env/dpdk: Detect DPDK's iova mode

Match DPDK's iova assignment strategy so there are never
any conflicts.

Change-Id: I3863487f9bd247c40edbf0d0d3a8c880bdad1708
Signed-off-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/477362
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Community-CI: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
This commit is contained in:
Ben Walker 2019-12-09 14:33:09 -07:00 committed by Tomasz Zawadzki
parent 801e76d8bf
commit dd7cd80c6d

View File

@ -1004,13 +1004,16 @@ spdk_vtophys_notify(void *cb_ctx, struct spdk_mem_map *map,
if (paddr == SPDK_VTOPHYS_ERROR) {
/* This is not an address that DPDK is managing. */
#if SPDK_VFIO_ENABLED
if (spdk_iommu_is_enabled()) {
/* We'll use the virtual address as the iova. DPDK
* currently uses physical addresses as the iovas (or counts
* up from 0 if it can't get physical addresses), so
* the range of user space virtual addresses and physical
* addresses will never overlap.
*/
enum rte_iova_mode iova_mode;
#if RTE_VERSION >= RTE_VERSION_NUM(19, 11, 0, 0)
iova_mode = rte_eal_iova_mode();
#else
iova_mode = rte_eal_get_configuration()->iova_mode;
#endif
if (spdk_iommu_is_enabled() && iova_mode == RTE_IOVA_VA) {
/* We'll use the virtual address as the iova to match DPDK. */
paddr = (uint64_t)vaddr;
rc = vtophys_iommu_map_dma((uint64_t)vaddr, paddr, len);
if (rc) {
@ -1062,6 +1065,17 @@ spdk_vtophys_notify(void *cb_ctx, struct spdk_mem_map *map,
DEBUG_PRINT("invalid paddr 0x%" PRIx64 " - must be 2MB aligned\n", paddr);
return -EINVAL;
}
#if SPDK_VFIO_ENABLED
/* If the IOMMU is on, but DPDK is using iova-mode=pa, we want to register this memory
* with the IOMMU using the physical address to match. */
if (spdk_iommu_is_enabled()) {
rc = vtophys_iommu_map_dma((uint64_t)vaddr, paddr, VALUE_2MB);
if (rc) {
DEBUG_PRINT("Unable to assign vaddr %p to paddr 0x%" PRIx64 "\n", vaddr, paddr);
return -EFAULT;
}
}
#endif
rc = spdk_mem_map_set_translation(map, (uint64_t)vaddr, VALUE_2MB, paddr);
if (rc != 0) {
@ -1101,13 +1115,50 @@ spdk_vtophys_notify(void *cb_ctx, struct spdk_mem_map *map,
*/
if (spdk_iommu_is_enabled()) {
uint64_t buffer_len = len;
paddr = spdk_mem_map_translate(map, (uint64_t)vaddr, &buffer_len);
if (buffer_len != len) {
return -EINVAL;
}
rc = vtophys_iommu_unmap_dma(paddr, len);
if (rc) {
return -EFAULT;
uint8_t *va = vaddr;
enum rte_iova_mode iova_mode;
#if RTE_VERSION >= RTE_VERSION_NUM(19, 11, 0, 0)
iova_mode = rte_eal_iova_mode();
#else
iova_mode = rte_eal_get_configuration()->iova_mode;
#endif
/*
* In virtual address mode, the region is contiguous and can be done in
* one unmap.
*/
if (iova_mode == RTE_IOVA_VA) {
paddr = spdk_mem_map_translate(map, (uint64_t)va, &buffer_len);
if (buffer_len != len || paddr != (uintptr_t)va) {
DEBUG_PRINT("Unmapping %p with length %lu failed because "
"translation had address 0x%" PRIx64 " and length %lu\n",
va, len, paddr, buffer_len);
return -EINVAL;
}
rc = vtophys_iommu_unmap_dma(paddr, len);
if (rc) {
DEBUG_PRINT("Failed to iommu unmap paddr 0x%" PRIx64 "\n", paddr);
return -EFAULT;
}
} else if (iova_mode == RTE_IOVA_PA) {
/* Get paddr for each 2MB chunk in this address range */
while (buffer_len > 0) {
paddr = spdk_mem_map_translate(map, (uint64_t)va, NULL);
if (paddr == SPDK_VTOPHYS_ERROR || buffer_len < VALUE_2MB) {
DEBUG_PRINT("could not get phys addr for %p\n", va);
return -EFAULT;
}
rc = vtophys_iommu_unmap_dma(paddr, VALUE_2MB);
if (rc) {
DEBUG_PRINT("Failed to iommu unmap paddr 0x%" PRIx64 "\n", paddr);
return -EFAULT;
}
va += VALUE_2MB;
buffer_len -= VALUE_2MB;
}
}
}
}