From: "Michael S. Tsirkin" <mst@redhat.com>
To: qemu-devel@nongnu.org
Cc: Peter Maydell <peter.maydell@linaro.org>,
Jason Wang <jasowang@redhat.com>,
Eduardo Habkost <ehabkost@redhat.com>
Subject: [Qemu-devel] [PULL 28/28] vhost: logs sharing
Date: Thu, 4 Jun 2015 13:11:41 +0200 [thread overview]
Message-ID: <1433416111-19022-29-git-send-email-mst@redhat.com> (raw)
In-Reply-To: <1433416111-19022-1-git-send-email-mst@redhat.com>
From: Jason Wang <jasowang@redhat.com>
Currently we allocate one vhost log per vhost device. This is sub
optimal when:
- Guest has several device with vhost as backend
- Guest has multiqueue devices
In the above cases, we can avoid the memory allocation by sharing a
single vhost log among all the vhost devices. This is done through:
- Introducing a new vhost_log structure with refcnt inside.
- Using a global pointer to vhost_log structure that will be used. And
introduce helper to get the log with expected log size and helper to
- drop the refcnt to the old log.
- Each vhost device still keep track of a pointer to the log that was
used.
With above, if no resize happens, all vhost device will share a single
vhost log. During resize, a new vhost_log structure will be allocated
and made for the global pointer. And each vhost devices will drop the
refcnt to the old log.
Tested by doing scp during migration for a 2 queues virtio-net-pci.
Cc: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
---
include/hw/virtio/vhost.h | 8 ++++-
hw/virtio/vhost.c | 77 ++++++++++++++++++++++++++++++++++++-----------
2 files changed, 66 insertions(+), 19 deletions(-)
diff --git a/include/hw/virtio/vhost.h b/include/hw/virtio/vhost.h
index 8f04888..816a2e8 100644
--- a/include/hw/virtio/vhost.h
+++ b/include/hw/virtio/vhost.h
@@ -28,6 +28,12 @@ typedef unsigned long vhost_log_chunk_t;
#define VHOST_LOG_CHUNK (VHOST_LOG_PAGE * VHOST_LOG_BITS)
#define VHOST_INVALID_FEATURE_BIT (0xff)
+struct vhost_log {
+ unsigned long long size;
+ int refcnt;
+ vhost_log_chunk_t log[0];
+};
+
struct vhost_memory;
struct vhost_dev {
MemoryListener memory_listener;
@@ -43,7 +49,6 @@ struct vhost_dev {
unsigned long long backend_features;
bool started;
bool log_enabled;
- vhost_log_chunk_t *log;
unsigned long long log_size;
Error *migration_blocker;
bool force;
@@ -52,6 +57,7 @@ struct vhost_dev {
hwaddr mem_changed_end_addr;
const VhostOps *vhost_ops;
void *opaque;
+ struct vhost_log *log;
};
int vhost_dev_init(struct vhost_dev *hdev, void *opaque,
diff --git a/hw/virtio/vhost.c b/hw/virtio/vhost.c
index 54851b7..01f1e04 100644
--- a/hw/virtio/vhost.c
+++ b/hw/virtio/vhost.c
@@ -22,15 +22,19 @@
#include "hw/virtio/virtio-bus.h"
#include "migration/migration.h"
+static struct vhost_log *vhost_log;
+
static void vhost_dev_sync_region(struct vhost_dev *dev,
MemoryRegionSection *section,
uint64_t mfirst, uint64_t mlast,
uint64_t rfirst, uint64_t rlast)
{
+ vhost_log_chunk_t *log = dev->log->log;
+
uint64_t start = MAX(mfirst, rfirst);
uint64_t end = MIN(mlast, rlast);
- vhost_log_chunk_t *from = dev->log + start / VHOST_LOG_CHUNK;
- vhost_log_chunk_t *to = dev->log + end / VHOST_LOG_CHUNK + 1;
+ vhost_log_chunk_t *from = log + start / VHOST_LOG_CHUNK;
+ vhost_log_chunk_t *to = log + end / VHOST_LOG_CHUNK + 1;
uint64_t addr = (start / VHOST_LOG_CHUNK) * VHOST_LOG_CHUNK;
if (end < start) {
@@ -280,22 +284,57 @@ static uint64_t vhost_get_log_size(struct vhost_dev *dev)
}
return log_size;
}
+static struct vhost_log *vhost_log_alloc(uint64_t size)
+{
+ struct vhost_log *log = g_malloc0(sizeof *log + size * sizeof(*(log->log)));
+
+ log->size = size;
+ log->refcnt = 1;
+
+ return log;
+}
+
+static struct vhost_log *vhost_log_get(uint64_t size)
+{
+ if (!vhost_log || vhost_log->size != size) {
+ vhost_log = vhost_log_alloc(size);
+ } else {
+ ++vhost_log->refcnt;
+ }
+
+ return vhost_log;
+}
+
+static void vhost_log_put(struct vhost_dev *dev, bool sync)
+{
+ struct vhost_log *log = dev->log;
+
+ if (!log) {
+ return;
+ }
+
+ --log->refcnt;
+ if (log->refcnt == 0) {
+ /* Sync only the range covered by the old log */
+ if (dev->log_size && sync) {
+ vhost_log_sync_range(dev, 0, dev->log_size * VHOST_LOG_CHUNK - 1);
+ }
+ if (vhost_log == log) {
+ vhost_log = NULL;
+ }
+ g_free(log);
+ }
+}
static inline void vhost_dev_log_resize(struct vhost_dev* dev, uint64_t size)
{
- vhost_log_chunk_t *log;
- uint64_t log_base;
+ struct vhost_log *log = vhost_log_get(size);
+ uint64_t log_base = (uintptr_t)log->log;
int r;
- log = g_malloc0(size * sizeof *log);
- log_base = (uintptr_t)log;
r = dev->vhost_ops->vhost_call(dev, VHOST_SET_LOG_BASE, &log_base);
assert(r >= 0);
- /* Sync only the range covered by the old log */
- if (dev->log_size) {
- vhost_log_sync_range(dev, 0, dev->log_size * VHOST_LOG_CHUNK - 1);
- }
- g_free(dev->log);
+ vhost_log_put(dev, true);
dev->log = log;
dev->log_size = size;
}
@@ -601,7 +640,7 @@ static int vhost_migration_log(MemoryListener *listener, int enable)
if (r < 0) {
return r;
}
- g_free(dev->log);
+ vhost_log_put(dev, false);
dev->log = NULL;
dev->log_size = 0;
} else {
@@ -1060,10 +1099,10 @@ int vhost_dev_start(struct vhost_dev *hdev, VirtIODevice *vdev)
uint64_t log_base;
hdev->log_size = vhost_get_log_size(hdev);
- hdev->log = hdev->log_size ?
- g_malloc0(hdev->log_size * sizeof *hdev->log) : NULL;
- log_base = (uintptr_t)hdev->log;
- r = hdev->vhost_ops->vhost_call(hdev, VHOST_SET_LOG_BASE, &log_base);
+ hdev->log = vhost_log_get(hdev->log_size);
+ log_base = (uintptr_t)hdev->log->log;
+ r = hdev->vhost_ops->vhost_call(hdev, VHOST_SET_LOG_BASE,
+ hdev->log_size ? &log_base : NULL);
if (r < 0) {
r = -errno;
goto fail_log;
@@ -1072,6 +1111,9 @@ int vhost_dev_start(struct vhost_dev *hdev, VirtIODevice *vdev)
return 0;
fail_log:
+ if (hdev->log_size) {
+ vhost_log_put(hdev, false);
+ }
fail_vq:
while (--i >= 0) {
vhost_virtqueue_stop(hdev,
@@ -1098,10 +1140,9 @@ void vhost_dev_stop(struct vhost_dev *hdev, VirtIODevice *vdev)
hdev->vqs + i,
hdev->vq_index + i);
}
- vhost_log_sync_range(hdev, 0, ~0x0ull);
+ vhost_log_put(hdev, true);
hdev->started = false;
- g_free(hdev->log);
hdev->log = NULL;
hdev->log_size = 0;
}
--
MST
next prev parent reply other threads:[~2015-06-04 11:11 UTC|newest]
Thread overview: 30+ messages / expand[flat|nested] mbox.gz Atom feed top
2015-06-04 11:10 [Qemu-devel] [PULL 00/28] pc, acpi, virtio, tpm Michael S. Tsirkin
2015-06-04 11:10 ` [Qemu-devel] [PULL 01/28] acpi: add missing ssdt Michael S. Tsirkin
2015-06-04 11:10 ` [Qemu-devel] [PULL 02/28] hw/q35: fix floppy controller definition in ich9 Michael S. Tsirkin
2015-06-04 11:10 ` [Qemu-devel] [PULL 03/28] virtio-pci: don't try to mask or unmask vqs without notifiers Michael S. Tsirkin
2015-06-04 11:10 ` [Qemu-devel] [PULL 04/28] TPM: fix build with tpm disabled Michael S. Tsirkin
2015-06-04 11:10 ` [Qemu-devel] [PULL 05/28] virtio: 64bit features fixups Michael S. Tsirkin
2015-06-04 11:10 ` [Qemu-devel] [PULL 06/28] acpi: add acpi_send_gpe_event() to rise sci for hotplug Michael S. Tsirkin
2015-06-04 11:10 ` [Qemu-devel] [PULL 07/28] acpi: add implementation of aml_while() term Michael S. Tsirkin
2015-06-04 11:10 ` [Qemu-devel] [PULL 08/28] hw/pci: made pci_bus_is_root a PCIBusClass method Michael S. Tsirkin
2015-06-04 11:10 ` [Qemu-devel] [PULL 09/28] hw/pci: made pci_bus_num " Michael S. Tsirkin
2015-06-04 11:10 ` [Qemu-devel] [PULL 10/28] hw/i386: query only for q35/pc when looking for pci host bridge Michael S. Tsirkin
2015-06-04 11:10 ` [Qemu-devel] [PULL 11/28] hw/pci: extend PCI config access to support devices behind PXB Michael S. Tsirkin
2015-06-04 11:10 ` [Qemu-devel] [PULL 12/28] hw/acpi: add support for i440fx 'snooping' root busses Michael S. Tsirkin
2015-06-04 11:10 ` [Qemu-devel] [PULL 13/28] hw/apci: add _PRT method for extra PCI " Michael S. Tsirkin
2015-06-04 11:10 ` [Qemu-devel] [PULL 14/28] hw/acpi: add _CRS method for extra " Michael S. Tsirkin
2015-06-04 11:10 ` [Qemu-devel] [PULL 15/28] hw/acpi: remove from root bus 0 the crs resources used by other buses Michael S. Tsirkin
2015-06-04 11:11 ` [Qemu-devel] [PULL 16/28] hw/pci: removed 'rootbus nr is 0' assumption from qmp_pci_query Michael S. Tsirkin
2015-06-04 11:11 ` [Qemu-devel] [PULL 17/28] hw/pci: introduce PCI Expander Bridge (PXB) Michael S. Tsirkin
2015-06-04 11:11 ` [Qemu-devel] [PULL 18/28] hw/pci: inform bios if the system has extra pci root buses Michael S. Tsirkin
2015-06-04 11:11 ` [Qemu-devel] [PULL 19/28] hw/pxb: add map_irq func Michael S. Tsirkin
2015-06-04 11:11 ` [Qemu-devel] [PULL 20/28] hw/pci: add support for NUMA nodes Michael S. Tsirkin
2015-06-04 11:11 ` [Qemu-devel] [PULL 21/28] hw/pxb: add numa_node parameter Michael S. Tsirkin
2015-06-04 11:11 ` [Qemu-devel] [PULL 22/28] apci: fix PXB behaviour if used with unsupported BIOS Michael S. Tsirkin
2015-06-04 11:11 ` [Qemu-devel] [PULL 23/28] docs: Add PXB documentation Michael S. Tsirkin
2015-06-04 11:11 ` [Qemu-devel] [PULL 24/28] pc-dimm: don't assert if pc-dimm alignment != hotpluggable mem range size Michael S. Tsirkin
2015-06-04 11:11 ` [Qemu-devel] [PULL 25/28] hw/acpi: acpi_pm1_cnt_init(): take "disable_s3" and "disable_s4" Michael S. Tsirkin
2015-06-04 11:11 ` [Qemu-devel] [PULL 26/28] hw/acpi: move "etc/system-states" fw_cfg file from PIIX4 to core Michael S. Tsirkin
2015-06-04 11:11 ` [Qemu-devel] [PULL 27/28] hw/acpi: piix4_pm_init(): take fw_cfg object no more Michael S. Tsirkin
2015-06-04 11:11 ` Michael S. Tsirkin [this message]
2015-06-05 11:03 ` [Qemu-devel] [PULL 00/28] pc, acpi, virtio, tpm Peter Maydell
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=1433416111-19022-29-git-send-email-mst@redhat.com \
--to=mst@redhat.com \
--cc=ehabkost@redhat.com \
--cc=jasowang@redhat.com \
--cc=peter.maydell@linaro.org \
--cc=qemu-devel@nongnu.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).