lib/accel: add dualcast function

Add to both accel framework and add sw engine implementation.
IDXD implementation and accel example app patches to follow.

Dual-cast copies the same source to two separate destination buffers.

Signed-off-by: paul luse <paul.e.luse@intel.com>
Change-Id: I7df28518505f0d4bbb32cc8e69eb8a231fb29452
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/2127
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Community-CI: Broadcom CI
Community-CI: Mellanox Build Bot
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>

include/spdk/accel_engine.h

@@ -135,6 +135,24 @@ uint64_t spdk_accel_get_capabilities(struct spdk_io_channel *ch);
 int spdk_accel_submit_copy(struct spdk_accel_task *accel_req, struct spdk_io_channel *ch, void *dst,
 			   void *src, uint64_t nbytes, spdk_accel_completion_cb cb);
 
+/**
+ * Submit a dual cast copy request.
+ *
+ * \param accel_req Accel request task.
+ * \param ch I/O channel to submit request to the accel engine. This channel can
+ * be obtained by the function spdk_accel_engine_get_io_channel().
+ * \param dst1 First destination to copy to (must be 4K aligned).
+ * \param dst2 Second destination to copy to (must be 4K aligned).
+ * \param src Source to copy from.
+ * \param nbytes Length in bytes to copy.
+ * \param cb Called when this copy operation completes.
+ *
+ * \return 0 on success, negative errno on failure.
+ */
+int spdk_accel_submit_dualcast(struct spdk_accel_task *accel_req, struct spdk_io_channel *ch,
+			       void *dst1, void *dst2, void *src, uint64_t nbytes,
+			       spdk_accel_completion_cb cb);
+
 /**
  * Submit a compare request.
  *
@@ -149,8 +167,8 @@ int spdk_accel_submit_copy(struct spdk_accel_task *accel_req, struct spdk_io_cha
  * \return 0 on success, any other value means there was a miscompare.
  */
 int spdk_accel_submit_compare(struct spdk_accel_task *accel_req, struct spdk_io_channel *ch,
-			      void *src1,
-			      void *src2, uint64_t nbytes, spdk_accel_completion_cb cb);
+			      void *src1, void *src2, uint64_t nbytes,
+			      spdk_accel_completion_cb cb);
 
 /**
  * Submit a fill request.

lib/accel/Makefile

@@ -36,6 +36,7 @@ include $(SPDK_ROOT_DIR)/mk/spdk.common.mk
 
 SO_VER := 3
 SO_MINOR := 0
+SO_SUFFIX := $(SO_VER).$(SO_MINOR)
 
 LIBNAME = accel
 C_SRCS = accel_engine.c

lib/accel/accel_engine.c

@@ -48,6 +48,8 @@
  * later in this file.
  */
 
+#define ALIGN_4K 0x1000
+
 /* Largest context size for all accel modules */
 static size_t g_max_accel_module_size = 0;
 
@@ -121,6 +123,25 @@ spdk_accel_submit_copy(struct spdk_accel_task *accel_req, struct spdk_io_channel
 				      _accel_engine_done);
 }
 
+/* Accel framework public API for dual cast copy function */
+int
+spdk_accel_submit_dualcast(struct spdk_accel_task *accel_req, struct spdk_io_channel *ch,
+			   void *dst1, void *dst2, void *src, uint64_t nbytes,
+			   spdk_accel_completion_cb cb)
+{
+	struct spdk_accel_task *req = accel_req;
+	struct accel_io_channel *accel_ch = spdk_io_channel_get_ctx(ch);
+
+	if ((uintptr_t)dst1 & (ALIGN_4K - 1) || (uintptr_t)dst2 & (ALIGN_4K - 1)) {
+		SPDK_ERRLOG("Dualcast requires 4K alignment on dst addresses\n");
+		return -EINVAL;
+	}
+
+	req->cb = cb;
+	return accel_ch->engine->dualcast(req->offload_ctx, accel_ch->ch, dst1, dst2, src, nbytes,
+					  _accel_engine_done);
+}
+
 /* Accel framework public API for compare function */
 int
 spdk_accel_submit_compare(struct spdk_accel_task *accel_req, struct spdk_io_channel *ch,
@@ -316,7 +337,8 @@ spdk_accel_engine_config_text(FILE *fp)
 static uint64_t
 sw_accel_get_capabilities(void)
 {
-	return ACCEL_COPY | ACCEL_FILL | ACCEL_CRC32C | ACCEL_COMPARE;
+	return ACCEL_COPY | ACCEL_FILL | ACCEL_CRC32C | ACCEL_COMPARE |
+	       ACCEL_DUALCAST;
 }
 
 static int
@@ -334,6 +356,21 @@ sw_accel_submit_copy(void *cb_arg, struct spdk_io_channel *ch, void *dst, void *
 	return 0;
 }
 
+static int
+sw_accel_submit_dualcast(void *cb_arg, struct spdk_io_channel *ch, void *dst1, void *dst2,
+			 void *src, uint64_t nbytes, spdk_accel_completion_cb cb)
+{
+	struct spdk_accel_task *accel_req;
+
+	memcpy(dst1, src, (size_t)nbytes);
+	memcpy(dst2, src, (size_t)nbytes);
+
+	accel_req = (struct spdk_accel_task *)((uintptr_t)cb_arg -
+					       offsetof(struct spdk_accel_task, offload_ctx));
+	cb(accel_req, 0);
+	return 0;
+}
+
 static int
 sw_accel_submit_compare(void *cb_arg, struct spdk_io_channel *ch, void *src1, void *src2,
 			uint64_t nbytes,
@@ -386,6 +423,7 @@ static struct spdk_io_channel *sw_accel_get_io_channel(void);
 static struct spdk_accel_engine sw_accel_engine = {
 	.get_capabilities	= sw_accel_get_capabilities,
 	.copy			= sw_accel_submit_copy,
+	.dualcast		= sw_accel_submit_dualcast,
 	.compare		= sw_accel_submit_compare,
 	.fill			= sw_accel_submit_fill,
 	.crc32c			= sw_accel_submit_crc32c,

lib/accel/spdk_accel.map

@@ -9,6 +9,7 @@
 	spdk_accel_engine_get_io_channel;
 	spdk_accel_get_capabilities;
 	spdk_accel_submit_copy;
+	spdk_accel_submit_dualcast;
 	spdk_accel_submit_compare;
 	spdk_accel_submit_fill;
 	spdk_accel_submit_crc32c;