pci: keep API compatibility with mmap values
[dpdk.git] / drivers / bus / pci / linux / pci_vfio.c
index ffd26f1..07e072e 100644 (file)
 #include <rte_log.h>
 #include <rte_pci.h>
 #include <rte_bus_pci.h>
-#include <rte_eal_memconfig.h>
+#include <rte_eal_paging.h>
 #include <rte_malloc.h>
 #include <rte_vfio.h>
 #include <rte_eal.h>
 #include <rte_bus.h>
 #include <rte_spinlock.h>
+#include <rte_tailq.h>
 
 #include "eal_filesystem.h"
 
@@ -149,6 +150,38 @@ pci_vfio_get_msix_bar(int fd, struct pci_msix_table *msix_table)
        return 0;
 }
 
+/* enable PCI bus memory space */
+static int
+pci_vfio_enable_bus_memory(int dev_fd)
+{
+       uint16_t cmd;
+       int ret;
+
+       ret = pread64(dev_fd, &cmd, sizeof(cmd),
+                     VFIO_GET_REGION_ADDR(VFIO_PCI_CONFIG_REGION_INDEX) +
+                     PCI_COMMAND);
+
+       if (ret != sizeof(cmd)) {
+               RTE_LOG(ERR, EAL, "Cannot read command from PCI config space!\n");
+               return -1;
+       }
+
+       if (cmd & PCI_COMMAND_MEMORY)
+               return 0;
+
+       cmd |= PCI_COMMAND_MEMORY;
+       ret = pwrite64(dev_fd, &cmd, sizeof(cmd),
+                      VFIO_GET_REGION_ADDR(VFIO_PCI_CONFIG_REGION_INDEX) +
+                      PCI_COMMAND);
+
+       if (ret != sizeof(cmd)) {
+               RTE_LOG(ERR, EAL, "Cannot write command to PCI config space!\n");
+               return -1;
+       }
+
+       return 0;
+}
+
 /* set PCI bus mastering */
 static int
 pci_vfio_set_bus_master(int dev_fd, bool op)
@@ -427,6 +460,11 @@ pci_rte_vfio_setup_device(struct rte_pci_device *dev, int vfio_dev_fd)
                return -1;
        }
 
+       if (pci_vfio_enable_bus_memory(vfio_dev_fd)) {
+               RTE_LOG(ERR, EAL, "Cannot enable bus memory!\n");
+               return -1;
+       }
+
        /* set bus mastering for the device */
        if (pci_vfio_set_bus_master(vfio_dev_fd, true)) {
                RTE_LOG(ERR, EAL, "Cannot set up bus mastering!\n");
@@ -451,15 +489,17 @@ pci_vfio_mmap_bar(int vfio_dev_fd, struct mapped_pci_resource *vfio_res,
                int bar_index, int additional_flags)
 {
        struct memreg {
-               unsigned long offset, size;
+               uint64_t offset;
+               size_t   size;
        } memreg[2] = {};
        void *bar_addr;
        struct pci_msix_table *msix_table = &vfio_res->msix_table;
        struct pci_map *bar = &vfio_res->maps[bar_index];
 
-       if (bar->size == 0)
-               /* Skip this BAR */
+       if (bar->size == 0) {
+               RTE_LOG(DEBUG, EAL, "Bar size is 0, skip BAR%d\n", bar_index);
                return 0;
+       }
 
        if (msix_table->bar_index == bar_index) {
                /*
@@ -468,8 +508,15 @@ pci_vfio_mmap_bar(int vfio_dev_fd, struct mapped_pci_resource *vfio_res,
                 */
                uint32_t table_start = msix_table->offset;
                uint32_t table_end = table_start + msix_table->size;
-               table_end = (table_end + ~PAGE_MASK) & PAGE_MASK;
-               table_start &= PAGE_MASK;
+               table_end = RTE_ALIGN(table_end, PAGE_SIZE);
+               table_start = RTE_ALIGN_FLOOR(table_start, PAGE_SIZE);
+
+               /* If page-aligned start of MSI-X table is less than the
+                * actual MSI-X table start address, reassign to the actual
+                * start address.
+                */
+               if (table_start < msix_table->offset)
+                       table_start = msix_table->offset;
 
                if (table_start == 0 && table_end >= bar->size) {
                        /* Cannot map this BAR */
@@ -481,13 +528,23 @@ pci_vfio_mmap_bar(int vfio_dev_fd, struct mapped_pci_resource *vfio_res,
 
                memreg[0].offset = bar->offset;
                memreg[0].size = table_start;
-               memreg[1].offset = bar->offset + table_end;
-               memreg[1].size = bar->size - table_end;
+               if (bar->size < table_end) {
+                       /*
+                        * If MSI-X table end is beyond BAR end, don't attempt
+                        * to perform second mapping.
+                        */
+                       memreg[1].offset = 0;
+                       memreg[1].size = 0;
+               } else {
+                       memreg[1].offset = bar->offset + table_end;
+                       memreg[1].size = bar->size - table_end;
+               }
 
                RTE_LOG(DEBUG, EAL,
                        "Trying to map BAR%d that contains the MSI-X "
                        "table. Trying offsets: "
-                       "0x%04lx:0x%04lx, 0x%04lx:0x%04lx\n", bar_index,
+                       "0x%04" PRIx64 ":0x%04zx, 0x%04" PRIx64 ":0x%04zx\n",
+                       bar_index,
                        memreg[0].offset, memreg[0].size,
                        memreg[1].offset, memreg[1].size);
        } else {
@@ -505,23 +562,23 @@ pci_vfio_mmap_bar(int vfio_dev_fd, struct mapped_pci_resource *vfio_res,
                        map_addr = pci_map_resource(bar_addr, vfio_dev_fd,
                                                        memreg[0].offset,
                                                        memreg[0].size,
-                                                       MAP_FIXED);
+                                                       RTE_MAP_FORCE_ADDRESS);
                }
 
                /* if there's a second part, try to map it */
                if (map_addr != MAP_FAILED
                        && memreg[1].offset && memreg[1].size) {
                        void *second_addr = RTE_PTR_ADD(bar_addr,
-                                                       memreg[1].offset -
-                                                       (uintptr_t)bar->offset);
+                                               (uintptr_t)(memreg[1].offset -
+                                               bar->offset));
                        map_addr = pci_map_resource(second_addr,
                                                        vfio_dev_fd,
                                                        memreg[1].offset,
                                                        memreg[1].size,
-                                                       MAP_FIXED);
+                                                       RTE_MAP_FORCE_ADDRESS);
                }
 
-               if (map_addr == MAP_FAILED || !map_addr) {
+               if (map_addr == NULL || map_addr == MAP_FAILED) {
                        munmap(bar_addr, bar->size);
                        bar_addr = MAP_FAILED;
                        RTE_LOG(ERR, EAL, "Failed to map pci BAR%d\n",
@@ -658,7 +715,7 @@ pci_vfio_map_resource_primary(struct rte_pci_device *dev)
        vfio_res = rte_zmalloc("VFIO_RES", sizeof(*vfio_res), 0);
        if (vfio_res == NULL) {
                RTE_LOG(ERR, EAL,
-                       "%s(): cannot store uio mmap details\n", __func__);
+                       "%s(): cannot store vfio mmap details\n", __func__);
                goto err_vfio_dev_fd;
        }
        memcpy(&vfio_res->pci_addr, &dev->addr, sizeof(vfio_res->pci_addr));
@@ -731,6 +788,9 @@ pci_vfio_map_resource_primary(struct rte_pci_device *dev)
                bar_addr = pci_map_addr;
                pci_map_addr = RTE_PTR_ADD(bar_addr, (size_t) reg->size);
 
+               pci_map_addr = RTE_PTR_ALIGN(pci_map_addr,
+                                       sysconf(_SC_PAGE_SIZE));
+
                maps[i].addr = bar_addr;
                maps[i].offset = reg->offset;
                maps[i].size = reg->size;
@@ -794,11 +854,6 @@ pci_vfio_map_resource_secondary(struct rte_pci_device *dev)
        snprintf(pci_addr, sizeof(pci_addr), PCI_PRI_FMT,
                        loc->domain, loc->bus, loc->devid, loc->function);
 
-       ret = rte_vfio_setup_device(rte_pci_get_sysfs_path(), pci_addr,
-                                       &vfio_dev_fd, &device_info);
-       if (ret)
-               return ret;
-
        /* if we're in a secondary process, just find our tailq entry */
        TAILQ_FOREACH(vfio_res, vfio_res_list, next) {
                if (rte_pci_addr_cmp(&vfio_res->pci_addr,
@@ -810,9 +865,14 @@ pci_vfio_map_resource_secondary(struct rte_pci_device *dev)
        if (vfio_res == NULL) {
                RTE_LOG(ERR, EAL, "  %s cannot find TAILQ entry for PCI device!\n",
                                pci_addr);
-               goto err_vfio_dev_fd;
+               return -1;
        }
 
+       ret = rte_vfio_setup_device(rte_pci_get_sysfs_path(), pci_addr,
+                                       &vfio_dev_fd, &device_info);
+       if (ret)
+               return ret;
+
        /* map BARs */
        maps = vfio_res->maps;