct msm_gem_object *msm_obj;
+	unsigned long freed = 0;
+	bool unlock;
+
+	if (!msm_gem_shrinker_lock(dev, &unlock))
+		return SHRINK_STOP;
+
+	list_for_each_entry(msm_obj, &priv->inactive_list, mm_list) {
+		if (freed >= sc->nr_to_scan)
+			break;
+		if (is_purgeable(msm_obj)) {
+			msm_gem_purge(&msm_obj->base);
+			freed += msm_obj->base.size >> PAGE_SHIFT;
+		}
+	}
+
+	if (unlock)
+		mutex_unlock(&dev->struct_mutex);
+
+	if (freed > 0)
+		pr_info("Purging %lu bytes\n", freed << PAGE_SHIFT);
+
+	return freed;
+}
+
+/**
+ * msm_gem_shrinker_init - Initialize msm shrinker
+ * @dev_priv: msm device
+ *
+ * This function registers and sets up the msm shrinker.
+ */
+void msm_gem_shrinker_init(struct drm_device *dev)
+{
+	struct msm_drm_private *priv = dev->dev_private;
+	priv->shrinker.count_objects = msm_gem_shrinker_count;
+	priv->shrinker.scan_objects = msm_gem_shrinker_scan;
+	priv->shrinker.seeks = DEFAULT_SEEKS;
+	WARN_ON(register_shrinker(&priv->shrinker));
+}
+
+/**
+ * msm_gem_shrinker_cleanup - Clean up msm shrinker
+ * @dev_priv: msm device
+ *
+ * This function unregisters the msm shrinker.
+ */
+void msm_gem_shrinker_cleanup(struct drm_device *dev)
+{
+	struct msm_drm_private *priv = dev->dev_private;
+	unregister_shrinker(&priv->shrinker);
+}
-- 
2.5.5

[PATCH 05/10] drm/msm: change gem->vmap() to get/put

[Rob Clark]

Before we can add vmap shrinking, we really need to know which vmap'ings
are currently being used.  So switch to get/put interface.  Stubbed put
fxns for now.

Signed-off-by: Rob Clark <robdclark@gmail.com>
---
 drivers/gpu/drm/msm/adreno/adreno_gpu.c |  6 +++++-
 drivers/gpu/drm/msm/dsi/dsi_host.c      |  5 ++++-
 drivers/gpu/drm/msm/msm_drv.h           |  6 ++++--
 drivers/gpu/drm/msm/msm_fbdev.c         |  3 ++-
 drivers/gpu/drm/msm/msm_gem.c           | 17 ++++++++++++++---
 drivers/gpu/drm/msm/msm_gem_prime.c     |  4 ++--
 drivers/gpu/drm/msm/msm_gem_submit.c    |  4 +++-
 drivers/gpu/drm/msm/msm_rd.c            |  4 +++-
 drivers/gpu/drm/msm/msm_ringbuffer.c    |  6 ++++--
 9 files changed, 41 insertions(+), 14 deletions(-)

diff --git a/drivers/gpu/drm/msm/adreno/adreno_gpu.c b/drivers/gpu/drm/msm/adreno/adreno_gpu.c
index 2aec27d..7dfe22e 100644
--- a/drivers/gpu/drm/msm/adreno/adreno_gpu.c
+++ b/drivers/gpu/drm/msm/adreno/adreno_gpu.c
@@ -407,7 +407,7 @@ int adreno_gpu_init(struct drm_device *drm, struct platform_device *pdev,
 		return ret;
 	}
 
-	adreno_gpu->memptrs = msm_gem_vaddr(adreno_gpu->memptrs_bo);
+	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;
@@ -426,8 +426,12 @@ int adreno_gpu_init(struct drm_device *drm, struct platform_device *pdev,
 void adreno_gpu_cleanup(struct adreno_gpu *gpu)
 {
 	if (gpu->memptrs_bo) {
+		if (gpu->memptrs)
+			msm_gem_put_vaddr(gpu->memptrs_bo);
+
 		if (gpu->memptrs_iova)
 			msm_gem_put_iova(gpu->memptrs_bo, gpu->base.id);
+
 		drm_gem_object_unreference_unlocked(gpu->memptrs_bo);
 	}
 	release_firmware(gpu->pm4);
diff --git a/drivers/gpu/drm/msm/dsi/dsi_host.c b/drivers/gpu/drm/msm/dsi/dsi_host.c
index a3e47ad8..fb8ba9d 100644
--- a/drivers/gpu/drm/msm/dsi/dsi_host.c
+++ b/drivers/gpu/drm/msm/dsi/dsi_host.c
@@ -1066,7 +1066,7 @@ static int dsi_cmd_dma_add(struct msm_dsi_host *msm_host,
 	}
 
 	if (cfg_hnd->major == MSM_DSI_VER_MAJOR_6G) {
-		data = msm_gem_vaddr(msm_host->tx_gem_obj);
+		data = msm_gem_get_vaddr(msm_host->tx_gem_obj);
 		if (IS_ERR(data)) {
 			ret = PTR_ERR(data);
 			pr_err("%s: get vaddr failed, %d\n", __func__, ret);
@@ -1094,6 +1094,9 @@ static int dsi_cmd_dma_add(struct msm_dsi_host *msm_host,
 	if (packet.size < len)
 		memset(data + packet.size, 0xff, len - packet.size);
 
+	if (cfg_hnd->major == MSM_DSI_VER_MAJOR_6G)
+		msm_gem_put_vaddr(msm_host->tx_gem_obj);
+
 	return len;
 }
 
diff --git a/drivers/gpu/drm/msm/msm_drv.h b/drivers/gpu/drm/msm/msm_drv.h
index bcf33cf..2c16385 100644
--- a/drivers/gpu/drm/msm/msm_drv.h
+++ b/drivers/gpu/drm/msm/msm_drv.h
@@ -196,8 +196,10 @@ struct drm_gem_object *msm_gem_prime_import_sg_table(struct drm_device *dev,
 		struct dma_buf_attachment *attach, struct sg_table *sg);
 int msm_gem_prime_pin(struct drm_gem_object *obj);
 void msm_gem_prime_unpin(struct drm_gem_object *obj);
-void *msm_gem_vaddr_locked(struct drm_gem_object *obj);
-void *msm_gem_vaddr(struct drm_gem_object *obj);
+void *msm_gem_get_vaddr_locked(struct drm_gem_object *obj);
+void *msm_gem_get_vaddr(struct drm_gem_object *obj);
+void msm_gem_put_vaddr_locked(struct drm_gem_object *obj);
+void msm_gem_put_vaddr(struct drm_gem_object *obj);
 int msm_gem_madvise(struct drm_gem_object *obj, unsigned madv);
 void msm_gem_purge(struct drm_gem_object *obj);
 int msm_gem_sync_object(struct drm_gem_object *obj,
diff --git a/drivers/gpu/drm/msm/msm_fbdev.c b/drivers/gpu/drm/msm/msm_fbdev.c
index a9223be..ffd4a33 100644
--- a/drivers/gpu/drm/msm/msm_fbdev.c
+++ b/drivers/gpu/drm/msm/msm_fbdev.c
@@ -158,7 +158,7 @@ static int msm_fbdev_create(struct drm_fb_helper *helper,
 
 	dev->mode_config.fb_base = paddr;
 
-	fbi->screen_base = msm_gem_vaddr_locked(fbdev->bo);
+	fbi->screen_base = msm_gem_get_vaddr_locked(fbdev->bo);
 	if (IS_ERR(fbi->screen_base)) {
 		ret = PTR_ERR(fbi->screen_base);
 		goto fail_unlock;
@@ -251,6 +251,7 @@ void msm_fbdev_free(struct drm_device *dev)
 
 	/* this will free the backing object */
 	if (fbdev->fb) {
+		msm_gem_put_vaddr(fbdev->bo);
 		drm_framebuffer_unregister_private(fbdev->fb);
 		drm_framebuffer_remove(fbdev->fb);
 	}
diff --git a/drivers/gpu/drm/msm/msm_gem.c b/drivers/gpu/drm/msm/msm_gem.c
index 444d0b5..c05fc1d 100644
--- a/drivers/gpu/drm/msm/msm_gem.c
+++ b/drivers/gpu/drm/msm/msm_gem.c
@@ -408,7 +408,7 @@ fail:
 	return ret;
 }
 
-void *msm_gem_vaddr_locked(struct drm_gem_object *obj)
+void *msm_gem_get_vaddr_locked(struct drm_gem_object *obj)
 {
 	struct msm_gem_object *msm_obj = to_msm_bo(obj);
 	WARN_ON(!mutex_is_locked(&obj->dev->struct_mutex));
@@ -424,15 +424,26 @@ void *msm_gem_vaddr_locked(struct drm_gem_object *obj)
 	return msm_obj->vaddr;
 }
 
-void *msm_gem_vaddr(struct drm_gem_object *obj)
+void *msm_gem_get_vaddr(struct drm_gem_object *obj)
 {
 	void *ret;
 	mutex_lock(&obj->dev->struct_mutex);
-	ret = msm_gem_vaddr_locked(obj);
+	ret = msm_gem_get_vaddr_locked(obj);
 	mutex_unlock(&obj->dev->struct_mutex);
 	return ret;
 }
 
+void msm_gem_put_vaddr_locked(struct drm_gem_object *obj)
+{
+	WARN_ON(!mutex_is_locked(&obj->dev->struct_mutex));
+	/* no-op for now */
+}
+
+void msm_gem_put_vaddr(struct drm_gem_object *obj)
+{
+	/* no-op for now */
+}
+
 /* Update madvise status, returns true if not purged, else
  * false or -errno.
  */
diff --git a/drivers/gpu/drm/msm/msm_gem_prime.c b/drivers/gpu/drm/msm/msm_gem_prime.c
index 6b90890..60bb290 100644
--- a/drivers/gpu/drm/msm/msm_gem_prime.c
+++ b/drivers/gpu/drm/msm/msm_gem_prime.c
@@ -33,12 +33,12 @@ struct sg_table *msm_gem_prime_get_sg_table(struct drm_gem_object *obj)
 
 void *msm_gem_prime_vmap(struct drm_gem_object *obj)
 {
-	return msm_gem_vaddr(obj);
+	return msm_gem_get_vaddr(obj);
 }
 
 void msm_gem_prime_vunmap(struct drm_gem_object *obj, void *vaddr)
 {
-	/* TODO msm_gem_vunmap() */
+	msm_gem_put_vaddr(obj);
 }
 
 int msm_gem_prime_mmap(struct drm_gem_object *obj, struct vm_area_struct *vma)
diff --git a/drivers/gpu/drm/msm/msm_gem_submit.c b/drivers/gpu/drm/msm/msm_gem_submit.c
index ea5f709..736f158 100644
--- a/drivers/gpu/drm/msm/msm_gem_submit.c
+++ b/drivers/gpu/drm/msm/msm_gem_submit.c
@@ -279,7 +279,7 @@ static int submit_reloc(struct msm_gem_submit *submit, struct msm_gem_object *ob
 	/* For now, just map the entire thing.  Eventually we probably
 	 * to do it page-by-page, w/ kmap() if not vmap()d..
 	 */
-	ptr = msm_gem_vaddr_locked(&obj->base);
+	ptr = msm_gem_get_vaddr_locked(&obj->base);
 
 	if (IS_ERR(ptr)) {
 		ret = PTR_ERR(ptr);
@@ -332,6 +332,8 @@ static int submit_reloc(struct msm_gem_submit *submit, struct msm_gem_object *ob
 		last_offset = off;
 	}
 
+	msm_gem_put_vaddr_locked(&obj->base);
+
 	return 0;
 }
 
diff --git a/drivers/gpu/drm/msm/msm_rd.c b/drivers/gpu/drm/msm/msm_rd.c
index 0857710..3eeb8af 100644
--- a/drivers/gpu/drm/msm/msm_rd.c
+++ b/drivers/gpu/drm/msm/msm_rd.c
@@ -310,7 +310,7 @@ void msm_rd_dump_submit(struct msm_gem_submit *submit)
 		uint32_t iova = submit->cmd[i].iova;
 		uint32_t szd  = submit->cmd[i].size; /* in dwords */
 		struct msm_gem_object *obj = submit->bos[idx].obj;
-		const char *buf = msm_gem_vaddr_locked(&obj->base);
+		const char *buf = msm_gem_get_vaddr_locked(&obj->base);
 
 		if (IS_ERR(buf))
 			continue;
@@ -335,6 +335,8 @@ void msm_rd_dump_submit(struct msm_gem_submit *submit)
 					(uint32_t[2]){ iova, szd }, 8);
 			break;
 		}
+
+		msm_gem_put_vaddr_locked(&obj->base);
 	}
 }
 #endif
diff --git a/drivers/gpu/drm/msm/msm_ringbuffer.c b/drivers/gpu/drm/msm/msm_ringbuffer.c
index 42f5359..f326cf6 100644
--- a/drivers/gpu/drm/msm/msm_ringbuffer.c
+++ b/drivers/gpu/drm/msm/msm_ringbuffer.c
@@ -39,7 +39,7 @@ struct msm_ringbuffer *msm_ringbuffer_new(struct msm_gpu *gpu, int size)
 		goto fail;
 	}
 
-	ring->start = msm_gem_vaddr_locked(ring->bo);
+	ring->start = msm_gem_get_vaddr_locked(ring->bo);
 	if (IS_ERR(ring->start)) {
 		ret = PTR_ERR(ring->start);
 		goto fail;
@@ -59,7 +59,9 @@ fail:
 
 void msm_ringbuffer_destroy(struct msm_ringbuffer *ring)
 {
-	if (ring->bo)
+	if (ring->bo) {
+		msm_gem_put_vaddr(ring->bo);
 		drm_gem_object_unreference_unlocked(ring->bo);
+	}
 	kfree(ring);
 }
-- 
2.5.5

_______________________________________________
Freedreno mailing list
Freedreno@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/freedreno

[PATCH 06/10] drm/msm: wire up vmap shrinker

[Rob Clark]

Signed-off-by: Rob Clark <robdclark@gmail.com>
---
 drivers/gpu/drm/msm/msm_drv.h          |  2 ++
 drivers/gpu/drm/msm/msm_gem.c          | 25 ++++++++++++++++-----
 drivers/gpu/drm/msm/msm_gem.h          | 10 +++++++++
 drivers/gpu/drm/msm/msm_gem_shrinker.c | 40 ++++++++++++++++++++++++++++++++++
 4 files changed, 72 insertions(+), 5 deletions(-)

diff --git a/drivers/gpu/drm/msm/msm_drv.h b/drivers/gpu/drm/msm/msm_drv.h
index 2c16385..3ab10a1 100644
--- a/drivers/gpu/drm/msm/msm_drv.h
+++ b/drivers/gpu/drm/msm/msm_drv.h
@@ -149,6 +149,7 @@ struct msm_drm_private {
 		struct drm_mm mm;
 	} vram;
 
+	struct notifier_block vmap_notifier;
 	struct shrinker shrinker;
 
 	struct msm_vblank_ctrl vblank_ctrl;
@@ -202,6 +203,7 @@ void msm_gem_put_vaddr_locked(struct drm_gem_object *obj);
 void msm_gem_put_vaddr(struct drm_gem_object *obj);
 int msm_gem_madvise(struct drm_gem_object *obj, unsigned madv);
 void msm_gem_purge(struct drm_gem_object *obj);
+void msm_gem_vunmap(struct drm_gem_object *obj);
 int msm_gem_sync_object(struct drm_gem_object *obj,
 		struct msm_fence_context *fctx, bool exclusive);
 void msm_gem_move_to_active(struct drm_gem_object *obj,
diff --git a/drivers/gpu/drm/msm/msm_gem.c b/drivers/gpu/drm/msm/msm_gem.c
index c05fc1d..886cfe0 100644
--- a/drivers/gpu/drm/msm/msm_gem.c
+++ b/drivers/gpu/drm/msm/msm_gem.c
@@ -421,6 +421,7 @@ void *msm_gem_get_vaddr_locked(struct drm_gem_object *obj)
 		if (msm_obj->vaddr == NULL)
 			return ERR_PTR(-ENOMEM);
 	}
+	msm_obj->vmap_count++;
 	return msm_obj->vaddr;
 }
 
@@ -435,13 +436,17 @@ void *msm_gem_get_vaddr(struct drm_gem_object *obj)
 
 void msm_gem_put_vaddr_locked(struct drm_gem_object *obj)
 {
+	struct msm_gem_object *msm_obj = to_msm_bo(obj);
 	WARN_ON(!mutex_is_locked(&obj->dev->struct_mutex));
-	/* no-op for now */
+	WARN_ON(msm_obj->vmap_count < 1);
+	msm_obj->vmap_count--;
 }
 
 void msm_gem_put_vaddr(struct drm_gem_object *obj)
 {
-	/* no-op for now */
+	mutex_lock(&obj->dev->struct_mutex);
+	msm_gem_put_vaddr_locked(obj);
+	mutex_unlock(&obj->dev->struct_mutex);
 }
 
 /* Update madvise status, returns true if not purged, else
@@ -470,8 +475,7 @@ void msm_gem_purge(struct drm_gem_object *obj)
 
 	put_iova(obj);
 
-	vunmap(msm_obj->vaddr);
-	msm_obj->vaddr = NULL;
+	msm_gem_vunmap(obj);
 
 	put_pages(obj);
 
@@ -491,6 +495,17 @@ void msm_gem_purge(struct drm_gem_object *obj)
 			0, (loff_t)-1);
 }
 
+void msm_gem_vunmap(struct drm_gem_object *obj)
+{
+	struct msm_gem_object *msm_obj = to_msm_bo(obj);
+
+	if (!msm_obj->vaddr || WARN_ON(!is_vunmapable(msm_obj)))
+		return;
+
+	vunmap(msm_obj->vaddr);
+	msm_obj->vaddr = NULL;
+}
+
 /* must be called before _move_to_active().. */
 int msm_gem_sync_object(struct drm_gem_object *obj,
 		struct msm_fence_context *fctx, bool exclusive)
@@ -694,7 +709,7 @@ void msm_gem_free_object(struct drm_gem_object *obj)
 
 		drm_prime_gem_destroy(obj, msm_obj->sgt);
 	} else {
-		vunmap(msm_obj->vaddr);
+		msm_gem_vunmap(obj);
 		put_pages(obj);
 	}
 
diff --git a/drivers/gpu/drm/msm/msm_gem.h b/drivers/gpu/drm/msm/msm_gem.h
index 631dab5..7a4da81 100644
--- a/drivers/gpu/drm/msm/msm_gem.h
+++ b/drivers/gpu/drm/msm/msm_gem.h
@@ -34,6 +34,11 @@ struct msm_gem_object {
 	 */
 	uint8_t madv;
 
+	/**
+	 * count of active vmap'ing
+	 */
+	uint8_t vmap_count;
+
 	/* And object is either:
 	 *  inactive - on priv->inactive_list
 	 *  active   - on one one of the gpu's active_list..  well, at
@@ -83,6 +88,11 @@ static inline bool is_purgeable(struct msm_gem_object *msm_obj)
 			!msm_obj->base.dma_buf && !msm_obj->base.import_attach;
 }
 
+static inline bool is_vunmapable(struct msm_gem_object *msm_obj)
+{
+	return (msm_obj->vmap_count == 0) && msm_obj->vaddr;
+}
+
 #define MAX_CMDS 4
 
 /* Created per submit-ioctl, to track bo's and cmdstream bufs, etc,
diff --git a/drivers/gpu/drm/msm/msm_gem_shrinker.c b/drivers/gpu/drm/msm/msm_gem_shrinker.c
index 2bbd6d0..8d13b9e 100644
--- a/drivers/gpu/drm/msm/msm_gem_shrinker.c
+++ b/drivers/gpu/drm/msm/msm_gem_shrinker.c
@@ -100,6 +100,42 @@ msm_gem_shrinker_scan(struct shrinker *shrinker, struct shrink_control *sc)
 	return freed;
 }
 
+static int
+msm_gem_shrinker_vmap(struct notifier_block *nb, unsigned long event, void *ptr)
+{
+	struct msm_drm_private *priv =
+		container_of(nb, struct msm_drm_private, vmap_notifier);
+	struct drm_device *dev = priv->dev;
+	struct msm_gem_object *msm_obj;
+	unsigned unmapped = 0;
+	bool unlock;
+
+	if (!msm_gem_shrinker_lock(dev, &unlock))
+		return NOTIFY_DONE;
+
+	list_for_each_entry(msm_obj, &priv->inactive_list, mm_list) {
+		if (is_vunmapable(msm_obj)) {
+			msm_gem_vunmap(&msm_obj->base);
+			/* since we don't know any better, lets bail after a few
+			 * and if necessary the shrinker will be invoked again.
+			 * Seems better than unmapping *everything*
+			 */
+			if (++unmapped >= 5)
+				break;
+		}
+	}
+
+	if (unlock)
+		mutex_unlock(&dev->struct_mutex);
+
+	*(unsigned long *)ptr += unmapped;
+
+	if (unmapped > 0)
+		pr_info("Purging %u vmaps\n", unmapped);
+
+	return NOTIFY_DONE;
+}
+
 /**
  * msm_gem_shrinker_init - Initialize msm shrinker
  * @dev_priv: msm device
@@ -113,6 +149,9 @@ void msm_gem_shrinker_init(struct drm_device *dev)
 	priv->shrinker.scan_objects = msm_gem_shrinker_scan;
 	priv->shrinker.seeks = DEFAULT_SEEKS;
 	WARN_ON(register_shrinker(&priv->shrinker));
+
+	priv->vmap_notifier.notifier_call = msm_gem_shrinker_vmap;
+	WARN_ON(register_vmap_purge_notifier(&priv->vmap_notifier));
 }
 
 /**
@@ -124,5 +163,6 @@ void msm_gem_shrinker_init(struct drm_device *dev)
 void msm_gem_shrinker_cleanup(struct drm_device *dev)
 {
 	struct msm_drm_private *priv = dev->dev_private;
+	WARN_ON(unregister_vmap_purge_notifier(&priv->vmap_notifier));
 	unregister_shrinker(&priv->shrinker);
 }
-- 
2.5.5

[PATCH 07/10] drm/msm: deal with arbitrary # of cmd buffers

[Rob Clark]

For some optimizations coming on the userspace side, splitting larger
draw or gmem cmds into multiple cmdstream buffers, we need to support
much more than the previous small/arbitrary limit.

Signed-off-by: Rob Clark <robdclark@gmail.com>
---
 drivers/gpu/drm/msm/msm_gem.h        |  4 +---
 drivers/gpu/drm/msm/msm_gem_submit.c | 11 +++++------
 2 files changed, 6 insertions(+), 9 deletions(-)

diff --git a/drivers/gpu/drm/msm/msm_gem.h b/drivers/gpu/drm/msm/msm_gem.h
index 7a4da81..b2f13cf 100644
--- a/drivers/gpu/drm/msm/msm_gem.h
+++ b/drivers/gpu/drm/msm/msm_gem.h
@@ -93,8 +93,6 @@ static inline bool is_vunmapable(struct msm_gem_object *msm_obj)
 	return (msm_obj->vmap_count == 0) && msm_obj->vaddr;
 }
 
-#define MAX_CMDS 4
-
 /* Created per submit-ioctl, to track bo's and cmdstream bufs, etc,
  * associated with the cmdstream submission for synchronization (and
  * make it easier to unwind when things go wrong, etc).  This only
@@ -116,7 +114,7 @@ struct msm_gem_submit {
 		uint32_t size;  /* in dwords */
 		uint32_t iova;
 		uint32_t idx;   /* cmdstream buffer idx in bos[] */
-	} cmd[MAX_CMDS];
+	} *cmd;  /* array of size nr_cmds */
 	struct {
 		uint32_t flags;
 		struct msm_gem_object *obj;
diff --git a/drivers/gpu/drm/msm/msm_gem_submit.c b/drivers/gpu/drm/msm/msm_gem_submit.c
index 736f158..9766f9a 100644
--- a/drivers/gpu/drm/msm/msm_gem_submit.c
+++ b/drivers/gpu/drm/msm/msm_gem_submit.c
@@ -29,10 +29,11 @@
 #define BO_PINNED   0x2000
 
 static struct msm_gem_submit *submit_create(struct drm_device *dev,
-		struct msm_gpu *gpu, int nr)
+		struct msm_gpu *gpu, int nr_bos, int nr_cmds)
 {
 	struct msm_gem_submit *submit;
-	int sz = sizeof(*submit) + (nr * sizeof(submit->bos[0]));
+	int sz = sizeof(*submit) + (nr_bos * sizeof(submit->bos[0])) +
+			(nr_cmds * sizeof(*submit->cmd));
 
 	submit = kmalloc(sz, GFP_TEMPORARY | __GFP_NOWARN | __GFP_NORETRY);
 	if (!submit)
@@ -42,6 +43,7 @@ static struct msm_gem_submit *submit_create(struct drm_device *dev,
 	submit->gpu = gpu;
 	submit->fence = NULL;
 	submit->pid = get_pid(task_pid(current));
+	submit->cmd = (void *)&submit->bos[nr_bos];
 
 	/* initially, until copy_from_user() and bo lookup succeeds: */
 	submit->nr_bos = 0;
@@ -371,14 +373,11 @@ int msm_ioctl_gem_submit(struct drm_device *dev, void *data,
 	if (args->pipe != MSM_PIPE_3D0)
 		return -EINVAL;
 
-	if (args->nr_cmds > MAX_CMDS)
-		return -EINVAL;
-
 	ret = mutex_lock_interruptible(&dev->struct_mutex);
 	if (ret)
 		return ret;
 
-	submit = submit_create(dev, gpu, args->nr_bos);
+	submit = submit_create(dev, gpu, args->nr_bos, args->nr_cmds);
 	if (!submit) {
 		ret = -ENOMEM;
 		goto out_unlock;
-- 
2.5.5

[PATCH 08/10] drm/msm: bump kernel api version

[Rob Clark]

Signed-off-by: Rob Clark <robdclark@gmail.com>
---
 drivers/gpu/drm/msm/msm_drv.c | 15 +++++++++++++--
 1 file changed, 13 insertions(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/msm/msm_drv.c b/drivers/gpu/drm/msm/msm_drv.c
index 3e15a50..e0d077d 100644
--- a/drivers/gpu/drm/msm/msm_drv.c
+++ b/drivers/gpu/drm/msm/msm_drv.c
@@ -21,6 +21,16 @@
 #include "msm_gpu.h"
 #include "msm_kms.h"
 
+
+/*
+ * MSM driver version:
+ * - 1.0.0 - initial interface
+ * - 1.1.0 - adds madvise, and support for submits with > 4 cmd buffers
+ */
+#define MSM_VERSION_MAJOR	1
+#define MSM_VERSION_MINOR	1
+#define MSM_VERSION_PATCHLEVEL	0
+
 static void msm_fb_output_poll_changed(struct drm_device *dev)
 {
 	struct msm_drm_private *priv = dev->dev_private;
@@ -808,8 +818,9 @@ static struct drm_driver msm_driver = {
 	.name               = "msm",
 	.desc               = "MSM Snapdragon DRM",
 	.date               = "20130625",
-	.major              = 1,
-	.minor              = 0,
+	.major              = MSM_VERSION_MAJOR,
+	.minor              = MSM_VERSION_MINOR,
+	.patchlevel         = MSM_VERSION_PATCHLEVEL,
 };
 
 #ifdef CONFIG_PM_SLEEP
-- 
2.5.5

[PATCH 09/10] drm/msm/rd: split out snapshot_buf helper

[Rob Clark]

(reduce the noise in next patch)

Signed-off-by: Rob Clark <robdclark@gmail.com>
---
 drivers/gpu/drm/msm/msm_rd.c | 36 ++++++++++++++++++++++--------------
 1 file changed, 22 insertions(+), 14 deletions(-)

diff --git a/drivers/gpu/drm/msm/msm_rd.c b/drivers/gpu/drm/msm/msm_rd.c
index 3eeb8af..fa02b5a 100644
--- a/drivers/gpu/drm/msm/msm_rd.c
+++ b/drivers/gpu/drm/msm/msm_rd.c
@@ -277,6 +277,26 @@ void msm_rd_debugfs_cleanup(struct drm_minor *minor)
 	kfree(rd);
 }
 
+static void snapshot_buf(struct msm_rd_state *rd,
+		struct msm_gem_submit *submit, int idx,
+		uint32_t iova, uint32_t size)
+{
+	struct msm_gem_object *obj = submit->bos[idx].obj;
+	const char *buf;
+
+	buf = msm_gem_get_vaddr_locked(&obj->base);
+	if (IS_ERR(buf))
+		return;
+
+	buf += iova - submit->bos[idx].iova;
+
+	rd_write_section(rd, RD_GPUADDR,
+			(uint32_t[2]){ iova, size }, 8);
+	rd_write_section(rd, RD_BUFFER_CONTENTS, buf, size);
+
+	msm_gem_put_vaddr_locked(&obj->base);
+}
+
 /* called under struct_mutex */
 void msm_rd_dump_submit(struct msm_gem_submit *submit)
 {
@@ -306,21 +326,11 @@ void msm_rd_dump_submit(struct msm_gem_submit *submit)
 	 */
 
 	for (i = 0; i < submit->nr_cmds; i++) {
-		uint32_t idx  = submit->cmd[i].idx;
 		uint32_t iova = submit->cmd[i].iova;
 		uint32_t szd  = submit->cmd[i].size; /* in dwords */
-		struct msm_gem_object *obj = submit->bos[idx].obj;
-		const char *buf = msm_gem_get_vaddr_locked(&obj->base);
-
-		if (IS_ERR(buf))
-			continue;
 
-		buf += iova - submit->bos[idx].iova;
-
-		rd_write_section(rd, RD_GPUADDR,
-				(uint32_t[2]){ iova, szd * 4 }, 8);
-		rd_write_section(rd, RD_BUFFER_CONTENTS,
-				buf, szd * 4);
+		snapshot_buf(rd, submit, submit->cmd[i].idx,
+				submit->cmd[i].iova, szd * 4);
 
 		switch (submit->cmd[i].type) {
 		case MSM_SUBMIT_CMD_IB_TARGET_BUF:
@@ -335,8 +345,6 @@ void msm_rd_dump_submit(struct msm_gem_submit *submit)
 					(uint32_t[2]){ iova, szd }, 8);
 			break;
 		}
-
-		msm_gem_put_vaddr_locked(&obj->base);
 	}
 }
 #endif
-- 
2.5.5

[PATCH 10/10] drm/msm/rd: add module param to dump all bo's

[Rob Clark]

By default, if using $debugfs/.../rd to log cmdstream, only the
cmdstream buffers themselves are logged.  But in some cases we want
to capture other buffers in the submit (to see VBO's or shaders).
So add a mod-param knob to control this.

Signed-off-by: Rob Clark <robdclark@gmail.com>
---
 drivers/gpu/drm/msm/msm_rd.c | 38 +++++++++++++++++++++++++++++++-------
 1 file changed, 31 insertions(+), 7 deletions(-)

diff --git a/drivers/gpu/drm/msm/msm_rd.c b/drivers/gpu/drm/msm/msm_rd.c
index fa02b5a..24254e0 100644
--- a/drivers/gpu/drm/msm/msm_rd.c
+++ b/drivers/gpu/drm/msm/msm_rd.c
@@ -27,6 +27,11 @@
  * This bypasses drm_debugfs_create_files() mainly because we need to use
  * our own fops for a bit more control.  In particular, we don't want to
  * do anything if userspace doesn't have the debugfs file open.
+ *
+ * The module-param "rd_full", which defaults to false, enables snapshotting
+ * all (non-written) buffers in the submit, rather than just cmdstream bo's.
+ * This is useful to capture the contents of (for example) vbo's or textures,
+ * or shader programs (if not emitted inline in cmdstream).
  */
 
 #ifdef CONFIG_DEBUG_FS
@@ -40,6 +45,10 @@
 #include "msm_gpu.h"
 #include "msm_gem.h"
 
+static bool rd_full = false;
+MODULE_PARM_DESC(rd_full, "If true, $debugfs/.../rd will snapshot all buffer contents");
+module_param_named(rd_full, rd_full, bool, 0600);
+
 enum rd_sect_type {
 	RD_NONE,
 	RD_TEST,       /* ascii text */
@@ -288,7 +297,12 @@ static void snapshot_buf(struct msm_rd_state *rd,
 	if (IS_ERR(buf))
 		return;
 
-	buf += iova - submit->bos[idx].iova;
+	if (iova) {
+		buf += iova - submit->bos[idx].iova;
+	} else {
+		iova = submit->bos[idx].iova;
+		size = obj->base.size;
+	}
 
 	rd_write_section(rd, RD_GPUADDR,
 			(uint32_t[2]){ iova, size }, 8);
@@ -320,17 +334,27 @@ void msm_rd_dump_submit(struct msm_gem_submit *submit)
 
 	rd_write_section(rd, RD_CMD, msg, ALIGN(n, 4));
 
-	/* could be nice to have an option (module-param?) to snapshot
-	 * all the bo's associated with the submit.  Handy to see vtx
-	 * buffers, etc.  For now just the cmdstream bo's is enough.
-	 */
+	if (rd_full) {
+		for (i = 0; i < submit->nr_bos; i++) {
+			/* buffers that are written to probably don't start out
+			 * with anything interesting:
+			 */
+			if (submit->bos[i].flags & MSM_SUBMIT_BO_WRITE)
+				continue;
+
+			snapshot_buf(rd, submit, i, 0, 0);
+		}
+	}
 
 	for (i = 0; i < submit->nr_cmds; i++) {
 		uint32_t iova = submit->cmd[i].iova;
 		uint32_t szd  = submit->cmd[i].size; /* in dwords */
 
-		snapshot_buf(rd, submit, submit->cmd[i].idx,
-				submit->cmd[i].iova, szd * 4);
+		/* snapshot cmdstream bo's (if we haven't already): */
+		if (!rd_full) {
+			snapshot_buf(rd, submit, submit->cmd[i].idx,
+					submit->cmd[i].iova, szd * 4);
+		}
 
 		switch (submit->cmd[i].type) {
 		case MSM_SUBMIT_CMD_IB_TARGET_BUF:
-- 
2.5.5