Merge tag 'pull-hv-balloon-20231106' of https://github.com/maciejsszm…

…igiero/qemu into staging Hyper-V Dynamic Memory protocol driver. This driver is like virtio-balloon on steroids for Windows guests: it allows both changing the guest memory allocation via ballooning and inserting pieces of extra RAM into it on demand from a provided memory backend via Windows-native Hyper-V Dynamic Memory protocol. * Preparatory patches to support empty memory devices and ones with large alignment requirements. * Revert of recently added "hw/virtio/virtio-pmem: Replace impossible check by assertion" commit 5960f25 since this series makes this situation possible again. * Protocol definitions. * Hyper-V DM protocol driver (hv-balloon) base (ballooning only). * Hyper-V DM protocol driver (hv-balloon) hot-add support. * qapi query-memory-devices support for the driver. * qapi HV_BALLOON_STATUS_REPORT event. * The relevant PC machine plumbing. * New MAINTAINERS entry for the above. # -----BEGIN PGP SIGNATURE----- # # iQGzBAABCAAdFiEE4ndqq6COJv9aG0oJUrHW6VHQzgcFAmVI81IACgkQUrHW6VHQ # zgdzTgv+I5eV2R01YLOBBJhBjzxZ4/BUqkuUHNxHpfjuCqEIzPb7FIfoZ4ZyXZFT # YJdSE4lPeTZLrmmi/Nt6G0rUKDvdCeIgkS2VLHFSsTV8IzcT71BTRGzV0zAjUF5v # yDH6uzo6e9gmaziIalRjibUxSDjCQmoCifms2rS2DwazADudUp+naGfm+3uyA0gM # raOfBfRkNZsDqhXg2ayuqPIES75xQONoON9xYPKDAthS48POEbqtWBKuFopr3kXY # y0eph+NAw+RajCyLYKM3poIgaSu3l4WegInuKQffzqKR8dxrbwPdCmtgo6NSHx0W # uDfl7FUBnGzrR18VU4ZfTSrF5SVscGwF9EL7uocJen15inJjl1q3G53uZgyGzHLC # cw8fKMjucmE8njQR2qiMyX0b+T4+9nKO1rykBgTG/+c9prRUVoxYpFCF117Ei0U8 # QzLGACW1oK+LV41bekWAye7w9pShUtFaxffhPbJeZDDGh7q0x61R3Z3yKkA07p46 # /YWWFWUD # =RAb0 # -----END PGP SIGNATURE----- # gpg: Signature made Mon 06 Nov 2023 22:08:18 HKT # gpg: using RSA key E2776AABA08E26FF5A1B4A0952B1D6E951D0CE07 # gpg: Good signature from "Maciej S. Szmigiero <mail@maciej.szmigiero.name>" [unknown] # gpg: WARNING: This key is not certified with a trusted signature! # gpg: There is no indication that the signature belongs to the owner. # Primary key fingerprint: 727A 0D4D DB9E D9F6 039B ECEF 847F 5E37 90CE 0977 # Subkey fingerprint: E277 6AAB A08E 26FF 5A1B 4A09 52B1 D6E9 51D0 CE07 * tag 'pull-hv-balloon-20231106' of https://github.com/maciejsszmigiero/qemu: MAINTAINERS: Add an entry for Hyper-V Dynamic Memory Protocol hw/i386/pc: Support hv-balloon qapi: Add HV_BALLOON_STATUS_REPORT event and its QMP query command qapi: Add query-memory-devices support to hv-balloon Add Hyper-V Dynamic Memory Protocol driver (hv-balloon) hot-add support Add Hyper-V Dynamic Memory Protocol driver (hv-balloon) base Add Hyper-V Dynamic Memory Protocol definitions memory-device: Drop size alignment check Revert "hw/virtio/virtio-pmem: Replace impossible check by assertion" memory-device: Support empty memory devices Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
qemu · Nov 7, 2023 · 17735e9 · 17735e9
2 parents 9f33cf2 + 00313b5
commit 17735e9
Show file tree

Hide file tree

Showing 26 changed files with 3,180 additions and 14 deletions.
diff --git a/Kconfig.host b/Kconfig.host
@@ -46,3 +46,6 @@ config FUZZ
 config VFIO_USER_SERVER_ALLOWED
     bool
     imply VFIO_USER_SERVER
+
+config HV_BALLOON_POSSIBLE
+    bool
diff --git a/MAINTAINERS b/MAINTAINERS
@@ -2656,6 +2656,14 @@ F: hw/usb/canokey.c
 F: hw/usb/canokey.h
 F: docs/system/devices/canokey.rst
 
+Hyper-V Dynamic Memory Protocol
+M: Maciej S. Szmigiero <maciej.szmigiero@oracle.com>
+S: Supported
+F: hw/hyperv/hv-balloon*.c
+F: hw/hyperv/hv-balloon*.h
+F: include/hw/hyperv/dynmem-proto.h
+F: include/hw/hyperv/hv-balloon.h
+
 Subsystems
 ----------
 Overall Audio backends

diff --git a/hw/core/machine-hmp-cmds.c b/hw/core/machine-hmp-cmds.c
@@ -253,6 +253,7 @@ void hmp_info_memory_devices(Monitor *mon, const QDict *qdict)
     MemoryDeviceInfo *value;
     PCDIMMDeviceInfo *di;
     SgxEPCDeviceInfo *se;
+    HvBalloonDeviceInfo *hi;
 
     for (info = info_list; info; info = info->next) {
         value = info->value;
@@ -310,6 +311,20 @@ void hmp_info_memory_devices(Monitor *mon, const QDict *qdict)
                 monitor_printf(mon, "  node: %" PRId64 "\n", se->node);
                 monitor_printf(mon, "  memdev: %s\n", se->memdev);
                 break;
+            case MEMORY_DEVICE_INFO_KIND_HV_BALLOON:
+                hi = value->u.hv_balloon.data;
+                monitor_printf(mon, "Memory device [%s]: \"%s\"\n",
+                               MemoryDeviceInfoKind_str(value->type),
+                               hi->id ? hi->id : "");
+                if (hi->has_memaddr) {
+                    monitor_printf(mon, "  memaddr: 0x%" PRIx64 "\n",
+                                   hi->memaddr);
+                }
+                monitor_printf(mon, "  max-size: %" PRIu64 "\n", hi->max_size);
+                if (hi->memdev) {
+                    monitor_printf(mon, "  memdev: %s\n", hi->memdev);
+                }
+                break;
             default:
                 g_assert_not_reached();
             }

diff --git a/hw/hyperv/Kconfig b/hw/hyperv/Kconfig
@@ -16,3 +16,13 @@ config SYNDBG
     bool
     default y
     depends on VMBUS
+
+config HV_BALLOON_SUPPORTED
+    bool
+
+config HV_BALLOON
+    bool
+    default y
+    depends on VMBUS
+    depends on HV_BALLOON_POSSIBLE
+    depends on HV_BALLOON_SUPPORTED
diff --git a/hw/hyperv/hv-balloon-internal.h b/hw/hyperv/hv-balloon-internal.h
@@ -0,0 +1,33 @@
+/*
+ * 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_INTERNAL_H
+#define HW_HYPERV_HV_BALLOON_INTERNAL_H
+
+#include "qemu/osdep.h"
+
+#define HV_BALLOON_PFN_SHIFT 12
+#define HV_BALLOON_PAGE_SIZE (1 << HV_BALLOON_PFN_SHIFT)
+
+#define SUM_OVERFLOW_U64(in1, in2) ((in1) > UINT64_MAX - (in2))
+#define SUM_SATURATE_U64(in1, in2)              \
+    ({                                          \
+        uint64_t _in1 = (in1), _in2 = (in2);    \
+        uint64_t _result;                       \
+                                                \
+        if (!SUM_OVERFLOW_U64(_in1, _in2)) {    \
+            _result = _in1 + _in2;              \
+        } else {                                \
+            _result = UINT64_MAX;               \
+        }                                       \
+                                                \
+        _result;                                \
+    })
+
+#endif
diff --git a/hw/hyperv/hv-balloon-our_range_memslots.c b/hw/hyperv/hv-balloon-our_range_memslots.c
@@ -0,0 +1,201 @@
+/*
+ * 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();
+}