[dpdk-dev] [PATCH v2 3/3] bus/dpaa: optimize physical to virtual address searching

[Shreyansh Jain]

With Hotplugging memory support, the order of memseg has been changed
from physically contiguous to virtual contiguous. DPAA bus and drivers
depend on PA to VA address conversion for I/O.

This patch creates a list of blocks requested to be pinned to the
DPAA mempool. For searching physical addresses, it is expected that
it would belong to this list (from hardware pool) and hence it is
less expensive than memseg walks. Though, there is a marginal drop
in performance vis-a-vis the legacy mode with physically contiguous
memsegs.

Signed-off-by: Shreyansh Jain <shreyansh.jain at nxp.com>

---
An optimized algorithm is being worked upon based on some recent patches
in hotplugging. That would improve/recover the performance. Until that
time, this patch is to be treated a stop-gap solution.
---
 drivers/bus/dpaa/rte_dpaa_bus.h     | 27 ++++++++++++++++++++++++++-
 drivers/mempool/dpaa/dpaa_mempool.c | 33 ++++++++++++++++++++++++++++++++-
 2 files changed, 58 insertions(+), 2 deletions(-)

diff --git a/drivers/bus/dpaa/rte_dpaa_bus.h b/drivers/bus/dpaa/rte_dpaa_bus.h
index 89aeac2d1..ca32b7f2f 100644
--- a/drivers/bus/dpaa/rte_dpaa_bus.h
+++ b/drivers/bus/dpaa/rte_dpaa_bus.h
@@ -95,9 +95,34 @@ struct dpaa_portal {
 	uint64_t tid;/**< Parent Thread id for this portal */
 };
 
-/* TODO - this is costly, need to write a fast coversion routine */
+/* Various structures representing contiguous memory maps */
+struct dpaa_memseg {
+	TAILQ_ENTRY(dpaa_memseg) next;
+	char *vaddr;
+	rte_iova_t iova;
+	size_t len;
+};
+
+TAILQ_HEAD(dpaa_memseg_list, dpaa_memseg);
+extern struct dpaa_memseg_list dpaa_memsegs;
+
+/* Either iterate over the list of internal memseg references or fallback to
+ * EAL memseg based iova2virt.
+ */
 static inline void *rte_dpaa_mem_ptov(phys_addr_t paddr)
 {
+	struct dpaa_memseg *ms;
+
+	/* Check if the address is already part of the memseg list internally
+	 * maintained by the dpaa driver.
+	 */
+	TAILQ_FOREACH(ms, &dpaa_memsegs, next) {
+		if (paddr >= ms->iova && paddr <
+			ms->iova + ms->len)
+			return RTE_PTR_ADD(ms->vaddr, (paddr - ms->iova));
+	}
+
+	/* If not, Fallback to full memseg list searching */
 	return rte_mem_iova2virt(paddr);
 }
 
diff --git a/drivers/mempool/dpaa/dpaa_mempool.c b/drivers/mempool/dpaa/dpaa_mempool.c
index 580e4640c..9d6277f82 100644
--- a/drivers/mempool/dpaa/dpaa_mempool.c
+++ b/drivers/mempool/dpaa/dpaa_mempool.c
@@ -27,6 +27,13 @@
 
 #include <dpaa_mempool.h>
 
+/* List of all the memseg information locally maintained in dpaa driver. This
+ * is to optimize the PA_to_VA searches until a better mechanism (algo) is
+ * available.
+ */
+struct dpaa_memseg_list dpaa_memsegs
+	= TAILQ_HEAD_INITIALIZER(dpaa_memsegs);
+
 struct dpaa_bp_info rte_dpaa_bpid_info[DPAA_MAX_BPOOLS];
 
 static int
@@ -287,10 +294,34 @@ dpaa_populate(struct rte_mempool *mp, unsigned int max_objs,
 	/* Detect pool area has sufficient space for elements in this memzone */
 	if (len >= total_elt_sz * mp->size)
 		bp_info->flags |= DPAA_MPOOL_SINGLE_SEGMENT;
+	struct dpaa_memseg *ms;
+
+	/* For each memory chunk pinned to the Mempool, a linked list of the
+	 * contained memsegs is created for searching when PA to VA
+	 * conversion is required.
+	 */
+	ms = rte_zmalloc(NULL, sizeof(struct dpaa_memseg), 0);
+	if (!ms) {
+		DPAA_MEMPOOL_ERR("Unable to allocate internal memory.");
+		DPAA_MEMPOOL_WARN("Fast Physical to Virtual Addr translation would not be available.");
+		/* If the element is not added, it would only lead to failure
+		 * in searching for the element and the logic would Fallback
+		 * to traditional DPDK memseg traversal code. So, this is not
+		 * a blocking error - but, error would be printed on screen.
+		 */
+		return 0;
+	}
+
+	ms->vaddr = vaddr;
+	ms->iova = paddr;
+	ms->len = len;
+	/* Head insertions are generally faster than tail insertions as the
+	 * buffers pinned are picked from rear end.
+	 */
+	TAILQ_INSERT_HEAD(&dpaa_memsegs, ms, next);
 
 	return rte_mempool_op_populate_default(mp, max_objs, vaddr, paddr, len,
 					       obj_cb, obj_cb_arg);
-
 }
 
 struct rte_mempool_ops dpaa_mpool_ops = {
-- 
2.14.1