This patch misses two things: 1. Documentation on pci_map_resource() mentions MAP_FAILED, while rte_mem_map() returns NULL on failure.
2. Code in drivers/bus/pci/linux passes MAP_FIXED to pci_map_resource(), which should be changed to corresponding RTE_MAP_FORCE_ADDRESS flag (it is renamed in v4 of MM patch series). On 2020-04-28 12:11 GMT+0300 tal...@mellanox.com wrote: [snip] > + if (mapaddr == NULL) { > RTE_LOG(ERR, EAL, > "%s(): cannot mmap(%d, %p, 0x%zx, 0x%llx): %s (%p)\n", > __func__, fd, requested_addr, size, > @@ -160,7 +160,7 @@ pci_unmap_resource(void *requested_addr, size_t size) > return; > > /* Unmap the PCI memory resource of device */ > - if (munmap(requested_addr, size)) { > + if (rte_mem_unmap(requested_addr, size)) { > RTE_LOG(ERR, EAL, "%s(): cannot munmap(%p, %#zx): %s\n", > __func__, requested_addr, size, > strerror(errno)); Logging can be removed here if ERR level is not a requirement: rte_mem_map() and rte_mem_unmap() already log OS errors on debug level. -- Dmitry Kozlyuk