Add Hyper-V Dynamic Memory Protocol driver (hv-balloon) hot-add support

One of advantages of using this protocol over ACPI-based PC DIMM hotplug is
that it allows hot-adding memory in much smaller granularity because the
ACPI DIMM slot limit does not apply.

In order to enable this functionality a new memory backend needs to be
created and provided to the driver via the "memdev" parameter.

This can be achieved by, for example, adding
"-object memory-backend-ram,id=mem1,size=32G" to the QEMU command line and
then instantiating the driver with "memdev=mem1" parameter.

The device will try to use multiple memslots to cover the memory backend in
order to reduce the size of metadata for the not-yet-hot-added part of the
memory backend.

Co-developed-by: David Hildenbrand <david@redhat.com>
Acked-by: David Hildenbrand <david@redhat.com>
Signed-off-by: Maciej S. Szmigiero <maciej.szmigiero@oracle.com>

hw/hyperv/hv-balloon-our_range_memslots.c

+/*
+ * QEMU Hyper-V Dynamic Memory Protocol driver
+ *
+ * Copyright (C) 2020-2023 Oracle and/or its affiliates.
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2 or later.
+ * See the COPYING file in the top-level directory.
+ */
+
+#include "hv-balloon-internal.h"
+#include "hv-balloon-our_range_memslots.h"
+#include "trace.h"
+
+/* OurRange */
+static void our_range_init(OurRange *our_range, uint64_t start, uint64_t count)
+{
+    assert(count <= UINT64_MAX - start);
+    our_range->range.start = start;
+    our_range->range.count = count;
+
+    hvb_page_range_tree_init(&our_range->removed_guest);
+    hvb_page_range_tree_init(&our_range->removed_both);
+
+    /* mark the whole range as unused but for potential use */
+    our_range->added = 0;
+    our_range->unusable_tail = 0;
+}
+
+static void our_range_destroy(OurRange *our_range)
+{
+    hvb_page_range_tree_destroy(&our_range->removed_guest);
+    hvb_page_range_tree_destroy(&our_range->removed_both);
+}
+
+void hvb_our_range_clear_removed_trees(OurRange *our_range)
+{
+    hvb_page_range_tree_destroy(&our_range->removed_guest);
+    hvb_page_range_tree_destroy(&our_range->removed_both);
+    hvb_page_range_tree_init(&our_range->removed_guest);
+    hvb_page_range_tree_init(&our_range->removed_both);
+}
+
+void hvb_our_range_mark_added(OurRange *our_range, uint64_t additional_size)
+{
+    assert(additional_size <= UINT64_MAX - our_range->added);
+
+    our_range->added += additional_size;
+
+    assert(our_range->added <= UINT64_MAX - our_range->unusable_tail);
+    assert(our_range->added + our_range->unusable_tail <=
+           our_range->range.count);
+}
+
+/* OurRangeMemslots */
+static void our_range_memslots_init_slots(OurRangeMemslots *our_range,
+                                          MemoryRegion *backing_mr,
+                                          Object *memslot_owner)
+{
+    OurRangeMemslotsSlots *memslots = &our_range->slots;
+    unsigned int idx;
+    uint64_t memslot_offset;
+
+    assert(memslots->count > 0);
+    memslots->slots = g_new0(MemoryRegion, memslots->count);
+
+    /* Initialize our memslots, but don't map them yet. */
+    assert(memslots->size_each > 0);
+    for (idx = 0, memslot_offset = 0; idx < memslots->count;
+         idx++, memslot_offset += memslots->size_each) {
+        uint64_t memslot_size;
+        g_autofree char *name = NULL;
+
+        /* The size of the last memslot might be smaller. */
+        if (idx == memslots->count - 1) {
+            uint64_t region_size;
+
+            assert(our_range->mr);
+            region_size = memory_region_size(our_range->mr);
+            memslot_size = region_size - memslot_offset;
+        } else {
+            memslot_size = memslots->size_each;
+        }
+
+        name = g_strdup_printf("memslot-%u", idx);
+        memory_region_init_alias(&memslots->slots[idx], memslot_owner, name,
+                                 backing_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(&memslots->slots[idx], true);
+    }
+
+    memslots->mapped_count = 0;
+}
+
+OurRangeMemslots *hvb_our_range_memslots_new(uint64_t addr,
+                                             MemoryRegion *parent_mr,
+                                             MemoryRegion *backing_mr,
+                                             Object *memslot_owner,
+                                             unsigned int memslot_count,
+                                             uint64_t memslot_size)
+{
+    OurRangeMemslots *our_range;
+
+    our_range = g_malloc(sizeof(*our_range));
+    our_range_init(&our_range->range,
+                   addr / HV_BALLOON_PAGE_SIZE,
+                   memory_region_size(parent_mr) / HV_BALLOON_PAGE_SIZE);
+    our_range->slots.size_each = memslot_size;
+    our_range->slots.count = memslot_count;
+    our_range->mr = parent_mr;
+    our_range_memslots_init_slots(our_range, backing_mr, memslot_owner);
+
+    return our_range;
+}
+
+static void our_range_memslots_free_memslots(OurRangeMemslots *our_range)
+{
+    OurRangeMemslotsSlots *memslots = &our_range->slots;
+    unsigned int idx;
+    uint64_t offset;
+
+    memory_region_transaction_begin();
+    for (idx = 0, offset = 0; idx < memslots->mapped_count;
+         idx++, offset += memslots->size_each) {
+        trace_hv_balloon_unmap_slot(idx, memslots->count, offset);
+        assert(memory_region_is_mapped(&memslots->slots[idx]));
+        memory_region_del_subregion(our_range->mr, &memslots->slots[idx]);
+    }
+    memory_region_transaction_commit();
+
+    for (idx = 0; idx < memslots->count; idx++) {
+        object_unparent(OBJECT(&memslots->slots[idx]));
+    }
+
+    g_clear_pointer(&our_range->slots.slots, g_free);
+}
+
+void hvb_our_range_memslots_free(OurRangeMemslots *our_range)
+{
+    OurRangeMemslotsSlots *memslots = &our_range->slots;
+    MemoryRegion *hostmem_mr;
+    RAMBlock *rb;
+
+    assert(our_range->slots.count > 0);
+    assert(our_range->slots.slots);
+
+    hostmem_mr = memslots->slots[0].alias;
+    rb = hostmem_mr->ram_block;
+    ram_block_discard_range(rb, 0, qemu_ram_get_used_length(rb));
+
+    our_range_memslots_free_memslots(our_range);
+    our_range_destroy(&our_range->range);
+    g_free(our_range);
+}
+
+void hvb_our_range_memslots_ensure_mapped_additional(OurRangeMemslots *our_range,
+                                                     uint64_t additional_map_size)
+{
+    OurRangeMemslotsSlots *memslots = &our_range->slots;
+    uint64_t total_map_size;
+    unsigned int idx;
+    uint64_t offset;
+
+    total_map_size = (our_range->range.added + additional_map_size) *
+        HV_BALLOON_PAGE_SIZE;
+    idx = memslots->mapped_count;
+    assert(memslots->size_each > 0);
+    offset = idx * memslots->size_each;
+
+    /*
+     * Activate all memslots covered by the newly added region in a single
+     * transaction.
+     */
+    memory_region_transaction_begin();
+    for ( ; idx < memslots->count;
+          idx++, offset += memslots->size_each) {
+        /*
+         * If this memslot starts beyond or at the end of the range to map so
+         * does every next one.
+         */
+        if (offset >= total_map_size) {
+            break;
+        }
+
+        /*
+         * Instead of enabling/disabling memslot, we add/remove them. This
+         * should make address space updates faster, because we don't have to
+         * loop over many disabled subregions.
+         */
+        trace_hv_balloon_map_slot(idx, memslots->count, offset);
+        assert(!memory_region_is_mapped(&memslots->slots[idx]));
+        memory_region_add_subregion(our_range->mr, offset,
+                                    &memslots->slots[idx]);
+
+        memslots->mapped_count++;
+    }
+    memory_region_transaction_commit();
+}

hw/hyperv/hv-balloon-our_range_memslots.h

+/*
+ * QEMU Hyper-V Dynamic Memory Protocol driver
+ *
+ * Copyright (C) 2020-2023 Oracle and/or its affiliates.
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2 or later.
+ * See the COPYING file in the top-level directory.
+ */
+
+#ifndef HW_HYPERV_HV_BALLOON_OUR_RANGE_MEMSLOTS_H
+#define HW_HYPERV_HV_BALLOON_OUR_RANGE_MEMSLOTS_H
+
+#include "qemu/osdep.h"
+
+#include "exec/memory.h"
+#include "qom/object.h"
+#include "hv-balloon-page_range_tree.h"
+
+/* OurRange */
+#define OUR_RANGE(ptr) ((OurRange *)(ptr))
+
+/* "our range" means the memory range owned by this driver (for hot-adding) */
+typedef struct OurRange {
+    PageRange range;
+
+    /* How many pages were hot-added to the guest */
+    uint64_t added;
+
+    /* Pages at the end not currently usable */
+    uint64_t unusable_tail;
+
+    /* Memory removed from the guest */
+    PageRangeTree removed_guest, removed_both;
+} OurRange;
+
+static inline uint64_t our_range_get_remaining_start(OurRange *our_range)
+{
+    return our_range->range.start + our_range->added;
+}
+
+static inline uint64_t our_range_get_remaining_size(OurRange *our_range)
+{
+    return our_range->range.count - our_range->added - our_range->unusable_tail;
+}
+
+void hvb_our_range_mark_added(OurRange *our_range, uint64_t additional_size);
+
+static inline void our_range_mark_remaining_unusable(OurRange *our_range)
+{
+    our_range->unusable_tail = our_range->range.count - our_range->added;
+}
+
+static inline PageRangeTree our_range_get_removed_tree(OurRange *our_range,
+                                                       bool both)
+{
+    if (both) {
+        return our_range->removed_both;
+    } else {
+        return our_range->removed_guest;
+    }
+}
+
+static inline bool our_range_is_removed_tree_empty(OurRange *our_range,
+                                                   bool both)
+{
+    if (both) {
+        return page_range_tree_is_empty(our_range->removed_both);
+    } else {
+        return page_range_tree_is_empty(our_range->removed_guest);
+    }
+}
+
+void hvb_our_range_clear_removed_trees(OurRange *our_range);
+
+/* OurRangeMemslots */
+typedef struct OurRangeMemslotsSlots {
+    /* Nominal size of each memslot (the last one might be smaller) */
+    uint64_t size_each;
+
+    /* Slots array and its element count */
+    MemoryRegion *slots;
+    unsigned int count;
+
+    /* How many slots are currently mapped */
+    unsigned int mapped_count;
+} OurRangeMemslotsSlots;
+
+typedef struct OurRangeMemslots {
+    OurRange range;
+
+    /* Memslots covering our range */
+    OurRangeMemslotsSlots slots;
+
+    MemoryRegion *mr;
+} OurRangeMemslots;
+
+OurRangeMemslots *hvb_our_range_memslots_new(uint64_t addr,
+                                             MemoryRegion *parent_mr,
+                                             MemoryRegion *backing_mr,
+                                             Object *memslot_owner,
+                                             unsigned int memslot_count,
+                                             uint64_t memslot_size);
+void hvb_our_range_memslots_free(OurRangeMemslots *our_range);
+
+G_DEFINE_AUTOPTR_CLEANUP_FUNC(OurRangeMemslots, hvb_our_range_memslots_free)
+
+void hvb_our_range_memslots_ensure_mapped_additional(OurRangeMemslots *our_range,
+                                                     uint64_t additional_map_size);
+
+#endif

hw/hyperv/meson.build

@@ -2,4 +2,4 @@ specific_ss.add(when: 'CONFIG_HYPERV', if_true: files('hyperv.c'))
 specific_ss.add(when: 'CONFIG_HYPERV_TESTDEV', if_true: files('hyperv_testdev.c'))
 specific_ss.add(when: 'CONFIG_VMBUS', if_true: files('vmbus.c'))
 specific_ss.add(when: 'CONFIG_SYNDBG', if_true: files('syndbg.c'))
-specific_ss.add(when: 'CONFIG_HV_BALLOON', if_true: files('hv-balloon.c', 'hv-balloon-page_range_tree.c'))
+specific_ss.add(when: 'CONFIG_HV_BALLOON', if_true: files('hv-balloon.c', 'hv-balloon-page_range_tree.c', 'hv-balloon-our_range_memslots.c'))

hw/hyperv/trace-events

@@ -23,9 +23,14 @@ hv_balloon_incoming_version(uint16_t major, uint16_t minor) "incoming proto vers
 hv_balloon_incoming_caps(uint32_t caps) "incoming caps 0x%x"
 hv_balloon_outgoing_unballoon(uint32_t trans_id, uint64_t count, uint64_t start, uint64_t rempages) "posting unballoon %"PRIu32" for %"PRIu64" @ 0x%"PRIx64", remaining %"PRIu64
 hv_balloon_incoming_unballoon(uint32_t trans_id) "incoming unballoon response %"PRIu32
+hv_balloon_outgoing_hot_add(uint32_t trans_id, uint64_t count, uint64_t start) "posting hot add %"PRIu32" for %"PRIu64" @ 0x%"PRIx64
+hv_balloon_incoming_hot_add(uint32_t trans_id, uint32_t result, uint32_t count) "incoming hot add response %"PRIu32", result %"PRIu32", count %"PRIu32
 hv_balloon_outgoing_balloon(uint32_t trans_id, uint64_t count, uint64_t rempages) "posting balloon %"PRIu32" for %"PRIu64", remaining %"PRIu64
 hv_balloon_incoming_balloon(uint32_t trans_id, uint32_t range_count, uint32_t more_pages) "incoming balloon response %"PRIu32", ranges %"PRIu32", more %"PRIu32
+hv_balloon_our_range_add(uint64_t count, uint64_t start) "adding our range %"PRIu64" @ 0x%"PRIx64
 hv_balloon_remove_response(uint64_t count, uint64_t start, unsigned int both) "processing remove response range %"PRIu64" @ 0x%"PRIx64", both %u"
 hv_balloon_remove_response_hole(uint64_t counthole, uint64_t starthole, uint64_t countrange, uint64_t startrange, uint64_t starthpr, unsigned int both) "response range hole %"PRIu64" @ 0x%"PRIx64" from range %"PRIu64" @ 0x%"PRIx64", before our start 0x%"PRIx64", both %u"
 hv_balloon_remove_response_common(uint64_t countcommon, uint64_t startcommon, uint64_t countrange, uint64_t startrange, uint64_t counthpr, uint64_t starthpr, uint64_t removed, unsigned int both) "response common range %"PRIu64" @ 0x%"PRIx64" from range %"PRIu64" @ 0x%"PRIx64" with our %"PRIu64" @ 0x%"PRIx64", removed %"PRIu64", both %u"
 hv_balloon_remove_response_remainder(uint64_t count, uint64_t start, unsigned int both) "remove response remaining range %"PRIu64" @ 0x%"PRIx64", both %u"
+hv_balloon_map_slot(unsigned int idx, unsigned int total_slots, uint64_t offset) "mapping memslot %u / %u @ 0x%"PRIx64
+hv_balloon_unmap_slot(unsigned int idx, unsigned int total_slots, uint64_t offset) "unmapping memslot %u / %u @ 0x%"PRIx64