diff options
Diffstat (limited to 'hw/virtio/virtio-mem.c')
| -rw-r--r-- | hw/virtio/virtio-mem.c | 330 |
1 files changed, 320 insertions, 10 deletions
diff --git a/hw/virtio/virtio-mem.c b/hw/virtio/virtio-mem.c index da5b09cefc..9dc3c61b5a 100644 --- a/hw/virtio/virtio-mem.c +++ b/hw/virtio/virtio-mem.c @@ -67,6 +67,13 @@ static uint32_t virtio_mem_default_thp_size(void) } /* + * The minimum memslot size depends on this setting ("sane default"), the + * device block size, and the memory backend page size. The last (or single) + * memslot might be smaller than this constant. + */ +#define VIRTIO_MEM_MIN_MEMSLOT_SIZE (1 * GiB) + +/* * We want to have a reasonable default block size such that * 1. We avoid splitting THPs when unplugging memory, which degrades * performance. @@ -177,10 +184,10 @@ static bool virtio_mem_is_busy(void) return migration_in_incoming_postcopy() || !migration_is_idle(); } -typedef int (*virtio_mem_range_cb)(const VirtIOMEM *vmem, void *arg, +typedef int (*virtio_mem_range_cb)(VirtIOMEM *vmem, void *arg, uint64_t offset, uint64_t size); -static int virtio_mem_for_each_unplugged_range(const VirtIOMEM *vmem, void *arg, +static int virtio_mem_for_each_unplugged_range(VirtIOMEM *vmem, void *arg, virtio_mem_range_cb cb) { unsigned long first_zero_bit, last_zero_bit; @@ -204,7 +211,7 @@ static int virtio_mem_for_each_unplugged_range(const VirtIOMEM *vmem, void *arg, return ret; } -static int virtio_mem_for_each_plugged_range(const VirtIOMEM *vmem, void *arg, +static int virtio_mem_for_each_plugged_range(VirtIOMEM *vmem, void *arg, virtio_mem_range_cb cb) { unsigned long first_bit, last_bit; @@ -483,6 +490,96 @@ static bool virtio_mem_valid_range(const VirtIOMEM *vmem, uint64_t gpa, return true; } +static void virtio_mem_activate_memslot(VirtIOMEM *vmem, unsigned int idx) +{ + const uint64_t memslot_offset = idx * vmem->memslot_size; + + assert(vmem->memslots); + + /* + * Instead of enabling/disabling memslots, we add/remove them. This should + * make address space updates faster, because we don't have to loop over + * many disabled subregions. + */ + if (memory_region_is_mapped(&vmem->memslots[idx])) { + return; + } + memory_region_add_subregion(vmem->mr, memslot_offset, &vmem->memslots[idx]); +} + +static void virtio_mem_deactivate_memslot(VirtIOMEM *vmem, unsigned int idx) +{ + assert(vmem->memslots); + + if (!memory_region_is_mapped(&vmem->memslots[idx])) { + return; + } + memory_region_del_subregion(vmem->mr, &vmem->memslots[idx]); +} + +static void virtio_mem_activate_memslots_to_plug(VirtIOMEM *vmem, + uint64_t offset, uint64_t size) +{ + const unsigned int start_idx = offset / vmem->memslot_size; + const unsigned int end_idx = (offset + size + vmem->memslot_size - 1) / + vmem->memslot_size; + unsigned int idx; + + if (!vmem->dynamic_memslots) { + return; + } + + /* Activate all involved memslots in a single transaction. */ + memory_region_transaction_begin(); + for (idx = start_idx; idx < end_idx; idx++) { + virtio_mem_activate_memslot(vmem, idx); + } + memory_region_transaction_commit(); +} + +static void virtio_mem_deactivate_unplugged_memslots(VirtIOMEM *vmem, + uint64_t offset, + uint64_t size) +{ + const uint64_t region_size = memory_region_size(&vmem->memdev->mr); + const unsigned int start_idx = offset / vmem->memslot_size; + const unsigned int end_idx = (offset + size + vmem->memslot_size - 1) / + vmem->memslot_size; + unsigned int idx; + + if (!vmem->dynamic_memslots) { + return; + } + + /* Deactivate all memslots with unplugged blocks in a single transaction. */ + memory_region_transaction_begin(); + for (idx = start_idx; idx < end_idx; idx++) { + const uint64_t memslot_offset = idx * vmem->memslot_size; + uint64_t memslot_size = vmem->memslot_size; + + /* The size of the last memslot might be smaller. */ + if (idx == vmem->nb_memslots - 1) { + memslot_size = region_size - memslot_offset; + } + + /* + * Partially covered memslots might still have some blocks plugged and + * have to remain active if that's the case. + */ + if (offset > memslot_offset || + offset + size < memslot_offset + memslot_size) { + const uint64_t gpa = vmem->addr + memslot_offset; + + if (!virtio_mem_is_range_unplugged(vmem, gpa, memslot_size)) { + continue; + } + } + + virtio_mem_deactivate_memslot(vmem, idx); + } + memory_region_transaction_commit(); +} + static int virtio_mem_set_block_state(VirtIOMEM *vmem, uint64_t start_gpa, uint64_t size, bool plug) { @@ -500,6 +597,8 @@ static int virtio_mem_set_block_state(VirtIOMEM *vmem, uint64_t start_gpa, } virtio_mem_notify_unplug(vmem, offset, size); virtio_mem_set_range_unplugged(vmem, start_gpa, size); + /* Deactivate completely unplugged memslots after updating the state. */ + virtio_mem_deactivate_unplugged_memslots(vmem, offset, size); return 0; } @@ -527,7 +626,20 @@ static int virtio_mem_set_block_state(VirtIOMEM *vmem, uint64_t start_gpa, } if (!ret) { + /* + * Activate before notifying and rollback in case of any errors. + * + * When activating a yet inactive memslot, memory notifiers will get + * notified about the added memory region and can register with the + * RamDiscardManager; this will traverse all plugged blocks and skip the + * blocks we are plugging here. The following notification will inform + * registered listeners about the blocks we're plugging. + */ + virtio_mem_activate_memslots_to_plug(vmem, offset, size); ret = virtio_mem_notify_plug(vmem, offset, size); + if (ret) { + virtio_mem_deactivate_unplugged_memslots(vmem, offset, size); + } } if (ret) { /* Could be preallocation or a notifier populated memory. */ @@ -620,6 +732,7 @@ static void virtio_mem_resize_usable_region(VirtIOMEM *vmem, static int virtio_mem_unplug_all(VirtIOMEM *vmem) { + const uint64_t region_size = memory_region_size(&vmem->memdev->mr); RAMBlock *rb = vmem->memdev->mr.ram_block; if (vmem->size) { @@ -634,6 +747,9 @@ static int virtio_mem_unplug_all(VirtIOMEM *vmem) bitmap_clear(vmem->bitmap, 0, vmem->bitmap_size); vmem->size = 0; notifier_list_notify(&vmem->size_change_notifiers, &vmem->size); + + /* Deactivate all memslots after updating the state. */ + virtio_mem_deactivate_unplugged_memslots(vmem, 0, region_size); } trace_virtio_mem_unplugged_all(); @@ -790,6 +906,49 @@ static void virtio_mem_system_reset(void *opaque) virtio_mem_unplug_all(vmem); } +static void virtio_mem_prepare_mr(VirtIOMEM *vmem) +{ + const uint64_t region_size = memory_region_size(&vmem->memdev->mr); + + assert(!vmem->mr && vmem->dynamic_memslots); + vmem->mr = g_new0(MemoryRegion, 1); + memory_region_init(vmem->mr, OBJECT(vmem), "virtio-mem", + region_size); + vmem->mr->align = memory_region_get_alignment(&vmem->memdev->mr); +} + +static void virtio_mem_prepare_memslots(VirtIOMEM *vmem) +{ + const uint64_t region_size = memory_region_size(&vmem->memdev->mr); + unsigned int idx; + + g_assert(!vmem->memslots && vmem->nb_memslots && vmem->dynamic_memslots); + vmem->memslots = g_new0(MemoryRegion, vmem->nb_memslots); + + /* Initialize our memslots, but don't map them yet. */ + for (idx = 0; idx < vmem->nb_memslots; idx++) { + const uint64_t memslot_offset = idx * vmem->memslot_size; + uint64_t memslot_size = vmem->memslot_size; + char name[20]; + + /* The size of the last memslot might be smaller. */ + if (idx == vmem->nb_memslots - 1) { + memslot_size = region_size - memslot_offset; + } + + snprintf(name, sizeof(name), "memslot-%u", idx); + memory_region_init_alias(&vmem->memslots[idx], OBJECT(vmem), name, + &vmem->memdev->mr, memslot_offset, + memslot_size); + /* + * We want to be able to atomically and efficiently activate/deactivate + * individual memslots without affecting adjacent memslots in memory + * notifiers. + */ + memory_region_set_unmergeable(&vmem->memslots[idx], true); + } +} + static void virtio_mem_device_realize(DeviceState *dev, Error **errp) { MachineState *ms = MACHINE(qdev_get_machine()); @@ -861,6 +1020,14 @@ static void virtio_mem_device_realize(DeviceState *dev, Error **errp) vmem->unplugged_inaccessible = ON_OFF_AUTO_ON; #endif /* VIRTIO_MEM_HAS_LEGACY_GUESTS */ + if (vmem->dynamic_memslots && + vmem->unplugged_inaccessible != ON_OFF_AUTO_ON) { + error_setg(errp, "'%s' property set to 'on' requires '%s' to be 'on'", + VIRTIO_MEM_DYNAMIC_MEMSLOTS_PROP, + VIRTIO_MEM_UNPLUGGED_INACCESSIBLE_PROP); + return; + } + /* * If the block size wasn't configured by the user, use a sane default. This * allows using hugetlbfs backends of any page size without manual @@ -930,6 +1097,25 @@ static void virtio_mem_device_realize(DeviceState *dev, Error **errp) virtio_init(vdev, VIRTIO_ID_MEM, sizeof(struct virtio_mem_config)); vmem->vq = virtio_add_queue(vdev, 128, virtio_mem_handle_request); + /* + * With "dynamic-memslots=off" (old behavior) we always map the whole + * RAM memory region directly. + */ + if (vmem->dynamic_memslots) { + if (!vmem->mr) { + virtio_mem_prepare_mr(vmem); + } + if (vmem->nb_memslots <= 1) { + vmem->nb_memslots = 1; + vmem->memslot_size = memory_region_size(&vmem->memdev->mr); + } + if (!vmem->memslots) { + virtio_mem_prepare_memslots(vmem); + } + } else { + assert(!vmem->mr && !vmem->nb_memslots && !vmem->memslots); + } + host_memory_backend_set_mapped(vmem->memdev, true); vmstate_register_ram(&vmem->memdev->mr, DEVICE(vmem)); if (vmem->early_migration) { @@ -969,7 +1155,7 @@ static void virtio_mem_device_unrealize(DeviceState *dev) ram_block_coordinated_discard_require(false); } -static int virtio_mem_discard_range_cb(const VirtIOMEM *vmem, void *arg, +static int virtio_mem_discard_range_cb(VirtIOMEM *vmem, void *arg, uint64_t offset, uint64_t size) { RAMBlock *rb = vmem->memdev->mr.ram_block; @@ -984,13 +1170,32 @@ static int virtio_mem_restore_unplugged(VirtIOMEM *vmem) virtio_mem_discard_range_cb); } -static int virtio_mem_post_load(void *opaque, int version_id) +static int virtio_mem_activate_memslot_range_cb(VirtIOMEM *vmem, void *arg, + uint64_t offset, uint64_t size) +{ + virtio_mem_activate_memslots_to_plug(vmem, offset, size); + return 0; +} + +static int virtio_mem_post_load_bitmap(VirtIOMEM *vmem) { - VirtIOMEM *vmem = VIRTIO_MEM(opaque); RamDiscardListener *rdl; int ret; /* + * We restored the bitmap and updated the requested size; activate all + * memslots (so listeners register) before notifying about plugged blocks. + */ + if (vmem->dynamic_memslots) { + /* + * We don't expect any active memslots at this point to deactivate: no + * memory was plugged on the migration destination. + */ + virtio_mem_for_each_plugged_range(vmem, NULL, + virtio_mem_activate_memslot_range_cb); + } + + /* * We started out with all memory discarded and our memory region is mapped * into an address space. Replay, now that we updated the bitmap. */ @@ -1001,6 +1206,20 @@ static int virtio_mem_post_load(void *opaque, int version_id) return ret; } } + return 0; +} + +static int virtio_mem_post_load(void *opaque, int version_id) +{ + VirtIOMEM *vmem = VIRTIO_MEM(opaque); + int ret; + + if (!vmem->early_migration) { + ret = virtio_mem_post_load_bitmap(vmem); + if (ret) { + return ret; + } + } /* * If shared RAM is migrated using the file content and not using QEMU, @@ -1021,7 +1240,7 @@ static int virtio_mem_post_load(void *opaque, int version_id) return virtio_mem_restore_unplugged(vmem); } -static int virtio_mem_prealloc_range_cb(const VirtIOMEM *vmem, void *arg, +static int virtio_mem_prealloc_range_cb(VirtIOMEM *vmem, void *arg, uint64_t offset, uint64_t size) { void *area = memory_region_get_ram_ptr(&vmem->memdev->mr) + offset; @@ -1043,7 +1262,7 @@ static int virtio_mem_post_load_early(void *opaque, int version_id) int ret; if (!vmem->prealloc) { - return 0; + goto post_load_bitmap; } /* @@ -1051,7 +1270,7 @@ static int virtio_mem_post_load_early(void *opaque, int version_id) * don't mess with preallocation and postcopy. */ if (migrate_ram_is_ignored(rb)) { - return 0; + goto post_load_bitmap; } /* @@ -1084,7 +1303,10 @@ static int virtio_mem_post_load_early(void *opaque, int version_id) return -EBUSY; } } - return 0; + +post_load_bitmap: + /* Finally, update any other state to be consistent with the new bitmap. */ + return virtio_mem_post_load_bitmap(vmem); } typedef struct VirtIOMEMMigSanityChecks { @@ -1235,11 +1457,79 @@ static MemoryRegion *virtio_mem_get_memory_region(VirtIOMEM *vmem, Error **errp) if (!vmem->memdev) { error_setg(errp, "'%s' property must be set", VIRTIO_MEM_MEMDEV_PROP); return NULL; + } else if (vmem->dynamic_memslots) { + if (!vmem->mr) { + virtio_mem_prepare_mr(vmem); + } + return vmem->mr; } return &vmem->memdev->mr; } +static void virtio_mem_decide_memslots(VirtIOMEM *vmem, unsigned int limit) +{ + uint64_t region_size, memslot_size, min_memslot_size; + unsigned int memslots; + RAMBlock *rb; + + if (!vmem->dynamic_memslots) { + return; + } + + /* We're called exactly once, before realizing the device. */ + assert(!vmem->nb_memslots); + + /* If realizing the device will fail, just assume a single memslot. */ + if (limit <= 1 || !vmem->memdev || !vmem->memdev->mr.ram_block) { + vmem->nb_memslots = 1; + return; + } + + rb = vmem->memdev->mr.ram_block; + region_size = memory_region_size(&vmem->memdev->mr); + + /* + * Determine the default block size now, to determine the minimum memslot + * size. We want the minimum slot size to be at least the device block size. + */ + if (!vmem->block_size) { + vmem->block_size = virtio_mem_default_block_size(rb); + } + /* If realizing the device will fail, just assume a single memslot. */ + if (vmem->block_size < qemu_ram_pagesize(rb) || + !QEMU_IS_ALIGNED(region_size, vmem->block_size)) { + vmem->nb_memslots = 1; + return; + } + + /* + * All memslots except the last one have a reasonable minimum size, and + * and all memslot sizes are aligned to the device block size. + */ + memslot_size = QEMU_ALIGN_UP(region_size / limit, vmem->block_size); + min_memslot_size = MAX(vmem->block_size, VIRTIO_MEM_MIN_MEMSLOT_SIZE); + memslot_size = MAX(memslot_size, min_memslot_size); + + memslots = QEMU_ALIGN_UP(region_size, memslot_size) / memslot_size; + if (memslots != 1) { + vmem->memslot_size = memslot_size; + } + vmem->nb_memslots = memslots; +} + +static unsigned int virtio_mem_get_memslots(VirtIOMEM *vmem) +{ + if (!vmem->dynamic_memslots) { + /* Exactly one static RAM memory region. */ + return 1; + } + + /* We're called after instructed to make a decision. */ + g_assert(vmem->nb_memslots); + return vmem->nb_memslots; +} + static void virtio_mem_add_size_change_notifier(VirtIOMEM *vmem, Notifier *notifier) { @@ -1377,6 +1667,21 @@ static void virtio_mem_instance_init(Object *obj) NULL, NULL); } +static void virtio_mem_instance_finalize(Object *obj) +{ + VirtIOMEM *vmem = VIRTIO_MEM(obj); + + /* + * Note: the core already dropped the references on all memory regions + * (it's passed as the owner to memory_region_init_*()) and finalized + * these objects. We can simply free the memory. + */ + g_free(vmem->memslots); + vmem->memslots = NULL; + g_free(vmem->mr); + vmem->mr = NULL; +} + static Property virtio_mem_properties[] = { DEFINE_PROP_UINT64(VIRTIO_MEM_ADDR_PROP, VirtIOMEM, addr, 0), DEFINE_PROP_UINT32(VIRTIO_MEM_NODE_PROP, VirtIOMEM, node, 0), @@ -1389,6 +1694,8 @@ static Property virtio_mem_properties[] = { #endif DEFINE_PROP_BOOL(VIRTIO_MEM_EARLY_MIGRATION_PROP, VirtIOMEM, early_migration, true), + DEFINE_PROP_BOOL(VIRTIO_MEM_DYNAMIC_MEMSLOTS_PROP, VirtIOMEM, + dynamic_memslots, false), DEFINE_PROP_END_OF_LIST(), }; @@ -1556,6 +1863,8 @@ static void virtio_mem_class_init(ObjectClass *klass, void *data) vmc->fill_device_info = virtio_mem_fill_device_info; vmc->get_memory_region = virtio_mem_get_memory_region; + vmc->decide_memslots = virtio_mem_decide_memslots; + vmc->get_memslots = virtio_mem_get_memslots; vmc->add_size_change_notifier = virtio_mem_add_size_change_notifier; vmc->remove_size_change_notifier = virtio_mem_remove_size_change_notifier; vmc->unplug_request_check = virtio_mem_unplug_request_check; @@ -1573,6 +1882,7 @@ static const TypeInfo virtio_mem_info = { .parent = TYPE_VIRTIO_DEVICE, .instance_size = sizeof(VirtIOMEM), .instance_init = virtio_mem_instance_init, + .instance_finalize = virtio_mem_instance_finalize, .class_init = virtio_mem_class_init, .class_size = sizeof(VirtIOMEMClass), .interfaces = (InterfaceInfo[]) { |