[PATCH v4 10/10] drm/msm: Add a way for userspace to allocate GPU iova

[Rob Clark]

From: Rob Clark <robdclark@xxxxxxxxxxxx>

The motivation at this point is mainly native userspace mesa driver in a
VM guest.  The one remaining synchronous "hotpath" is buffer allocation,
because guest needs to wait to know the bo's iova before it can start
emitting cmdstream/state that references the new bo.  By allocating the
iova in the guest userspace, we no longer need to wait for a response
from the host, but can just rely on the allocation request being
processed before the cmdstream submission.  Allocation failures (OoM,
etc) would just be treated as context-lost (ie. GL_GUILTY_CONTEXT_RESET)
or subsequent allocations (or readpix, etc) can raise GL_OUT_OF_MEMORY.

v2: Fix inuse check
v3: Change mismatched iova case to -EBUSY

Signed-off-by: Rob Clark <robdclark@xxxxxxxxxxxx>
Reviewed-by: Dmitry Baryshkov <dmitry.baryshkov@xxxxxxxxxx>
Reviewed-by: Dmitry Osipenko <dmitry.osipenko@xxxxxxxxxxxxx>
---
 drivers/gpu/drm/msm/adreno/adreno_gpu.c | 10 ++++++
 drivers/gpu/drm/msm/msm_drv.c           | 21 +++++++++++
 drivers/gpu/drm/msm/msm_gem.c           | 48 +++++++++++++++++++++++++
 drivers/gpu/drm/msm/msm_gem.h           |  8 +++++
 drivers/gpu/drm/msm/msm_gem_vma.c       |  2 ++
 include/uapi/drm/msm_drm.h              |  3 ++
 6 files changed, 92 insertions(+)

diff --git a/drivers/gpu/drm/msm/adreno/adreno_gpu.c b/drivers/gpu/drm/msm/adreno/adreno_gpu.c
index 6385ab06632f..4caae0229518 100644
--- a/drivers/gpu/drm/msm/adreno/adreno_gpu.c
+++ b/drivers/gpu/drm/msm/adreno/adreno_gpu.c
@@ -281,6 +281,16 @@ int adreno_get_param(struct msm_gpu *gpu, struct msm_file_private *ctx,
 	case MSM_PARAM_SUSPENDS:
 		*value = gpu->suspend_count;
 		return 0;
+	case MSM_PARAM_VA_START:
+		if (ctx->aspace == gpu->aspace)
+			return -EINVAL;
+		*value = ctx->aspace->va_start;
+		return 0;
+	case MSM_PARAM_VA_SIZE:
+		if (ctx->aspace == gpu->aspace)
+			return -EINVAL;
+		*value = ctx->aspace->va_size;
+		return 0;
 	default:
 		DBG("%s: invalid param: %u", gpu->name, param);
 		return -EINVAL;
diff --git a/drivers/gpu/drm/msm/msm_drv.c b/drivers/gpu/drm/msm/msm_drv.c
index d618953d33ea..34e2169308b4 100644
--- a/drivers/gpu/drm/msm/msm_drv.c
+++ b/drivers/gpu/drm/msm/msm_drv.c
@@ -722,6 +722,23 @@ static int msm_ioctl_gem_info_iova(struct drm_device *dev,
 	return msm_gem_get_iova(obj, ctx->aspace, iova);
 }
 
+static int msm_ioctl_gem_info_set_iova(struct drm_device *dev,
+		struct drm_file *file, struct drm_gem_object *obj,
+		uint64_t iova)
+{
+	struct msm_drm_private *priv = dev->dev_private;
+	struct msm_file_private *ctx = file->driver_priv;
+
+	if (!priv->gpu)
+		return -EINVAL;
+
+	/* Only supported if per-process address space is supported: */
+	if (priv->gpu->aspace == ctx->aspace)
+		return -EOPNOTSUPP;
+
+	return msm_gem_set_iova(obj, ctx->aspace, iova);
+}
+
 static int msm_ioctl_gem_info(struct drm_device *dev, void *data,
 		struct drm_file *file)
 {
@@ -736,6 +753,7 @@ static int msm_ioctl_gem_info(struct drm_device *dev, void *data,
 	switch (args->info) {
 	case MSM_INFO_GET_OFFSET:
 	case MSM_INFO_GET_IOVA:
+	case MSM_INFO_SET_IOVA:
 		/* value returned as immediate, not pointer, so len==0: */
 		if (args->len)
 			return -EINVAL;
@@ -760,6 +778,9 @@ static int msm_ioctl_gem_info(struct drm_device *dev, void *data,
 	case MSM_INFO_GET_IOVA:
 		ret = msm_ioctl_gem_info_iova(dev, file, obj, &args->value);
 		break;
+	case MSM_INFO_SET_IOVA:
+		ret = msm_ioctl_gem_info_set_iova(dev, file, obj, args->value);
+		break;
 	case MSM_INFO_SET_NAME:
 		/* length check should leave room for terminating null: */
 		if (args->len >= sizeof(msm_obj->name)) {
diff --git a/drivers/gpu/drm/msm/msm_gem.c b/drivers/gpu/drm/msm/msm_gem.c
index bf4af17e2f1e..3ee30b8a76bd 100644
--- a/drivers/gpu/drm/msm/msm_gem.c
+++ b/drivers/gpu/drm/msm/msm_gem.c
@@ -525,6 +525,54 @@ int msm_gem_get_iova(struct drm_gem_object *obj,
 	return ret;
 }
 
+static int clear_iova(struct drm_gem_object *obj,
+		      struct msm_gem_address_space *aspace)
+{
+	struct msm_gem_vma *vma = lookup_vma(obj, aspace);
+
+	if (!vma)
+		return 0;
+
+	if (msm_gem_vma_inuse(vma))
+		return -EBUSY;
+
+	msm_gem_purge_vma(vma->aspace, vma);
+	msm_gem_close_vma(vma->aspace, vma);
+	del_vma(vma);
+
+	return 0;
+}
+
+/*
+ * Get the requested iova but don't pin it.  Fails if the requested iova is
+ * not available.  Doesn't need a put because iovas are currently valid for
+ * the life of the object.
+ *
+ * Setting an iova of zero will clear the vma.
+ */
+int msm_gem_set_iova(struct drm_gem_object *obj,
+		     struct msm_gem_address_space *aspace, uint64_t iova)
+{
+	int ret = 0;
+
+	msm_gem_lock(obj);
+	if (!iova) {
+		ret = clear_iova(obj, aspace);
+	} else {
+		struct msm_gem_vma *vma;
+		vma = get_vma_locked(obj, aspace, iova, iova + obj->size);
+		if (IS_ERR(vma)) {
+			ret = PTR_ERR(vma);
+		} else if (GEM_WARN_ON(vma->iova != iova)) {
+			clear_iova(obj, aspace);
+			ret = -EBUSY;
+		}
+	}
+	msm_gem_unlock(obj);
+
+	return ret;
+}
+
 /*
  * Unpin a iova by updating the reference counts. The memory isn't actually
  * purged until something else (shrinker, mm_notifier, destroy, etc) decides
diff --git a/drivers/gpu/drm/msm/msm_gem.h b/drivers/gpu/drm/msm/msm_gem.h
index 580b6eb95edd..c75d3b879a53 100644
--- a/drivers/gpu/drm/msm/msm_gem.h
+++ b/drivers/gpu/drm/msm/msm_gem.h
@@ -38,6 +38,12 @@ struct msm_gem_address_space {
 
 	/* @faults: the number of GPU hangs associated with this address space */
 	int faults;
+
+	/** @va_start: lowest possible address to allocate */
+	uint64_t va_start;
+
+	/** @va_size: the size of the address space (in bytes) */
+	uint64_t va_size;
 };
 
 struct msm_gem_address_space *
@@ -144,6 +150,8 @@ struct msm_gem_vma *msm_gem_get_vma_locked(struct drm_gem_object *obj,
 					   struct msm_gem_address_space *aspace);
 int msm_gem_get_iova(struct drm_gem_object *obj,
 		struct msm_gem_address_space *aspace, uint64_t *iova);
+int msm_gem_set_iova(struct drm_gem_object *obj,
+		struct msm_gem_address_space *aspace, uint64_t iova);
 int msm_gem_get_and_pin_iova_range(struct drm_gem_object *obj,
 		struct msm_gem_address_space *aspace, uint64_t *iova,
 		u64 range_start, u64 range_end);
diff --git a/drivers/gpu/drm/msm/msm_gem_vma.c b/drivers/gpu/drm/msm/msm_gem_vma.c
index 0cd6770faf41..3c1dc9241831 100644
--- a/drivers/gpu/drm/msm/msm_gem_vma.c
+++ b/drivers/gpu/drm/msm/msm_gem_vma.c
@@ -184,6 +184,8 @@ msm_gem_address_space_create(struct msm_mmu *mmu, const char *name,
 	spin_lock_init(&aspace->lock);
 	aspace->name = name;
 	aspace->mmu = mmu;
+	aspace->va_start = va_start;
+	aspace->va_size  = size;
 
 	drm_mm_init(&aspace->mm, va_start, size);
 
diff --git a/include/uapi/drm/msm_drm.h b/include/uapi/drm/msm_drm.h
index 794ad1948497..3c7b097c4e3d 100644
--- a/include/uapi/drm/msm_drm.h
+++ b/include/uapi/drm/msm_drm.h
@@ -84,6 +84,8 @@ struct drm_msm_timespec {
 #define MSM_PARAM_SYSPROF    0x0b  /* WO: 1 preserves perfcntrs, 2 also disables suspend */
 #define MSM_PARAM_COMM       0x0c  /* WO: override for task->comm */
 #define MSM_PARAM_CMDLINE    0x0d  /* WO: override for task cmdline */
+#define MSM_PARAM_VA_START   0x0e  /* RO: start of valid GPU iova range */
+#define MSM_PARAM_VA_SIZE    0x0f  /* RO: size of valid GPU iova range (bytes) */
 
 /* For backwards compat.  The original support for preemption was based on
  * a single ring per priority level so # of priority levels equals the #
@@ -135,6 +137,7 @@ struct drm_msm_gem_new {
 #define MSM_INFO_GET_IOVA	0x01   /* get iova, returned by value */
 #define MSM_INFO_SET_NAME	0x02   /* set the debug name (by pointer) */
 #define MSM_INFO_GET_NAME	0x03   /* get debug name, returned by pointer */
+#define MSM_INFO_SET_IOVA	0x04   /* set the iova, passed by value */
 
 struct drm_msm_gem_info {
 	__u32 handle;         /* in */
-- 
2.35.1