#include <rte_log.h>
#include <rte_memory.h>
#include <rte_memzone.h>
-#include <rte_tailq.h>
#include <rte_eal.h>
#include <rte_eal_memconfig.h>
#include <rte_per_lcore.h>
#include <rte_string_fns.h>
#include <rte_common.h>
+#include "malloc_heap.h"
+#include "malloc_elem.h"
#include "eal_private.h"
-/* internal copy of free memory segments */
-static struct rte_memseg *free_memseg = NULL;
-
static inline const struct rte_memzone *
memzone_lookup_thread_unsafe(const char *name)
{
const struct rte_mem_config *mcfg;
+ const struct rte_memzone *mz;
unsigned i = 0;
/* get pointer to global configuration */
* the algorithm is not optimal (linear), but there are few
* zones and this function should be called at init only
*/
- for (i = 0; i < RTE_MAX_MEMZONE && mcfg->memzone[i].addr != NULL; i++) {
- if (!strncmp(name, mcfg->memzone[i].name, RTE_MEMZONE_NAMESIZE))
+ for (i = 0; i < RTE_MAX_MEMZONE; i++) {
+ mz = &mcfg->memzone[i];
+ if (mz->addr != NULL && !strncmp(name, mz->name, RTE_MEMZONE_NAMESIZE))
return &mcfg->memzone[i];
}
return NULL;
}
-/*
- * Return a pointer to a correctly filled memzone descriptor. If the
- * allocation cannot be done, return NULL.
- */
-const struct rte_memzone *
-rte_memzone_reserve(const char *name, size_t len, int socket_id,
- unsigned flags)
+static inline struct rte_memzone *
+get_next_free_memzone(void)
{
- return rte_memzone_reserve_aligned(name,
- len, socket_id, flags, CACHE_LINE_SIZE);
-}
+ struct rte_mem_config *mcfg;
+ unsigned i = 0;
-/*
- * Helper function for memzone_reserve_aligned_thread_unsafe().
- * Calculate address offset from the start of the segment.
- * Align offset in that way that it satisfy istart alignmnet and
- * buffer of the requested length would not cross specified boundary.
- */
-static inline phys_addr_t
-align_phys_boundary(const struct rte_memseg *ms, size_t len, size_t align,
- size_t bound)
-{
- phys_addr_t addr_offset, bmask, end, start;
- size_t step;
+ /* get pointer to global configuration */
+ mcfg = rte_eal_get_configuration()->mem_config;
- step = RTE_MAX(align, bound);
- bmask = ~((phys_addr_t)bound - 1);
+ for (i = 0; i < RTE_MAX_MEMZONE; i++) {
+ if (mcfg->memzone[i].addr == NULL)
+ return &mcfg->memzone[i];
+ }
- /* calculate offset to closest alignment */
- start = RTE_ALIGN_CEIL(ms->phys_addr, align);
- addr_offset = start - ms->phys_addr;
+ return NULL;
+}
- while (addr_offset + len < ms->len) {
+/* This function will return the greatest free block if a heap has been
+ * specified. If no heap has been specified, it will return the heap and
+ * length of the greatest free block available in all heaps */
+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;
+ size_t len = 0;
- /* check, do we meet boundary condition */
- end = start + len - (len != 0);
- if ((start & bmask) == (end & bmask))
- break;
+ /* 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))
+ continue;
- /* calculate next offset */
- start = RTE_ALIGN_CEIL(start + 1, step);
- addr_offset = start - ms->phys_addr;
+ malloc_heap_get_stats(&mcfg->malloc_heaps[i], &stats);
+ if (stats.greatest_free_size > len) {
+ len = stats.greatest_free_size;
+ *s = i;
+ }
}
- return (addr_offset);
+ if (len < MALLOC_ELEM_OVERHEAD + align)
+ return 0;
+
+ return len - MALLOC_ELEM_OVERHEAD - align;
}
static const struct rte_memzone *
memzone_reserve_aligned_thread_unsafe(const char *name, size_t len,
int socket_id, unsigned flags, unsigned align, unsigned bound)
{
+ struct rte_memzone *mz;
struct rte_mem_config *mcfg;
- unsigned i = 0;
- int memseg_idx = -1;
- uint64_t addr_offset, seg_offset = 0;
size_t requested_len;
- size_t memseg_len = 0;
- phys_addr_t memseg_physaddr;
- void *memseg_addr;
+ int socket, i;
/* get pointer to global configuration */
mcfg = rte_eal_get_configuration()->mem_config;
/* no more room in config */
- if (mcfg->memzone_idx >= RTE_MAX_MEMZONE) {
+ if (mcfg->memzone_cnt >= RTE_MAX_MEMZONE) {
RTE_LOG(ERR, EAL, "%s(): No more room in config\n", __func__);
rte_errno = ENOSPC;
return NULL;
return NULL;
}
+ if (strlen(name) >= sizeof(mz->name) - 1) {
+ RTE_LOG(DEBUG, EAL, "%s(): memzone <%s>: name too long\n",
+ __func__, name);
+ rte_errno = EEXIST;
+ return NULL;
+ }
+
/* if alignment is not a power of two */
- if (!rte_is_power_of_2(align)) {
+ if (align && !rte_is_power_of_2(align)) {
RTE_LOG(ERR, EAL, "%s(): Invalid alignment: %u\n", __func__,
align);
rte_errno = EINVAL;
}
/* alignment less than cache size is not allowed */
- if (align < CACHE_LINE_SIZE)
- align = CACHE_LINE_SIZE;
-
+ if (align < RTE_CACHE_LINE_SIZE)
+ align = RTE_CACHE_LINE_SIZE;
/* align length on cache boundary. Check for overflow before doing so */
- if (len > SIZE_MAX - CACHE_LINE_MASK) {
+ if (len > SIZE_MAX - RTE_CACHE_LINE_MASK) {
rte_errno = EINVAL; /* requested size too big */
return NULL;
}
- len += CACHE_LINE_MASK;
- len &= ~((size_t) CACHE_LINE_MASK);
+ len += RTE_CACHE_LINE_MASK;
+ len &= ~((size_t) RTE_CACHE_LINE_MASK);
/* save minimal requested length */
- requested_len = RTE_MAX((size_t)CACHE_LINE_SIZE, len);
+ requested_len = RTE_MAX((size_t)RTE_CACHE_LINE_SIZE, len);
/* check that boundary condition is valid */
- if (bound != 0 &&
- (requested_len > bound || !rte_is_power_of_2(bound))) {
+ if (bound != 0 && (requested_len > bound || !rte_is_power_of_2(bound))) {
rte_errno = EINVAL;
return NULL;
}
- /* find the smallest segment matching requirements */
- for (i = 0; i < RTE_MAX_MEMSEG; i++) {
- /* last segment */
- if (free_memseg[i].addr == NULL)
- break;
-
- /* empty segment, skip it */
- if (free_memseg[i].len == 0)
- continue;
-
- /* bad socket ID */
- if (socket_id != SOCKET_ID_ANY &&
- free_memseg[i].socket_id != SOCKET_ID_ANY &&
- socket_id != free_memseg[i].socket_id)
- continue;
-
- /*
- * calculate offset to closest alignment that
- * meets boundary conditions.
- */
- addr_offset = align_phys_boundary(free_memseg + i,
- requested_len, align, bound);
-
- /* check len */
- if ((requested_len + addr_offset) > free_memseg[i].len)
- continue;
+ if ((socket_id != SOCKET_ID_ANY) && (socket_id >= RTE_MAX_NUMA_NODES)) {
+ rte_errno = EINVAL;
+ return NULL;
+ }
- /* check flags for hugepage sizes */
- if ((flags & RTE_MEMZONE_2MB) &&
- free_memseg[i].hugepage_sz == RTE_PGSIZE_1G)
- continue;
- if ((flags & RTE_MEMZONE_1GB) &&
- free_memseg[i].hugepage_sz == RTE_PGSIZE_2M)
- continue;
- if ((flags & RTE_MEMZONE_16MB) &&
- free_memseg[i].hugepage_sz == RTE_PGSIZE_16G)
- continue;
- if ((flags & RTE_MEMZONE_16GB) &&
- free_memseg[i].hugepage_sz == RTE_PGSIZE_16M)
- continue;
+ if (!rte_eal_has_hugepages())
+ socket_id = SOCKET_ID_ANY;
- /* this segment is the best until now */
- if (memseg_idx == -1) {
- memseg_idx = i;
- memseg_len = free_memseg[i].len;
- seg_offset = addr_offset;
- }
- /* find the biggest contiguous zone */
- else if (len == 0) {
- if (free_memseg[i].len > memseg_len) {
- memseg_idx = i;
- memseg_len = free_memseg[i].len;
- seg_offset = addr_offset;
+ if (len == 0) {
+ if (bound != 0)
+ requested_len = bound;
+ else {
+ requested_len = find_heap_max_free_elem(&socket_id, align);
+ if (requested_len == 0) {
+ rte_errno = ENOMEM;
+ return NULL;
}
}
- /*
- * find the smallest (we already checked that current
- * zone length is > len
- */
- else if (free_memseg[i].len + align < memseg_len ||
- (free_memseg[i].len <= memseg_len + align &&
- addr_offset < seg_offset)) {
- memseg_idx = i;
- memseg_len = free_memseg[i].len;
- seg_offset = addr_offset;
- }
}
- /* no segment found */
- if (memseg_idx == -1) {
- /*
- * If RTE_MEMZONE_SIZE_HINT_ONLY flag is specified,
- * try allocating again without the size parameter otherwise -fail.
- */
- if ((flags & RTE_MEMZONE_SIZE_HINT_ONLY) &&
- ((flags & RTE_MEMZONE_1GB) || (flags & RTE_MEMZONE_2MB)
- || (flags & RTE_MEMZONE_16MB) || (flags & RTE_MEMZONE_16GB)))
- return memzone_reserve_aligned_thread_unsafe(name,
- len, socket_id, 0, align, bound);
+ if (socket_id == SOCKET_ID_ANY)
+ socket = malloc_get_numa_socket();
+ else
+ socket = socket_id;
+
+ /* allocate memory on heap */
+ void *mz_addr = malloc_heap_alloc(&mcfg->malloc_heaps[socket], NULL,
+ requested_len, flags, align, bound);
+
+ if ((mz_addr == NULL) && (socket_id == SOCKET_ID_ANY)) {
+ /* try other heaps */
+ for (i = 0; i < RTE_MAX_NUMA_NODES; i++) {
+ if (socket == i)
+ continue;
+
+ mz_addr = malloc_heap_alloc(&mcfg->malloc_heaps[i],
+ NULL, requested_len, flags, align, bound);
+ if (mz_addr != NULL)
+ break;
+ }
+ }
+ if (mz_addr == NULL) {
rte_errno = ENOMEM;
return NULL;
}
- /* save aligned physical and virtual addresses */
- memseg_physaddr = free_memseg[memseg_idx].phys_addr + seg_offset;
- memseg_addr = RTE_PTR_ADD(free_memseg[memseg_idx].addr,
- (uintptr_t) seg_offset);
+ const struct malloc_elem *elem = malloc_elem_from_data(mz_addr);
- /* if we are looking for a biggest memzone */
- if (len == 0) {
- if (bound == 0)
- requested_len = memseg_len - seg_offset;
- else
- requested_len = RTE_ALIGN_CEIL(memseg_physaddr + 1,
- bound) - memseg_physaddr;
- }
-
- /* set length to correct value */
- len = (size_t)seg_offset + requested_len;
+ /* fill the zone in config */
+ mz = get_next_free_memzone();
- /* update our internal state */
- free_memseg[memseg_idx].len -= len;
- free_memseg[memseg_idx].phys_addr += len;
- free_memseg[memseg_idx].addr =
- (char *)free_memseg[memseg_idx].addr + len;
+ if (mz == NULL) {
+ RTE_LOG(ERR, EAL, "%s(): Cannot find free memzone but there is room "
+ "in config!\n", __func__);
+ rte_errno = ENOSPC;
+ return NULL;
+ }
- /* fill the zone in config */
- struct rte_memzone *mz = &mcfg->memzone[mcfg->memzone_idx++];
+ mcfg->memzone_cnt++;
snprintf(mz->name, sizeof(mz->name), "%s", name);
- mz->phys_addr = memseg_physaddr;
- mz->addr = memseg_addr;
- mz->len = requested_len;
- mz->hugepage_sz = free_memseg[memseg_idx].hugepage_sz;
- mz->socket_id = free_memseg[memseg_idx].socket_id;
+ mz->phys_addr = rte_malloc_virt2phy(mz_addr);
+ mz->addr = mz_addr;
+ mz->len = (requested_len == 0 ? elem->size : requested_len);
+ mz->hugepage_sz = elem->ms->hugepage_sz;
+ mz->socket_id = elem->ms->socket_id;
mz->flags = 0;
- mz->memseg_id = memseg_idx;
+ mz->memseg_id = elem->ms - rte_eal_get_configuration()->mem_config->memseg;
return mz;
}
-/*
- * Return a pointer to a correctly filled memzone descriptor (with a
- * specified alignment). If the allocation cannot be done, return NULL.
- */
-const struct rte_memzone *
-rte_memzone_reserve_aligned(const char *name, size_t len,
- int socket_id, unsigned flags, unsigned align)
+static const struct rte_memzone *
+rte_memzone_reserve_thread_safe(const char *name, size_t len,
+ int socket_id, unsigned flags, unsigned align,
+ unsigned bound)
{
struct rte_mem_config *mcfg;
const struct rte_memzone *mz = NULL;
- /* both sizes cannot be explicitly called for */
- if (((flags & RTE_MEMZONE_1GB) && (flags & RTE_MEMZONE_2MB))
- || ((flags & RTE_MEMZONE_16MB) && (flags & RTE_MEMZONE_16GB))) {
- rte_errno = EINVAL;
- return NULL;
- }
-
/* get pointer to global configuration */
mcfg = rte_eal_get_configuration()->mem_config;
rte_rwlock_write_lock(&mcfg->mlock);
mz = memzone_reserve_aligned_thread_unsafe(
- name, len, socket_id, flags, align, 0);
+ name, len, socket_id, flags, align, bound);
rte_rwlock_write_unlock(&mcfg->mlock);
/*
* Return a pointer to a correctly filled memzone descriptor (with a
- * specified alignment and boundary).
- * If the allocation cannot be done, return NULL.
+ * specified alignment and boundary). If the allocation cannot be done,
+ * return NULL.
*/
const struct rte_memzone *
-rte_memzone_reserve_bounded(const char *name, size_t len,
- int socket_id, unsigned flags, unsigned align, unsigned bound)
+rte_memzone_reserve_bounded(const char *name, size_t len, int socket_id,
+ unsigned flags, unsigned align, unsigned bound)
+{
+ return rte_memzone_reserve_thread_safe(name, len, socket_id, flags,
+ align, bound);
+}
+
+/*
+ * Return a pointer to a correctly filled memzone descriptor (with a
+ * specified alignment). If the allocation cannot be done, return NULL.
+ */
+const struct rte_memzone *
+rte_memzone_reserve_aligned(const char *name, size_t len, int socket_id,
+ unsigned flags, unsigned align)
+{
+ return rte_memzone_reserve_thread_safe(name, len, socket_id, flags,
+ align, 0);
+}
+
+/*
+ * Return a pointer to a correctly filled memzone descriptor. If the
+ * allocation cannot be done, return NULL.
+ */
+const struct rte_memzone *
+rte_memzone_reserve(const char *name, size_t len, int socket_id,
+ unsigned flags)
+{
+ return rte_memzone_reserve_thread_safe(name, len, socket_id,
+ flags, RTE_CACHE_LINE_SIZE, 0);
+}
+
+int
+rte_memzone_free(const struct rte_memzone *mz)
{
struct rte_mem_config *mcfg;
- const struct rte_memzone *mz = NULL;
+ int ret = 0;
+ void *addr;
+ unsigned idx;
- /* both sizes cannot be explicitly called for */
- if (((flags & RTE_MEMZONE_1GB) && (flags & RTE_MEMZONE_2MB))
- || ((flags & RTE_MEMZONE_16MB) && (flags & RTE_MEMZONE_16GB))) {
- rte_errno = EINVAL;
- return NULL;
- }
+ if (mz == NULL)
+ return -EINVAL;
- /* get pointer to global configuration */
mcfg = rte_eal_get_configuration()->mem_config;
rte_rwlock_write_lock(&mcfg->mlock);
- mz = memzone_reserve_aligned_thread_unsafe(
- name, len, socket_id, flags, align, bound);
+ idx = ((uintptr_t)mz - (uintptr_t)mcfg->memzone);
+ idx = idx / sizeof(struct rte_memzone);
+
+#ifdef RTE_LIBRTE_IVSHMEM
+ /*
+ * If ioremap_addr is set, it's an IVSHMEM memzone and we cannot
+ * free it.
+ */
+ if (mcfg->memzone[idx].ioremap_addr != 0) {
+ rte_rwlock_write_unlock(&mcfg->mlock);
+ return -EINVAL;
+ }
+#endif
+
+ addr = mcfg->memzone[idx].addr;
+
+ if (addr == NULL)
+ ret = -EINVAL;
+ else if (mcfg->memzone_cnt == 0) {
+ rte_panic("%s(): memzone address not NULL but memzone_cnt is 0!\n",
+ __func__);
+ } else {
+ memset(&mcfg->memzone[idx], 0, sizeof(mcfg->memzone[idx]));
+ mcfg->memzone_cnt--;
+ }
rte_rwlock_write_unlock(&mcfg->mlock);
- return mz;
-}
+ rte_free(addr);
+ return ret;
+}
/*
* Lookup for the memzone identified by the given name
rte_rwlock_read_unlock(&mcfg->mlock);
}
-/*
- * called by init: modify the free memseg list to have cache-aligned
- * addresses and cache-aligned lengths
- */
-static int
-memseg_sanitize(struct rte_memseg *memseg)
-{
- unsigned phys_align;
- unsigned virt_align;
- unsigned off;
-
- phys_align = memseg->phys_addr & CACHE_LINE_MASK;
- virt_align = (unsigned long)memseg->addr & CACHE_LINE_MASK;
-
- /*
- * sanity check: phys_addr and addr must have the same
- * alignment
- */
- if (phys_align != virt_align)
- return -1;
-
- /* memseg is really too small, don't bother with it */
- if (memseg->len < (2 * CACHE_LINE_SIZE)) {
- memseg->len = 0;
- return 0;
- }
-
- /* align start address */
- off = (CACHE_LINE_SIZE - phys_align) & CACHE_LINE_MASK;
- memseg->phys_addr += off;
- memseg->addr = (char *)memseg->addr + off;
- memseg->len -= off;
-
- /* align end address */
- memseg->len &= ~((uint64_t)CACHE_LINE_MASK);
-
- return 0;
-}
-
/*
* Init the memzone subsystem
*/
{
struct rte_mem_config *mcfg;
const struct rte_memseg *memseg;
- unsigned i = 0;
/* get pointer to global configuration */
mcfg = rte_eal_get_configuration()->mem_config;
- /* mirror the runtime memsegs from config */
- free_memseg = mcfg->free_memseg;
-
/* secondary processes don't need to initialise anything */
if (rte_eal_process_type() == RTE_PROC_SECONDARY)
return 0;
rte_rwlock_write_lock(&mcfg->mlock);
- /* fill in uninitialized free_memsegs */
- for (i = 0; i < RTE_MAX_MEMSEG; i++) {
- if (memseg[i].addr == NULL)
- break;
- if (free_memseg[i].addr != NULL)
- continue;
- memcpy(&free_memseg[i], &memseg[i], sizeof(struct rte_memseg));
- }
-
- /* make all zones cache-aligned */
- for (i = 0; i < RTE_MAX_MEMSEG; i++) {
- if (free_memseg[i].addr == NULL)
- break;
- if (memseg_sanitize(&free_memseg[i]) < 0) {
- RTE_LOG(ERR, EAL, "%s(): Sanity check failed\n", __func__);
- rte_rwlock_write_unlock(&mcfg->mlock);
- return -1;
- }
- }
-
/* delete all zones */
- mcfg->memzone_idx = 0;
+ mcfg->memzone_cnt = 0;
memset(mcfg->memzone, 0, sizeof(mcfg->memzone));
rte_rwlock_write_unlock(&mcfg->mlock);
- return 0;
+ return rte_eal_malloc_heap_init();
}
/* Walk all reserved memory zones */