habanalabs: add command submission module

This patch adds the main flow for the user to submit work to the device.

Each work is described by a command submission object (CS). The CS contains
3 arrays of command buffers: One for execution, and two for context-switch
(store and restore).

For each CB, the user specifies on which queue to put that CB. In case of
an internal queue, the entry doesn't contain a pointer to the CB but the
address in the on-chip memory that the CB resides at.

The driver parses some of the CBs to enforce security restrictions.

The user receives a sequence number that represents the CS object. The user
can then query the driver regarding the status of the CS, using that
sequence number.

In case the CS doesn't finish before the timeout expires, the driver will
perform a soft-reset of the device.

Reviewed-by: Mike Rapoport <rppt@linux.ibm.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

1 files changed, 48 insertions, 4 deletions

diff --git a/drivers/misc/habanalabs/context.c b/drivers/misc/habanalabs/context.c
index de1258e7a6e6..c3854714b46c 100644
--- a/drivers/misc/habanalabs/context.c
+++ b/drivers/misc/habanalabs/context.c
@@ -12,6 +12,18 @@
 static void hl_ctx_fini(struct hl_ctx *ctx)
 {
 	struct hl_device *hdev = ctx->hdev;
+	int i;
+
+	/*
+	 * If we arrived here, there are no jobs waiting for this context
+	 * on its queues so we can safely remove it.
+	 * This is because for each CS, we increment the ref count and for
+	 * every CS that was finished we decrement it and we won't arrive
+	 * to this function unless the ref count is 0
+	 */
+
+	for (i = 0 ; i < HL_MAX_PENDING_CS ; i++)
+		dma_fence_put(ctx->cs_pending[i]);
 
 	if (ctx->asid != HL_KERNEL_ASID_ID)
 		hl_asid_free(hdev, ctx->asid);
@@ -23,8 +35,6 @@ void hl_ctx_do_release(struct kref *ref)
 
 	ctx = container_of(ref, struct hl_ctx, refcount);
 
-	dev_dbg(ctx->hdev->dev, "Now really releasing context %d\n", ctx->asid);
-
 	hl_ctx_fini(ctx);
 
 	if (ctx->hpriv)
@@ -90,6 +100,11 @@ int hl_ctx_init(struct hl_device *hdev, struct hl_ctx *ctx, bool is_kernel_ctx)
 
 	kref_init(&ctx->refcount);
 
+	ctx->cs_sequence = 1;
+	spin_lock_init(&ctx->cs_lock);
+	atomic_set(&ctx->thread_restore_token, 1);
+	ctx->thread_restore_wait_token = 0;
+
 	if (is_kernel_ctx) {
 		ctx->asid = HL_KERNEL_ASID_ID; /* KMD gets ASID 0 */
 	} else {
@@ -100,8 +115,6 @@ int hl_ctx_init(struct hl_device *hdev, struct hl_ctx *ctx, bool is_kernel_ctx)
 		}
 	}
 
-	dev_dbg(hdev->dev, "Created context with ASID %u\n", ctx->asid);
-
 	return 0;
 }
 
@@ -115,6 +128,37 @@ int hl_ctx_put(struct hl_ctx *ctx)
 	return kref_put(&ctx->refcount, hl_ctx_do_release);
 }
 
+struct dma_fence *hl_ctx_get_fence(struct hl_ctx *ctx, u64 seq)
+{
+	struct hl_device *hdev = ctx->hdev;
+	struct dma_fence *fence;
+
+	spin_lock(&ctx->cs_lock);
+
+	if (seq >= ctx->cs_sequence) {
+		dev_notice(hdev->dev,
+			"Can't wait on seq %llu because current CS is at seq %llu\n",
+			seq, ctx->cs_sequence);
+		spin_unlock(&ctx->cs_lock);
+		return ERR_PTR(-EINVAL);
+	}
+
+
+	if (seq + HL_MAX_PENDING_CS < ctx->cs_sequence) {
+		dev_dbg(hdev->dev,
+			"Can't wait on seq %llu because current CS is at seq %llu (Fence is gone)\n",
+			seq, ctx->cs_sequence);
+		spin_unlock(&ctx->cs_lock);
+		return NULL;
+	}
+
+	fence = dma_fence_get(
+			ctx->cs_pending[seq & (HL_MAX_PENDING_CS - 1)]);
+	spin_unlock(&ctx->cs_lock);
+
+	return fence;
+}
+
 /*
  * hl_ctx_mgr_init - initialize the context manager
  *