qemu/oslib-posix-optimise-vm-startup-time-for-1G-hugepage.patch

58 lines
2.0 KiB
Diff
Raw Permalink Normal View History

QEMU update to version 8.2.0-5 - vfio/migration: Add support for manual clear vfio dirty log - vfio: Maintain DMA mapping range for the container - linux-headers: update against 5.10 and manual clear vfio dirty log series - arm/acpi: Fix when make qemu-system-aarch64 at x86_64 host bios_tables_test fail reason: __aarch64__ macro let build_pptt at x86_64 and aarch64 host build different function that let bios_tables_test fail. - pl031: support rtc-timer property for pl031 - feature: Add logs for vm start and destroy - feature: Add log for each modules - log: Add log at boot & cpu init for aarch64 - bugfix: irq: Avoid covering object refcount of qemu_irq - i386: cache passthrough: Update AMD 8000_001D.EAX[25:14] based on vCPU topo - freeclock: set rtc_date_diff for X86 - freeclock: set rtc_date_diff for arm - freeclock: add qmp command to get time offset of vm in seconds - tests: Disable filemonitor testcase - shadow_dev: introduce shadow dev for virtio-net device - pl011: reset read FIFO when UARTTIMSC=0 & UARTICR=0xffff - tests: virt: Update expected ACPI tables for virt test(update BinDir) - arm64: Add the cpufreq device to show cpufreq info to guest - hw/arm64: add vcpu cache info support - tests: virt: Allow changes to PPTT test table - cpu: add Cortex-A72 processor kvm target support - cpu: add Kunpeng-920 cpu support - net: eepro100: validate various address valuesi(CVE-2021-20255) - ide: ahci: add check to avoid null dereference (CVE-2019-12067) - vdpa: set vring enable only if the vring address has already been set - docs: Add generic vhost-vdpa device documentation - vdpa: don't suspend/resume device when vdpa device not started - vdpa: correct param passed in when unregister save - vdpa: suspend function return 0 when the vdpa device is stopped - vdpa: support vdpa device suspend/resume - vdpa: move memory listener to the realize stage - vdpa: implement vdpa device migration - vhost: implement migration state notifier for vdpa device - vhost: implement post resume bh - vhost: implement savevm_handler for vdpa device - vhost: implement vhost_vdpa_device_suspend/resume - vhost: implement vhost-vdpa suspend/resume - vhost: add vhost_dev_suspend/resume_op - vhost: introduce bytemap for vhost backend logging - vhost-vdpa: add migration log ops for VhostOps - vhost-vdpa: add VHOST_BACKEND_F_BYTEMAPLOG - hw/usb: reduce the vpcu cost of UHCI when VNC disconnect - virtio-net: update the default and max of rx/tx_queue_size - virtio-net: set the max of queue size to 4096 - virtio-net: fix max vring buf size when set ring num - virtio-net: bugfix: do not delete netdev before virtio net - monitor: Discard BLOCK_IO_ERROR event when VM rebooted - vhost-user: add unregister_savevm when vhost-user cleanup - vhost-user: add vhost_set_mem_table when vm load_setup at destination - vhost-user: quit infinite loop while used memslots is more than the backend limit - fix qemu-core when vhost-user-net config with server mode - vhost-user: Add support reconnect vhost-user socket - vhost-user: Set the acked_features to vm's featrue - i6300esb watchdog: bugfix: Add a runstate transition - hw/net/rocker_of_dpa: fix double free bug of rocker device - net/dump.c: Suppress spurious compiler warning - pcie: Add pcie-root-port fast plug/unplug feature - pcie: Compat with devices which do not support Link Width, such as ioh3420 - qdev/monitors: Fix reundant error_setg of qdev_add_device - qemu-nbd: set timeout to qemu-nbd socket - qemu-nbd: make native as the default aio mode - nbd/server.c: fix invalid read after client was already free - virtio-scsi: bugfix: fix qemu crash for hotplug scsi disk with dataplane - virtio: bugfix: check the value of caches before accessing it - virtio: print the guest virtio_net features that host does not support - virtio: bugfix: add rcu_read_lock when vring_avail_idx is called - virtio: check descriptor numbers - migration: report multiFd related thread pid to libvirt - migration: report migration related thread pid to libvirt - cpu/features: fix bug for memory leakage - doc: Update multi-thread compression doc - migration: Add compress_level sanity check - migration: Add zstd support in multi-thread compression - migration: Add multi-thread compress ops - migration: Refactoring multi-thread compress migration - migration: Add multi-thread compress method - migration: skip cache_drop for bios bootloader and nvram template - oslib-posix: optimise vm startup time for 1G hugepage - monitor/qmp: drop inflight rsp if qmp client broken - ps2: fix oob in ps2 kbd - Currently, while kvm and qemu can not handle some kvm exit, qemu will do vm_stop, which will make vm in pause state. This action make vm unrecoverable, so send guest panic to libvirt instead. - vhost: cancel migration when vhost-user restarted during migraiton Signed-off-by: Jiabo Feng <fengjiabo1@huawei.com>
2024-04-07 10:21:31 +08:00
From b6c45f5ea5d1a379ac0a507cf59345c573b27cc8 Mon Sep 17 00:00:00 2001
From: Chuan Zheng <zhengchuan@huawei.com>
Date: Wed, 9 Feb 2022 14:21:39 +0800
Subject: [PATCH] oslib-posix: optimise vm startup time for 1G hugepage
It takes quit a long time to clear 1G-hugepage, which makes glibc
pthread_create quit slow.
Create touch_pages threads in advance, and then handle the touch_pages
callback. Only read lock is held here.
---
util/oslib-posix.c | 9 ++++++++-
1 file changed, 8 insertions(+), 1 deletion(-)
diff --git a/util/oslib-posix.c b/util/oslib-posix.c
index e86fd64e09..9ca3fee2b8 100644
--- a/util/oslib-posix.c
+++ b/util/oslib-posix.c
@@ -88,6 +88,8 @@ static QemuMutex sigbus_mutex;
static QemuMutex page_mutex;
static QemuCond page_cond;
+static int started_num_threads;
+
int qemu_get_thread_id(void)
{
#if defined(__linux__)
@@ -344,6 +346,10 @@ static void *do_touch_pages(void *arg)
}
qemu_mutex_unlock(&page_mutex);
+ while (started_num_threads != memset_args->context.num_threads) {
+ smp_mb();
+ }
+
/* unblock SIGBUS */
sigemptyset(&set);
sigaddset(&set, SIGBUS);
@@ -448,7 +454,7 @@ static int touch_all_pages(char *area, size_t hpagesize, size_t numpages,
context.threads = g_new0(MemsetThread, context.num_threads);
numpages_per_thread = numpages / context.num_threads;
leftover = numpages % context.num_threads;
- for (i = 0; i < context.num_threads; i++) {
+ for (i = 0, started_num_threads = 0; i < context.num_threads; i++) {
context.threads[i].addr = addr;
context.threads[i].numpages = numpages_per_thread + (i < leftover);
context.threads[i].hpagesize = hpagesize;
@@ -464,6 +470,7 @@ static int touch_all_pages(char *area, size_t hpagesize, size_t numpages,
QEMU_THREAD_JOINABLE);
}
addr += context.threads[i].numpages * hpagesize;
+ started_num_threads++;
}
if (!use_madv_populate_write) {
--
2.27.0