eal: hide shared memory config
[dpdk.git] / lib / librte_eal / linux / eal / eal.c
index 3e1d6eb..228d6b3 100644 (file)
 #include "eal_internal_cfg.h"
 #include "eal_filesystem.h"
 #include "eal_hugepages.h"
+#include "eal_memcfg.h"
 #include "eal_options.h"
 #include "eal_vfio.h"
+#include "hotplug_mp.h"
 
 #define MEMSIZE_IF_NO_HUGE_PAGE (64ULL * 1024ULL * 1024ULL)
 
@@ -300,7 +302,7 @@ eal_parse_sysfs_value(const char *filename, unsigned long *val)
  * We also don't lock the whole file, so that in future we can use read-locks
  * on other parts, e.g. memzones, to detect if there are running secondary
  * processes. */
-static void
+static int
 rte_eal_config_create(void)
 {
        void *rte_mem_cfg_addr;
@@ -309,7 +311,7 @@ rte_eal_config_create(void)
        const char *pathname = eal_runtime_config_path();
 
        if (internal_config.no_shconf)
-               return;
+               return 0;
 
        /* map the config before hugepage address so that we don't waste a page */
        if (internal_config.base_virtaddr != 0)
@@ -321,29 +323,41 @@ rte_eal_config_create(void)
 
        if (mem_cfg_fd < 0){
                mem_cfg_fd = open(pathname, O_RDWR | O_CREAT, 0600);
-               if (mem_cfg_fd < 0)
-                       rte_panic("Cannot open '%s' for rte_mem_config\n", pathname);
+               if (mem_cfg_fd < 0) {
+                       RTE_LOG(ERR, EAL, "Cannot open '%s' for rte_mem_config\n",
+                               pathname);
+                       return -1;
+               }
        }
 
        retval = ftruncate(mem_cfg_fd, sizeof(*rte_config.mem_config));
        if (retval < 0){
                close(mem_cfg_fd);
-               rte_panic("Cannot resize '%s' for rte_mem_config\n", pathname);
+               mem_cfg_fd = -1;
+               RTE_LOG(ERR, EAL, "Cannot resize '%s' for rte_mem_config\n",
+                       pathname);
+               return -1;
        }
 
        retval = fcntl(mem_cfg_fd, F_SETLK, &wr_lock);
        if (retval < 0){
                close(mem_cfg_fd);
-               rte_exit(EXIT_FAILURE, "Cannot create lock on '%s'. Is another primary "
-                               "process running?\n", pathname);
+               mem_cfg_fd = -1;
+               RTE_LOG(ERR, EAL, "Cannot create lock on '%s'. Is another primary "
+                       "process running?\n", pathname);
+               return -1;
        }
 
        rte_mem_cfg_addr = mmap(rte_mem_cfg_addr, sizeof(*rte_config.mem_config),
                                PROT_READ | PROT_WRITE, MAP_SHARED, mem_cfg_fd, 0);
 
        if (rte_mem_cfg_addr == MAP_FAILED){
-               rte_panic("Cannot mmap memory for rte_config\n");
+               close(mem_cfg_fd);
+               mem_cfg_fd = -1;
+               RTE_LOG(ERR, EAL, "Cannot mmap memory for rte_config\n");
+               return -1;
        }
+
        memcpy(rte_mem_cfg_addr, &early_mem_config, sizeof(early_mem_config));
        rte_config.mem_config = rte_mem_cfg_addr;
 
@@ -353,10 +367,11 @@ rte_eal_config_create(void)
 
        rte_config.mem_config->dma_maskbits = 0;
 
+       return 0;
 }
 
 /* attach to an existing shared memory config */
-static void
+static int
 rte_eal_config_attach(void)
 {
        struct rte_mem_config *mem_config;
@@ -364,33 +379,42 @@ rte_eal_config_attach(void)
        const char *pathname = eal_runtime_config_path();
 
        if (internal_config.no_shconf)
-               return;
+               return 0;
 
        if (mem_cfg_fd < 0){
                mem_cfg_fd = open(pathname, O_RDWR);
-               if (mem_cfg_fd < 0)
-                       rte_panic("Cannot open '%s' for rte_mem_config\n", pathname);
+               if (mem_cfg_fd < 0) {
+                       RTE_LOG(ERR, EAL, "Cannot open '%s' for rte_mem_config\n",
+                               pathname);
+                       return -1;
+               }
        }
 
        /* map it as read-only first */
        mem_config = (struct rte_mem_config *) mmap(NULL, sizeof(*mem_config),
                        PROT_READ, MAP_SHARED, mem_cfg_fd, 0);
-       if (mem_config == MAP_FAILED)
-               rte_panic("Cannot mmap memory for rte_config! error %i (%s)\n",
-                         errno, strerror(errno));
+       if (mem_config == MAP_FAILED) {
+               close(mem_cfg_fd);
+               mem_cfg_fd = -1;
+               RTE_LOG(ERR, EAL, "Cannot mmap memory for rte_config! error %i (%s)\n",
+                       errno, strerror(errno));
+               return -1;
+       }
 
        rte_config.mem_config = mem_config;
+
+       return 0;
 }
 
 /* reattach the shared config at exact memory location primary process has it */
-static void
+static int
 rte_eal_config_reattach(void)
 {
        struct rte_mem_config *mem_config;
        void *rte_mem_cfg_addr;
 
        if (internal_config.no_shconf)
-               return;
+               return 0;
 
        /* save the address primary process has mapped shared config to */
        rte_mem_cfg_addr = (void *) (uintptr_t) rte_config.mem_config->mem_cfg_addr;
@@ -402,19 +426,26 @@ rte_eal_config_reattach(void)
        mem_config = (struct rte_mem_config *) mmap(rte_mem_cfg_addr,
                        sizeof(*mem_config), PROT_READ | PROT_WRITE, MAP_SHARED,
                        mem_cfg_fd, 0);
+
+       close(mem_cfg_fd);
+       mem_cfg_fd = -1;
+
        if (mem_config == MAP_FAILED || mem_config != rte_mem_cfg_addr) {
-               if (mem_config != MAP_FAILED)
+               if (mem_config != MAP_FAILED) {
                        /* errno is stale, don't use */
-                       rte_panic("Cannot mmap memory for rte_config at [%p], got [%p]"
-                                 " - please use '--base-virtaddr' option\n",
-                                 rte_mem_cfg_addr, mem_config);
-               else
-                       rte_panic("Cannot mmap memory for rte_config! error %i (%s)\n",
-                                 errno, strerror(errno));
+                       RTE_LOG(ERR, EAL, "Cannot mmap memory for rte_config at [%p], got [%p]"
+                               " - please use '--base-virtaddr' option\n",
+                               rte_mem_cfg_addr, mem_config);
+                       return -1;
+               }
+               RTE_LOG(ERR, EAL, "Cannot mmap memory for rte_config! error %i (%s)\n",
+                       errno, strerror(errno));
+               return -1;
        }
-       close(mem_cfg_fd);
 
        rte_config.mem_config = mem_config;
+
+       return 0;
 }
 
 /* Detect if we are a primary or a secondary process */
@@ -461,26 +492,33 @@ eal_update_internal_config(void)
 }
 
 /* Sets up rte_config structure with the pointer to shared memory config.*/
-static void
+static int
 rte_config_init(void)
 {
        rte_config.process_type = internal_config.process_type;
 
        switch (rte_config.process_type){
        case RTE_PROC_PRIMARY:
-               rte_eal_config_create();
+               if (rte_eal_config_create() < 0)
+                       return -1;
                eal_update_mem_config();
                break;
        case RTE_PROC_SECONDARY:
-               rte_eal_config_attach();
+               if (rte_eal_config_attach() < 0)
+                       return -1;
                rte_eal_mcfg_wait_complete(rte_config.mem_config);
-               rte_eal_config_reattach();
+               if (rte_eal_config_reattach() < 0)
+                       return -1;
                eal_update_internal_config();
                break;
        case RTE_PROC_AUTO:
        case RTE_PROC_INVALID:
-               rte_panic("Invalid process type\n");
+               RTE_LOG(ERR, EAL, "Invalid process type %d\n",
+                       rte_config.process_type);
+               return -1;
        }
+
+       return 0;
 }
 
 /* Unlocks hugepage directories that were locked by eal_hugepage_info_init */
@@ -948,6 +986,7 @@ rte_eal_init(int argc, char **argv)
        static char logid[PATH_MAX];
        char cpuset[RTE_CPU_AFFINITY_STR_LEN];
        char thread_name[RTE_MAX_THREAD_NAME_LEN];
+       bool phys_addrs;
 
        /* checks if the machine is adequate */
        if (!rte_cpu_is_supported()) {
@@ -998,7 +1037,10 @@ rte_eal_init(int argc, char **argv)
                return -1;
        }
 
-       rte_config_init();
+       if (rte_config_init() < 0) {
+               rte_eal_init_alert("Cannot init config");
+               return -1;
+       }
 
        if (rte_eal_intr_init() < 0) {
                rte_eal_init_alert("Cannot init interrupt-handling thread");
@@ -1006,7 +1048,7 @@ rte_eal_init(int argc, char **argv)
        }
 
        if (rte_eal_alarm_init() < 0) {
-               rte_eal_init_alert("Cannot init interrupt-handling thread");
+               rte_eal_init_alert("Cannot init alarm");
                /* rte_eal_alarm_init sets rte_errno on failure. */
                return -1;
        }
@@ -1023,7 +1065,7 @@ rte_eal_init(int argc, char **argv)
        }
 
        /* register multi-process action callbacks for hotplug */
-       if (rte_mp_dev_hotplug_init() < 0) {
+       if (eal_mp_dev_hotplug_init() < 0) {
                rte_eal_init_alert("failed to register mp callback for hotplug");
                return -1;
        }
@@ -1035,25 +1077,46 @@ rte_eal_init(int argc, char **argv)
                return -1;
        }
 
+       phys_addrs = rte_eal_using_phys_addrs() != 0;
+
        /* if no EAL option "--iova-mode=<pa|va>", use bus IOVA scheme */
        if (internal_config.iova_mode == RTE_IOVA_DC) {
-               /* autodetect the IOVA mapping mode (default is RTE_IOVA_PA) */
-               rte_eal_get_configuration()->iova_mode =
-                       rte_bus_get_iommu_class();
+               /* autodetect the IOVA mapping mode */
+               enum rte_iova_mode iova_mode = rte_bus_get_iommu_class();
 
+               if (iova_mode == RTE_IOVA_DC) {
+                       iova_mode = phys_addrs ? RTE_IOVA_PA : RTE_IOVA_VA;
+                       RTE_LOG(DEBUG, EAL,
+                               "Buses did not request a specific IOVA mode, using '%s' based on physical addresses availability.\n",
+                               phys_addrs ? "PA" : "VA");
+               }
+#ifdef RTE_LIBRTE_KNI
                /* Workaround for KNI which requires physical address to work */
-               if (rte_eal_get_configuration()->iova_mode == RTE_IOVA_VA &&
+               if (iova_mode == RTE_IOVA_VA &&
                                rte_eal_check_module("rte_kni") == 1) {
-                       rte_eal_get_configuration()->iova_mode = RTE_IOVA_PA;
-                       RTE_LOG(WARNING, EAL,
-                               "Some devices want IOVA as VA but PA will be used because.. "
-                               "KNI module inserted\n");
+                       if (phys_addrs) {
+                               iova_mode = RTE_IOVA_PA;
+                               RTE_LOG(WARNING, EAL, "Forcing IOVA as 'PA' because KNI module is loaded\n");
+                       } else {
+                               RTE_LOG(DEBUG, EAL, "KNI can not work since physical addresses are unavailable\n");
+                       }
                }
+#endif
+               rte_eal_get_configuration()->iova_mode = iova_mode;
        } else {
                rte_eal_get_configuration()->iova_mode =
                        internal_config.iova_mode;
        }
 
+       if (rte_eal_iova_mode() == RTE_IOVA_PA && !phys_addrs) {
+               rte_eal_init_alert("Cannot use IOVA as 'PA' since physical addresses are not available");
+               rte_errno = EINVAL;
+               return -1;
+       }
+
+       RTE_LOG(INFO, EAL, "Selected IOVA mode '%s'\n",
+               rte_eal_iova_mode() == RTE_IOVA_PA ? "PA" : "VA");
+
        if (internal_config.no_hugetlbfs == 0) {
                /* rte_config isn't initialized yet */
                ret = internal_config.process_type == RTE_PROC_PRIMARY ?
@@ -1083,8 +1146,6 @@ rte_eal_init(int argc, char **argv)
 #endif
        }
 
-       rte_srand(rte_rdtsc());
-
        if (rte_eal_log_init(logid, internal_config.syslog_facility) < 0) {
                rte_eal_init_alert("Cannot init logging.");
                rte_errno = ENOMEM;