drm/amdkfd: New IOCTL to allocate queue GWS

Add a new kfd ioctl to allocate queue GWS. Queue
GWS is released on queue destroy.

Signed-off-by: Oak Zeng <Oak.Zeng@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
index c92e931..aab2aa6 100644 (file)
--- a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
@@ -1567,6 +1567,31 @@ copy_from_user_failed:
        return err;
 }
 
+static int kfd_ioctl_alloc_queue_gws(struct file *filep,
+               struct kfd_process *p, void *data)
+{
+       int retval;
+       struct kfd_ioctl_alloc_queue_gws_args *args = data;
+       struct kfd_dev *dev = NULL;
+
+       if (!hws_gws_support ||
+               dev->dqm->sched_policy == KFD_SCHED_POLICY_NO_HWS)
+               return -EINVAL;
+
+       dev = kfd_device_by_id(args->gpu_id);
+       if (!dev) {
+               pr_debug("Could not find gpu id 0x%x\n", args->gpu_id);
+               return -EINVAL;
+       }
+
+       mutex_lock(&p->mutex);
+       retval = pqm_set_gws(&p->pqm, args->queue_id, args->num_gws ? dev->gws : NULL);
+       mutex_unlock(&p->mutex);
+
+       args->first_gws = 0;
+       return retval;
+}
+
 static int kfd_ioctl_get_dmabuf_info(struct file *filep,
                struct kfd_process *p, void *data)
 {
@@ -1769,6 +1794,8 @@ static const struct amdkfd_ioctl_desc amdkfd_ioctls[] = {
        AMDKFD_IOCTL_DEF(AMDKFD_IOC_IMPORT_DMABUF,
                                kfd_ioctl_import_dmabuf, 0),
 
+       AMDKFD_IOCTL_DEF(AMDKFD_IOC_ALLOC_QUEUE_GWS,
+                       kfd_ioctl_alloc_queue_gws, 0),
 };
 
 #define AMDKFD_CORE_IOCTL_COUNT        ARRAY_SIZE(amdkfd_ioctls)

diff --git a/include/uapi/linux/kfd_ioctl.h b/include/uapi/linux/kfd_ioctl.h
index 20917c5..070d1bc 100644 (file)
--- a/include/uapi/linux/kfd_ioctl.h
+++ b/include/uapi/linux/kfd_ioctl.h
@@ -410,6 +410,21 @@ struct kfd_ioctl_unmap_memory_from_gpu_args {
        __u32 n_success;                /* to/from KFD */
 };
 
+/* Allocate GWS for specific queue
+ *
+ * @gpu_id:      device identifier
+ * @queue_id:    queue's id that GWS is allocated for
+ * @num_gws:     how many GWS to allocate
+ * @first_gws:   index of the first GWS allocated.
+ *               only support contiguous GWS allocation
+ */
+struct kfd_ioctl_alloc_queue_gws_args {
+       __u32 gpu_id;           /* to KFD */
+       __u32 queue_id;         /* to KFD */
+       __u32 num_gws;          /* to KFD */
+       __u32 first_gws;        /* from KFD */
+};
+
 struct kfd_ioctl_get_dmabuf_info_args {
        __u64 size;             /* from KFD */
        __u64 metadata_ptr;     /* to KFD */
@@ -529,7 +544,10 @@ enum kfd_mmio_remap {
 #define AMDKFD_IOC_IMPORT_DMABUF               \
                AMDKFD_IOWR(0x1D, struct kfd_ioctl_import_dmabuf_args)
 
+#define AMDKFD_IOC_ALLOC_QUEUE_GWS             \
+               AMDKFD_IOWR(0x1E, struct kfd_ioctl_alloc_queue_gws_args)
+
 #define AMDKFD_COMMAND_START           0x01
-#define AMDKFD_COMMAND_END             0x1E
+#define AMDKFD_COMMAND_END             0x1F
 
 #endif