[dpdk-dev] [PATCH 1/3] mem: provide thread-unsafe contig walk variant

[dpdk-dev] [PATCH 1/3] mem: provide thread-unsafe contig walk variant

[Anatoly Burakov]

Sometimes, user code needs to walk memseg list while being inside
a memory-related callback. Rather than making everyone copy around
the same iteration code and depending on DPDK internals, provide an
official way to do memseg_contig_walk() inside callbacks.

Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com>
---
 lib/librte_eal/common/eal_common_memory.c  | 28 ++++++++++++----------
 lib/librte_eal/common/include/rte_memory.h | 18 ++++++++++++++
 lib/librte_eal/rte_eal_version.map         |  1 +
 3 files changed, 35 insertions(+), 12 deletions(-)

diff --git a/lib/librte_eal/common/eal_common_memory.c b/lib/librte_eal/common/eal_common_memory.c
index 4f0688f9d..e3320a746 100644
--- a/lib/librte_eal/common/eal_common_memory.c
+++ b/lib/librte_eal/common/eal_common_memory.c
@@ -788,14 +788,11 @@ rte_mem_lock_page(const void *virt)
 }
 
 int __rte_experimental
-rte_memseg_contig_walk(rte_memseg_contig_walk_t func, void *arg)
+rte_memseg_contig_walk_thread_unsafe(rte_memseg_contig_walk_t func, void *arg)
 {
 	struct rte_mem_config *mcfg = rte_eal_get_configuration()->mem_config;
 	int i, ms_idx, ret = 0;
 
-	/* do not allow allocations/frees/init while we iterate */
-	rte_rwlock_read_lock(&mcfg->memory_hotplug_lock);
-
 	for (i = 0; i < RTE_MAX_MEMSEG_LISTS; i++) {
 		struct rte_memseg_list *msl = &mcfg->memsegs[i];
 		const struct rte_memseg *ms;
@@ -820,19 +817,26 @@ rte_memseg_contig_walk(rte_memseg_contig_walk_t func, void *arg)
 			len = n_segs * msl->page_sz;
 
 			ret = func(msl, ms, len, arg);
-			if (ret < 0) {
-				ret = -1;
-				goto out;
-			} else if (ret > 0) {
-				ret = 1;
-				goto out;
-			}
+			if (ret)
+				return ret;
 			ms_idx = rte_fbarray_find_next_used(arr,
 					ms_idx + n_segs);
 		}
 	}
-out:
+	return 0;
+}
+
+int __rte_experimental
+rte_memseg_contig_walk(rte_memseg_contig_walk_t func, void *arg)
+{
+	struct rte_mem_config *mcfg = rte_eal_get_configuration()->mem_config;
+	int ret = 0;
+
+	/* do not allow allocations/frees/init while we iterate */
+	rte_rwlock_read_lock(&mcfg->memory_hotplug_lock);
+	ret = rte_memseg_contig_walk_thread_unsafe(func, arg);
 	rte_rwlock_read_unlock(&mcfg->memory_hotplug_lock);
+
 	return ret;
 }
 
diff --git a/lib/librte_eal/common/include/rte_memory.h b/lib/librte_eal/common/include/rte_memory.h
index aab9f6fe5..aeba38bfa 100644
--- a/lib/librte_eal/common/include/rte_memory.h
+++ b/lib/librte_eal/common/include/rte_memory.h
@@ -263,6 +263,24 @@ rte_memseg_contig_walk(rte_memseg_contig_walk_t func, void *arg);
 int __rte_experimental
 rte_memseg_list_walk(rte_memseg_list_walk_t func, void *arg);
 
+/**
+ * Walk each VA-contiguous area without performing any locking.
+ *
+ * @note This function does not perform any locking, and is only safe to call
+ *       from within memory-related callback functions.
+ *
+ * @param func
+ *   Iterator function
+ * @param arg
+ *   Argument passed to iterator
+ * @return
+ *   0 if walked over the entire list
+ *   1 if stopped by the user
+ *   -1 if user function reported error
+ */
+int __rte_experimental
+rte_memseg_contig_walk_thread_unsafe(rte_memseg_contig_walk_t func, void *arg);
+
 /**
  * Dump the physical memory layout to a file.
  *
diff --git a/lib/librte_eal/rte_eal_version.map b/lib/librte_eal/rte_eal_version.map
index f7dd0e7bc..98bfbe796 100644
--- a/lib/librte_eal/rte_eal_version.map
+++ b/lib/librte_eal/rte_eal_version.map
@@ -286,6 +286,7 @@ EXPERIMENTAL {
 	rte_mem_virt2memseg;
 	rte_mem_virt2memseg_list;
 	rte_memseg_contig_walk;
+	rte_memseg_contig_walk_thread_unsafe;
 	rte_memseg_list_walk;
 	rte_memseg_walk;
 	rte_mp_action_register;
-- 
2.17.1

[dpdk-dev] [PATCH 2/3] mem: provide thread-unsafe memseg walk variant

[Anatoly Burakov]

Sometimes, user code needs to walk memseg list while being inside
a memory-related callback. Rather than making everyone copy around
the same iteration code and depending on DPDK internals, provide an
official way to do memseg_walk() inside callbacks.

Also, remove existing reimplementation from sPAPR VFIO code and use
the new API instead.

Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com>
---
 lib/librte_eal/common/eal_common_memory.c  | 28 ++++++++------
 lib/librte_eal/common/include/rte_memory.h | 18 +++++++++
 lib/librte_eal/linuxapp/eal/eal_vfio.c     | 43 +++-------------------
 lib/librte_eal/rte_eal_version.map         |  1 +
 4 files changed, 40 insertions(+), 50 deletions(-)

diff --git a/lib/librte_eal/common/eal_common_memory.c b/lib/librte_eal/common/eal_common_memory.c
index e3320a746..afe0d5b57 100644
--- a/lib/librte_eal/common/eal_common_memory.c
+++ b/lib/librte_eal/common/eal_common_memory.c
@@ -841,14 +841,11 @@ rte_memseg_contig_walk(rte_memseg_contig_walk_t func, void *arg)
 }
 
 int __rte_experimental
-rte_memseg_walk(rte_memseg_walk_t func, void *arg)
+rte_memseg_walk_thread_unsafe(rte_memseg_walk_t func, void *arg)
 {
 	struct rte_mem_config *mcfg = rte_eal_get_configuration()->mem_config;
 	int i, ms_idx, ret = 0;
 
-	/* do not allow allocations/frees/init while we iterate */
-	rte_rwlock_read_lock(&mcfg->memory_hotplug_lock);
-
 	for (i = 0; i < RTE_MAX_MEMSEG_LISTS; i++) {
 		struct rte_memseg_list *msl = &mcfg->memsegs[i];
 		const struct rte_memseg *ms;
@@ -863,18 +860,25 @@ rte_memseg_walk(rte_memseg_walk_t func, void *arg)
 		while (ms_idx >= 0) {
 			ms = rte_fbarray_get(arr, ms_idx);
 			ret = func(msl, ms, arg);
-			if (ret < 0) {
-				ret = -1;
-				goto out;
-			} else if (ret > 0) {
-				ret = 1;
-				goto out;
-			}
+			if (ret)
+				return ret;
 			ms_idx = rte_fbarray_find_next_used(arr, ms_idx + 1);
 		}
 	}
-out:
+	return 0;
+}
+
+int __rte_experimental
+rte_memseg_walk(rte_memseg_walk_t func, void *arg)
+{
+	struct rte_mem_config *mcfg = rte_eal_get_configuration()->mem_config;
+	int ret = 0;
+
+	/* do not allow allocations/frees/init while we iterate */
+	rte_rwlock_read_lock(&mcfg->memory_hotplug_lock);
+	ret = rte_memseg_walk_thread_unsafe(func, arg);
 	rte_rwlock_read_unlock(&mcfg->memory_hotplug_lock);
+
 	return ret;
 }
 
diff --git a/lib/librte_eal/common/include/rte_memory.h b/lib/librte_eal/common/include/rte_memory.h
index aeba38bfa..c5a84c333 100644
--- a/lib/librte_eal/common/include/rte_memory.h
+++ b/lib/librte_eal/common/include/rte_memory.h
@@ -263,6 +263,24 @@ rte_memseg_contig_walk(rte_memseg_contig_walk_t func, void *arg);
 int __rte_experimental
 rte_memseg_list_walk(rte_memseg_list_walk_t func, void *arg);
 
+/**
+ * Walk list of all memsegs without performing any locking.
+ *
+ * @note This function does not perform any locking, and is only safe to call
+ *       from within memory-related callback functions.
+ *
+ * @param func
+ *   Iterator function
+ * @param arg
+ *   Argument passed to iterator
+ * @return
+ *   0 if walked over the entire list
+ *   1 if stopped by the user
+ *   -1 if user function reported error
+ */
+int __rte_experimental
+rte_memseg_walk_thread_unsafe(rte_memseg_walk_t func, void *arg);
+
 /**
  * Walk each VA-contiguous area without performing any locking.
  *
diff --git a/lib/librte_eal/linuxapp/eal/eal_vfio.c b/lib/librte_eal/linuxapp/eal/eal_vfio.c
index a2bbdfbf4..14c9332e9 100644
--- a/lib/librte_eal/linuxapp/eal/eal_vfio.c
+++ b/lib/librte_eal/linuxapp/eal/eal_vfio.c
@@ -87,42 +87,6 @@ static const struct vfio_iommu_type iommu_types[] = {
 	},
 };
 
-/* for sPAPR IOMMU, we will need to walk memseg list, but we cannot use
- * rte_memseg_walk() because by the time we enter callback we will be holding a
- * write lock, so regular rte-memseg_walk will deadlock. copying the same
- * iteration code everywhere is not ideal as well. so, use a lockless copy of
- * memseg walk here.
- */
-static int
-memseg_walk_thread_unsafe(rte_memseg_walk_t func, void *arg)
-{
-	struct rte_mem_config *mcfg = rte_eal_get_configuration()->mem_config;
-	int i, ms_idx, ret = 0;
-
-	for (i = 0; i < RTE_MAX_MEMSEG_LISTS; i++) {
-		struct rte_memseg_list *msl = &mcfg->memsegs[i];
-		const struct rte_memseg *ms;
-		struct rte_fbarray *arr;
-
-		if (msl->memseg_arr.count == 0)
-			continue;
-
-		arr = &msl->memseg_arr;
-
-		ms_idx = rte_fbarray_find_next_used(arr, 0);
-		while (ms_idx >= 0) {
-			ms = rte_fbarray_get(arr, ms_idx);
-			ret = func(msl, ms, arg);
-			if (ret < 0)
-				return -1;
-			if (ret > 0)
-				return 1;
-			ms_idx = rte_fbarray_find_next_used(arr, ms_idx + 1);
-		}
-	}
-	return 0;
-}
-
 static int
 is_null_map(const struct user_mem_map *map)
 {
@@ -1357,7 +1321,8 @@ vfio_spapr_dma_mem_map(int vfio_container_fd, uint64_t vaddr, uint64_t iova,
 	/* check if window size needs to be adjusted */
 	memset(&param, 0, sizeof(param));
 
-	if (memseg_walk_thread_unsafe(vfio_spapr_window_size_walk,
+	/* we're inside a callback so use thread-unsafe version */
+	if (rte_memseg_walk_thread_unsafe(vfio_spapr_window_size_walk,
 				&param) < 0) {
 		RTE_LOG(ERR, EAL, "Could not get window size\n");
 		ret = -1;
@@ -1386,7 +1351,9 @@ vfio_spapr_dma_mem_map(int vfio_container_fd, uint64_t vaddr, uint64_t iova,
 				ret = -1;
 				goto out;
 			}
-			if (memseg_walk_thread_unsafe(vfio_spapr_map_walk,
+			/* we're inside a callback, so use thread-unsafe version
+			 */
+			if (rte_memseg_walk_thread_unsafe(vfio_spapr_map_walk,
 					&vfio_container_fd) < 0) {
 				RTE_LOG(ERR, EAL, "Could not recreate DMA maps\n");
 				ret = -1;
diff --git a/lib/librte_eal/rte_eal_version.map b/lib/librte_eal/rte_eal_version.map
index 98bfbe796..72d32fc39 100644
--- a/lib/librte_eal/rte_eal_version.map
+++ b/lib/librte_eal/rte_eal_version.map
@@ -289,6 +289,7 @@ EXPERIMENTAL {
 	rte_memseg_contig_walk_thread_unsafe;
 	rte_memseg_list_walk;
 	rte_memseg_walk;
+	rte_memseg_walk_thread_unsafe;
 	rte_mp_action_register;
 	rte_mp_action_unregister;
 	rte_mp_reply;
-- 
2.17.1

[dpdk-dev] [PATCH 3/3] mem: provide thread-unsafe memseg list walk variant

[Anatoly Burakov]

Sometimes, user code needs to walk memseg list while being inside
a memory-related callback. Rather than making everyone copy around
the same iteration code and depending on DPDK internals, provide an
official way to do memseg_list_walk() inside callbacks.

Also, remove existing reimplementation from memalloc code and use
the new API instead.

Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com>
---
 lib/librte_eal/common/eal_common_memory.c  | 29 +++++++++--------
 lib/librte_eal/common/include/rte_memory.h | 18 +++++++++++
 lib/librte_eal/linuxapp/eal/eal_memalloc.c | 37 +++++-----------------
 lib/librte_eal/rte_eal_version.map         |  1 +
 4 files changed, 43 insertions(+), 42 deletions(-)

diff --git a/lib/librte_eal/common/eal_common_memory.c b/lib/librte_eal/common/eal_common_memory.c
index afe0d5b57..6c4a8d40b 100644
--- a/lib/librte_eal/common/eal_common_memory.c
+++ b/lib/librte_eal/common/eal_common_memory.c
@@ -883,14 +883,11 @@ rte_memseg_walk(rte_memseg_walk_t func, void *arg)
 }
 
 int __rte_experimental
-rte_memseg_list_walk(rte_memseg_list_walk_t func, void *arg)
+rte_memseg_list_walk_thread_unsafe(rte_memseg_list_walk_t func, void *arg)
 {
 	struct rte_mem_config *mcfg = rte_eal_get_configuration()->mem_config;
 	int i, ret = 0;
 
-	/* do not allow allocations/frees/init while we iterate */
-	rte_rwlock_read_lock(&mcfg->memory_hotplug_lock);
-
 	for (i = 0; i < RTE_MAX_MEMSEG_LISTS; i++) {
 		struct rte_memseg_list *msl = &mcfg->memsegs[i];
 
@@ -898,17 +895,23 @@ rte_memseg_list_walk(rte_memseg_list_walk_t func, void *arg)
 			continue;
 
 		ret = func(msl, arg);
-		if (ret < 0) {
-			ret = -1;
-			goto out;
-		}
-		if (ret > 0) {
-			ret = 1;
-			goto out;
-		}
+		if (ret)
+			return ret;
 	}
-out:
+	return 0;
+}
+
+int __rte_experimental
+rte_memseg_list_walk(rte_memseg_list_walk_t func, void *arg)
+{
+	struct rte_mem_config *mcfg = rte_eal_get_configuration()->mem_config;
+	int ret = 0;
+
+	/* do not allow allocations/frees/init while we iterate */
+	rte_rwlock_read_lock(&mcfg->memory_hotplug_lock);
+	ret = rte_memseg_list_walk_thread_unsafe(func, arg);
 	rte_rwlock_read_unlock(&mcfg->memory_hotplug_lock);
+
 	return ret;
 }
 
diff --git a/lib/librte_eal/common/include/rte_memory.h b/lib/librte_eal/common/include/rte_memory.h
index c5a84c333..c4b7f4cff 100644
--- a/lib/librte_eal/common/include/rte_memory.h
+++ b/lib/librte_eal/common/include/rte_memory.h
@@ -299,6 +299,24 @@ rte_memseg_walk_thread_unsafe(rte_memseg_walk_t func, void *arg);
 int __rte_experimental
 rte_memseg_contig_walk_thread_unsafe(rte_memseg_contig_walk_t func, void *arg);
 
+/**
+ * Walk each allocated memseg list without performing any locking.
+ *
+ * @note This function does not perform any locking, and is only safe to call
+ *       from within memory-related callback functions.
+ *
+ * @param func
+ *   Iterator function
+ * @param arg
+ *   Argument passed to iterator
+ * @return
+ *   0 if walked over the entire list
+ *   1 if stopped by the user
+ *   -1 if user function reported error
+ */
+int __rte_experimental
+rte_memseg_list_walk_thread_unsafe(rte_memseg_list_walk_t func, void *arg);
+
 /**
  * Dump the physical memory layout to a file.
  *
diff --git a/lib/librte_eal/linuxapp/eal/eal_memalloc.c b/lib/librte_eal/linuxapp/eal/eal_memalloc.c
index 8c11f98c9..1ebc4b571 100644
--- a/lib/librte_eal/linuxapp/eal/eal_memalloc.c
+++ b/lib/librte_eal/linuxapp/eal/eal_memalloc.c
@@ -171,32 +171,6 @@ get_file_size(int fd)
 	return st.st_size;
 }
 
-/* we cannot use rte_memseg_list_walk() here because we will be holding a
- * write lock whenever we enter every function in this file, however copying
- * the same iteration code everywhere is not ideal as well. so, use a lockless
- * copy of memseg list walk here.
- */
-static int
-memseg_list_walk_thread_unsafe(rte_memseg_list_walk_t func, void *arg)
-{
-	struct rte_mem_config *mcfg = rte_eal_get_configuration()->mem_config;
-	int i, ret = 0;
-
-	for (i = 0; i < RTE_MAX_MEMSEG_LISTS; i++) {
-		struct rte_memseg_list *msl = &mcfg->memsegs[i];
-
-		if (msl->base_va == NULL)
-			continue;
-
-		ret = func(msl, arg);
-		if (ret < 0)
-			return -1;
-		if (ret > 0)
-			return 1;
-	}
-	return 0;
-}
-
 /* returns 1 on successful lock, 0 on unsuccessful lock, -1 on error */
 static int lock(int fd, int type)
 {
@@ -878,7 +852,8 @@ eal_memalloc_alloc_seg_bulk(struct rte_memseg **ms, int n_segs, size_t page_sz,
 	wa.socket = socket;
 	wa.segs_allocated = 0;
 
-	ret = memseg_list_walk_thread_unsafe(alloc_seg_walk, &wa);
+	/* memalloc is locked, so it's safe to use thread-unsafe version */
+	ret = rte_memseg_list_walk_thread_unsafe(alloc_seg_walk, &wa);
 	if (ret == 0) {
 		RTE_LOG(ERR, EAL, "%s(): couldn't find suitable memseg_list\n",
 			__func__);
@@ -943,7 +918,10 @@ eal_memalloc_free_seg_bulk(struct rte_memseg **ms, int n_segs)
 		wa.ms = cur;
 		wa.hi = hi;
 
-		walk_res = memseg_list_walk_thread_unsafe(free_seg_walk, &wa);
+		/* memalloc is locked, so it's safe to use thread-unsafe version
+		 */
+		walk_res = rte_memseg_list_walk_thread_unsafe(free_seg_walk,
+				&wa);
 		if (walk_res == 1)
 			continue;
 		if (walk_res == 0)
@@ -1230,7 +1208,8 @@ eal_memalloc_sync_with_primary(void)
 	if (rte_eal_process_type() == RTE_PROC_PRIMARY)
 		return 0;
 
-	if (memseg_list_walk_thread_unsafe(sync_walk, NULL))
+	/* memalloc is locked, so it's safe to call thread-unsafe version */
+	if (rte_memseg_list_walk_thread_unsafe(sync_walk, NULL))
 		return -1;
 	return 0;
 }
diff --git a/lib/librte_eal/rte_eal_version.map b/lib/librte_eal/rte_eal_version.map
index 72d32fc39..592ffb867 100644
--- a/lib/librte_eal/rte_eal_version.map
+++ b/lib/librte_eal/rte_eal_version.map
@@ -288,6 +288,7 @@ EXPERIMENTAL {
 	rte_memseg_contig_walk;
 	rte_memseg_contig_walk_thread_unsafe;
 	rte_memseg_list_walk;
+	rte_memseg_list_walk_thread_unsafe;
 	rte_memseg_walk;
 	rte_memseg_walk_thread_unsafe;
 	rte_mp_action_register;
-- 
2.17.1

Re: [dpdk-dev] [PATCH 1/3] mem: provide thread-unsafe contig walk variant

[Thomas Monjalon]

12/06/2018 11:46, Anatoly Burakov:
> Sometimes, user code needs to walk memseg list while being inside
> a memory-related callback. Rather than making everyone copy around
> the same iteration code and depending on DPDK internals, provide an
> official way to do memseg_contig_walk() inside callbacks.
> 
> Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com>

Series applied, thanks