gpu: nvgpu: report number of syncpoints in nvgpu_as_get_sync_ro_map_arg

Add reporting for the number of syncpoints when mapping the RO
shim. This allows the userspace to perform boundary condition checks
when computing the GPU VA for a syncpoint.

JIRA GCSS-1579

Change-Id: Ia6c9eee917d2c1e08f9905701e03f2b09e01ba60
Signed-off-by: Sami Kiminki <skiminki@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2533981
Reviewed-by: Seshendra Gadagottu <sgadagottu@nvidia.com>
Reviewed-by: Lakshmanan M <lm@nvidia.com>
Reviewed-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
GVS: Gerrit_Virtual_Submit

drivers/gpu/nvgpu/hal/sync/syncpt_cmdbuf_gv11b.h

@@ -52,7 +52,7 @@ int gv11b_syncpt_alloc_buf(struct nvgpu_channel *c,
 		u32 syncpt_id, struct nvgpu_mem *syncpt_buf);
 
 int gv11b_syncpt_get_sync_ro_map(struct vm_gk20a *vm,
-		u64 *base_gpuva, u32 *sync_size);
+		u64 *base_gpuva, u32 *sync_size, u32 *num_syncpoints);
 
 #else
 

drivers/gpu/nvgpu/hal/sync/syncpt_cmdbuf_gv11b_fusa.c

@@ -105,10 +105,11 @@ void gv11b_syncpt_free_buf(struct nvgpu_channel *c,
 }
 
 int gv11b_syncpt_get_sync_ro_map(struct vm_gk20a *vm,
-		u64 *base_gpuva, u32 *sync_size)
+		u64 *base_gpuva, u32 *sync_size, u32 *num_syncpoints)
 {
 	struct gk20a *g = gk20a_from_vm(vm);
 	int err;
+	size_t tmp;
 
 	nvgpu_mutex_acquire(&vm->syncpt_ro_map_lock);
 	err = set_syncpt_ro_map_gpu_va_locked(vm);
@@ -120,5 +121,8 @@ int gv11b_syncpt_get_sync_ro_map(struct vm_gk20a *vm,
 	*base_gpuva = vm->syncpt_ro_map_gpu_va;
 	*sync_size = g->syncpt_size;
 
+	tmp = g->syncpt_size ? (g->syncpt_unit_size / g->syncpt_size) : 0U;
+	*num_syncpoints = (tmp <= U32_MAX) ? tmp : U32_MAX;
+
 	return 0;
 }

drivers/gpu/nvgpu/hal/vgpu/sync/syncpt_cmdbuf_gv11b_vgpu.c

@@ -131,10 +131,11 @@ void vgpu_gv11b_syncpt_free_buf(struct nvgpu_channel *c,
 }
 
 int vgpu_gv11b_syncpt_get_sync_ro_map(struct vm_gk20a *vm,
-	u64 *base_gpuva, u32 *sync_size)
+	u64 *base_gpuva, u32 *sync_size, u32 *num_syncpoints)
 {
 	struct gk20a *g = gk20a_from_vm(vm);
 	int err;
+	size_t tmp;
 
 	nvgpu_mutex_acquire(&vm->syncpt_ro_map_lock);
 	err = set_syncpt_ro_map_gpu_va_locked(vm);
@@ -146,6 +147,9 @@ int vgpu_gv11b_syncpt_get_sync_ro_map(struct vm_gk20a *vm,
 	*base_gpuva = vm->syncpt_ro_map_gpu_va;
 	*sync_size = g->syncpt_size;
 
+	tmp = g->syncpt_size ? (g->syncpt_unit_size / g->syncpt_size) : 0U;
+	*num_syncpoints = (tmp <= U32_MAX) ? tmp : U32_MAX;
+
 	return 0;
 }
 #endif /* CONFIG_TEGRA_GK20A_NVHOST */

drivers/gpu/nvgpu/hal/vgpu/sync/syncpt_cmdbuf_gv11b_vgpu.h

@@ -35,5 +35,5 @@ int vgpu_gv11b_syncpt_alloc_buf(struct nvgpu_channel *c,
 void vgpu_gv11b_syncpt_free_buf(struct nvgpu_channel *c,
 					struct nvgpu_mem *syncpt_buf);
 int vgpu_gv11b_syncpt_get_sync_ro_map(struct vm_gk20a *vm,
-	u64 *base_gpuva, u32 *sync_size);
+	u64 *base_gpuva, u32 *sync_size, u32 *num_syncpoints);
 #endif /* NVGPU_SYNCPT_CMDBUF_GV11B_VGPU_H */

drivers/gpu/nvgpu/include/nvgpu/gops/sync.h

@@ -44,8 +44,9 @@ struct gops_sync_syncpt {
 	 * @param vm [in]		VM area for channel.
 	 * @param base_gpu [out]	Base GPU VA for mapped
 	 * 				syncpoint aperture.
-	 * @param sync_size [out]	Size of syncpoint aperture
-	 * 				in bytes.
+	 * @param sync_size [out]	Size per syncpoint in bytes.
+	 * @param num_syncpoints [out]	Number of syncpoints in the
+	 * 				aperture.
 	 *
 	 * Map syncpoint aperture in GPU virtual memory as read-only:
 	 * - Acquire syncpoint read-only map lock.
@@ -53,12 +54,14 @@ struct gops_sync_syncpt {
 	 *   if not already mapped. Map as read-only.
 	 * - Release syncpoint read-only map lock.
 	 *
+	 * The syncpoint shim mapping size is sync_size * num_syncpoints.
+	 *
 	 * @return 0 in case of success, < 0 in case of failure.
 	 * @retval -ENOMEM if syncpoint aperture could not be
 	 *         mapped to GPU virtual memory.
 	 */
 	int (*get_sync_ro_map)(struct vm_gk20a *vm,
-			u64 *base_gpuva, u32 *sync_size);
+			u64 *base_gpuva, u32 *sync_size, u32 *num_syncpoints);
 
 /** @cond DOXYGEN_SHOULD_SKIP_THIS */
 	int (*alloc_buf)(struct nvgpu_channel *c,

drivers/gpu/nvgpu/os/linux/ioctl_as.c

@@ -281,6 +281,7 @@ static int nvgpu_as_ioctl_get_sync_ro_map(
 	struct gk20a *g = gk20a_from_vm(vm);
 	u64 base_gpuva;
 	u32 sync_size;
+	u32 num_syncpoints;
 	int err = 0;
 
 	if (g->ops.sync.syncpt.get_sync_ro_map == NULL)
@@ -289,12 +290,14 @@ static int nvgpu_as_ioctl_get_sync_ro_map(
 	if (!nvgpu_has_syncpoints(g))
 		return -EINVAL;
 
-	err = g->ops.sync.syncpt.get_sync_ro_map(vm, &base_gpuva, &sync_size);
+	err = g->ops.sync.syncpt.get_sync_ro_map(vm, &base_gpuva, &sync_size,
+						 &num_syncpoints);
 	if (err)
 		return err;
 
 	args->base_gpuva = base_gpuva;
 	args->sync_size = sync_size;
+	args->num_syncpoints = num_syncpoints;
 
 	return err;
 #else

include/uapi/linux/nvgpu-as.h

@@ -375,7 +375,7 @@ struct nvgpu_as_map_buffer_batch_args {
 struct nvgpu_as_get_sync_ro_map_args {
 	__u64 base_gpuva;
 	__u32 sync_size;
-	__u32 padding;
+	__u32 num_syncpoints;
 };
 
 /*

userspace/units/sync/nvgpu-sync.c

@@ -341,6 +341,7 @@ int test_sync_get_ro_map(struct unit_module *m, struct gk20a *g, void *args)
 {
 	u64 base_gpuva = 0U;
 	u32 sync_size = 0U;
+	u32 num_syncpoints = 0U;
 	u32 branches;
 
 	int err = 0;
@@ -366,7 +367,7 @@ int test_sync_get_ro_map(struct unit_module *m, struct gk20a *g, void *args)
 		unit_info(m, "%s branch: %s\n", __func__, f_sync_get_ro_map[branches]);
 
 		err = g->ops.sync.syncpt.get_sync_ro_map(ch->vm,
-			&base_gpuva, &sync_size);
+				&base_gpuva, &sync_size, &num_syncpoints);
 
 		if (branches < F_SYNC_GET_RO_MAP_FAIL) {
 			if(err != 0) {