From 2e378ff29740013d3b0e2bbae1f31c9509d0dfd2 Mon Sep 17 00:00:00 2001 From: Anatoly Burakov Date: Wed, 11 Apr 2018 13:30:43 +0100 Subject: [PATCH] mem: add validator callback This API will enable application to register for notifications on page allocations that are about to happen, giving the application a chance to allow or deny the allocation when total memory utilization as a result would be above specified limit on specified socket. Signed-off-by: Anatoly Burakov Tested-by: Santosh Shukla Tested-by: Hemant Agrawal Tested-by: Gowrishankar Muthukrishnan --- lib/librte_eal/common/eal_common_memalloc.c | 138 +++++++++++++++++++- lib/librte_eal/common/eal_common_memory.c | 26 ++++ lib/librte_eal/common/eal_memalloc.h | 10 ++ lib/librte_eal/common/include/rte_memory.h | 63 +++++++++ lib/librte_eal/rte_eal_version.map | 2 + 5 files changed, 238 insertions(+), 1 deletion(-) diff --git a/lib/librte_eal/common/eal_common_memalloc.c b/lib/librte_eal/common/eal_common_memalloc.c index 2d2d46f0ab..49fd53c689 100644 --- a/lib/librte_eal/common/eal_common_memalloc.c +++ b/lib/librte_eal/common/eal_common_memalloc.c @@ -22,14 +22,26 @@ struct mem_event_callback_entry { rte_mem_event_callback_t clb; }; +struct mem_alloc_validator_entry { + TAILQ_ENTRY(mem_alloc_validator_entry) next; + char name[RTE_MEM_ALLOC_VALIDATOR_NAME_LEN]; + rte_mem_alloc_validator_t clb; + int socket_id; + size_t limit; +}; + /** Double linked list of actions. */ TAILQ_HEAD(mem_event_callback_entry_list, mem_event_callback_entry); +TAILQ_HEAD(mem_alloc_validator_entry_list, mem_alloc_validator_entry); static struct mem_event_callback_entry_list mem_event_callback_list = TAILQ_HEAD_INITIALIZER(mem_event_callback_list); - static rte_rwlock_t mem_event_rwlock = RTE_RWLOCK_INITIALIZER; +static struct mem_alloc_validator_entry_list mem_alloc_validator_list = + TAILQ_HEAD_INITIALIZER(mem_alloc_validator_list); +static rte_rwlock_t mem_alloc_validator_rwlock = RTE_RWLOCK_INITIALIZER; + static struct mem_event_callback_entry * find_mem_event_callback(const char *name) { @@ -42,6 +54,18 @@ find_mem_event_callback(const char *name) return r; } +static struct mem_alloc_validator_entry * +find_mem_alloc_validator(const char *name, int socket_id) +{ + struct mem_alloc_validator_entry *r; + + TAILQ_FOREACH(r, &mem_alloc_validator_list, next) { + if (!strcmp(r->name, name) && r->socket_id == socket_id) + break; + } + return r; +} + bool eal_memalloc_is_contig(const struct rte_memseg_list *msl, void *start, size_t len) @@ -221,3 +245,115 @@ eal_memalloc_mem_event_notify(enum rte_mem_event event, const void *start, rte_rwlock_read_unlock(&mem_event_rwlock); } + +int +eal_memalloc_mem_alloc_validator_register(const char *name, + rte_mem_alloc_validator_t clb, int socket_id, size_t limit) +{ + struct mem_alloc_validator_entry *entry; + int ret, len; + if (name == NULL || clb == NULL || socket_id < 0) { + rte_errno = EINVAL; + return -1; + } + len = strnlen(name, RTE_MEM_ALLOC_VALIDATOR_NAME_LEN); + if (len == 0) { + rte_errno = EINVAL; + return -1; + } else if (len == RTE_MEM_ALLOC_VALIDATOR_NAME_LEN) { + rte_errno = ENAMETOOLONG; + return -1; + } + rte_rwlock_write_lock(&mem_alloc_validator_rwlock); + + entry = find_mem_alloc_validator(name, socket_id); + if (entry != NULL) { + rte_errno = EEXIST; + ret = -1; + goto unlock; + } + + entry = malloc(sizeof(*entry)); + if (entry == NULL) { + rte_errno = ENOMEM; + ret = -1; + goto unlock; + } + + /* callback successfully created and is valid, add it to the list */ + entry->clb = clb; + entry->socket_id = socket_id; + entry->limit = limit; + snprintf(entry->name, RTE_MEM_ALLOC_VALIDATOR_NAME_LEN, "%s", name); + TAILQ_INSERT_TAIL(&mem_alloc_validator_list, entry, next); + + ret = 0; + + RTE_LOG(DEBUG, EAL, "Mem alloc validator '%s' on socket %i with limit %zu registered\n", + name, socket_id, limit); + +unlock: + rte_rwlock_write_unlock(&mem_alloc_validator_rwlock); + return ret; +} + +int +eal_memalloc_mem_alloc_validator_unregister(const char *name, int socket_id) +{ + struct mem_alloc_validator_entry *entry; + int ret, len; + + if (name == NULL || socket_id < 0) { + rte_errno = EINVAL; + return -1; + } + len = strnlen(name, RTE_MEM_ALLOC_VALIDATOR_NAME_LEN); + if (len == 0) { + rte_errno = EINVAL; + return -1; + } else if (len == RTE_MEM_ALLOC_VALIDATOR_NAME_LEN) { + rte_errno = ENAMETOOLONG; + return -1; + } + rte_rwlock_write_lock(&mem_alloc_validator_rwlock); + + entry = find_mem_alloc_validator(name, socket_id); + if (entry == NULL) { + rte_errno = ENOENT; + ret = -1; + goto unlock; + } + TAILQ_REMOVE(&mem_alloc_validator_list, entry, next); + free(entry); + + ret = 0; + + RTE_LOG(DEBUG, EAL, "Mem alloc validator '%s' on socket %i unregistered\n", + name, socket_id); + +unlock: + rte_rwlock_write_unlock(&mem_alloc_validator_rwlock); + return ret; +} + +int +eal_memalloc_mem_alloc_validate(int socket_id, size_t new_len) +{ + struct mem_alloc_validator_entry *entry; + int ret = 0; + + rte_rwlock_read_lock(&mem_alloc_validator_rwlock); + + TAILQ_FOREACH(entry, &mem_alloc_validator_list, next) { + if (entry->socket_id != socket_id || entry->limit > new_len) + continue; + RTE_LOG(DEBUG, EAL, "Calling mem alloc validator '%s' on socket %i\n", + entry->name, entry->socket_id); + if (entry->clb(socket_id, entry->limit, new_len) < 0) + ret = -1; + } + + rte_rwlock_read_unlock(&mem_alloc_validator_rwlock); + + return ret; +} diff --git a/lib/librte_eal/common/eal_common_memory.c b/lib/librte_eal/common/eal_common_memory.c index 1f15ff77c0..24a9ed5225 100644 --- a/lib/librte_eal/common/eal_common_memory.c +++ b/lib/librte_eal/common/eal_common_memory.c @@ -652,6 +652,32 @@ rte_mem_event_callback_unregister(const char *name) return eal_memalloc_mem_event_callback_unregister(name); } +int __rte_experimental +rte_mem_alloc_validator_register(const char *name, + rte_mem_alloc_validator_t clb, int socket_id, size_t limit) +{ + /* FreeBSD boots with legacy mem enabled by default */ + if (internal_config.legacy_mem) { + RTE_LOG(DEBUG, EAL, "Registering mem alloc validators not supported\n"); + rte_errno = ENOTSUP; + return -1; + } + return eal_memalloc_mem_alloc_validator_register(name, clb, socket_id, + limit); +} + +int __rte_experimental +rte_mem_alloc_validator_unregister(const char *name, int socket_id) +{ + /* FreeBSD boots with legacy mem enabled by default */ + if (internal_config.legacy_mem) { + RTE_LOG(DEBUG, EAL, "Registering mem alloc validators not supported\n"); + rte_errno = ENOTSUP; + return -1; + } + return eal_memalloc_mem_alloc_validator_unregister(name, socket_id); +} + /* Dump the physical memory layout on console */ void rte_dump_physmem_layout(FILE *f) diff --git a/lib/librte_eal/common/eal_memalloc.h b/lib/librte_eal/common/eal_memalloc.h index 98b31da2cb..6736fa3d4f 100644 --- a/lib/librte_eal/common/eal_memalloc.h +++ b/lib/librte_eal/common/eal_memalloc.h @@ -66,4 +66,14 @@ void eal_memalloc_mem_event_notify(enum rte_mem_event event, const void *start, size_t len); +int +eal_memalloc_mem_alloc_validator_register(const char *name, + rte_mem_alloc_validator_t clb, int socket_id, size_t limit); + +int +eal_memalloc_mem_alloc_validator_unregister(const char *name, int socket_id); + +int +eal_memalloc_mem_alloc_validate(int socket_id, size_t new_len); + #endif /* EAL_MEMALLOC_H */ diff --git a/lib/librte_eal/common/include/rte_memory.h b/lib/librte_eal/common/include/rte_memory.h index 3e4c09f551..01e7548dc4 100644 --- a/lib/librte_eal/common/include/rte_memory.h +++ b/lib/librte_eal/common/include/rte_memory.h @@ -364,6 +364,69 @@ rte_mem_event_callback_register(const char *name, rte_mem_event_callback_t clb); int __rte_experimental rte_mem_event_callback_unregister(const char *name); + +#define RTE_MEM_ALLOC_VALIDATOR_NAME_LEN 64 +/**< maximum length of alloc validator name */ +/** + * Function typedef used to register memory allocation validation callbacks. + * + * Returning 0 will allow allocation attempt to continue. Returning -1 will + * prevent allocation from succeeding. + */ +typedef int (*rte_mem_alloc_validator_t)(int socket_id, + size_t cur_limit, size_t new_len); + +/** + * @brief Register validator callback for memory allocations. + * + * Callbacks registered by this function will be called right before memory + * allocator is about to trigger allocation of more pages from the system if + * said allocation will bring total memory usage above specified limit on + * specified socket. User will be able to cancel pending allocation if callback + * returns -1. + * + * @note callbacks will happen while memory hotplug subsystem is write-locked, + * therefore some functions (e.g. `rte_memseg_walk()`) will cause a + * deadlock when called from within such callbacks. + * + * @param name + * Name associated with specified callback to be added to the list. + * + * @param clb + * Callback function pointer. + * + * @param socket_id + * Socket ID on which to watch for allocations. + * + * @param limit + * Limit above which to trigger callbacks. + * + * @return + * 0 on successful callback register + * -1 on unsuccessful callback register, with rte_errno value indicating + * reason for failure. + */ +int __rte_experimental +rte_mem_alloc_validator_register(const char *name, + rte_mem_alloc_validator_t clb, int socket_id, size_t limit); + +/** + * @brief Unregister validator callback for memory allocations. + * + * @param name + * Name associated with specified callback to be removed from the list. + * + * @param socket_id + * Socket ID on which to watch for allocations. + * + * @return + * 0 on successful callback unregister + * -1 on unsuccessful callback unregister, with rte_errno value indicating + * reason for failure. + */ +int __rte_experimental +rte_mem_alloc_validator_unregister(const char *name, int socket_id); + #ifdef __cplusplus } #endif diff --git a/lib/librte_eal/rte_eal_version.map b/lib/librte_eal/rte_eal_version.map index d1ac9ea7a6..2b5b1dcf50 100644 --- a/lib/librte_eal/rte_eal_version.map +++ b/lib/librte_eal/rte_eal_version.map @@ -238,6 +238,8 @@ EXPERIMENTAL { rte_fbarray_set_used; rte_log_register_type_and_pick_level; rte_malloc_dump_heaps; + rte_mem_alloc_validator_register; + rte_mem_alloc_validator_unregister; rte_mem_event_callback_register; rte_mem_event_callback_unregister; rte_mem_iova2virt; -- 2.20.1