gpu: nvgpu: add ioctl for querying memory state

Add NVGPU_GPU_IOCTL_GET_MEMORY_STATE to read the amount of free
device-local video memory, if applicable.

Some reserved fields are added to support different types of queries in
the future (e.g. context-local free amount).

Bug 1787771
Bug 200233138

Change-Id: Id5ffd02ad4d6ed3a6dc196541938573c27b340ac
Signed-off-by: Konsta Holtta <kholtta@nvidia.com>
Reviewed-on: http://git-master/r/1223762
(cherry picked from commit 96221d96c7972c6387944603e974f7639d6dbe70)
Reviewed-on: http://git-master/r/1235980
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>

4 files changed, 60 insertions, 1 deletions

diff --git a/drivers/gpu/nvgpu/gk20a/ctrl_gk20a.c b/drivers/gpu/nvgpu/gk20a/ctrl_gk20a.c
index 6b832670..2fdf719a 100644
--- a/drivers/gpu/nvgpu/gk20a/ctrl_gk20a.c
+++ b/drivers/gpu/nvgpu/gk20a/ctrl_gk20a.c
@@ -747,6 +747,24 @@ static int nvgpu_gpu_alloc_vidmem(struct gk20a *g,
         return 0;
 }
 
+static int nvgpu_gpu_get_memory_state(struct gk20a *g,
+                        struct nvgpu_gpu_get_memory_state_args *args)
+{
+        int err;
+
+        gk20a_dbg_fn("");
+
+        if (args->reserved[0] || args->reserved[1] ||
+            args->reserved[2] || args->reserved[3])
+                return -EINVAL;
+
+        err = gk20a_vidmem_get_space(g, &args->total_free_bytes);
+
+        gk20a_dbg_fn("done, err=%d, bytes=%lld", err, args->total_free_bytes);
+
+        return err;
+}
+
 long gk20a_ctrl_dev_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
 {
         struct device *dev = filp->private_data;
@@ -999,6 +1017,11 @@ long gk20a_ctrl_dev_ioctl(struct file *filp, unsigned int cmd, unsigned long arg
                         (struct nvgpu_gpu_alloc_vidmem_args *)buf);
                 break;
 
+        case NVGPU_GPU_IOCTL_GET_MEMORY_STATE:
+                err = nvgpu_gpu_get_memory_state(g,
+                        (struct nvgpu_gpu_get_memory_state_args *)buf);
+                break;
+
         default:
                 dev_dbg(dev_from_gk20a(g), "unrecognized gpu ioctl cmd: 0x%x", cmd);
                 err = -ENOTTY;
diff --git a/drivers/gpu/nvgpu/gk20a/mm_gk20a.c b/drivers/gpu/nvgpu/gk20a/mm_gk20a.c
index f46f7a81..ff9bb5e2 100644
--- a/drivers/gpu/nvgpu/gk20a/mm_gk20a.c
+++ b/drivers/gpu/nvgpu/gk20a/mm_gk20a.c
@@ -2276,6 +2276,26 @@ err_kfree:
 #endif
 }
 
+int gk20a_vidmem_get_space(struct gk20a *g, u64 *space)
+{
+#if defined(CONFIG_GK20A_VIDMEM)
+        struct gk20a_allocator *allocator = &g->mm.vidmem.allocator;
+
+        gk20a_dbg_fn("");
+
+        if (!gk20a_alloc_initialized(allocator))
+                return -ENOSYS;
+
+        mutex_lock(&g->mm.vidmem.clear_list_mutex);
+        *space = gk20a_alloc_space(allocator) +
+                atomic64_read(&g->mm.vidmem.bytes_pending);
+        mutex_unlock(&g->mm.vidmem.clear_list_mutex);
+        return 0;
+#else
+        return -ENOSYS;
+#endif
+}
+
 static u64 gk20a_mm_get_align(struct gk20a *g, struct scatterlist *sgl,
                               enum gk20a_aperture aperture)
 {
diff --git a/drivers/gpu/nvgpu/gk20a/mm_gk20a.h b/drivers/gpu/nvgpu/gk20a/mm_gk20a.h
index b25a7789..e8f7da98 100644
--- a/drivers/gpu/nvgpu/gk20a/mm_gk20a.h
+++ b/drivers/gpu/nvgpu/gk20a/mm_gk20a.h
@@ -755,6 +755,7 @@ void gk20a_vm_mapping_batch_finish_locked(
 
 
 int gk20a_vidmem_buf_alloc(struct gk20a *g, size_t bytes);
+int gk20a_vidmem_get_space(struct gk20a *g, u64 *space);
 
 /* Note: batch may be NULL if map op is not part of a batch */
 int gk20a_vm_map_buffer(struct vm_gk20a *vm,
diff --git a/include/uapi/linux/nvgpu.h b/include/uapi/linux/nvgpu.h
index 84732452..0bf88628 100644
--- a/include/uapi/linux/nvgpu.h
+++ b/include/uapi/linux/nvgpu.h
@@ -510,6 +510,18 @@ struct nvgpu_gpu_alloc_vidmem_args {
         };
 };
 
+struct nvgpu_gpu_get_memory_state_args {
+        /*
+         * Current free space for this device; may change even when any
+         * kernel-managed metadata (e.g., page tables or channels) is allocated
+         * or freed. For an idle gpu, an allocation of this size would succeed.
+         */
+        __u64 total_free_bytes;
+
+        /* For future use; must be set to 0. */
+        __u64 reserved[4];
+};
+
 #define NVGPU_GPU_IOCTL_ZCULL_GET_CTX_SIZE \
         _IOR(NVGPU_GPU_IOCTL_MAGIC, 1, struct nvgpu_gpu_zcull_get_ctx_size_args)
 #define NVGPU_GPU_IOCTL_ZCULL_GET_INFO \
@@ -568,8 +580,11 @@ struct nvgpu_gpu_alloc_vidmem_args {
 #define NVGPU_GPU_IOCTL_ALLOC_VIDMEM \
         _IOWR(NVGPU_GPU_IOCTL_MAGIC, 27, \
                         struct nvgpu_gpu_alloc_vidmem_args)
+#define NVGPU_GPU_IOCTL_GET_MEMORY_STATE \
+        _IOWR(NVGPU_GPU_IOCTL_MAGIC, 33, \
+                        struct nvgpu_gpu_get_memory_state_args)
 #define NVGPU_GPU_IOCTL_LAST            \
-        _IOC_NR(NVGPU_GPU_IOCTL_ALLOC_VIDMEM)
+        _IOC_NR(NVGPU_GPU_IOCTL_GET_MEMORY_STATE)
 #define NVGPU_GPU_IOCTL_MAX_ARG_SIZE    \
         sizeof(struct nvgpu_gpu_get_cpu_time_correlation_info_args)