308 lines
7.4 KiB
C
Raw Normal View History

/* SPDX-License-Identifier: BSD-3-Clause
* Copyright(c) 2010-2014 Intel Corporation
*/
#include <sys/mman.h>
#include <unistd.h>
#include <sys/types.h>
#include <sys/sysctl.h>
#include <inttypes.h>
mem: replace memseg with memseg lists Before, we were aggregating multiple pages into one memseg, so the number of memsegs was small. Now, each page gets its own memseg, so the list of memsegs is huge. To accommodate the new memseg list size and to keep the under-the-hood workings sane, the memseg list is now not just a single list, but multiple lists. To be precise, each hugepage size available on the system gets one or more memseg lists, per socket. In order to support dynamic memory allocation, we reserve all memory in advance (unless we're in 32-bit legacy mode, in which case we do not preallocate memory). As in, we do an anonymous mmap() of the entire maximum size of memory per hugepage size, per socket (which is limited to either RTE_MAX_MEMSEG_PER_TYPE pages or RTE_MAX_MEM_MB_PER_TYPE megabytes worth of memory, whichever is the smaller one), split over multiple lists (which are limited to either RTE_MAX_MEMSEG_PER_LIST memsegs or RTE_MAX_MEM_MB_PER_LIST megabytes per list, whichever is the smaller one). There is also a global limit of CONFIG_RTE_MAX_MEM_MB megabytes, which is mainly used for 32-bit targets to limit amounts of preallocated memory, but can be used to place an upper limit on total amount of VA memory that can be allocated by DPDK application. So, for each hugepage size, we get (by default) up to 128G worth of memory, per socket, split into chunks of up to 32G in size. The address space is claimed at the start, in eal_common_memory.c. The actual page allocation code is in eal_memalloc.c (Linux-only), and largely consists of copied EAL memory init code. Pages in the list are also indexed by address. That is, in order to figure out where the page belongs, one can simply look at base address for a memseg list. Similarly, figuring out IOVA address of a memzone is a matter of finding the right memseg list, getting offset and dividing by page size to get the appropriate memseg. This commit also removes rte_eal_dump_physmem_layout() call, according to deprecation notice [1], and removes that deprecation notice as well. On 32-bit targets due to limited VA space, DPDK will no longer spread memory to different sockets like before. Instead, it will (by default) allocate all of the memory on socket where master lcore is. To override this behavior, --socket-mem must be used. The rest of the changes are really ripple effects from the memseg change - heap changes, compile fixes, and rewrites to support fbarray-backed memseg lists. Due to earlier switch to _walk() functions, most of the changes are simple fixes, however some of the _walk() calls were switched to memseg list walk, where it made sense to do so. Additionally, we are also switching locks from flock() to fcntl(). Down the line, we will be introducing single-file segments option, and we cannot use flock() locks to lock parts of the file. Therefore, we will use fcntl() locks for legacy mem as well, in case someone is unfortunate enough to accidentally start legacy mem primary process alongside an already working non-legacy mem-based primary process. [1] http://dpdk.org/dev/patchwork/patch/34002/ Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> Tested-by: Santosh Shukla <santosh.shukla@caviumnetworks.com> Tested-by: Hemant Agrawal <hemant.agrawal@nxp.com> Tested-by: Gowrishankar Muthukrishnan <gowrishankar.m@linux.vnet.ibm.com>
2018-04-11 13:30:24 +01:00
#include <errno.h>
#include <string.h>
#include <fcntl.h>
#include <rte_eal.h>
#include <rte_eal_memconfig.h>
#include <rte_log.h>
#include <rte_string_fns.h>
#include "eal_private.h"
#include "eal_internal_cfg.h"
#include "eal_filesystem.h"
#define EAL_PAGE_SIZE (sysconf(_SC_PAGESIZE))
/*
* Get physical address of any mapped virtual address in the current process.
*/
phys_addr_t
rte_mem_virt2phy(const void *virtaddr)
{
/* XXX not implemented. This function is only used by
* rte_mempool_virt2iova() when hugepages are disabled. */
(void)virtaddr;
return RTE_BAD_IOVA;
}
rte_iova_t
rte_mem_virt2iova(const void *virtaddr)
{
return rte_mem_virt2phy(virtaddr);
}
int
rte_eal_hugepage_init(void)
{
struct rte_mem_config *mcfg;
uint64_t total_mem = 0;
void *addr;
mem: replace memseg with memseg lists Before, we were aggregating multiple pages into one memseg, so the number of memsegs was small. Now, each page gets its own memseg, so the list of memsegs is huge. To accommodate the new memseg list size and to keep the under-the-hood workings sane, the memseg list is now not just a single list, but multiple lists. To be precise, each hugepage size available on the system gets one or more memseg lists, per socket. In order to support dynamic memory allocation, we reserve all memory in advance (unless we're in 32-bit legacy mode, in which case we do not preallocate memory). As in, we do an anonymous mmap() of the entire maximum size of memory per hugepage size, per socket (which is limited to either RTE_MAX_MEMSEG_PER_TYPE pages or RTE_MAX_MEM_MB_PER_TYPE megabytes worth of memory, whichever is the smaller one), split over multiple lists (which are limited to either RTE_MAX_MEMSEG_PER_LIST memsegs or RTE_MAX_MEM_MB_PER_LIST megabytes per list, whichever is the smaller one). There is also a global limit of CONFIG_RTE_MAX_MEM_MB megabytes, which is mainly used for 32-bit targets to limit amounts of preallocated memory, but can be used to place an upper limit on total amount of VA memory that can be allocated by DPDK application. So, for each hugepage size, we get (by default) up to 128G worth of memory, per socket, split into chunks of up to 32G in size. The address space is claimed at the start, in eal_common_memory.c. The actual page allocation code is in eal_memalloc.c (Linux-only), and largely consists of copied EAL memory init code. Pages in the list are also indexed by address. That is, in order to figure out where the page belongs, one can simply look at base address for a memseg list. Similarly, figuring out IOVA address of a memzone is a matter of finding the right memseg list, getting offset and dividing by page size to get the appropriate memseg. This commit also removes rte_eal_dump_physmem_layout() call, according to deprecation notice [1], and removes that deprecation notice as well. On 32-bit targets due to limited VA space, DPDK will no longer spread memory to different sockets like before. Instead, it will (by default) allocate all of the memory on socket where master lcore is. To override this behavior, --socket-mem must be used. The rest of the changes are really ripple effects from the memseg change - heap changes, compile fixes, and rewrites to support fbarray-backed memseg lists. Due to earlier switch to _walk() functions, most of the changes are simple fixes, however some of the _walk() calls were switched to memseg list walk, where it made sense to do so. Additionally, we are also switching locks from flock() to fcntl(). Down the line, we will be introducing single-file segments option, and we cannot use flock() locks to lock parts of the file. Therefore, we will use fcntl() locks for legacy mem as well, in case someone is unfortunate enough to accidentally start legacy mem primary process alongside an already working non-legacy mem-based primary process. [1] http://dpdk.org/dev/patchwork/patch/34002/ Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> Tested-by: Santosh Shukla <santosh.shukla@caviumnetworks.com> Tested-by: Hemant Agrawal <hemant.agrawal@nxp.com> Tested-by: Gowrishankar Muthukrishnan <gowrishankar.m@linux.vnet.ibm.com>
2018-04-11 13:30:24 +01:00
unsigned int i, j, seg_idx = 0;
/* get pointer to global configuration */
mcfg = rte_eal_get_configuration()->mem_config;
/* for debug purposes, hugetlbfs can be disabled */
if (internal_config.no_hugetlbfs) {
mem: replace memseg with memseg lists Before, we were aggregating multiple pages into one memseg, so the number of memsegs was small. Now, each page gets its own memseg, so the list of memsegs is huge. To accommodate the new memseg list size and to keep the under-the-hood workings sane, the memseg list is now not just a single list, but multiple lists. To be precise, each hugepage size available on the system gets one or more memseg lists, per socket. In order to support dynamic memory allocation, we reserve all memory in advance (unless we're in 32-bit legacy mode, in which case we do not preallocate memory). As in, we do an anonymous mmap() of the entire maximum size of memory per hugepage size, per socket (which is limited to either RTE_MAX_MEMSEG_PER_TYPE pages or RTE_MAX_MEM_MB_PER_TYPE megabytes worth of memory, whichever is the smaller one), split over multiple lists (which are limited to either RTE_MAX_MEMSEG_PER_LIST memsegs or RTE_MAX_MEM_MB_PER_LIST megabytes per list, whichever is the smaller one). There is also a global limit of CONFIG_RTE_MAX_MEM_MB megabytes, which is mainly used for 32-bit targets to limit amounts of preallocated memory, but can be used to place an upper limit on total amount of VA memory that can be allocated by DPDK application. So, for each hugepage size, we get (by default) up to 128G worth of memory, per socket, split into chunks of up to 32G in size. The address space is claimed at the start, in eal_common_memory.c. The actual page allocation code is in eal_memalloc.c (Linux-only), and largely consists of copied EAL memory init code. Pages in the list are also indexed by address. That is, in order to figure out where the page belongs, one can simply look at base address for a memseg list. Similarly, figuring out IOVA address of a memzone is a matter of finding the right memseg list, getting offset and dividing by page size to get the appropriate memseg. This commit also removes rte_eal_dump_physmem_layout() call, according to deprecation notice [1], and removes that deprecation notice as well. On 32-bit targets due to limited VA space, DPDK will no longer spread memory to different sockets like before. Instead, it will (by default) allocate all of the memory on socket where master lcore is. To override this behavior, --socket-mem must be used. The rest of the changes are really ripple effects from the memseg change - heap changes, compile fixes, and rewrites to support fbarray-backed memseg lists. Due to earlier switch to _walk() functions, most of the changes are simple fixes, however some of the _walk() calls were switched to memseg list walk, where it made sense to do so. Additionally, we are also switching locks from flock() to fcntl(). Down the line, we will be introducing single-file segments option, and we cannot use flock() locks to lock parts of the file. Therefore, we will use fcntl() locks for legacy mem as well, in case someone is unfortunate enough to accidentally start legacy mem primary process alongside an already working non-legacy mem-based primary process. [1] http://dpdk.org/dev/patchwork/patch/34002/ Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> Tested-by: Santosh Shukla <santosh.shukla@caviumnetworks.com> Tested-by: Hemant Agrawal <hemant.agrawal@nxp.com> Tested-by: Gowrishankar Muthukrishnan <gowrishankar.m@linux.vnet.ibm.com>
2018-04-11 13:30:24 +01:00
struct rte_memseg_list *msl;
struct rte_fbarray *arr;
struct rte_memseg *ms;
uint64_t page_sz;
int n_segs, cur_seg;
/* create a memseg list */
msl = &mcfg->memsegs[0];
page_sz = RTE_PGSIZE_4K;
n_segs = internal_config.memory / page_sz;
if (rte_fbarray_init(&msl->memseg_arr, "nohugemem", n_segs,
sizeof(struct rte_memseg))) {
RTE_LOG(ERR, EAL, "Cannot allocate memseg list\n");
return -1;
}
addr = mmap(NULL, internal_config.memory,
PROT_READ | PROT_WRITE,
MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
if (addr == MAP_FAILED) {
RTE_LOG(ERR, EAL, "%s: mmap() failed: %s\n", __func__,
strerror(errno));
return -1;
}
msl->base_va = addr;
msl->page_sz = page_sz;
msl->socket_id = 0;
/* populate memsegs. each memseg is 1 page long */
for (cur_seg = 0; cur_seg < n_segs; cur_seg++) {
arr = &msl->memseg_arr;
ms = rte_fbarray_get(arr, cur_seg);
if (rte_eal_iova_mode() == RTE_IOVA_VA)
ms->iova = (uintptr_t)addr;
else
ms->iova = RTE_BAD_IOVA;
ms->addr = addr;
ms->hugepage_sz = page_sz;
ms->len = page_sz;
ms->socket_id = 0;
rte_fbarray_set_used(arr, cur_seg);
addr = RTE_PTR_ADD(addr, page_sz);
}
return 0;
}
/* map all hugepages and sort them */
for (i = 0; i < internal_config.num_hugepage_sizes; i ++){
struct hugepage_info *hpi;
mem: replace memseg with memseg lists Before, we were aggregating multiple pages into one memseg, so the number of memsegs was small. Now, each page gets its own memseg, so the list of memsegs is huge. To accommodate the new memseg list size and to keep the under-the-hood workings sane, the memseg list is now not just a single list, but multiple lists. To be precise, each hugepage size available on the system gets one or more memseg lists, per socket. In order to support dynamic memory allocation, we reserve all memory in advance (unless we're in 32-bit legacy mode, in which case we do not preallocate memory). As in, we do an anonymous mmap() of the entire maximum size of memory per hugepage size, per socket (which is limited to either RTE_MAX_MEMSEG_PER_TYPE pages or RTE_MAX_MEM_MB_PER_TYPE megabytes worth of memory, whichever is the smaller one), split over multiple lists (which are limited to either RTE_MAX_MEMSEG_PER_LIST memsegs or RTE_MAX_MEM_MB_PER_LIST megabytes per list, whichever is the smaller one). There is also a global limit of CONFIG_RTE_MAX_MEM_MB megabytes, which is mainly used for 32-bit targets to limit amounts of preallocated memory, but can be used to place an upper limit on total amount of VA memory that can be allocated by DPDK application. So, for each hugepage size, we get (by default) up to 128G worth of memory, per socket, split into chunks of up to 32G in size. The address space is claimed at the start, in eal_common_memory.c. The actual page allocation code is in eal_memalloc.c (Linux-only), and largely consists of copied EAL memory init code. Pages in the list are also indexed by address. That is, in order to figure out where the page belongs, one can simply look at base address for a memseg list. Similarly, figuring out IOVA address of a memzone is a matter of finding the right memseg list, getting offset and dividing by page size to get the appropriate memseg. This commit also removes rte_eal_dump_physmem_layout() call, according to deprecation notice [1], and removes that deprecation notice as well. On 32-bit targets due to limited VA space, DPDK will no longer spread memory to different sockets like before. Instead, it will (by default) allocate all of the memory on socket where master lcore is. To override this behavior, --socket-mem must be used. The rest of the changes are really ripple effects from the memseg change - heap changes, compile fixes, and rewrites to support fbarray-backed memseg lists. Due to earlier switch to _walk() functions, most of the changes are simple fixes, however some of the _walk() calls were switched to memseg list walk, where it made sense to do so. Additionally, we are also switching locks from flock() to fcntl(). Down the line, we will be introducing single-file segments option, and we cannot use flock() locks to lock parts of the file. Therefore, we will use fcntl() locks for legacy mem as well, in case someone is unfortunate enough to accidentally start legacy mem primary process alongside an already working non-legacy mem-based primary process. [1] http://dpdk.org/dev/patchwork/patch/34002/ Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> Tested-by: Santosh Shukla <santosh.shukla@caviumnetworks.com> Tested-by: Hemant Agrawal <hemant.agrawal@nxp.com> Tested-by: Gowrishankar Muthukrishnan <gowrishankar.m@linux.vnet.ibm.com>
2018-04-11 13:30:24 +01:00
uint64_t page_sz, mem_needed;
unsigned int n_pages, max_pages;
hpi = &internal_config.hugepage_info[i];
mem: replace memseg with memseg lists Before, we were aggregating multiple pages into one memseg, so the number of memsegs was small. Now, each page gets its own memseg, so the list of memsegs is huge. To accommodate the new memseg list size and to keep the under-the-hood workings sane, the memseg list is now not just a single list, but multiple lists. To be precise, each hugepage size available on the system gets one or more memseg lists, per socket. In order to support dynamic memory allocation, we reserve all memory in advance (unless we're in 32-bit legacy mode, in which case we do not preallocate memory). As in, we do an anonymous mmap() of the entire maximum size of memory per hugepage size, per socket (which is limited to either RTE_MAX_MEMSEG_PER_TYPE pages or RTE_MAX_MEM_MB_PER_TYPE megabytes worth of memory, whichever is the smaller one), split over multiple lists (which are limited to either RTE_MAX_MEMSEG_PER_LIST memsegs or RTE_MAX_MEM_MB_PER_LIST megabytes per list, whichever is the smaller one). There is also a global limit of CONFIG_RTE_MAX_MEM_MB megabytes, which is mainly used for 32-bit targets to limit amounts of preallocated memory, but can be used to place an upper limit on total amount of VA memory that can be allocated by DPDK application. So, for each hugepage size, we get (by default) up to 128G worth of memory, per socket, split into chunks of up to 32G in size. The address space is claimed at the start, in eal_common_memory.c. The actual page allocation code is in eal_memalloc.c (Linux-only), and largely consists of copied EAL memory init code. Pages in the list are also indexed by address. That is, in order to figure out where the page belongs, one can simply look at base address for a memseg list. Similarly, figuring out IOVA address of a memzone is a matter of finding the right memseg list, getting offset and dividing by page size to get the appropriate memseg. This commit also removes rte_eal_dump_physmem_layout() call, according to deprecation notice [1], and removes that deprecation notice as well. On 32-bit targets due to limited VA space, DPDK will no longer spread memory to different sockets like before. Instead, it will (by default) allocate all of the memory on socket where master lcore is. To override this behavior, --socket-mem must be used. The rest of the changes are really ripple effects from the memseg change - heap changes, compile fixes, and rewrites to support fbarray-backed memseg lists. Due to earlier switch to _walk() functions, most of the changes are simple fixes, however some of the _walk() calls were switched to memseg list walk, where it made sense to do so. Additionally, we are also switching locks from flock() to fcntl(). Down the line, we will be introducing single-file segments option, and we cannot use flock() locks to lock parts of the file. Therefore, we will use fcntl() locks for legacy mem as well, in case someone is unfortunate enough to accidentally start legacy mem primary process alongside an already working non-legacy mem-based primary process. [1] http://dpdk.org/dev/patchwork/patch/34002/ Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> Tested-by: Santosh Shukla <santosh.shukla@caviumnetworks.com> Tested-by: Hemant Agrawal <hemant.agrawal@nxp.com> Tested-by: Gowrishankar Muthukrishnan <gowrishankar.m@linux.vnet.ibm.com>
2018-04-11 13:30:24 +01:00
page_sz = hpi->hugepage_sz;
max_pages = hpi->num_pages[0];
mem_needed = RTE_ALIGN_CEIL(internal_config.memory - total_mem,
page_sz);
n_pages = RTE_MIN(mem_needed / page_sz, max_pages);
for (j = 0; j < n_pages; j++) {
struct rte_memseg_list *msl;
struct rte_fbarray *arr;
struct rte_memseg *seg;
mem: replace memseg with memseg lists Before, we were aggregating multiple pages into one memseg, so the number of memsegs was small. Now, each page gets its own memseg, so the list of memsegs is huge. To accommodate the new memseg list size and to keep the under-the-hood workings sane, the memseg list is now not just a single list, but multiple lists. To be precise, each hugepage size available on the system gets one or more memseg lists, per socket. In order to support dynamic memory allocation, we reserve all memory in advance (unless we're in 32-bit legacy mode, in which case we do not preallocate memory). As in, we do an anonymous mmap() of the entire maximum size of memory per hugepage size, per socket (which is limited to either RTE_MAX_MEMSEG_PER_TYPE pages or RTE_MAX_MEM_MB_PER_TYPE megabytes worth of memory, whichever is the smaller one), split over multiple lists (which are limited to either RTE_MAX_MEMSEG_PER_LIST memsegs or RTE_MAX_MEM_MB_PER_LIST megabytes per list, whichever is the smaller one). There is also a global limit of CONFIG_RTE_MAX_MEM_MB megabytes, which is mainly used for 32-bit targets to limit amounts of preallocated memory, but can be used to place an upper limit on total amount of VA memory that can be allocated by DPDK application. So, for each hugepage size, we get (by default) up to 128G worth of memory, per socket, split into chunks of up to 32G in size. The address space is claimed at the start, in eal_common_memory.c. The actual page allocation code is in eal_memalloc.c (Linux-only), and largely consists of copied EAL memory init code. Pages in the list are also indexed by address. That is, in order to figure out where the page belongs, one can simply look at base address for a memseg list. Similarly, figuring out IOVA address of a memzone is a matter of finding the right memseg list, getting offset and dividing by page size to get the appropriate memseg. This commit also removes rte_eal_dump_physmem_layout() call, according to deprecation notice [1], and removes that deprecation notice as well. On 32-bit targets due to limited VA space, DPDK will no longer spread memory to different sockets like before. Instead, it will (by default) allocate all of the memory on socket where master lcore is. To override this behavior, --socket-mem must be used. The rest of the changes are really ripple effects from the memseg change - heap changes, compile fixes, and rewrites to support fbarray-backed memseg lists. Due to earlier switch to _walk() functions, most of the changes are simple fixes, however some of the _walk() calls were switched to memseg list walk, where it made sense to do so. Additionally, we are also switching locks from flock() to fcntl(). Down the line, we will be introducing single-file segments option, and we cannot use flock() locks to lock parts of the file. Therefore, we will use fcntl() locks for legacy mem as well, in case someone is unfortunate enough to accidentally start legacy mem primary process alongside an already working non-legacy mem-based primary process. [1] http://dpdk.org/dev/patchwork/patch/34002/ Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> Tested-by: Santosh Shukla <santosh.shukla@caviumnetworks.com> Tested-by: Hemant Agrawal <hemant.agrawal@nxp.com> Tested-by: Gowrishankar Muthukrishnan <gowrishankar.m@linux.vnet.ibm.com>
2018-04-11 13:30:24 +01:00
int msl_idx, ms_idx;
rte_iova_t physaddr;
int error;
size_t sysctl_size = sizeof(physaddr);
char physaddr_str[64];
mem: replace memseg with memseg lists Before, we were aggregating multiple pages into one memseg, so the number of memsegs was small. Now, each page gets its own memseg, so the list of memsegs is huge. To accommodate the new memseg list size and to keep the under-the-hood workings sane, the memseg list is now not just a single list, but multiple lists. To be precise, each hugepage size available on the system gets one or more memseg lists, per socket. In order to support dynamic memory allocation, we reserve all memory in advance (unless we're in 32-bit legacy mode, in which case we do not preallocate memory). As in, we do an anonymous mmap() of the entire maximum size of memory per hugepage size, per socket (which is limited to either RTE_MAX_MEMSEG_PER_TYPE pages or RTE_MAX_MEM_MB_PER_TYPE megabytes worth of memory, whichever is the smaller one), split over multiple lists (which are limited to either RTE_MAX_MEMSEG_PER_LIST memsegs or RTE_MAX_MEM_MB_PER_LIST megabytes per list, whichever is the smaller one). There is also a global limit of CONFIG_RTE_MAX_MEM_MB megabytes, which is mainly used for 32-bit targets to limit amounts of preallocated memory, but can be used to place an upper limit on total amount of VA memory that can be allocated by DPDK application. So, for each hugepage size, we get (by default) up to 128G worth of memory, per socket, split into chunks of up to 32G in size. The address space is claimed at the start, in eal_common_memory.c. The actual page allocation code is in eal_memalloc.c (Linux-only), and largely consists of copied EAL memory init code. Pages in the list are also indexed by address. That is, in order to figure out where the page belongs, one can simply look at base address for a memseg list. Similarly, figuring out IOVA address of a memzone is a matter of finding the right memseg list, getting offset and dividing by page size to get the appropriate memseg. This commit also removes rte_eal_dump_physmem_layout() call, according to deprecation notice [1], and removes that deprecation notice as well. On 32-bit targets due to limited VA space, DPDK will no longer spread memory to different sockets like before. Instead, it will (by default) allocate all of the memory on socket where master lcore is. To override this behavior, --socket-mem must be used. The rest of the changes are really ripple effects from the memseg change - heap changes, compile fixes, and rewrites to support fbarray-backed memseg lists. Due to earlier switch to _walk() functions, most of the changes are simple fixes, however some of the _walk() calls were switched to memseg list walk, where it made sense to do so. Additionally, we are also switching locks from flock() to fcntl(). Down the line, we will be introducing single-file segments option, and we cannot use flock() locks to lock parts of the file. Therefore, we will use fcntl() locks for legacy mem as well, in case someone is unfortunate enough to accidentally start legacy mem primary process alongside an already working non-legacy mem-based primary process. [1] http://dpdk.org/dev/patchwork/patch/34002/ Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> Tested-by: Santosh Shukla <santosh.shukla@caviumnetworks.com> Tested-by: Hemant Agrawal <hemant.agrawal@nxp.com> Tested-by: Gowrishankar Muthukrishnan <gowrishankar.m@linux.vnet.ibm.com>
2018-04-11 13:30:24 +01:00
for (msl_idx = 0; msl_idx < RTE_MAX_MEMSEG_LISTS;
msl_idx++) {
bool empty;
msl = &mcfg->memsegs[msl_idx];
arr = &msl->memseg_arr;
if (msl->page_sz != page_sz)
continue;
empty = arr->count == 0;
/* we need 1, plus hole if not empty */
ms_idx = rte_fbarray_find_next_n_free(arr,
0, 1 + (empty ? 1 : 0));
/* memseg list is full? */
if (ms_idx < 0)
continue;
/* leave some space between memsegs, they are
* not IOVA contiguous, so they shouldn't be VA
* contiguous either.
*/
if (!empty)
ms_idx++;
break;
}
if (msl_idx == RTE_MAX_MEMSEG_LISTS) {
RTE_LOG(ERR, EAL, "Could not find space for memseg. Please increase %s and/or %s in configuration.\n",
RTE_STR(CONFIG_RTE_MAX_MEMSEG_PER_TYPE),
RTE_STR(CONFIG_RTE_MAX_MEM_PER_TYPE));
return -1;
}
arr = &msl->memseg_arr;
seg = rte_fbarray_get(arr, ms_idx);
addr = RTE_PTR_ADD(msl->base_va,
(size_t)msl->page_sz * ms_idx);
/* address is already mapped in memseg list, so using
* MAP_FIXED here is safe.
*/
addr = mmap(addr, page_sz, PROT_READ|PROT_WRITE,
MAP_SHARED | MAP_FIXED,
hpi->lock_descriptor,
j * EAL_PAGE_SIZE);
if (addr == MAP_FAILED) {
RTE_LOG(ERR, EAL, "Failed to mmap buffer %u from %s\n",
j, hpi->hugedir);
return -1;
}
snprintf(physaddr_str, sizeof(physaddr_str), "hw.contigmem"
".physaddr.%d", j);
error = sysctlbyname(physaddr_str, &physaddr, &sysctl_size,
NULL, 0);
if (error < 0) {
RTE_LOG(ERR, EAL, "Failed to get physical addr for buffer %u "
"from %s\n", j, hpi->hugedir);
return -1;
}
seg->addr = addr;
seg->iova = physaddr;
mem: replace memseg with memseg lists Before, we were aggregating multiple pages into one memseg, so the number of memsegs was small. Now, each page gets its own memseg, so the list of memsegs is huge. To accommodate the new memseg list size and to keep the under-the-hood workings sane, the memseg list is now not just a single list, but multiple lists. To be precise, each hugepage size available on the system gets one or more memseg lists, per socket. In order to support dynamic memory allocation, we reserve all memory in advance (unless we're in 32-bit legacy mode, in which case we do not preallocate memory). As in, we do an anonymous mmap() of the entire maximum size of memory per hugepage size, per socket (which is limited to either RTE_MAX_MEMSEG_PER_TYPE pages or RTE_MAX_MEM_MB_PER_TYPE megabytes worth of memory, whichever is the smaller one), split over multiple lists (which are limited to either RTE_MAX_MEMSEG_PER_LIST memsegs or RTE_MAX_MEM_MB_PER_LIST megabytes per list, whichever is the smaller one). There is also a global limit of CONFIG_RTE_MAX_MEM_MB megabytes, which is mainly used for 32-bit targets to limit amounts of preallocated memory, but can be used to place an upper limit on total amount of VA memory that can be allocated by DPDK application. So, for each hugepage size, we get (by default) up to 128G worth of memory, per socket, split into chunks of up to 32G in size. The address space is claimed at the start, in eal_common_memory.c. The actual page allocation code is in eal_memalloc.c (Linux-only), and largely consists of copied EAL memory init code. Pages in the list are also indexed by address. That is, in order to figure out where the page belongs, one can simply look at base address for a memseg list. Similarly, figuring out IOVA address of a memzone is a matter of finding the right memseg list, getting offset and dividing by page size to get the appropriate memseg. This commit also removes rte_eal_dump_physmem_layout() call, according to deprecation notice [1], and removes that deprecation notice as well. On 32-bit targets due to limited VA space, DPDK will no longer spread memory to different sockets like before. Instead, it will (by default) allocate all of the memory on socket where master lcore is. To override this behavior, --socket-mem must be used. The rest of the changes are really ripple effects from the memseg change - heap changes, compile fixes, and rewrites to support fbarray-backed memseg lists. Due to earlier switch to _walk() functions, most of the changes are simple fixes, however some of the _walk() calls were switched to memseg list walk, where it made sense to do so. Additionally, we are also switching locks from flock() to fcntl(). Down the line, we will be introducing single-file segments option, and we cannot use flock() locks to lock parts of the file. Therefore, we will use fcntl() locks for legacy mem as well, in case someone is unfortunate enough to accidentally start legacy mem primary process alongside an already working non-legacy mem-based primary process. [1] http://dpdk.org/dev/patchwork/patch/34002/ Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> Tested-by: Santosh Shukla <santosh.shukla@caviumnetworks.com> Tested-by: Hemant Agrawal <hemant.agrawal@nxp.com> Tested-by: Gowrishankar Muthukrishnan <gowrishankar.m@linux.vnet.ibm.com>
2018-04-11 13:30:24 +01:00
seg->hugepage_sz = page_sz;
seg->len = page_sz;
seg->nchannel = mcfg->nchannel;
seg->nrank = mcfg->nrank;
seg->socket_id = 0;
mem: replace memseg with memseg lists Before, we were aggregating multiple pages into one memseg, so the number of memsegs was small. Now, each page gets its own memseg, so the list of memsegs is huge. To accommodate the new memseg list size and to keep the under-the-hood workings sane, the memseg list is now not just a single list, but multiple lists. To be precise, each hugepage size available on the system gets one or more memseg lists, per socket. In order to support dynamic memory allocation, we reserve all memory in advance (unless we're in 32-bit legacy mode, in which case we do not preallocate memory). As in, we do an anonymous mmap() of the entire maximum size of memory per hugepage size, per socket (which is limited to either RTE_MAX_MEMSEG_PER_TYPE pages or RTE_MAX_MEM_MB_PER_TYPE megabytes worth of memory, whichever is the smaller one), split over multiple lists (which are limited to either RTE_MAX_MEMSEG_PER_LIST memsegs or RTE_MAX_MEM_MB_PER_LIST megabytes per list, whichever is the smaller one). There is also a global limit of CONFIG_RTE_MAX_MEM_MB megabytes, which is mainly used for 32-bit targets to limit amounts of preallocated memory, but can be used to place an upper limit on total amount of VA memory that can be allocated by DPDK application. So, for each hugepage size, we get (by default) up to 128G worth of memory, per socket, split into chunks of up to 32G in size. The address space is claimed at the start, in eal_common_memory.c. The actual page allocation code is in eal_memalloc.c (Linux-only), and largely consists of copied EAL memory init code. Pages in the list are also indexed by address. That is, in order to figure out where the page belongs, one can simply look at base address for a memseg list. Similarly, figuring out IOVA address of a memzone is a matter of finding the right memseg list, getting offset and dividing by page size to get the appropriate memseg. This commit also removes rte_eal_dump_physmem_layout() call, according to deprecation notice [1], and removes that deprecation notice as well. On 32-bit targets due to limited VA space, DPDK will no longer spread memory to different sockets like before. Instead, it will (by default) allocate all of the memory on socket where master lcore is. To override this behavior, --socket-mem must be used. The rest of the changes are really ripple effects from the memseg change - heap changes, compile fixes, and rewrites to support fbarray-backed memseg lists. Due to earlier switch to _walk() functions, most of the changes are simple fixes, however some of the _walk() calls were switched to memseg list walk, where it made sense to do so. Additionally, we are also switching locks from flock() to fcntl(). Down the line, we will be introducing single-file segments option, and we cannot use flock() locks to lock parts of the file. Therefore, we will use fcntl() locks for legacy mem as well, in case someone is unfortunate enough to accidentally start legacy mem primary process alongside an already working non-legacy mem-based primary process. [1] http://dpdk.org/dev/patchwork/patch/34002/ Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> Tested-by: Santosh Shukla <santosh.shukla@caviumnetworks.com> Tested-by: Hemant Agrawal <hemant.agrawal@nxp.com> Tested-by: Gowrishankar Muthukrishnan <gowrishankar.m@linux.vnet.ibm.com>
2018-04-11 13:30:24 +01:00
rte_fbarray_set_used(arr, ms_idx);
RTE_LOG(INFO, EAL, "Mapped memory segment %u @ %p: physaddr:0x%"
PRIx64", len %zu\n",
mem: replace memseg with memseg lists Before, we were aggregating multiple pages into one memseg, so the number of memsegs was small. Now, each page gets its own memseg, so the list of memsegs is huge. To accommodate the new memseg list size and to keep the under-the-hood workings sane, the memseg list is now not just a single list, but multiple lists. To be precise, each hugepage size available on the system gets one or more memseg lists, per socket. In order to support dynamic memory allocation, we reserve all memory in advance (unless we're in 32-bit legacy mode, in which case we do not preallocate memory). As in, we do an anonymous mmap() of the entire maximum size of memory per hugepage size, per socket (which is limited to either RTE_MAX_MEMSEG_PER_TYPE pages or RTE_MAX_MEM_MB_PER_TYPE megabytes worth of memory, whichever is the smaller one), split over multiple lists (which are limited to either RTE_MAX_MEMSEG_PER_LIST memsegs or RTE_MAX_MEM_MB_PER_LIST megabytes per list, whichever is the smaller one). There is also a global limit of CONFIG_RTE_MAX_MEM_MB megabytes, which is mainly used for 32-bit targets to limit amounts of preallocated memory, but can be used to place an upper limit on total amount of VA memory that can be allocated by DPDK application. So, for each hugepage size, we get (by default) up to 128G worth of memory, per socket, split into chunks of up to 32G in size. The address space is claimed at the start, in eal_common_memory.c. The actual page allocation code is in eal_memalloc.c (Linux-only), and largely consists of copied EAL memory init code. Pages in the list are also indexed by address. That is, in order to figure out where the page belongs, one can simply look at base address for a memseg list. Similarly, figuring out IOVA address of a memzone is a matter of finding the right memseg list, getting offset and dividing by page size to get the appropriate memseg. This commit also removes rte_eal_dump_physmem_layout() call, according to deprecation notice [1], and removes that deprecation notice as well. On 32-bit targets due to limited VA space, DPDK will no longer spread memory to different sockets like before. Instead, it will (by default) allocate all of the memory on socket where master lcore is. To override this behavior, --socket-mem must be used. The rest of the changes are really ripple effects from the memseg change - heap changes, compile fixes, and rewrites to support fbarray-backed memseg lists. Due to earlier switch to _walk() functions, most of the changes are simple fixes, however some of the _walk() calls were switched to memseg list walk, where it made sense to do so. Additionally, we are also switching locks from flock() to fcntl(). Down the line, we will be introducing single-file segments option, and we cannot use flock() locks to lock parts of the file. Therefore, we will use fcntl() locks for legacy mem as well, in case someone is unfortunate enough to accidentally start legacy mem primary process alongside an already working non-legacy mem-based primary process. [1] http://dpdk.org/dev/patchwork/patch/34002/ Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> Tested-by: Santosh Shukla <santosh.shukla@caviumnetworks.com> Tested-by: Hemant Agrawal <hemant.agrawal@nxp.com> Tested-by: Gowrishankar Muthukrishnan <gowrishankar.m@linux.vnet.ibm.com>
2018-04-11 13:30:24 +01:00
seg_idx, addr, physaddr, page_sz);
total_mem += seg->len;
}
mem: replace memseg with memseg lists Before, we were aggregating multiple pages into one memseg, so the number of memsegs was small. Now, each page gets its own memseg, so the list of memsegs is huge. To accommodate the new memseg list size and to keep the under-the-hood workings sane, the memseg list is now not just a single list, but multiple lists. To be precise, each hugepage size available on the system gets one or more memseg lists, per socket. In order to support dynamic memory allocation, we reserve all memory in advance (unless we're in 32-bit legacy mode, in which case we do not preallocate memory). As in, we do an anonymous mmap() of the entire maximum size of memory per hugepage size, per socket (which is limited to either RTE_MAX_MEMSEG_PER_TYPE pages or RTE_MAX_MEM_MB_PER_TYPE megabytes worth of memory, whichever is the smaller one), split over multiple lists (which are limited to either RTE_MAX_MEMSEG_PER_LIST memsegs or RTE_MAX_MEM_MB_PER_LIST megabytes per list, whichever is the smaller one). There is also a global limit of CONFIG_RTE_MAX_MEM_MB megabytes, which is mainly used for 32-bit targets to limit amounts of preallocated memory, but can be used to place an upper limit on total amount of VA memory that can be allocated by DPDK application. So, for each hugepage size, we get (by default) up to 128G worth of memory, per socket, split into chunks of up to 32G in size. The address space is claimed at the start, in eal_common_memory.c. The actual page allocation code is in eal_memalloc.c (Linux-only), and largely consists of copied EAL memory init code. Pages in the list are also indexed by address. That is, in order to figure out where the page belongs, one can simply look at base address for a memseg list. Similarly, figuring out IOVA address of a memzone is a matter of finding the right memseg list, getting offset and dividing by page size to get the appropriate memseg. This commit also removes rte_eal_dump_physmem_layout() call, according to deprecation notice [1], and removes that deprecation notice as well. On 32-bit targets due to limited VA space, DPDK will no longer spread memory to different sockets like before. Instead, it will (by default) allocate all of the memory on socket where master lcore is. To override this behavior, --socket-mem must be used. The rest of the changes are really ripple effects from the memseg change - heap changes, compile fixes, and rewrites to support fbarray-backed memseg lists. Due to earlier switch to _walk() functions, most of the changes are simple fixes, however some of the _walk() calls were switched to memseg list walk, where it made sense to do so. Additionally, we are also switching locks from flock() to fcntl(). Down the line, we will be introducing single-file segments option, and we cannot use flock() locks to lock parts of the file. Therefore, we will use fcntl() locks for legacy mem as well, in case someone is unfortunate enough to accidentally start legacy mem primary process alongside an already working non-legacy mem-based primary process. [1] http://dpdk.org/dev/patchwork/patch/34002/ Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> Tested-by: Santosh Shukla <santosh.shukla@caviumnetworks.com> Tested-by: Hemant Agrawal <hemant.agrawal@nxp.com> Tested-by: Gowrishankar Muthukrishnan <gowrishankar.m@linux.vnet.ibm.com>
2018-04-11 13:30:24 +01:00
if (total_mem >= internal_config.memory)
break;
}
if (total_mem < internal_config.memory) {
RTE_LOG(ERR, EAL, "Couldn't reserve requested memory, "
"requested: %" PRIu64 "M "
"available: %" PRIu64 "M\n",
internal_config.memory >> 20, total_mem >> 20);
return -1;
}
return 0;
}
mem: replace memseg with memseg lists Before, we were aggregating multiple pages into one memseg, so the number of memsegs was small. Now, each page gets its own memseg, so the list of memsegs is huge. To accommodate the new memseg list size and to keep the under-the-hood workings sane, the memseg list is now not just a single list, but multiple lists. To be precise, each hugepage size available on the system gets one or more memseg lists, per socket. In order to support dynamic memory allocation, we reserve all memory in advance (unless we're in 32-bit legacy mode, in which case we do not preallocate memory). As in, we do an anonymous mmap() of the entire maximum size of memory per hugepage size, per socket (which is limited to either RTE_MAX_MEMSEG_PER_TYPE pages or RTE_MAX_MEM_MB_PER_TYPE megabytes worth of memory, whichever is the smaller one), split over multiple lists (which are limited to either RTE_MAX_MEMSEG_PER_LIST memsegs or RTE_MAX_MEM_MB_PER_LIST megabytes per list, whichever is the smaller one). There is also a global limit of CONFIG_RTE_MAX_MEM_MB megabytes, which is mainly used for 32-bit targets to limit amounts of preallocated memory, but can be used to place an upper limit on total amount of VA memory that can be allocated by DPDK application. So, for each hugepage size, we get (by default) up to 128G worth of memory, per socket, split into chunks of up to 32G in size. The address space is claimed at the start, in eal_common_memory.c. The actual page allocation code is in eal_memalloc.c (Linux-only), and largely consists of copied EAL memory init code. Pages in the list are also indexed by address. That is, in order to figure out where the page belongs, one can simply look at base address for a memseg list. Similarly, figuring out IOVA address of a memzone is a matter of finding the right memseg list, getting offset and dividing by page size to get the appropriate memseg. This commit also removes rte_eal_dump_physmem_layout() call, according to deprecation notice [1], and removes that deprecation notice as well. On 32-bit targets due to limited VA space, DPDK will no longer spread memory to different sockets like before. Instead, it will (by default) allocate all of the memory on socket where master lcore is. To override this behavior, --socket-mem must be used. The rest of the changes are really ripple effects from the memseg change - heap changes, compile fixes, and rewrites to support fbarray-backed memseg lists. Due to earlier switch to _walk() functions, most of the changes are simple fixes, however some of the _walk() calls were switched to memseg list walk, where it made sense to do so. Additionally, we are also switching locks from flock() to fcntl(). Down the line, we will be introducing single-file segments option, and we cannot use flock() locks to lock parts of the file. Therefore, we will use fcntl() locks for legacy mem as well, in case someone is unfortunate enough to accidentally start legacy mem primary process alongside an already working non-legacy mem-based primary process. [1] http://dpdk.org/dev/patchwork/patch/34002/ Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> Tested-by: Santosh Shukla <santosh.shukla@caviumnetworks.com> Tested-by: Hemant Agrawal <hemant.agrawal@nxp.com> Tested-by: Gowrishankar Muthukrishnan <gowrishankar.m@linux.vnet.ibm.com>
2018-04-11 13:30:24 +01:00
struct attach_walk_args {
int fd_hugepage;
int seg_idx;
};
static int
attach_segment(const struct rte_memseg_list *msl __rte_unused,
const struct rte_memseg *ms, void *arg)
{
struct attach_walk_args *wa = arg;
void *addr;
addr = mmap(ms->addr, ms->len, PROT_READ | PROT_WRITE,
MAP_SHARED | MAP_FIXED, wa->fd_hugepage,
wa->seg_idx * EAL_PAGE_SIZE);
if (addr == MAP_FAILED || addr != ms->addr)
return -1;
wa->seg_idx++;
return 0;
}
int
rte_eal_hugepage_attach(void)
{
const struct hugepage_info *hpi;
int fd_hugepage_info, fd_hugepage = -1;
mem: replace memseg with memseg lists Before, we were aggregating multiple pages into one memseg, so the number of memsegs was small. Now, each page gets its own memseg, so the list of memsegs is huge. To accommodate the new memseg list size and to keep the under-the-hood workings sane, the memseg list is now not just a single list, but multiple lists. To be precise, each hugepage size available on the system gets one or more memseg lists, per socket. In order to support dynamic memory allocation, we reserve all memory in advance (unless we're in 32-bit legacy mode, in which case we do not preallocate memory). As in, we do an anonymous mmap() of the entire maximum size of memory per hugepage size, per socket (which is limited to either RTE_MAX_MEMSEG_PER_TYPE pages or RTE_MAX_MEM_MB_PER_TYPE megabytes worth of memory, whichever is the smaller one), split over multiple lists (which are limited to either RTE_MAX_MEMSEG_PER_LIST memsegs or RTE_MAX_MEM_MB_PER_LIST megabytes per list, whichever is the smaller one). There is also a global limit of CONFIG_RTE_MAX_MEM_MB megabytes, which is mainly used for 32-bit targets to limit amounts of preallocated memory, but can be used to place an upper limit on total amount of VA memory that can be allocated by DPDK application. So, for each hugepage size, we get (by default) up to 128G worth of memory, per socket, split into chunks of up to 32G in size. The address space is claimed at the start, in eal_common_memory.c. The actual page allocation code is in eal_memalloc.c (Linux-only), and largely consists of copied EAL memory init code. Pages in the list are also indexed by address. That is, in order to figure out where the page belongs, one can simply look at base address for a memseg list. Similarly, figuring out IOVA address of a memzone is a matter of finding the right memseg list, getting offset and dividing by page size to get the appropriate memseg. This commit also removes rte_eal_dump_physmem_layout() call, according to deprecation notice [1], and removes that deprecation notice as well. On 32-bit targets due to limited VA space, DPDK will no longer spread memory to different sockets like before. Instead, it will (by default) allocate all of the memory on socket where master lcore is. To override this behavior, --socket-mem must be used. The rest of the changes are really ripple effects from the memseg change - heap changes, compile fixes, and rewrites to support fbarray-backed memseg lists. Due to earlier switch to _walk() functions, most of the changes are simple fixes, however some of the _walk() calls were switched to memseg list walk, where it made sense to do so. Additionally, we are also switching locks from flock() to fcntl(). Down the line, we will be introducing single-file segments option, and we cannot use flock() locks to lock parts of the file. Therefore, we will use fcntl() locks for legacy mem as well, in case someone is unfortunate enough to accidentally start legacy mem primary process alongside an already working non-legacy mem-based primary process. [1] http://dpdk.org/dev/patchwork/patch/34002/ Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> Tested-by: Santosh Shukla <santosh.shukla@caviumnetworks.com> Tested-by: Hemant Agrawal <hemant.agrawal@nxp.com> Tested-by: Gowrishankar Muthukrishnan <gowrishankar.m@linux.vnet.ibm.com>
2018-04-11 13:30:24 +01:00
unsigned int i;
/* Obtain a file descriptor for hugepage_info */
fd_hugepage_info = open(eal_hugepage_info_path(), O_RDONLY);
if (fd_hugepage_info < 0) {
RTE_LOG(ERR, EAL, "Could not open %s\n", eal_hugepage_info_path());
return -1;
}
/* Map the shared hugepage_info into the process address spaces */
mem: replace memseg with memseg lists Before, we were aggregating multiple pages into one memseg, so the number of memsegs was small. Now, each page gets its own memseg, so the list of memsegs is huge. To accommodate the new memseg list size and to keep the under-the-hood workings sane, the memseg list is now not just a single list, but multiple lists. To be precise, each hugepage size available on the system gets one or more memseg lists, per socket. In order to support dynamic memory allocation, we reserve all memory in advance (unless we're in 32-bit legacy mode, in which case we do not preallocate memory). As in, we do an anonymous mmap() of the entire maximum size of memory per hugepage size, per socket (which is limited to either RTE_MAX_MEMSEG_PER_TYPE pages or RTE_MAX_MEM_MB_PER_TYPE megabytes worth of memory, whichever is the smaller one), split over multiple lists (which are limited to either RTE_MAX_MEMSEG_PER_LIST memsegs or RTE_MAX_MEM_MB_PER_LIST megabytes per list, whichever is the smaller one). There is also a global limit of CONFIG_RTE_MAX_MEM_MB megabytes, which is mainly used for 32-bit targets to limit amounts of preallocated memory, but can be used to place an upper limit on total amount of VA memory that can be allocated by DPDK application. So, for each hugepage size, we get (by default) up to 128G worth of memory, per socket, split into chunks of up to 32G in size. The address space is claimed at the start, in eal_common_memory.c. The actual page allocation code is in eal_memalloc.c (Linux-only), and largely consists of copied EAL memory init code. Pages in the list are also indexed by address. That is, in order to figure out where the page belongs, one can simply look at base address for a memseg list. Similarly, figuring out IOVA address of a memzone is a matter of finding the right memseg list, getting offset and dividing by page size to get the appropriate memseg. This commit also removes rte_eal_dump_physmem_layout() call, according to deprecation notice [1], and removes that deprecation notice as well. On 32-bit targets due to limited VA space, DPDK will no longer spread memory to different sockets like before. Instead, it will (by default) allocate all of the memory on socket where master lcore is. To override this behavior, --socket-mem must be used. The rest of the changes are really ripple effects from the memseg change - heap changes, compile fixes, and rewrites to support fbarray-backed memseg lists. Due to earlier switch to _walk() functions, most of the changes are simple fixes, however some of the _walk() calls were switched to memseg list walk, where it made sense to do so. Additionally, we are also switching locks from flock() to fcntl(). Down the line, we will be introducing single-file segments option, and we cannot use flock() locks to lock parts of the file. Therefore, we will use fcntl() locks for legacy mem as well, in case someone is unfortunate enough to accidentally start legacy mem primary process alongside an already working non-legacy mem-based primary process. [1] http://dpdk.org/dev/patchwork/patch/34002/ Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> Tested-by: Santosh Shukla <santosh.shukla@caviumnetworks.com> Tested-by: Hemant Agrawal <hemant.agrawal@nxp.com> Tested-by: Gowrishankar Muthukrishnan <gowrishankar.m@linux.vnet.ibm.com>
2018-04-11 13:30:24 +01:00
hpi = mmap(NULL, sizeof(internal_config.hugepage_info),
PROT_READ, MAP_PRIVATE, fd_hugepage_info, 0);
if (hpi == MAP_FAILED) {
RTE_LOG(ERR, EAL, "Could not mmap %s\n", eal_hugepage_info_path());
goto error;
}
mem: replace memseg with memseg lists Before, we were aggregating multiple pages into one memseg, so the number of memsegs was small. Now, each page gets its own memseg, so the list of memsegs is huge. To accommodate the new memseg list size and to keep the under-the-hood workings sane, the memseg list is now not just a single list, but multiple lists. To be precise, each hugepage size available on the system gets one or more memseg lists, per socket. In order to support dynamic memory allocation, we reserve all memory in advance (unless we're in 32-bit legacy mode, in which case we do not preallocate memory). As in, we do an anonymous mmap() of the entire maximum size of memory per hugepage size, per socket (which is limited to either RTE_MAX_MEMSEG_PER_TYPE pages or RTE_MAX_MEM_MB_PER_TYPE megabytes worth of memory, whichever is the smaller one), split over multiple lists (which are limited to either RTE_MAX_MEMSEG_PER_LIST memsegs or RTE_MAX_MEM_MB_PER_LIST megabytes per list, whichever is the smaller one). There is also a global limit of CONFIG_RTE_MAX_MEM_MB megabytes, which is mainly used for 32-bit targets to limit amounts of preallocated memory, but can be used to place an upper limit on total amount of VA memory that can be allocated by DPDK application. So, for each hugepage size, we get (by default) up to 128G worth of memory, per socket, split into chunks of up to 32G in size. The address space is claimed at the start, in eal_common_memory.c. The actual page allocation code is in eal_memalloc.c (Linux-only), and largely consists of copied EAL memory init code. Pages in the list are also indexed by address. That is, in order to figure out where the page belongs, one can simply look at base address for a memseg list. Similarly, figuring out IOVA address of a memzone is a matter of finding the right memseg list, getting offset and dividing by page size to get the appropriate memseg. This commit also removes rte_eal_dump_physmem_layout() call, according to deprecation notice [1], and removes that deprecation notice as well. On 32-bit targets due to limited VA space, DPDK will no longer spread memory to different sockets like before. Instead, it will (by default) allocate all of the memory on socket where master lcore is. To override this behavior, --socket-mem must be used. The rest of the changes are really ripple effects from the memseg change - heap changes, compile fixes, and rewrites to support fbarray-backed memseg lists. Due to earlier switch to _walk() functions, most of the changes are simple fixes, however some of the _walk() calls were switched to memseg list walk, where it made sense to do so. Additionally, we are also switching locks from flock() to fcntl(). Down the line, we will be introducing single-file segments option, and we cannot use flock() locks to lock parts of the file. Therefore, we will use fcntl() locks for legacy mem as well, in case someone is unfortunate enough to accidentally start legacy mem primary process alongside an already working non-legacy mem-based primary process. [1] http://dpdk.org/dev/patchwork/patch/34002/ Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> Tested-by: Santosh Shukla <santosh.shukla@caviumnetworks.com> Tested-by: Hemant Agrawal <hemant.agrawal@nxp.com> Tested-by: Gowrishankar Muthukrishnan <gowrishankar.m@linux.vnet.ibm.com>
2018-04-11 13:30:24 +01:00
for (i = 0; i < internal_config.num_hugepage_sizes; i++) {
const struct hugepage_info *cur_hpi = &hpi[i];
struct attach_walk_args wa;
mem: replace memseg with memseg lists Before, we were aggregating multiple pages into one memseg, so the number of memsegs was small. Now, each page gets its own memseg, so the list of memsegs is huge. To accommodate the new memseg list size and to keep the under-the-hood workings sane, the memseg list is now not just a single list, but multiple lists. To be precise, each hugepage size available on the system gets one or more memseg lists, per socket. In order to support dynamic memory allocation, we reserve all memory in advance (unless we're in 32-bit legacy mode, in which case we do not preallocate memory). As in, we do an anonymous mmap() of the entire maximum size of memory per hugepage size, per socket (which is limited to either RTE_MAX_MEMSEG_PER_TYPE pages or RTE_MAX_MEM_MB_PER_TYPE megabytes worth of memory, whichever is the smaller one), split over multiple lists (which are limited to either RTE_MAX_MEMSEG_PER_LIST memsegs or RTE_MAX_MEM_MB_PER_LIST megabytes per list, whichever is the smaller one). There is also a global limit of CONFIG_RTE_MAX_MEM_MB megabytes, which is mainly used for 32-bit targets to limit amounts of preallocated memory, but can be used to place an upper limit on total amount of VA memory that can be allocated by DPDK application. So, for each hugepage size, we get (by default) up to 128G worth of memory, per socket, split into chunks of up to 32G in size. The address space is claimed at the start, in eal_common_memory.c. The actual page allocation code is in eal_memalloc.c (Linux-only), and largely consists of copied EAL memory init code. Pages in the list are also indexed by address. That is, in order to figure out where the page belongs, one can simply look at base address for a memseg list. Similarly, figuring out IOVA address of a memzone is a matter of finding the right memseg list, getting offset and dividing by page size to get the appropriate memseg. This commit also removes rte_eal_dump_physmem_layout() call, according to deprecation notice [1], and removes that deprecation notice as well. On 32-bit targets due to limited VA space, DPDK will no longer spread memory to different sockets like before. Instead, it will (by default) allocate all of the memory on socket where master lcore is. To override this behavior, --socket-mem must be used. The rest of the changes are really ripple effects from the memseg change - heap changes, compile fixes, and rewrites to support fbarray-backed memseg lists. Due to earlier switch to _walk() functions, most of the changes are simple fixes, however some of the _walk() calls were switched to memseg list walk, where it made sense to do so. Additionally, we are also switching locks from flock() to fcntl(). Down the line, we will be introducing single-file segments option, and we cannot use flock() locks to lock parts of the file. Therefore, we will use fcntl() locks for legacy mem as well, in case someone is unfortunate enough to accidentally start legacy mem primary process alongside an already working non-legacy mem-based primary process. [1] http://dpdk.org/dev/patchwork/patch/34002/ Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> Tested-by: Santosh Shukla <santosh.shukla@caviumnetworks.com> Tested-by: Hemant Agrawal <hemant.agrawal@nxp.com> Tested-by: Gowrishankar Muthukrishnan <gowrishankar.m@linux.vnet.ibm.com>
2018-04-11 13:30:24 +01:00
memset(&wa, 0, sizeof(wa));
mem: replace memseg with memseg lists Before, we were aggregating multiple pages into one memseg, so the number of memsegs was small. Now, each page gets its own memseg, so the list of memsegs is huge. To accommodate the new memseg list size and to keep the under-the-hood workings sane, the memseg list is now not just a single list, but multiple lists. To be precise, each hugepage size available on the system gets one or more memseg lists, per socket. In order to support dynamic memory allocation, we reserve all memory in advance (unless we're in 32-bit legacy mode, in which case we do not preallocate memory). As in, we do an anonymous mmap() of the entire maximum size of memory per hugepage size, per socket (which is limited to either RTE_MAX_MEMSEG_PER_TYPE pages or RTE_MAX_MEM_MB_PER_TYPE megabytes worth of memory, whichever is the smaller one), split over multiple lists (which are limited to either RTE_MAX_MEMSEG_PER_LIST memsegs or RTE_MAX_MEM_MB_PER_LIST megabytes per list, whichever is the smaller one). There is also a global limit of CONFIG_RTE_MAX_MEM_MB megabytes, which is mainly used for 32-bit targets to limit amounts of preallocated memory, but can be used to place an upper limit on total amount of VA memory that can be allocated by DPDK application. So, for each hugepage size, we get (by default) up to 128G worth of memory, per socket, split into chunks of up to 32G in size. The address space is claimed at the start, in eal_common_memory.c. The actual page allocation code is in eal_memalloc.c (Linux-only), and largely consists of copied EAL memory init code. Pages in the list are also indexed by address. That is, in order to figure out where the page belongs, one can simply look at base address for a memseg list. Similarly, figuring out IOVA address of a memzone is a matter of finding the right memseg list, getting offset and dividing by page size to get the appropriate memseg. This commit also removes rte_eal_dump_physmem_layout() call, according to deprecation notice [1], and removes that deprecation notice as well. On 32-bit targets due to limited VA space, DPDK will no longer spread memory to different sockets like before. Instead, it will (by default) allocate all of the memory on socket where master lcore is. To override this behavior, --socket-mem must be used. The rest of the changes are really ripple effects from the memseg change - heap changes, compile fixes, and rewrites to support fbarray-backed memseg lists. Due to earlier switch to _walk() functions, most of the changes are simple fixes, however some of the _walk() calls were switched to memseg list walk, where it made sense to do so. Additionally, we are also switching locks from flock() to fcntl(). Down the line, we will be introducing single-file segments option, and we cannot use flock() locks to lock parts of the file. Therefore, we will use fcntl() locks for legacy mem as well, in case someone is unfortunate enough to accidentally start legacy mem primary process alongside an already working non-legacy mem-based primary process. [1] http://dpdk.org/dev/patchwork/patch/34002/ Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> Tested-by: Santosh Shukla <santosh.shukla@caviumnetworks.com> Tested-by: Hemant Agrawal <hemant.agrawal@nxp.com> Tested-by: Gowrishankar Muthukrishnan <gowrishankar.m@linux.vnet.ibm.com>
2018-04-11 13:30:24 +01:00
/* Obtain a file descriptor for contiguous memory */
fd_hugepage = open(cur_hpi->hugedir, O_RDWR);
if (fd_hugepage < 0) {
RTE_LOG(ERR, EAL, "Could not open %s\n",
cur_hpi->hugedir);
goto error;
}
wa.fd_hugepage = fd_hugepage;
wa.seg_idx = 0;
mem: replace memseg with memseg lists Before, we were aggregating multiple pages into one memseg, so the number of memsegs was small. Now, each page gets its own memseg, so the list of memsegs is huge. To accommodate the new memseg list size and to keep the under-the-hood workings sane, the memseg list is now not just a single list, but multiple lists. To be precise, each hugepage size available on the system gets one or more memseg lists, per socket. In order to support dynamic memory allocation, we reserve all memory in advance (unless we're in 32-bit legacy mode, in which case we do not preallocate memory). As in, we do an anonymous mmap() of the entire maximum size of memory per hugepage size, per socket (which is limited to either RTE_MAX_MEMSEG_PER_TYPE pages or RTE_MAX_MEM_MB_PER_TYPE megabytes worth of memory, whichever is the smaller one), split over multiple lists (which are limited to either RTE_MAX_MEMSEG_PER_LIST memsegs or RTE_MAX_MEM_MB_PER_LIST megabytes per list, whichever is the smaller one). There is also a global limit of CONFIG_RTE_MAX_MEM_MB megabytes, which is mainly used for 32-bit targets to limit amounts of preallocated memory, but can be used to place an upper limit on total amount of VA memory that can be allocated by DPDK application. So, for each hugepage size, we get (by default) up to 128G worth of memory, per socket, split into chunks of up to 32G in size. The address space is claimed at the start, in eal_common_memory.c. The actual page allocation code is in eal_memalloc.c (Linux-only), and largely consists of copied EAL memory init code. Pages in the list are also indexed by address. That is, in order to figure out where the page belongs, one can simply look at base address for a memseg list. Similarly, figuring out IOVA address of a memzone is a matter of finding the right memseg list, getting offset and dividing by page size to get the appropriate memseg. This commit also removes rte_eal_dump_physmem_layout() call, according to deprecation notice [1], and removes that deprecation notice as well. On 32-bit targets due to limited VA space, DPDK will no longer spread memory to different sockets like before. Instead, it will (by default) allocate all of the memory on socket where master lcore is. To override this behavior, --socket-mem must be used. The rest of the changes are really ripple effects from the memseg change - heap changes, compile fixes, and rewrites to support fbarray-backed memseg lists. Due to earlier switch to _walk() functions, most of the changes are simple fixes, however some of the _walk() calls were switched to memseg list walk, where it made sense to do so. Additionally, we are also switching locks from flock() to fcntl(). Down the line, we will be introducing single-file segments option, and we cannot use flock() locks to lock parts of the file. Therefore, we will use fcntl() locks for legacy mem as well, in case someone is unfortunate enough to accidentally start legacy mem primary process alongside an already working non-legacy mem-based primary process. [1] http://dpdk.org/dev/patchwork/patch/34002/ Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> Tested-by: Santosh Shukla <santosh.shukla@caviumnetworks.com> Tested-by: Hemant Agrawal <hemant.agrawal@nxp.com> Tested-by: Gowrishankar Muthukrishnan <gowrishankar.m@linux.vnet.ibm.com>
2018-04-11 13:30:24 +01:00
/* Map the contiguous memory into each memory segment */
if (rte_memseg_walk(attach_segment, &wa) < 0) {
RTE_LOG(ERR, EAL, "Failed to mmap buffer %u from %s\n",
mem: replace memseg with memseg lists Before, we were aggregating multiple pages into one memseg, so the number of memsegs was small. Now, each page gets its own memseg, so the list of memsegs is huge. To accommodate the new memseg list size and to keep the under-the-hood workings sane, the memseg list is now not just a single list, but multiple lists. To be precise, each hugepage size available on the system gets one or more memseg lists, per socket. In order to support dynamic memory allocation, we reserve all memory in advance (unless we're in 32-bit legacy mode, in which case we do not preallocate memory). As in, we do an anonymous mmap() of the entire maximum size of memory per hugepage size, per socket (which is limited to either RTE_MAX_MEMSEG_PER_TYPE pages or RTE_MAX_MEM_MB_PER_TYPE megabytes worth of memory, whichever is the smaller one), split over multiple lists (which are limited to either RTE_MAX_MEMSEG_PER_LIST memsegs or RTE_MAX_MEM_MB_PER_LIST megabytes per list, whichever is the smaller one). There is also a global limit of CONFIG_RTE_MAX_MEM_MB megabytes, which is mainly used for 32-bit targets to limit amounts of preallocated memory, but can be used to place an upper limit on total amount of VA memory that can be allocated by DPDK application. So, for each hugepage size, we get (by default) up to 128G worth of memory, per socket, split into chunks of up to 32G in size. The address space is claimed at the start, in eal_common_memory.c. The actual page allocation code is in eal_memalloc.c (Linux-only), and largely consists of copied EAL memory init code. Pages in the list are also indexed by address. That is, in order to figure out where the page belongs, one can simply look at base address for a memseg list. Similarly, figuring out IOVA address of a memzone is a matter of finding the right memseg list, getting offset and dividing by page size to get the appropriate memseg. This commit also removes rte_eal_dump_physmem_layout() call, according to deprecation notice [1], and removes that deprecation notice as well. On 32-bit targets due to limited VA space, DPDK will no longer spread memory to different sockets like before. Instead, it will (by default) allocate all of the memory on socket where master lcore is. To override this behavior, --socket-mem must be used. The rest of the changes are really ripple effects from the memseg change - heap changes, compile fixes, and rewrites to support fbarray-backed memseg lists. Due to earlier switch to _walk() functions, most of the changes are simple fixes, however some of the _walk() calls were switched to memseg list walk, where it made sense to do so. Additionally, we are also switching locks from flock() to fcntl(). Down the line, we will be introducing single-file segments option, and we cannot use flock() locks to lock parts of the file. Therefore, we will use fcntl() locks for legacy mem as well, in case someone is unfortunate enough to accidentally start legacy mem primary process alongside an already working non-legacy mem-based primary process. [1] http://dpdk.org/dev/patchwork/patch/34002/ Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> Tested-by: Santosh Shukla <santosh.shukla@caviumnetworks.com> Tested-by: Hemant Agrawal <hemant.agrawal@nxp.com> Tested-by: Gowrishankar Muthukrishnan <gowrishankar.m@linux.vnet.ibm.com>
2018-04-11 13:30:24 +01:00
wa.seg_idx, cur_hpi->hugedir);
goto error;
}
mem: replace memseg with memseg lists Before, we were aggregating multiple pages into one memseg, so the number of memsegs was small. Now, each page gets its own memseg, so the list of memsegs is huge. To accommodate the new memseg list size and to keep the under-the-hood workings sane, the memseg list is now not just a single list, but multiple lists. To be precise, each hugepage size available on the system gets one or more memseg lists, per socket. In order to support dynamic memory allocation, we reserve all memory in advance (unless we're in 32-bit legacy mode, in which case we do not preallocate memory). As in, we do an anonymous mmap() of the entire maximum size of memory per hugepage size, per socket (which is limited to either RTE_MAX_MEMSEG_PER_TYPE pages or RTE_MAX_MEM_MB_PER_TYPE megabytes worth of memory, whichever is the smaller one), split over multiple lists (which are limited to either RTE_MAX_MEMSEG_PER_LIST memsegs or RTE_MAX_MEM_MB_PER_LIST megabytes per list, whichever is the smaller one). There is also a global limit of CONFIG_RTE_MAX_MEM_MB megabytes, which is mainly used for 32-bit targets to limit amounts of preallocated memory, but can be used to place an upper limit on total amount of VA memory that can be allocated by DPDK application. So, for each hugepage size, we get (by default) up to 128G worth of memory, per socket, split into chunks of up to 32G in size. The address space is claimed at the start, in eal_common_memory.c. The actual page allocation code is in eal_memalloc.c (Linux-only), and largely consists of copied EAL memory init code. Pages in the list are also indexed by address. That is, in order to figure out where the page belongs, one can simply look at base address for a memseg list. Similarly, figuring out IOVA address of a memzone is a matter of finding the right memseg list, getting offset and dividing by page size to get the appropriate memseg. This commit also removes rte_eal_dump_physmem_layout() call, according to deprecation notice [1], and removes that deprecation notice as well. On 32-bit targets due to limited VA space, DPDK will no longer spread memory to different sockets like before. Instead, it will (by default) allocate all of the memory on socket where master lcore is. To override this behavior, --socket-mem must be used. The rest of the changes are really ripple effects from the memseg change - heap changes, compile fixes, and rewrites to support fbarray-backed memseg lists. Due to earlier switch to _walk() functions, most of the changes are simple fixes, however some of the _walk() calls were switched to memseg list walk, where it made sense to do so. Additionally, we are also switching locks from flock() to fcntl(). Down the line, we will be introducing single-file segments option, and we cannot use flock() locks to lock parts of the file. Therefore, we will use fcntl() locks for legacy mem as well, in case someone is unfortunate enough to accidentally start legacy mem primary process alongside an already working non-legacy mem-based primary process. [1] http://dpdk.org/dev/patchwork/patch/34002/ Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> Tested-by: Santosh Shukla <santosh.shukla@caviumnetworks.com> Tested-by: Hemant Agrawal <hemant.agrawal@nxp.com> Tested-by: Gowrishankar Muthukrishnan <gowrishankar.m@linux.vnet.ibm.com>
2018-04-11 13:30:24 +01:00
close(fd_hugepage);
fd_hugepage = -1;
}
/* hugepage_info is no longer required */
mem: replace memseg with memseg lists Before, we were aggregating multiple pages into one memseg, so the number of memsegs was small. Now, each page gets its own memseg, so the list of memsegs is huge. To accommodate the new memseg list size and to keep the under-the-hood workings sane, the memseg list is now not just a single list, but multiple lists. To be precise, each hugepage size available on the system gets one or more memseg lists, per socket. In order to support dynamic memory allocation, we reserve all memory in advance (unless we're in 32-bit legacy mode, in which case we do not preallocate memory). As in, we do an anonymous mmap() of the entire maximum size of memory per hugepage size, per socket (which is limited to either RTE_MAX_MEMSEG_PER_TYPE pages or RTE_MAX_MEM_MB_PER_TYPE megabytes worth of memory, whichever is the smaller one), split over multiple lists (which are limited to either RTE_MAX_MEMSEG_PER_LIST memsegs or RTE_MAX_MEM_MB_PER_LIST megabytes per list, whichever is the smaller one). There is also a global limit of CONFIG_RTE_MAX_MEM_MB megabytes, which is mainly used for 32-bit targets to limit amounts of preallocated memory, but can be used to place an upper limit on total amount of VA memory that can be allocated by DPDK application. So, for each hugepage size, we get (by default) up to 128G worth of memory, per socket, split into chunks of up to 32G in size. The address space is claimed at the start, in eal_common_memory.c. The actual page allocation code is in eal_memalloc.c (Linux-only), and largely consists of copied EAL memory init code. Pages in the list are also indexed by address. That is, in order to figure out where the page belongs, one can simply look at base address for a memseg list. Similarly, figuring out IOVA address of a memzone is a matter of finding the right memseg list, getting offset and dividing by page size to get the appropriate memseg. This commit also removes rte_eal_dump_physmem_layout() call, according to deprecation notice [1], and removes that deprecation notice as well. On 32-bit targets due to limited VA space, DPDK will no longer spread memory to different sockets like before. Instead, it will (by default) allocate all of the memory on socket where master lcore is. To override this behavior, --socket-mem must be used. The rest of the changes are really ripple effects from the memseg change - heap changes, compile fixes, and rewrites to support fbarray-backed memseg lists. Due to earlier switch to _walk() functions, most of the changes are simple fixes, however some of the _walk() calls were switched to memseg list walk, where it made sense to do so. Additionally, we are also switching locks from flock() to fcntl(). Down the line, we will be introducing single-file segments option, and we cannot use flock() locks to lock parts of the file. Therefore, we will use fcntl() locks for legacy mem as well, in case someone is unfortunate enough to accidentally start legacy mem primary process alongside an already working non-legacy mem-based primary process. [1] http://dpdk.org/dev/patchwork/patch/34002/ Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> Tested-by: Santosh Shukla <santosh.shukla@caviumnetworks.com> Tested-by: Hemant Agrawal <hemant.agrawal@nxp.com> Tested-by: Gowrishankar Muthukrishnan <gowrishankar.m@linux.vnet.ibm.com>
2018-04-11 13:30:24 +01:00
munmap((void *)(uintptr_t)hpi, sizeof(internal_config.hugepage_info));
close(fd_hugepage_info);
return 0;
error:
if (fd_hugepage_info >= 0)
close(fd_hugepage_info);
if (fd_hugepage >= 0)
close(fd_hugepage);
return -1;
}
int
rte_eal_using_phys_addrs(void)
{
return 0;
}