Support externally allocated memory in DPDK

When we allocate and use DPDK memory, we need to be able to differentiate between DPDK hugepage segments and segments that were made part of DPDK but are externally allocated. Add such a property to memseg lists.

Signed-off-by Anatoly Burakov <[email protected]>

---
 lib/librte_eal/common/eal_common_memory.c     | 51 ++++++++++++++++---
 .../common/include/rte_eal_memconfig.h        |  1 +
 lib/librte_eal/common/malloc_heap.c           |  2 +-
 3 files changed, 47 insertions(+), 7 deletions(-)

lib/librte_eal/common/eal_common_memory.c +3 -0

index 4f0688f9d..835bbffb6 100644
@@ -24,6 +24,21 @@
 #include "eal_private.h"
 #include "eal_internal_cfg.h"
 
+/* forward declarations for memseg walk functions. we support external
 segments,
+ * but for some functionality to work, we need to either skip or not skip
+ * external segments. for example, while we expect for virt2memseg to

return a
+ * valid memseg even though it's an external memseg, for regular memseg walk we
+ * want to skip those because the expectation is that we will only walk the
+ * DPDK allocated memory.
+ */

lib/librte_eal/common/eal_common_memory.c +2 -0

+static int
+memseg_list_walk(rte_memseg_list_walk_t func, void *arg, bool

skip_external);

lib/librte_eal/common/eal_common_memory.c +10 -4

+static int
+memseg_walk(rte_memseg_walk_t func, void *arg, bool skip_external);
+static int
+memseg_contig_walk(rte_memseg_contig_walk_t func, void *arg,
+               bool skip_external);
+
 /*
  * Try to mmap *size bytes in /dev/zero. If it is successful, return the
  * pointer to the mmap'd area and keep *size unmodified. Else, retry

@@ -621,9 +636,9 @@ rte_mem_iova2virt(rte_iova_t iova)
         * as we know they are PA-contiguous as well
         */
        if (internal_config.legacy_mem)
-               rte_memseg_contig_walk(find_virt_legacy, &vi);
+               memseg_contig_walk(find_virt_legacy, &vi, false);
        else
-               rte_memseg_walk(find_virt, &vi);
+               memseg_walk(find_virt, &vi, false);
 
        return vi.virt;
 }

@@ -787,8 +802,8 @@ rte_mem_lock_page(const void *virt)
        return mlock((void *)aligned, page_size);
 }
 
-int __rte_experimental
-rte_memseg_contig_walk(rte_memseg_contig_walk_t func, void *arg)
+static int
+memseg_contig_walk(rte_memseg_contig_walk_t func, void *arg, bool
 skip_external)
 {
        struct rte_mem_config *mcfg = rte_eal_get_configuration()->

mem_config;
        int i, ms_idx, ret = 0;

lib/librte_eal/common/eal_common_memory.c +9 -1

@@ -803,6 +818,8 @@ rte_memseg_contig_walk(rte_memseg_contig_walk_t func,
 void *arg)
 
                if (msl->memseg_arr.count == 0)
                        continue;
+               if (skip_external && msl->external)
+                       continue;
 
                arr = &msl->memseg_arr;

@@ -837,7 +854,13 @@ 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_contig_walk(rte_memseg_contig_walk_t func, void *arg)
+{
+       return memseg_contig_walk(func, arg, true);
+}
+
+static int
+memseg_walk(rte_memseg_walk_t func, void *arg, bool skip_external)
 {
        struct rte_mem_config *mcfg = rte_eal_get_configuration()->

mem_config;
        int i, ms_idx, ret = 0;

lib/librte_eal/common/eal_common_memory.c +9 -1

@@ -852,6 +875,8 @@ rte_memseg_walk(rte_memseg_walk_t func, void *arg)
 
                if (msl->memseg_arr.count == 0)
                        continue;
+               if (skip_external && msl->external)
+                       continue;
 
                arr = &msl->memseg_arr;
 

@@ -875,7 +900,13 @@ 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_walk(rte_memseg_walk_t func, void *arg)
+{
+       return memseg_walk(func, arg, true);
+}
+
+static int
+memseg_list_walk(rte_memseg_list_walk_t func, void *arg, bool
 skip_external)
 {
        struct rte_mem_config *mcfg = rte_eal_get_configuration()->

mem_config;
        int i, ret = 0;

lib/librte_eal/common/eal_common_memory.c +2 -0

@@ -888,6 +919,8 @@ rte_memseg_list_walk(rte_memseg_list_walk_t func,
 void *arg)
 
                if (msl->base_va == NULL)
                        continue;
+               if (skip_external && msl->external)
+                       continue;
 
                ret = func(msl, arg);

                if (ret < 0) {

lib/librte_eal/common/eal_common_memory.c +6 -0

@@ -904,6 +937,12 @@ rte_memseg_list_walk(rte_memseg_list_walk_t func,
 void *arg)
        return ret;
 }
 
+int __rte_experimental
+rte_memseg_list_walk(rte_memseg_list_walk_t func, void *arg)
+{
+       return memseg_list_walk(func, arg, true);
+}
+
 /* init memory subsystem */
 int

rte_eal_memory_init(void)

lib/librte_eal/common/include/rte_eal_memconfig.h +1 -0

index aff0688dd..4e8720ba6 100644
@@ -30,6 +30,7 @@ struct rte_memseg_list {
                uint64_t addr_64;
                /**< Makes sure addr is always 64-bits */
        };
+       bool external; /**< true if this list points to external memory */
        int socket_id; /**< Socket ID for all memsegs in this list. */
        uint64_t page_sz; /**< Page size for all memsegs in this list. */
        volatile uint32_t version; /**< version number for multiprocess

sync. */

lib/librte_eal/common/ renamed from lib/librte_eal/common/malloc_heap.c +0 -0

malloc_heap.c
index d6cf3af81..8a1f54905 100644

lib/librte_eal/common/malloc_heap.c +1 -2

@@ -631,7 +631,7 @@ malloc_heap_free(struct malloc_elem *elem)
        ret = 0;
 
        /* ...of which we can't avail if we are in legacy mode */
-       if (internal_config.legacy_mem)
+       if (internal_config.legacy_mem || msl->external)
                goto free_unlock;
 
        /* check if we can free any memory back to the system */
--

2.17.1

Quoted 31 lines · Anatoly Burakov

When we allocate and use DPDK memory, we need to be able to differentiate between DPDK hugepage segments and segments that were made part of DPDK but are externally allocated. Add such a property to memseg lists.

Signed-off-by Anatoly Burakov <[email protected]>

---
  lib/librte_eal/common/eal_common_memory.c     | 51 ++++++++++++++++---
  .../common/include/rte_eal_memconfig.h        |  1 +
  lib/librte_eal/common/malloc_heap.c           |  2 +-
  3 files changed, 47 insertions(+), 7 deletions(-)

lib/librte_eal/common/eal_common_memory.c +3 -0

index 4f0688f9d..835bbffb6 100644
@@ -24,6 +24,21 @@
  #include "eal_private.h"
  #include "eal_internal_cfg.h"
 
+/* forward declarations for memseg walk functions. we support external
 segments,
+ * but for some functionality to work, we need to either skip or not skip
+ * external segments. for example, while we expect for virt2memseg to

return a
+ * valid memseg even though it's an external memseg, for regular memseg walk we
+ * want to skip those because the expectation is that we will only walk the
+ * DPDK allocated memory.

Quoted 2 lines

+ */

I do not clear understand when external segments can be used along with hugepages segments, but if this is a possibility, should not we support memseg walk for just external segments and also to allow any walk type as an API? If I'm right, it seems just memseg walk skipping external memsegs can be invoked from other files.

We are preparing to switch to index heap based on heap indexes rather than by NUMA nodes. First few indexes will be equal to NUMA node ID indexes. For example, currently on a machine with NUMA nodes [0, 8], heaps 0 and 8 will be active, while we want to make it so that heaps 0 and 1 are active. However, currently we don't have a function to map a specific NUMA node to a node index, so add it in this patch.

Signed-off-by Anatoly Burakov <[email protected]>

---
 lib/librte_eal/common/eal_common_lcore.c  | 15 +++++++++++++++
 lib/librte_eal/common/include/rte_lcore.h | 19 ++++++++++++++++++-
 lib/librte_eal/rte_eal_version.map        |  1 +
 3 files changed, 34 insertions(+), 1 deletion(-)

lib/librte_eal/common/eal_common_lcore.c +15 -0

index 3167e9d79..579f5a0a1 100644
@@ -132,3 +132,18 @@ rte_socket_id_by_idx(unsigned int idx)
        }
        return config->numa_nodes[idx];
 }
+
+int __rte_experimental
+rte_socket_idx_by_id(unsigned int socket)
+{
+       const struct rte_config *config = rte_eal_get_configuration();
+       int i;
+
+       for (i = 0; i < (int) config->numa_node_count; i++) {
+               unsigned int cur_socket = config->numa_nodes[i];
+               if (cur_socket == socket)
+                       return i;
+       }
+       rte_errno = EINVAL;
+       return -1;
+}

lib/librte_eal/common/include/rte_lcore.h +5 -1

index 6e09d9181..f58cda09a 100644
@@ -156,11 +156,28 @@ rte_socket_count(void);
  *
  * @return
  *   - physical socket id as recognized by EAL
- *   - -1 on error, with errno set to EINVAL
+ *   - -1 on error, with rte_errno set to EINVAL
  */
 int __rte_experimental
 rte_socket_id_by_idx(unsigned int idx);
 
+/**
+ * Return index for a particular socket id.
+ *
+ * This will return position in list of all detected physical socket id's

for a
+ * given socket. For example, on a machine with sockets [0, 8], passing
+ * 8 as a parameter will return 1.
+ *
+ * @param socket
+ * physical socket id to return index for
+ *
+ * @return
+ * - index of physical socket id as recognized by EAL
+ * - -1 on error, with rte_errno set to EINVAL
+ */

lib/librte_eal/common/include/rte_lcore.h +3 -0

+int __rte_experimental
+rte_socket_idx_by_id(unsigned int socket);
+
 /**
  * Get the ID of the physical socket of the specified lcore
  *

lib/librte_eal/rte_eal_ renamed from lib/librte_eal/rte_eal_version.map +0 -0

version.map
index f7dd0e7bc..e7fb37b2a 100644

lib/librte_eal/rte_eal_version.map +1 -1

@@ -296,6 +296,7 @@ EXPERIMENTAL {
        rte_mp_sendmsg;
        rte_socket_count;
        rte_socket_id_by_idx;
+       rte_socket_idx_by_id;
        rte_vfio_dma_map;
        rte_vfio_dma_unmap;
        rte_vfio_get_container_fd;
--

2.17.1

Switch over all parts of EAL to use heap ID instead of NUMA node ID to identify heaps. Heap ID for DPDK-internal heaps is NUMA node's index within the detected NUMA node list.

Signed-off-by Anatoly Burakov <[email protected]>

---
 config/common_base                            |  1 +
 lib/librte_eal/common/eal_common_memzone.c    | 46 ++++++++++------
 .../common/include/rte_eal_memconfig.h        |  4 +-
 lib/librte_eal/common/malloc_heap.c           | 53 ++++++++++++-------
 lib/librte_eal/common/rte_malloc.c            | 28 ++++++----
 5 files changed, 84 insertions(+), 48 deletions(-)

config/common_base +1 -0

index fcf3a1f6f..b0e3937e0 100644
@@ -61,6 +61,7 @@ CONFIG_RTE_CACHE_LINE_SIZE=64
 CONFIG_RTE_LIBRTE_EAL=y
 CONFIG_RTE_MAX_LCORE=128
 CONFIG_RTE_MAX_NUMA_NODES=8
+CONFIG_RTE_MAX_HEAPS=32
 CONFIG_RTE_MAX_MEMSEG_LISTS=64
 # each memseg list will be limited to either RTE_MAX_MEMSEG_PER_LIST pages
 # or RTE_MAX_MEM_MB_PER_LIST megabytes worth of memory, whichever is

smaller

lib/librte_eal/common/eal_common_memzone.c +13 -0

index faa3b0615..25c56052c 100644
@@ -52,6 +52,26 @@ memzone_lookup_thread_unsafe(const char *name)
        return NULL;
 }
 
+static size_t
+heap_max_free_elem(unsigned int heap_idx, unsigned int align)
+{
+       struct rte_malloc_socket_stats stats;
+       struct rte_mem_config *mcfg;
+       size_t len;
+
+       /* get pointer to global configuration */
+       mcfg = rte_eal_get_configuration()->mem_config;
+
+       malloc_heap_get_stats(&mcfg->malloc_heaps[heap_idx], &stats);
+
+       len = stats.greatest_free_size;

lib/librte_eal/common/eal_common_memzone.c +8 -12

@@ -59,29 +79,23 @@ memzone_lookup_thread_unsafe(const char *name)
 static size_t
 find_heap_max_free_elem(int *s, unsigned align)
 {
-       struct rte_mem_config *mcfg;
-       struct rte_malloc_socket_stats stats;
-       int i, socket = *s;
+       unsigned int idx;
+       int socket = *s;
        size_t len = 0;
 
-       /* get pointer to global configuration */
-       mcfg = rte_eal_get_configuration()->mem_config;
-
-       for (i = 0; i < RTE_MAX_NUMA_NODES; i++) {
-               if ((socket != SOCKET_ID_ANY) && (socket != i))
+       for (idx = 0; idx < rte_socket_count(); idx++) {
+               int cur_socket = rte_socket_id_by_idx(idx);
+               if ((socket != SOCKET_ID_ANY) && (socket != cur_socket))
                        continue;
 
-               malloc_heap_get_stats(&mcfg->malloc_heaps[i], &stats);
-               if (stats.greatest_free_size > len) {
-                       len = stats.greatest_free_size;
-                       *s = i;
+               size_t cur_len = heap_max_free_elem(idx, align);
+               if (cur_len > len) {
+                       len = cur_len;

lib/librte_eal/common/eal_common_memzone.c +0 -0

 }
 

static const struct rte_memzone *

lib/librte_eal/common/include/rte_eal_memconfig.h +0 -0

index 4e8720ba6..7e03196a6 100644
@@ -71,8 +71,8 @@ struct rte_mem_config {
 
        struct rte_tailq_head tailq_head[RTE_MAX_TAILQ]; /**< Tailqs for

objects */

lib/librte_eal/common/include/rte_eal_memconfig.h +2 -2

-       /* Heaps of Malloc per socket */
-       struct malloc_heap malloc_heaps[RTE_MAX_NUMA_NODES];
+       /* Heaps of Malloc */
+       struct malloc_heap malloc_heaps[RTE_MAX_HEAPS];
 
        /* address of mem_config in primary process. used to map shared

config into
* exact same address the primary process maps it.

lib/librte_eal/common/ renamed from lib/librte_eal/common/malloc_heap.c +0 -0

malloc_heap.c
index 8a1f54905..e7e1838b1 100644

lib/librte_eal/common/malloc_heap.c +3 -2

@@ -93,9 +93,10 @@ malloc_add_seg(const struct rte_memseg_list *msl,
        struct rte_mem_config *mcfg = rte_eal_get_configuration()->
 mem_config;
        struct rte_memseg_list *found_msl;
        struct malloc_heap *heap;
-       int msl_idx;
+       int msl_idx, heap_idx;
 
-       heap = &mcfg->malloc_heaps[msl->socket_id];
+       heap_idx = rte_socket_idx_by_id(msl->socket_id);
+       heap = &mcfg->malloc_heaps[heap_idx];
 
        /* msl is const, so find it */

        msl_idx = msl - mcfg->memsegs;

lib/librte_eal/common/malloc_heap.c +9 -3

@@ -494,14 +495,20 @@ alloc_more_mem_on_socket(struct malloc_heap *heap,
 size_t size, int socket,
 
 /* this will try lower page sizes first */
 static void *
-heap_alloc_on_socket(const char *type, size_t size, int socket,
-               unsigned int flags, size_t align, size_t bound, bool
 contig)
+malloc_heap_alloc_on_heap_id(const char *type, size_t size,
+               unsigned int heap_id, unsigned int flags, size_t align,
+               size_t bound, bool contig)
 {
        struct rte_mem_config *mcfg = rte_eal_get_configuration()->
 mem_config;
-       struct malloc_heap *heap = &mcfg->malloc_heaps[socket];
+       struct malloc_heap *heap = &mcfg->malloc_heaps[heap_id];
        unsigned int size_flags = flags & ~RTE_MEMZONE_SIZE_HINT_ONLY;
+       int socket_id;
        void *ret;
 
+       /* return NULL if size is 0 or alignment is not power-of-2 */
+       if (size == 0 || (align && !rte_is_power_of_2(align)))
+               return NULL;
+

        rte_spinlock_lock(&(heap->lock));

        align = align == 0 ? 1 : align;

lib/librte_eal/common/malloc_heap.c +0 -0

@@ -521,8 +528,13 @@ heap_alloc_on_socket(const char *type, size_t size,
 int socket,
        if (ret != NULL)

lib/librte_eal/common/malloc_heap.c +3 -0

+
+       if (!alloc_more_mem_on_socket(heap, size, socket_id, flags, align,
+                       bound, contig)) {

ret = heap_alloc(heap, type, size, flags, align, bound,
contig);

                /* this should have succeeded */

lib/librte_eal/common/malloc_heap.c +1 -5

@@ -538,13 +550,9 @@ void *
 malloc_heap_alloc(const char *type, size_t size, int socket_arg,
                unsigned int flags, size_t align, size_t bound, bool
 contig)
 {
-       int socket, i, cur_socket;
+       int socket, heap_id, i;
        void *ret;
 
-       /* return NULL if size is 0 or alignment is not power-of-2 */
-       if (size == 0 || (align && !rte_is_power_of_2(align)))
-               return NULL;
-
        if (!rte_eal_has_hugepages())
                socket_arg = SOCKET_ID_ANY;

@@ -557,18 +565,23 @@ malloc_heap_alloc(const char *type, size_t size, int
 socket_arg,

lib/librte_eal/common/malloc_heap.c +6 -4

+               return NULL;
+
+       ret = malloc_heap_alloc_on_heap_id(type, size, heap_id, flags,
 align,
+                       bound, contig);
        if (ret != NULL || socket_arg != SOCKET_ID_ANY)
                return ret;
 
        /* try other heaps */
        for (i = 0; i < (int) rte_socket_count(); i++) {
-               cur_socket = rte_socket_id_by_idx(i);
-               if (cur_socket == socket)
+               if (i == heap_id)
                        continue;
-               ret = heap_alloc_on_socket(type, size, cur_socket, flags,
-                               align, bound, contig);
+               ret = malloc_heap_alloc_on_heap_id(type, size, i, flags,

align,

lib/librte_eal/common/malloc_heap.c +2 -1

+                               bound, contig);
                if (ret != NULL)
                        return ret;
        }

@@ -788,7 +801,7 @@ malloc_heap_resize(struct malloc_elem *elem, size_t
 size)
 }
 
 /*
- * Function to retrieve data for heap on given socket
+ * Function to retrieve data for a given heap
  */
 int

malloc_heap_get_stats(struct malloc_heap *heap,

lib/librte_eal/common/malloc_heap.c +1 -1

@@ -826,7 +839,7 @@ malloc_heap_get_stats(struct malloc_heap *heap,
 }
 
 /*
- * Function to retrieve data for heap on given socket
+ * Function to retrieve data for a given heap
  */
 void
 malloc_heap_dump(struct malloc_heap *heap, FILE *f)

lib/librte_eal/common/rte_malloc.c +7 -1

index b51a6d111..4387bc494 100644
@@ -152,11 +152,17 @@ rte_malloc_get_socket_stats(int socket,
                struct rte_malloc_socket_stats *socket_stats)
 {
        struct rte_mem_config *mcfg = rte_eal_get_configuration()->
 mem_config;
+       int heap_idx;
 
        if (socket >= RTE_MAX_NUMA_NODES || socket < 0)
                return -1;
 
-       return malloc_heap_get_stats(&mcfg->malloc_heaps[socket],
 socket_stats);
+       heap_idx = rte_socket_idx_by_id(socket);
+       if (heap_idx < 0)
+               return -1;
+
+       return malloc_heap_get_stats(&mcfg->malloc_heaps[heap_idx],
+                       socket_stats);
 }

/*

lib/librte_eal/common/rte_malloc.c +3 -4

@@ -168,10 +174,9 @@ rte_malloc_dump_heaps(FILE *f)
        struct rte_mem_config *mcfg = rte_eal_get_configuration()->
 mem_config;
        unsigned int idx;
 
-       for (idx = 0; idx < rte_socket_count(); idx++) {
-               unsigned int socket = rte_socket_id_by_idx(idx);
-               fprintf(f, "Heap on socket %i:\n", socket);
-               malloc_heap_dump(&mcfg->malloc_heaps[socket], f);
+       for (idx = 0; idx < RTE_MAX_HEAPS; idx++) {
+               fprintf(f, "Heap id: %u\n", idx);
+               malloc_heap_dump(&mcfg->malloc_heaps[idx], f);
        }
 

}

lib/librte_eal/common/rte_malloc.c +8 -5

@@ -182,14 +187,17 @@ rte_malloc_dump_heaps(FILE *f)
 void
 rte_malloc_dump_stats(FILE *f, __rte_unused const char *type)
 {
-       unsigned int socket;
+       struct rte_mem_config *mcfg = rte_eal_get_configuration()->
 mem_config;
+       unsigned int heap_id;
        struct rte_malloc_socket_stats sock_stats;
+
        /* Iterate through all initialised heaps */
-       for (socket=0; socket< RTE_MAX_NUMA_NODES; socket++) {
-               if ((rte_malloc_get_socket_stats(socket, &sock_stats) <
 0))
-                       continue;
+       for (heap_id = 0; heap_id < RTE_MAX_HEAPS; heap_id++) {
+               struct malloc_heap *heap = &mcfg->malloc_heaps[heap_id];
 
-               fprintf(f, "Socket:%u\n", socket);
+               malloc_heap_get_stats(heap, &sock_stats);
+
+               fprintf(f, "Heap id:%u\n", heap_id);
                fprintf(f, "\tHeap_size:%zu,\n",

sock_stats.heap_totalsz_bytes);
                fprintf(f, "\tFree_size:%zu,\n",
sock_stats.heap_freesz_bytes);
                fprintf(f, "\tAlloc_size:%zu,\n",

sock_stats.heap_allocsz_bytes);
--
2.17.1

Quoted 21 lines

lib/librte_eal/common/eal_common_memzone.c +7 -9

-       /* get pointer to global configuration */
-       mcfg = rte_eal_get_configuration()->mem_config;
-
-       for (i = 0; i < RTE_MAX_NUMA_NODES; i++) {
-               if ((socket != SOCKET_ID_ANY) && (socket != i))
+       for (idx = 0; idx < rte_socket_count(); idx++) {
+               int cur_socket = rte_socket_id_by_idx(idx);
+               if ((socket != SOCKET_ID_ANY) && (socket != cur_socket))
                         continue;
 
-               malloc_heap_get_stats(&mcfg->malloc_heaps[i], &stats);
-               if (stats.greatest_free_size > len) {
-                       len = stats.greatest_free_size;
-                       *s = i;
+               size_t cur_len = heap_max_free_elem(idx, align);
+               if (cur_len > len) {
+                       len = cur_len;
+                       *s = cur_socket;
                 }
         }

lib/librte_eal/common/eal_common_memzone.c memzone_lookup_thread_unsafe(const char *name) · line 98

Quoted 6 lines

lib/librte_eal/common/eal_common_memzone.c +1 -4

-       if (len < MALLOC_ELEM_OVERHEAD + align)
-               return 0;
-
-       return len - MALLOC_ELEM_OVERHEAD - align;
+       return len;

Is it worth to set *s to some safe value if no space at all?

We will need to refer to external heaps in some way. While we use heap ID's internally, for external API use it has to be something more user-friendly. So, we will be using a string to uniquely identify a heap.

Signed-off-by Anatoly Burakov <[email protected]>

---
 lib/librte_eal/common/include/rte_malloc_heap.h |  2 ++
 lib/librte_eal/common/malloc_heap.c             | 13 +++++++++++++
 lib/librte_eal/common/rte_malloc.c              |  1 +
 3 files changed, 16 insertions(+)

lib/librte_eal/common/include/rte_malloc_heap.h +2 -0

index d43fa9097..bd64dff03 100644
@@ -12,6 +12,7 @@
 
 /* Number of free lists per heap, grouped by size. */
 #define RTE_HEAP_NUM_FREELISTS  13
+#define RTE_HEAP_NAME_MAX_LEN 32
 
 /* dummy definition, for pointers */
 struct malloc_elem;

@@ -27,6 +28,7 @@ struct malloc_heap {
 
        unsigned alloc_count;
        size_t total_size;
+       char name[RTE_HEAP_NAME_MAX_LEN];
 } __rte_cache_aligned;
 
 #endif /* _RTE_MALLOC_HEAP_H_ */

lib/librte_eal/common/ renamed from lib/librte_eal/common/malloc_heap.c +0 -0

malloc_heap.c
index e7e1838b1..8f22c062b 100644

lib/librte_eal/common/malloc_heap.c +13 -0

@@ -848,6 +848,7 @@ malloc_heap_dump(struct malloc_heap *heap, FILE *f)
 
        rte_spinlock_lock(&heap->lock);
 
+       fprintf(f, "Heap name: %s\n", heap->name);
        fprintf(f, "Heap size: 0x%zx\n", heap->total_size);
        fprintf(f, "Heap alloc count: %u\n", heap->alloc_count);
 

@@ -864,6 +865,18 @@ int
 rte_eal_malloc_heap_init(void)
 {
        struct rte_mem_config *mcfg = rte_eal_get_configuration()->
 mem_config;
+       unsigned int i;
+
+       /* assign names to default DPDK heaps */
+       for (i = 0; i < rte_socket_count(); i++) {
+               struct malloc_heap *heap = &mcfg->malloc_heaps[i];
+               char heap_name[RTE_HEAP_NAME_MAX_LEN];
+               int socket_id = rte_socket_id_by_idx(i);
+
+               snprintf(heap_name, sizeof(heap_name) - 1,
+                               "socket_%i", socket_id);
+               strlcpy(heap->name, heap_name, RTE_HEAP_NAME_MAX_LEN);
+       }
 
        if (register_mp_requests()) {

RTE_LOG(ERR, EAL, "Couldn't register malloc multiprocess
actions\n");

lib/librte_eal/common/rte_malloc.c +1 -0

index 4387bc494..75d6e0b4d 100644
@@ -198,6 +198,7 @@ rte_malloc_dump_stats(FILE *f, __rte_unused const char
 *type)
                malloc_heap_get_stats(heap, &sock_stats);
 
                fprintf(f, "Heap id:%u\n", heap_id);
+               fprintf(f, "\tHeap name:%s\n", heap->name);
                fprintf(f, "\tHeap_size:%zu,\n",
 sock_stats.heap_totalsz_bytes);

                fprintf(f, "\tFree_size:%zu,\n",
sock_stats.heap_freesz_bytes);
                fprintf(f, "\tAlloc_size:%zu,\n",

sock_stats.heap_allocsz_bytes);
--
2.17.1

Add internal functions to look up heap by name, and enable dumping statistics for a specified named heap.

Signed-off-by Anatoly Burakov <[email protected]>

---
 lib/librte_eal/common/include/rte_malloc.h | 19 +++++++++++--
 lib/librte_eal/common/malloc_heap.c        | 31 ++++++++++++++++++++++
 lib/librte_eal/common/malloc_heap.h        |  6 +++++
 lib/librte_eal/common/rte_malloc.c         | 17 ++++++++++++
 lib/librte_eal/rte_eal_version.map         |  1 +
 5 files changed, 72 insertions(+), 2 deletions(-)

lib/librte_eal/common/include/rte_malloc.h +17 -2

index a9fb7e452..7cbcd3184 100644
@@ -256,13 +256,28 @@ rte_malloc_validate(const void *ptr, size_t *size);
  * @param socket_stats
  *   A structure which provides memory to store statistics
  * @return
- *   Null on error
- *   Pointer to structure storing statistics on success
+ *   0 on success
+ *   -1 on error
  */
 int
 rte_malloc_get_socket_stats(int socket,
                struct rte_malloc_socket_stats *socket_stats);
 
+/**
+ * Get heap statistics for the specified heap.
+ *
+ * @param socket
+ *   An unsigned integer specifying the socket to get heap statistics for
+ * @param socket_stats
+ *   A structure which provides memory to store statistics
+ * @return
+ *   0 on success
+ *   -1 on error
+ */
+int __rte_experimental
+rte_malloc_get_stats_from_heap(const char *heap_name,
+               struct rte_malloc_socket_stats *socket_stats);
+
 /**
  * Dump statistics.
  *

lib/librte_eal/common/malloc_heap.c +3 -0

index 8f22c062b..8437d33b3 100644
@@ -614,6 +614,37 @@ malloc_heap_free_pages(void *aligned_start, size_t
 aligned_len)
        return 0;
 }
 
+int
+malloc_heap_find_named_heap_idx(const char *heap_name)
+{

lib/librte_eal/common/malloc_heap.c +3 -0

+                       heap_idx++) {
+               struct malloc_heap *heap = &mcfg->malloc_heaps[heap_idx];
+               if (strncmp(heap->name, heap_name, RTE_HEAP_NAME_MAX_LEN)

== 0)

lib/librte_eal/common/malloc_heap.c +9 -0

+                       return heap_idx;
+       }
+       return -1;
+}
+
+struct malloc_heap *
+malloc_heap_find_named_heap(const char *heap_name)
+{
+       struct rte_mem_config *mcfg = rte_eal_get_configuration()->m

em_config;

lib/librte_eal/common/malloc_heap.c +8 -0

+       int heap_idx;
+
+       heap_idx = malloc_heap_find_named_heap_idx(heap_name);
+       if (heap_idx < 0)
+               return NULL;
+       return &mcfg->malloc_heaps[heap_idx];
+}
+
 int
 malloc_heap_free(struct malloc_elem *elem)
 {

lib/librte_eal/common/malloc_heap.h +6 -0

index 03b801415..4f3137260 100644
@@ -29,6 +29,12 @@ void *
 malloc_heap_alloc(const char *type, size_t size, int socket, unsigned int
 flags,
                size_t align, size_t bound, bool contig);
 
+int
+malloc_heap_find_named_heap_idx(const char *name);
+
+struct malloc_heap *
+malloc_heap_find_named_heap(const char *name);
+
 int
 malloc_heap_free(struct malloc_elem *elem);

lib/librte_eal/common/rte_malloc.c +17 -0

index 75d6e0b4d..2508abdb1 100644
@@ -165,6 +165,23 @@ rte_malloc_get_socket_stats(int socket,
                        socket_stats);
 }
 
+/*
+ * Function to retrieve data for heap on given socket
+ */
+int __rte_experimental
+rte_malloc_get_stats_from_heap(const char *heap_name,
+               struct rte_malloc_socket_stats *socket_stats)
+{
+       struct malloc_heap *heap;
+
+       heap = malloc_heap_find_named_heap(heap_name);
+
+       if (heap == NULL)
+               return -1;
+
+       return malloc_heap_get_stats(heap, socket_stats);
+}
+
 /*
  * Function to dump contents of all heaps
  */

lib/librte_eal/rte_eal_version.map +1 -1

index e7fb37b2a..786df1e39 100644
@@ -278,6 +278,7 @@ EXPERIMENTAL {
        rte_fbarray_set_used;
        rte_log_register_type_and_pick_level;
        rte_malloc_dump_heaps;
+       rte_malloc_get_stats_from_heap;
        rte_mem_alloc_validator_register;
        rte_mem_alloc_validator_unregister;
        rte_mem_event_callback_register;
--

2.17.1

Add new malloc API to allocate memory from heap referenced to by specified name.

Signed-off-by Anatoly Burakov <[email protected]>

---
 lib/librte_eal/common/include/rte_malloc.h | 25 ++++++++++++++++++++++
 lib/librte_eal/common/malloc_heap.c        |  2 +-
 lib/librte_eal/common/malloc_heap.h        |  6 ++++++
 lib/librte_eal/common/rte_malloc.c         | 11 ++++++++++
 lib/librte_eal/rte_eal_version.map         |  1 +
 5 files changed, 44 insertions(+), 1 deletion(-)

lib/librte_eal/common/include/rte_malloc.h +8 -0

index 7cbcd3184..f1bcd9b65 100644
@@ -213,6 +213,31 @@ rte_zmalloc_socket(const char *type, size_t size,
 unsigned align, int socket);
 void *
 rte_calloc_socket(const char *type, size_t num, size_t size, unsigned
 align, int socket);
 
+/**
+ * This function allocates memory from a specified named heap.
+ *
+ * @param name
+ *   Name of the heap to allocate from.
+ * @param type
+ *   A string identifying the type of allocated objects (useful for debug
+ *   purposes, such as identifying the cause of a memory leak). Can be

NULL.
+ * @param size
+ * Size (in bytes) to be allocated.
+ * @param align
+ * If 0, the return is a pointer that is suitably aligned for any kind
of

+ *   variable (in the same manner as malloc()).
+ *   Otherwise, the return is a pointer that is a multiple of *align*. In
+ *   this case, it must be a power of two. (Minimum alignment is the
+ *   cacheline size, i.e. 64-bytes)

+ * @return
+ * - NULL on error. Not enough memory, or invalid arguments (size is 0,
+ * align is not a power of two).
+ * - Otherwise, the pointer to the allocated object.
+ */

lib/librte_eal/common/include/rte_malloc.h +4 -0

+__rte_experimental void *
+rte_malloc_from_heap(const char *heap_name, const char *type, size_t size,
+               unsigned int align);
+
 /**
  * Frees the memory space pointed to by the provided pointer.
  *

lib/librte_eal/common/ renamed from lib/librte_eal/common/malloc_heap.c +0 -0

malloc_heap.c
index 8437d33b3..a33acc252 100644

lib/librte_eal/common/malloc_heap.c +1 -1

@@ -494,7 +494,7 @@ alloc_more_mem_on_socket(struct malloc_heap *heap,
 size_t size, int socket,
 }
 
 /* this will try lower page sizes first */
-static void *
+void *
 malloc_heap_alloc_on_heap_id(const char *type, size_t size,
                unsigned int heap_id, unsigned int flags, size_t align,

size_t bound, bool contig)

lib/librte_eal/common/ renamed from lib/librte_eal/common/malloc_heap.h +0 -0

malloc_heap.h
index 4f3137260..a7e408c99 100644

lib/librte_eal/common/malloc_heap.h +6 -0

@@ -29,6 +29,12 @@ void *
 malloc_heap_alloc(const char *type, size_t size, int socket, unsigned int
 flags,
                size_t align, size_t bound, bool contig);
 
+/* allocate from specified heap id */
+void *
+malloc_heap_alloc_on_heap_id(const char *type, size_t size,
+               unsigned int heap_id, unsigned int flags, size_t align,
+               size_t bound, bool contig);
+
 int
 malloc_heap_find_named_heap_idx(const char *name);

lib/librte_eal/common/rte_malloc.c +0 -0

index 2508abdb1..215876a59 100644
@@ -55,6 +55,17 @@ rte_malloc_socket(const char *type, size_t size,
 unsigned int align,
                        align == 0 ? 1 : align, 0, false);

lib/librte_eal/common/rte_malloc.c +4 -0

+       return malloc_heap_alloc_on_heap_id(type, size, heap_idx, 0,
+                       align == 0 ? 1 : align, 0, false);
+}
+
 /*

* Allocate memory on default heap.
*/

lib/librte_eal/rte_eal_ renamed from lib/librte_eal/rte_eal_version.map +0 -0

version.map
index 786df1e39..716a7585d 100644

lib/librte_eal/rte_eal_version.map +1 -1

@@ -278,6 +278,7 @@ EXPERIMENTAL {
        rte_fbarray_set_used;
        rte_log_register_type_and_pick_level;
        rte_malloc_dump_heaps;
+       rte_malloc_from_heap;
        rte_malloc_get_stats_from_heap;
        rte_mem_alloc_validator_register;
        rte_mem_alloc_validator_unregister;
--

2.17.1

Add an API to add externally allocated memory to malloc heap. The memory will be stored in memseg lists like regular DPDK memory. Multiple segments are allowed within a heap. If IOVA table is not provided, IOVA addresses are filled in with RTE_BAD_IOVA.

Signed-off-by Anatoly Burakov <[email protected]>

---
 lib/librte_eal/common/include/rte_malloc.h | 44 ++++++++++++++
 lib/librte_eal/common/malloc_heap.c        | 70 ++++++++++++++++++++++
 lib/librte_eal/common/malloc_heap.h        |  4 ++
 lib/librte_eal/common/rte_malloc.c         | 39 ++++++++++++
 lib/librte_eal/rte_eal_version.map         |  1 +
 5 files changed, 158 insertions(+)

lib/librte_eal/common/include/rte_malloc.h +4 -0

index fa6de073a..5f933993b 100644
@@ -274,6 +274,50 @@ rte_free(void *ptr);
 int __rte_experimental
 rte_malloc_heap_create(const char *heap_name);
 
+/**
+ * Add more memory to heap with specified name.
+ *
+ * @note Concurrently adding memory to or removing memory from different

heaps
+ * is not safe.
+ *
+ * @note This function does not need to be called in multiple processes, as
+ * multiprocess synchronization will happen automatically as far as
heap data
+ * is concerned. However, before accessing pointers to memory in this
heap, it
+ * is responsibility of the user to ensure that the heap memory is
accessible
+ * in all processes.
+ *
+ * @note Memory must be previously allocated for DPDK to be able to use it as a
+ * malloc heap. Failing to do so will result in undefined behavior, up
to and
+ * including crashes.

+ *
+ * @note Adding memory to heap may fail in multiple processes scenario, as
+ * attaching to ``rte_fbarray`` structures may not always be successful
in
+ * secondary processes.
+ *
+ * @param heap_name
+ * Name of the heap to create.

+ * @param va_addr
+ * Start of virtual area to add to the heap.
+ * @param len
+ * Length of virtual area to add to the heap.
+ * @param iova_addrs
+ * Array of page IOVA addresses corresponding to each page in this
memory
+ * area. Can be NULL, in which case page IOVA addresses will be set to
+ * RTE_BAD_IOVA.
+ * @param n_pages
+ * Number of elements in the iova_addrs array. Must be zero of
``iova_addrs``
+ * is NULL.
+ * @param page_sz
+ * Page size of the underlying memory.
+ *
+ * @return
+ * - 0 on successful creation.
+ * - -1 on error.
+ */

lib/librte_eal/common/include/rte_malloc.h +2 -0

+int __rte_experimental
+rte_malloc_heap_add_memory(const char *heap_name, void *va_addr, size_t

len,
+ rte_iova_t iova_addrs[], unsigned int n_pages, size_t
page_sz);

lib/librte_eal/common/include/rte_malloc.h +1 -0

+
 /**
  * If malloc debug is enabled, check a memory block for header
  * and trailer markers to indicate that all is well with the block.

lib/librte_eal/common/malloc_heap.c +44 -0

index f5d103626..29446cac9 100644
@@ -892,6 +892,76 @@ malloc_heap_dump(struct malloc_heap *heap, FILE *f)
        rte_spinlock_unlock(&heap->lock);
 }
 
+int
+malloc_heap_add_external_memory(struct malloc_heap *heap, void *va_addr,
+               rte_iova_t iova_addrs[], unsigned int n_pages, size_t
 page_sz)
+{
+       struct rte_mem_config *mcfg = rte_eal_get_configuration()->m
 em_config;
+       char fbarray_name[RTE_FBARRAY_NAME_LEN];
+       struct rte_memseg_list *msl = NULL;
+       struct rte_fbarray *arr;
+       size_t seg_len = n_pages * page_sz;
+       unsigned int i;
+
+       /* first, find a free memseg list */
+       for (i = 0; i < RTE_MAX_MEMSEG_LISTS; i++) {
+               struct rte_memseg_list *tmp = &mcfg->memsegs[i];
+               if (tmp->base_va == NULL) {
+                       msl = tmp;
+                       break;
+               }
+       }
+       if (msl == NULL) {
+               RTE_LOG(ERR, EAL, "Couldn't find empty memseg list\n");
+               rte_errno = ENOSPC;
+               return -1;
+       }
+
+       snprintf(fbarray_name, sizeof(fbarray_name) - 1, "%s_%p",
+                       heap->name, va_addr);
+
+       /* create the backing fbarray */
+       if (rte_fbarray_init(&msl->memseg_arr, fbarray_name, n_pages,
+                       sizeof(struct rte_memseg)) < 0) {
+               RTE_LOG(ERR, EAL, "Couldn't create fbarray backing the
 memseg list\n");
+               return -1;
+       }
+       arr = &msl->memseg_arr;
+
+       /* fbarray created, fill it up */
+       for (i = 0; i < n_pages; i++) {
+               struct rte_memseg *ms;
+
+               rte_fbarray_set_used(arr, i);
+               ms = rte_fbarray_get(arr, i);
+               ms->addr = RTE_PTR_ADD(va_addr, n_pages * page_sz);
+               ms->iova = iova_addrs == NULL ? RTE_BAD_IOVA :

iova_addrs[i];

lib/librte_eal/common/malloc_heap.c +26 -0

+               ms->hugepage_sz = page_sz;
+               ms->len = page_sz;
+               ms->nchannel = rte_memory_get_nchannel();
+               ms->nrank = rte_memory_get_nrank();
+               ms->socket_id = -1; /* invalid socket ID */
+       }
+
+       /* set up the memseg list */
+       msl->base_va = va_addr;
+       msl->page_sz = page_sz;
+       msl->socket_id = -1; /* invalid socket ID */
+       msl->version = 0;
+       msl->external = true;
+
+       /* now, add newly minted memory to the malloc heap */
+       malloc_heap_add_memory(heap, msl, va_addr, seg_len);
+
+       heap->total_size += seg_len;
+
+       /* all done! */
+       RTE_LOG(DEBUG, EAL, "Added segment for heap %s starting at %p\n",
+                       heap->name, va_addr);
+
+       return 0;
+}
+
 int
 malloc_heap_create(struct malloc_heap *heap, const char *heap_name)
 {

lib/librte_eal/common/malloc_heap.h +4 -0

index aa819ef65..3be4656d0 100644
@@ -38,6 +38,10 @@ malloc_heap_alloc_on_heap_id(const char *type, size_t
 size,
 int
 malloc_heap_create(struct malloc_heap *heap, const char *heap_name);
 
+int
+malloc_heap_add_external_memory(struct malloc_heap *heap, void *va_addr,
+               rte_iova_t iova_addrs[], unsigned int n_pages, size_t
 page_sz);
+
 int

malloc_heap_find_named_heap_idx(const char *name);

lib/librte_eal/common/rte_malloc.c +39 -0

index e000dc5b7..db0f604ad 100644
@@ -274,6 +274,45 @@ rte_malloc_virt2iova(const void *addr)
        return ms->iova + RTE_PTR_DIFF(addr, ms->addr);
 }
 
+int
+rte_malloc_heap_add_memory(const char *heap_name, void *va_addr, size_t
 len,
+               rte_iova_t iova_addrs[], unsigned int n_pages, size_t
 page_sz)
+{
+       struct malloc_heap *heap = NULL;
+       unsigned int n;
+       int ret;
+
+       /* iova_addrs is allowed to be NULL */
+       if (heap_name == NULL || va_addr == NULL ||
+                       /* n_pages can be 0 if iova_addrs is NULL */
+                       ((iova_addrs != NULL) == (n_pages == 0)) ||
+                       page_sz == 0 || !rte_is_power_of_2(page_sz) ||
+                       strnlen(heap_name, RTE_HEAP_NAME_MAX_LEN) == 0 ||
+                       strnlen(heap_name, RTE_HEAP_NAME_MAX_LEN) ==
+                               RTE_HEAP_NAME_MAX_LEN) {
+               rte_errno = EINVAL;
+               return -1;
+       }
+       /* find our heap */
+       heap = malloc_heap_find_named_heap(heap_name);
+       if (heap == NULL) {
+               rte_errno = EINVAL;
+               return -1;
+       }
+       n = len / page_sz;
+       if (n != n_pages && iova_addrs != NULL) {
+               rte_errno = EINVAL;
+               return -1;
+       }
+
+       rte_spinlock_lock(&heap->lock);
+       ret = malloc_heap_add_external_memory(heap, va_addr, iova_addrs,
 n,
+                       page_sz);
+       rte_spinlock_unlock(&heap->lock);
+
+       return ret;
+}
+

int
rte_malloc_heap_create(const char *heap_name)
{

lib/librte_eal/rte_eal_version.map +1 -1

index f3c375156..6290cc910 100644
@@ -280,6 +280,7 @@ EXPERIMENTAL {
        rte_malloc_dump_heaps;
        rte_malloc_from_heap;
        rte_malloc_get_stats_from_heap;
+       rte_malloc_heap_add_memory;
        rte_malloc_heap_create;
        rte_mem_alloc_validator_register;
        rte_mem_alloc_validator_unregister;
--

2.17.1

This is a proposal to enable using externally allocated memory in DPDK.

In a nutshell, here is what is being done here:

- Index malloc heaps by NUMA node index, rather than NUMA node itself
- Add identifier string to malloc heap, to uniquely identify it
- Allow creating named heaps and add/remove memory to/from those heaps
- Allocate memseg lists at runtime, to keep track of IOVA addresses of externally allocated memory
- If IOVA addresses aren't provided, use RTE_BAD_IOVA
- Allow malloc and memzones to allocate from named heaps

The responsibility to ensure memory is accessible before using it is on the shoulders of the user - there is no checking done with regards to validity of the memory (nor could there be...).

The following limitations are present:

- No multiprocess support
- No thread safety

There is currently no way to allocate memory during initialization stage, so even if multiprocess support is added, it is not guaranteed to work because of underlying issues with mapping fbarrays in secondary processes. This is not an issue in single process scenario, but it may be an issue in a multiprocess scenario in case where primary doesn't intend to share the externally allocated memory, yet adding such memory could fail because some other process failed to attach to this shared memory when it wasn't needed.

Anatoly Burakov (11):
mem: allow memseg lists to be marked as external
eal: add function to rerieve socket index by socket ID
malloc: index heaps using heap ID rather than NUMA node
malloc: add name to malloc heaps
malloc: enable retrieving statistics from named heaps
malloc: enable allocating from named heaps
malloc: enable creating new malloc heaps
malloc: allow adding memory to named heaps
malloc: allow removing memory from named heaps
malloc: allow destroying heaps
memzone: enable reserving memory from named heaps

  config/common_base                            |   1 +
  lib/librte_eal/common/eal_common_lcore.c      |  15 +
  lib/librte_eal/common/eal_common_memory.c     |  51 +++-
  lib/librte_eal/common/eal_common_memzone.c    | 283 ++++++++++++++----
  .../common/include/rte_eal_memconfig.h        |   5 +-
  lib/librte_eal/common/include/rte_lcore.h     |  19 +-
  lib/librte_eal/common/include/rte_malloc.h    | 158 +++++++++-
  .../common/include/rte_malloc_heap.h          |   2 +
  lib/librte_eal/common/include/rte_memzone.h   | 183 +++++++++++
  lib/librte_eal/common/malloc_heap.c           | 277 +++++++++++++++--
  lib/librte_eal/common/malloc_heap.h           |  26 ++
  lib/librte_eal/common/rte_malloc.c            | 197 +++++++++++-
  lib/librte_eal/rte_eal_version.map            |  10 +
  13 files changed, 1118 insertions(+), 109 deletions(-)

Quoted 36 lines · Anatoly Burakov

This is a proposal to enable using externally allocated memory in DPDK.
In a nutshell, here is what is being done here:
- Index malloc heaps by NUMA node index, rather than NUMA node itself
- Add identifier string to malloc heap, to uniquely identify it
- Allow creating named heaps and add/remove memory to/from those heaps
- Allocate memseg lists at runtime, to keep track of IOVA addresses of externally allocated memory
- If IOVA addresses aren't provided, use RTE_BAD_IOVA
- Allow malloc and memzones to allocate from named heaps The responsibility to ensure memory is accessible before using it is on the shoulders of the user - there is no checking done with regards to validity of the memory (nor could there be...).
The following limitations are present:
- No multiprocess support
- No thread safety
There is currently no way to allocate memory during initialization stage, so even if multiprocess support is added, it is not guaranteed to work because of underlying issues with mapping fbarrays in secondary processes. This is not an issue in single process scenario, but it may be an issue in a multiprocess scenario in case where primary doesn't intend to share the externally allocated memory, yet adding such memory could fail because some other process failed to attach to this shared memory when it wasn't needed.
Anatoly Burakov (11):
mem: allow memseg lists to be marked as external
eal: add function to rerieve socket index by socket ID
malloc: index heaps using heap ID rather than NUMA node
malloc: add name to malloc heaps
malloc: enable retrieving statistics from named heaps
malloc: enable allocating from named heaps
malloc: enable creating new malloc heaps
malloc: allow adding memory to named heaps
malloc: allow removing memory from named heaps
malloc: allow destroying heaps
memzone: enable reserving memory from named heaps

Quoted 14 lines

 config/common_base                            |   1 +
 lib/librte_eal/common/eal_common_lcore.c      |  15 +
 lib/librte_eal/common/eal_common_memory.c     |  51 +++-
 lib/librte_eal/common/eal_common_memzone.c    | 283 ++++++++++++++----
 .../common/include/rte_eal_memconfig.h        |   5 +-
 lib/librte_eal/common/include/rte_lcore.h     |  19 +-
 lib/librte_eal/common/include/rte_malloc.h    | 158 +++++++++-
 .../common/include/rte_malloc_heap.h          |   2 +
 lib/librte_eal/common/include/rte_memzone.h   | 183 +++++++++++
 lib/librte_eal/common/malloc_heap.c           | 277 +++++++++++++++--
 lib/librte_eal/common/malloc_heap.h           |  26 ++
 lib/librte_eal/common/rte_malloc.c            | 197 +++++++++++-
 lib/librte_eal/rte_eal_version.map            |  10 +
 13 files changed, 1118 insertions(+), 109 deletions(-)

So, now that the RFC is out, i would like to ask a general question.

One other thing that this patchset is missing, is the ability for data structures (e.g. hash, mempool, etc.) to be allocated from external heaps. Currently, we can kinda sorta do that with various _init() API's (initializing a data structure over already allocated memzone), but this is not ideal and is a hassle for anyone using external memory in DPDK.

There are basically four ways to approach this problem (that i can see).

First way is to change "socket ID" to mean "heap ID" everywhere. This has an upside of having a consistent API to allocate from internal and external heaps, with little to no API additions, only internal changes to account for the fact that "socket ID" is now "heap ID".

However, there is a massive downside to this approach: it is a *giant* API change, and it's also a giant *ABI-compatible* API change. Meaning, replacing socket ID with heap ID will not cause compile failures for old code, which would result in many subtle bugs in already existing codebases. So, while in the perfect world this would've been my preferred approach, realistically i think this is a very, very bad idea.

Second one is to add a separate "heap name" API's to everything. This has an upside of clean separation between allocation from internal and external heaps. (well, whether it's an upside is debatable...) This is the approach i expected to take when i was creating this patchset.

The downside is that we have to add new API's to every library and every DPDK data structure, to allow explicit allocation from external heaps. We will have to maintain both, and things like hardware drivers will need to have a way to indicate the need to allocate things from a particular external heap.

The third way is to expose the "heap ID" externally, and allow a single, unified API to reserve memory. That is, create an API that would map either a NUMA node ID or a heap name to an ID, and allow reserving memory through that ID regardless of whether it's internal or external memory. This would also allow to gradually phase out socket-based ID's in favor of heap ID API, should we choose to do so.

The downside for this is, it adds a layer of indirection between socket ID and reserving memory on a particular NUMA node, and it makes it hard to produce a single value of "heap ID" in such a way as to replicate current functionality of allocating with SOCKET_ID_ANY. Most likely user will have to explicitly try to allocate on all sockets, unless we keep old API's around in parallel.

Finally, a fourth way would be to abuse the socket ID to also mean something else, which is an approach i've seen numerous times already, and one that i don't like. We could register new heaps as a new, fake socket ID, and use that to address external heaps (each heap would get its own socket). So, keep current socket ID behavior, but for non-existent sockets it would be possible to be registered as a fake socket pointing to an external heap.

The upside for this approach would be that no API changes are required whatsoever to existing libraries - this scheme is compatible with both internal and external heaps without adding a separate API.

The downside is bad semantics - "special" sockets, handling of SOCKET_ID_ANY, handling of "invalid socket" vs. "invalid socket that happens to correspond to an existing external heap", and many other things that can be confusing. I don't like this option, but it's an option :)

Thoughts? Comments?

I myself still favor the second way, however there are good arguments to be made for each of these options.

Signature · 2 lines

Thanks,
Anatoly

Quoted 37 lines · Anatoly

Quoted 33 lines · Anatoly Burakov

This is a proposal to enable using externally allocated memory in DPDK.
In a nutshell, here is what is being done here:
- Index malloc heaps by NUMA node index, rather than NUMA node itself
- Add identifier string to malloc heap, to uniquely identify it
- Allow creating named heaps and add/remove memory to/from those heaps
- Allocate memseg lists at runtime, to keep track of IOVA addresses of externally allocated memory
- If IOVA addresses aren't provided, use RTE_BAD_IOVA
- Allow malloc and memzones to allocate from named heaps The responsibility to ensure memory is accessible before using it is on the shoulders of the user - there is no checking done with regards to validity of the memory (nor could there be...).
The following limitations are present:
- No multiprocess support
- No thread safety
There is currently no way to allocate memory during initialization stage, so even if multiprocess support is added, it is not guaranteed to work because of underlying issues with mapping fbarrays in secondary processes. This is not an issue in single process scenario, but it may be an issue in a multiprocess scenario in case where primary doesn't intend to share the externally allocated memory, yet adding such memory could fail because some other process failed to attach to this shared memory when it wasn't needed. Anatoly Burakov (11):
mem: allow memseg lists to be marked as external
eal: add function to rerieve socket index by socket ID
malloc: index heaps using heap ID rather than NUMA node
malloc: add name to malloc heaps
malloc: enable retrieving statistics from named heaps
malloc: enable allocating from named heaps
malloc: enable creating new malloc heaps

Quoted 8 lines

Quoted 8 lines

malloc: allow adding memory to named heaps
malloc: allow removing memory from named heaps
malloc: allow destroying heaps
memzone: enable reserving memory from named heaps

 config/common_base                            |   1 +
 lib/librte_eal/common/eal_common_lcore.c      |  15 +
 lib/librte_eal/common/eal_common_memory.c     |  51 +++-
 lib/librte_eal/common/eal_common_memzone.c    | 283

++++++++++++++----

Quoted 11 lines

Quoted 10 lines

 .../common/include/rte_eal_memconfig.h        |   5 +-
 lib/librte_eal/common/include/rte_lcore.h     |  19 +-
 lib/librte_eal/common/include/rte_malloc.h    | 158 +++++++++-
 .../common/include/rte_malloc_heap.h          |   2 +
 lib/librte_eal/common/include/rte_memzone.h   | 183 +++++++++++
 lib/librte_eal/common/malloc_heap.c           | 277 +++++++++++++++--
 lib/librte_eal/common/malloc_heap.h           |  26 ++
 lib/librte_eal/common/rte_malloc.c            | 197 +++++++++++-
 lib/librte_eal/rte_eal_version.map            |  10 +
 13 files changed, 1118 insertions(+), 109 deletions(-)

Quoted 3 lines

So, now that the RFC is out, i would like to ask a general question.

One other thing that this patchset is missing, is the ability for data

structures (e.g. hash, mempool, etc.) to be allocated from external heaps. Currently, we can kinda sorta do that with various _init() API's (initializing a data structure over already allocated memzone), but this is not ideal and is a hassle for anyone using external memory in DPDK.

Quoted 4 lines

There are basically four ways to approach this problem (that i can see).

First way is to change "socket ID" to mean "heap ID" everywhere. This has

an upside of having a consistent API to allocate from internal and external heaps, with little to no API additions, only internal changes to account for the fact that "socket ID" is now "heap ID".

Quoted 2 lines

However, there is a massive downside to this approach: it is a *giant* API

change, and it's also a giant *ABI-compatible* API change. Meaning, replacing socket ID with heap ID will not cause compile failures for old code, which would result in many subtle bugs in already existing codebases. So, while in the perfect world this would've been my preferred approach, realistically i think this is a very, very bad idea.

Quoted 2 lines

Second one is to add a separate "heap name" API's to everything. This has

an upside of clean separation between allocation from internal and external heaps. (well, whether it's an upside is debatable...) This is the approach i expected to take when i was creating this patchset.

Quoted 2 lines

The downside is that we have to add new API's to every library and every

DPDK data structure, to allow explicit allocation from external heaps. We will have to maintain both, and things like hardware drivers will need to have a way to indicate the need to allocate things from a particular external heap.

Quoted 2 lines

The third way is to expose the "heap ID" externally, and allow a single,

unified API to reserve memory. That is, create an API that would map either a NUMA node ID or a heap name to an ID, and allow reserving memory through that ID regardless of whether it's internal or external memory. This would also allow to gradually phase out socket-based ID's in favor of heap ID API, should we choose to do so.

Quoted 2 lines

The downside for this is, it adds a layer of indirection between socket ID

and reserving memory on a particular NUMA node, and it makes it hard to produce a single value of "heap ID" in such a way as to replicate current functionality of allocating with SOCKET_ID_ANY. Most likely user will have to explicitly try to allocate on all sockets, unless we keep old API's around in parallel.

Quoted 2 lines

Finally, a fourth way would be to abuse the socket ID to also mean

something else, which is an approach i've seen numerous times already, and one that i don't like. We could register new heaps as a new, fake socket ID, and use that to address external heaps (each heap would get its own socket). So, keep current socket ID behavior, but for non-existent sockets it would be possible to be registered as a fake socket pointing to an external heap.

Quoted 2 lines

The upside for this approach would be that no API changes are required

whatsoever to existing libraries - this scheme is compatible with both internal and external heaps without adding a separate API.

Quoted 8 lines

The downside is bad semantics - "special" sockets, handling of SOCKET_ID_ANY, handling of "invalid socket" vs. "invalid socket that happens to correspond to an existing external heap", and many other things that can be confusing. I don't like this option, but it's an option :)

Thoughts? Comments?

#1 is super clean, but very disruptive to everyone. Very Bad IMO #2 is also clean, but adds a lot of new APIs that everyone needs to use or at least in the external heap cases.
#3 not sure I fully understand it, but reproducing heap IDs for testing is a problem and requires new/old APIs

#4 Very easy to add, IMO it is clean and very small disruption to developers. It does require the special handling, but I feel it is OK and can be explained in the docs. Having a socket id as an ‘int’ gives us a lot room e.g. id < 64K is normal socket and > 64K is external id.

My vote would be #4, as it seems the least risk and work. :-)

Quoted 2 lines

I myself still favor the second way, however there are good arguments to

be made for each of these options.

Quoted 4 lines

--
Thanks,
Anatoly

Regards,
Keith

Quoted 135 lines · Keith

Quoted 37 lines · Anatoly

Quoted 33 lines · Anatoly Burakov

This is a proposal to enable using externally allocated memory in DPDK.
In a nutshell, here is what is being done here:
- Index malloc heaps by NUMA node index, rather than NUMA node itself
- Add identifier string to malloc heap, to uniquely identify it
- Allow creating named heaps and add/remove memory to/from those heaps
- Allocate memseg lists at runtime, to keep track of IOVA addresses of externally allocated memory
- If IOVA addresses aren't provided, use RTE_BAD_IOVA
- Allow malloc and memzones to allocate from named heaps The responsibility to ensure memory is accessible before using it is on the shoulders of the user - there is no checking done with regards to validity of the memory (nor could there be...).
The following limitations are present:
- No multiprocess support
- No thread safety
There is currently no way to allocate memory during initialization stage, so even if multiprocess support is added, it is not guaranteed to work because of underlying issues with mapping fbarrays in secondary processes. This is not an issue in single process scenario, but it may be an issue in a multiprocess scenario in case where primary doesn't intend to share the externally allocated memory, yet adding such memory could fail because some other process failed to attach to this shared memory when it wasn't needed. Anatoly Burakov (11):
mem: allow memseg lists to be marked as external
eal: add function to rerieve socket index by socket ID
malloc: index heaps using heap ID rather than NUMA node
malloc: add name to malloc heaps
malloc: enable retrieving statistics from named heaps
malloc: enable allocating from named heaps
malloc: enable creating new malloc heaps

Quoted 8 lines

Quoted 8 lines

malloc: allow adding memory to named heaps
malloc: allow removing memory from named heaps
malloc: allow destroying heaps
memzone: enable reserving memory from named heaps

  config/common_base                            |   1 +
  lib/librte_eal/common/eal_common_lcore.c      |  15 +
  lib/librte_eal/common/eal_common_memory.c     |  51 +++-
  lib/librte_eal/common/eal_common_memzone.c    | 283

++++++++++++++----

Quoted 11 lines

Quoted 10 lines

  .../common/include/rte_eal_memconfig.h        |   5 +-
  lib/librte_eal/common/include/rte_lcore.h     |  19 +-
  lib/librte_eal/common/include/rte_malloc.h    | 158 +++++++++-
  .../common/include/rte_malloc_heap.h          |   2 +
  lib/librte_eal/common/include/rte_memzone.h   | 183 +++++++++++
  lib/librte_eal/common/malloc_heap.c           | 277 +++++++++++++++--
  lib/librte_eal/common/malloc_heap.h           |  26 ++
  lib/librte_eal/common/rte_malloc.c            | 197 +++++++++++-
  lib/librte_eal/rte_eal_version.map            |  10 +
  13 files changed, 1118 insertions(+), 109 deletions(-)

Quoted 3 lines

So, now that the RFC is out, i would like to ask a general question.

One other thing that this patchset is missing, is the ability for data

structures (e.g. hash, mempool, etc.) to be allocated from external heaps. Currently, we can kinda sorta do that with various _init() API's (initializing a data structure over already allocated memzone), but this is not ideal and is a hassle for anyone using external memory in DPDK.

Quoted 4 lines

There are basically four ways to approach this problem (that i can see).

First way is to change "socket ID" to mean "heap ID" everywhere. This has

an upside of having a consistent API to allocate from internal and external heaps, with little to no API additions, only internal changes to account for the fact that "socket ID" is now "heap ID".

Quoted 2 lines

However, there is a massive downside to this approach: it is a *giant* API

change, and it's also a giant *ABI-compatible* API change. Meaning, replacing socket ID with heap ID will not cause compile failures for old code, which would result in many subtle bugs in already existing codebases. So, while in the perfect world this would've been my preferred approach, realistically i think this is a very, very bad idea.

Quoted 2 lines

Second one is to add a separate "heap name" API's to everything. This has

an upside of clean separation between allocation from internal and external heaps. (well, whether it's an upside is debatable...) This is the approach i expected to take when i was creating this patchset.

Quoted 2 lines

The downside is that we have to add new API's to every library and every

DPDK data structure, to allow explicit allocation from external heaps. We will have to maintain both, and things like hardware drivers will need to have a way to indicate the need to allocate things from a particular external heap.

Quoted 2 lines

The third way is to expose the "heap ID" externally, and allow a single,

unified API to reserve memory. That is, create an API that would map either a NUMA node ID or a heap name to an ID, and allow reserving memory through that ID regardless of whether it's internal or external memory. This would also allow to gradually phase out socket-based ID's in favor of heap ID API, should we choose to do so.

Quoted 2 lines

The downside for this is, it adds a layer of indirection between socket ID

and reserving memory on a particular NUMA node, and it makes it hard to produce a single value of "heap ID" in such a way as to replicate current functionality of allocating with SOCKET_ID_ANY. Most likely user will have to explicitly try to allocate on all sockets, unless we keep old API's around in parallel.

Quoted 2 lines

Finally, a fourth way would be to abuse the socket ID to also mean

something else, which is an approach i've seen numerous times already, and one that i don't like. We could register new heaps as a new, fake socket ID, and use that to address external heaps (each heap would get its own socket). So, keep current socket ID behavior, but for non-existent sockets it would be possible to be registered as a fake socket pointing to an external heap.

Quoted 2 lines

The upside for this approach would be that no API changes are required

whatsoever to existing libraries - this scheme is compatible with both internal and external heaps without adding a separate API.

Quoted 8 lines

The downside is bad semantics - "special" sockets, handling of SOCKET_ID_ANY, handling of "invalid socket" vs. "invalid socket that happens to correspond to an existing external heap", and many other things that can be confusing. I don't like this option, but it's an option :)

Thoughts? Comments?

#1 is super clean, but very disruptive to everyone. Very Bad IMO #2 is also clean, but adds a lot of new APIs that everyone needs to use or at least in the external heap cases.
#3 not sure I fully understand it, but reproducing heap IDs for testing is a problem and requires new/old APIs

#4 Very easy to add, IMO it is clean and very small disruption to developers. It does require the special handling, but I feel it is OK and can be explained in the docs. Having a socket id as an ‘int’ gives us a lot room e.g. id < 64K is normal socket and > 64K is external id.

Quoted 2 lines

My vote would be #4, as it seems the least risk and work. :-)

We are living with #4 (overloaded socket_ids) since ~5 years now but it indeed generates some confusion and it is a kind of hack so it may not be the best choice going forward in official releases but for sure is the easiest/simplest solution requiring the least modifications.
Using an overloaded socket_id is especially disturbing in the dump memory config printout where the user will see multiple socket ids on a single socket system or more than the available real number of sockets, however it could still be explained in the notes and the documentation.
The allocation behavior with SOCKET_ID_ANY is also a question as I think it shouldn’t roll over to allocate memory in the external heap, we especially disabled this feature in our implementation. The reason behind is that the external memory may be a limited resource where only explicit allocation requests would be allowed and also in a multi-process environment we may not want all external heaps to be mapped into all other processes address space meaning that not all heaps are accessible from every process (I’m not sure if it is planned to be supported though but it would be an important and useful feature based on our experiences).

Anyway I think the confusion with this option comes due to the misleading “socket_id” name which would not really mean socket id anymore. So we should probably just document it as pseudo socket_id and problem solved with #4 :)

The cleanest solution in my opinion would be #1 which could be combined with #4 so that the physical socket_id could be directly passed as the heap_id (or rather call it allocation id or just location?) so that backward compatibility could also be kept.
Meaning to apply #1, change “socket_id” to “heap_id” (or “alloc_id”?) in all functions which are today expecting the socket_id to indicate the location of the allocations but keep a direct mapping from socket_id to heap_id, e.g. as Keith suggested lower range of heap_id would be equivalent to the socket_id and upper range would be the external id, this way even existing applications would still work without changing the code just by passing the socket_id in the heap_id parameter. However it is a question what would happen with the socket_id stored in data structures such as struct rte_mempool where socket_id is stored with a meaning “Socket id passed at create.” SOCKET_ID_ANY would only mean lower range of heap_ids (physical socket ids) but not the external heap and if needed a new HEAP_ID_ANY could be introduced.

If changing heap_id to socket_id in existing functions is a big issue then one option would be to keep the original API and introduce new equivalent functions allowing to use the heap_id instead of the socket_id, e.g. rte_mempool_create would have an equivalent function to use with the heap_id instead of the socket_id. Socket_id could then be converted to heap_id with a new function which should always be possible and can still use to direct mapping approach with lower/upper range convention.
The socket_id based functions would then just be wrappers calling the heap_id equivalent function after converting the socket_id to heap_id.
Using socket_id to indicate the location could still be relevant so the old socket_id based functions may not even need to be deprecated unless it would become hard to maintain.

Irrespective of the chosen option, external heaps should be registered/identified by name and there could be a function to fetch/lookup the id (heap_id or pseudo socket_id) by registered heap name which then could be used in the related API calls.

It would also be another work item to update all the data structures which are storing the socket_id to use it as the location identifier and I think few of them may need to store both the real physical socket_id and the heap_id, e.g. in struct lcore_config where the user may want to know the real physical socket id but want to set specific heap_id as the default allocation location for the given lcore.

Quoted 11 lines

Quoted 3 lines

I myself still favor the second way, however there are good arguments to

be made for each of these options.

Quoted 4 lines

--
Thanks,
Anatoly

Regards,
Keith

Thanks,
Laszlo