[libvirt] [PATCH] qemu: Expose rx/tx_queue_size in qemu.conf too
Michal Privoznik
mprivozn at redhat.com
Thu Feb 1 13:27:54 UTC 2018
On 01/31/2018 03:28 PM, John Ferlan wrote:
>
>
> On 01/19/2018 07:50 AM, Michal Privoznik wrote:
>> In 2074ef6cd4a2 and c56cdf259 (and friends) we've added two
>> attributes to virtio NICs: rx_queue_size and tx_queue_size.
>> However, sysadmins might want to set these on per-host basis but
>> don't necessarily have an access to domain XML (e.g. because they
>> are generated by some other app). So let's expose them under
>> qemu.conf (the settings from domain XML still take precedence as
>> they are more specific ones).
>
> This wording says to me domain XML takes precedence; however,... [1]
>
>>
>> Signed-off-by: Michal Privoznik <mprivozn at redhat.com>
>> ---
>> docs/formatdomain.html.in | 12 +++++++++--
>> src/qemu/libvirtd_qemu.aug | 4 ++++
>> src/qemu/qemu.conf | 7 +++++++
>> src/qemu/qemu_command.c | 42 ++++++++++++++++++++++++++++++--------
>> src/qemu/qemu_command.h | 3 ++-
>> src/qemu/qemu_conf.c | 4 ++++
>> src/qemu/qemu_conf.h | 3 +++
>> src/qemu/qemu_hotplug.c | 2 +-
>> src/qemu/test_libvirtd_qemu.aug.in | 2 ++
>> 9 files changed, 66 insertions(+), 13 deletions(-)
>>
>> diff --git a/docs/formatdomain.html.in b/docs/formatdomain.html.in
>> index d272cc1ba..c0107ab4b 100644
>> --- a/docs/formatdomain.html.in
>> +++ b/docs/formatdomain.html.in
>> @@ -5373,7 +5373,11 @@ qemu-kvm -net nic,model=? /dev/null
>> some restrictions on actual value. For instance, latest
>> QEMU (as of 2016-09-01) requires value to be a power of two
>> from [256, 1024] range.
>> - <span class="since">Since 2.3.0 (QEMU and KVM only)</span><br/><br/>
>> + <span class="since">Since 2.3.0 (QEMU and KVM only)</span>
>> + Then, <span class="since">Since 4.1.0</span> the default value can be
>> + set in <code>qemu.conf</code> file and thus overrides hypervisor
>> + default.
>
> [1] ...this and...
>
>> + <br/><br/>
>>
>> <b>In general you should leave this option alone, unless you
>> are very certain you know what you are doing.</b>
>> @@ -5389,7 +5393,11 @@ qemu-kvm -net nic,model=? /dev/null
>> range. In addition to that, this may work only for a subset of
>> interface types, e.g. aforementioned QEMU enables this option
>> only for <code>vhostuser</code> type.
>> - <span class="since">Since 3.7.0 (QEMU and KVM only)</span><br/><br/>
>> + <span class="since">Since 3.7.0 (QEMU and KVM only)</span>
>> + Then, <span class="since">Since 4.1.0</span> the default value can be
>> + set in <code>qemu.conf</code> file and thus overrides hypervisor
>> + default.
>> + <br/><br/>
>
> [1] ... this seems to imply the conf value overrides XML (or as written
> hypervisor default)...
No. hypervisor default has the least precedence. It's like this:
hv default < qemu.conf < domain XML.
>
> Then the code does something even more interesting... [2]
>
> BTW: If you look at the generated output you see "Then, Since 4.1.0...."
>
> Personally, "Then," is probably not the best transition "word"... Also
> it's not clear that "and thus overrides hypervisor default." is actually
> what's being done in the code.
>
> My suggestion:
>
> "Additionally, since 4.1.0 the value can be set in the qemu.conf file in
> order to override the domain XML setting."
Okay, I'm no good with documentation.
>
>>
>> <b>In general you should leave this option alone, unless you
>> are very certain you know what you are doing.</b>
>> diff --git a/src/qemu/libvirtd_qemu.aug b/src/qemu/libvirtd_qemu.aug
>> index c19bf3a43..084290296 100644
>> --- a/src/qemu/libvirtd_qemu.aug
>> +++ b/src/qemu/libvirtd_qemu.aug
>> @@ -118,6 +118,9 @@ module Libvirtd_qemu =
>> let vxhs_entry = bool_entry "vxhs_tls"
>> | str_entry "vxhs_tls_x509_cert_dir"
>>
>> + let virtio_entry = int_entry "rx_queue_size"
>> + | int_entry "tx_queue_size"
>> +
>> (* Each entry in the config is one of the following ... *)
>> let entry = default_tls_entry
>> | vnc_entry
>> @@ -137,6 +140,7 @@ module Libvirtd_qemu =
>> | gluster_debug_level_entry
>> | memory_entry
>> | vxhs_entry
>> + | virtio_entry
>>
>> let comment = [ label "#comment" . del /#[ \t]*/ "# " . store /([^ \t\n][^\n]*)?/ . del /\n/ "\n" ]
>> let empty = [ label "#empty" . eol ]
>> diff --git a/src/qemu/qemu.conf b/src/qemu/qemu.conf
>> index 43dd561cc..a945ebdd5 100644
>> --- a/src/qemu/qemu.conf
>> +++ b/src/qemu/qemu.conf
>> @@ -775,3 +775,10 @@
>> # This directory is used for memoryBacking source if configured as file.
>> # NOTE: big files will be stored here
>> #memory_backing_dir = "/var/lib/libvirt/qemu/ram"
>> +
>> +# The following two values set the default RX/TX ring buffer size for virtio
>> +# interfaces. These values are taken unless overridden in domain XML. Please
>> +# note that QEMU accepts 256, 512 and 1024 only. These values correspond to
>> +# those from domain XML.
>
>
> [1]... and this says, XML overrides values - which to a degree is what
> the code does.
>
> Interesting to note the QEMU valid values... I think we'd be better off
> indicating that valid values are described in formatdomain.html, but a
> hyperlink to the docs isn't something that's already in the qemu.conf
> file and I'm not sure/clear if it's "good" or "valid" to put it there.
Sure. I can drop the sentence mentioning values completely. The next one
refers to the docs anyway.
>
> In the long run, I think it'd be nice to have one place to describe when
> the feature is supported and what the valid values are so that if/when
> they change in the future it's only one place to change.
>
> Someone could set this, but if their QEMU isn't new enough, then domains
> will fail to start and it may not be obvious why.
No, if qemu doesn't support the feature these settings don't get
applied. Starting fails only if the values are requested in domain XML,
not qemu.conf.
>
>> +#rx_queue_size = 1024
>> +#tx_queue_size = 1024
>> diff --git a/src/qemu/qemu_command.c b/src/qemu/qemu_command.c
>> index b8aede32d..771c12445 100644
>> --- a/src/qemu/qemu_command.c
>> +++ b/src/qemu/qemu_command.c
>> @@ -3693,7 +3693,8 @@ qemuBuildNicStr(virDomainNetDefPtr net,
>>
>>
>> char *
>> -qemuBuildNicDevStr(virDomainDefPtr def,
>> +qemuBuildNicDevStr(virQEMUDriverConfigPtr cfg,
>> + virDomainDefPtr def,
>> virDomainNetDefPtr net,
>> int vlan,
>> unsigned int bootindex,
>> @@ -3813,21 +3814,40 @@ qemuBuildNicDevStr(virDomainDefPtr def,
>> virBufferAsprintf(&buf, ",mq=on,vectors=%zu", 2 * vhostfdSize + 2);
>> }
>> }
>> - if (usingVirtio && net->driver.virtio.rx_queue_size) {
>> - if (!virQEMUCapsGet(qemuCaps, QEMU_CAPS_VIRTIO_NET_RX_QUEUE_SIZE)) {
>> + if (usingVirtio) {
>> + unsigned int rx_queue_size = net->driver.virtio.rx_queue_size;
>> +
>> + if (rx_queue_size == 0 &&
>> + virQEMUCapsGet(qemuCaps, QEMU_CAPS_VIRTIO_NET_RX_QUEUE_SIZE))
>> + rx_queue_size = cfg->rx_queue_size;
>> +
>> +
>> + if (rx_queue_size &&
>> + virQEMUCapsGet(qemuCaps, QEMU_CAPS_VIRTIO_NET_RX_QUEUE_SIZE)) {
>> + net->driver.virtio.rx_queue_size = rx_queue_size;
>
> [2] ...if someone sets the qemu.conf variable it overwrites the domain
> value? Is that really something we want to do? Is this is only the
> running XML or is it the config XML?
It is just live XML and we need to do this to keep ABI stability during
migration. Also, it's nice to see what values were applied, therefore
you can see them in 'dumpxml' for a live domain. It's not unusual that
we do this, for instance: qemuDomainPrepareChardevSourceTLS().
>
>> + virBufferAsprintf(&buf, ",rx_queue_size=%u", rx_queue_size);
>> + } else if (rx_queue_size) {
>> virReportError(VIR_ERR_CONFIG_UNSUPPORTED, "%s",
>> _("virtio rx_queue_size option is not supported with this QEMU binary"));
>> goto error;
>> }
>> - virBufferAsprintf(&buf, ",rx_queue_size=%u", net->driver.virtio.rx_queue_size);
>> }
>> - if (usingVirtio && net->driver.virtio.tx_queue_size) {
>> - if (!virQEMUCapsGet(qemuCaps, QEMU_CAPS_VIRTIO_NET_TX_QUEUE_SIZE)) {
>> + if (usingVirtio) {
>> + unsigned int tx_queue_size = net->driver.virtio.tx_queue_size;
>> +
>> + if (tx_queue_size == 0 &&
>> + virQEMUCapsGet(qemuCaps, QEMU_CAPS_VIRTIO_NET_TX_QUEUE_SIZE))
>> + tx_queue_size = cfg->tx_queue_size;
>> +
>> + if (tx_queue_size &&
>> + virQEMUCapsGet(qemuCaps, QEMU_CAPS_VIRTIO_NET_TX_QUEUE_SIZE)) {
>> + net->driver.virtio.tx_queue_size = tx_queue_size;
>> + virBufferAsprintf(&buf, ",tx_queue_size=%u", tx_queue_size);
>> + } else if (tx_queue_size) {
>> virReportError(VIR_ERR_CONFIG_UNSUPPORTED, "%s",
>> _("virtio tx_queue_size option is not supported with this QEMU binary"));
>> goto error;
>> }
>> - virBufferAsprintf(&buf, ",tx_queue_size=%u", net->driver.virtio.tx_queue_size);
>> }
>>
>> if (usingVirtio && net->mtu) {
>> @@ -8489,7 +8509,7 @@ qemuBuildVhostuserCommandLine(virQEMUDriverPtr driver,
>> virCommandAddArg(cmd, netdev);
>> VIR_FREE(netdev);
>>
>> - if (!(nic = qemuBuildNicDevStr(def, net, -1, bootindex,
>> + if (!(nic = qemuBuildNicDevStr(cfg, def, net, -1, bootindex,
>> queues, qemuCaps))) {
>> virReportError(VIR_ERR_INTERNAL_ERROR,
>> "%s", _("Error generating NIC -device string"));
>> @@ -8526,6 +8546,7 @@ qemuBuildInterfaceCommandLine(virQEMUDriverPtr driver,
>> int **nicindexes,
>> bool chardevStdioLogd)
>> {
>> + virQEMUDriverConfigPtr cfg;
>
> Today this is fine, worried about some future adjustment which goes to
> cleanup before cfg is set... So if we set it to NULL we avoid that
> adjustment causing some unforeseen issue.
Okay, consider done.
>
>> int ret = -1;
>> char *nic = NULL, *host = NULL;
>> int *tapfd = NULL;
>> @@ -8587,6 +8608,8 @@ qemuBuildInterfaceCommandLine(virQEMUDriverPtr driver,
>> return -1;
>> }
>>
>> + cfg = virQEMUDriverGetConfig(driver);
>> +
>> switch (actualType) {
>> case VIR_DOMAIN_NET_TYPE_NETWORK:
>> case VIR_DOMAIN_NET_TYPE_BRIDGE:
>> @@ -8782,7 +8805,7 @@ qemuBuildInterfaceCommandLine(virQEMUDriverPtr driver,
>> virCommandAddArgList(cmd, "-netdev", host, NULL);
>> }
>> if (qemuDomainSupportsNicdev(def, net)) {
>> - if (!(nic = qemuBuildNicDevStr(def, net, vlan, bootindex,
>> + if (!(nic = qemuBuildNicDevStr(cfg, def, net, vlan, bootindex,
>> vhostfdSize, qemuCaps)))
>> goto cleanup;
>> virCommandAddArgList(cmd, "-device", nic, NULL);
>> @@ -8826,6 +8849,7 @@ qemuBuildInterfaceCommandLine(virQEMUDriverPtr driver,
>> VIR_FREE(host);
>> VIR_FREE(tapfdName);
>> VIR_FREE(vhostfdName);
>> + virObjectUnref(cfg);
>> return ret;
>> }
>>
>> diff --git a/src/qemu/qemu_command.h b/src/qemu/qemu_command.h
>> index bdde6f918..85f7bd1b4 100644
>> --- a/src/qemu/qemu_command.h
>> +++ b/src/qemu/qemu_command.h
>> @@ -90,7 +90,8 @@ char *qemuBuildNicStr(virDomainNetDefPtr net,
>> int vlan);
>>
>> /* Current, best practice */
>> -char *qemuBuildNicDevStr(virDomainDefPtr def,
>> +char *qemuBuildNicDevStr(virQEMUDriverConfigPtr cfg,
>> + virDomainDefPtr def,
>> virDomainNetDefPtr net,
>> int vlan,
>> unsigned int bootindex,
>> diff --git a/src/qemu/qemu_conf.c b/src/qemu/qemu_conf.c
>> index af503d31c..2fa96431f 100644
>> --- a/src/qemu/qemu_conf.c
>> +++ b/src/qemu/qemu_conf.c
>> @@ -912,6 +912,10 @@ int virQEMUDriverConfigLoadFile(virQEMUDriverConfigPtr cfg,
>> if (virConfGetValueString(conf, "memory_backing_dir", &cfg->memoryBackingDir) < 0)
>> goto cleanup;
>>
>> + if (virConfGetValueUInt(conf, "rx_queue_size", &cfg->rx_queue_size) < 0 ||
>> + virConfGetValueUInt(conf, "tx_queue_size", &cfg->tx_queue_size) < 0)
>> + goto cleanup;
>> +
>
> Once the domain capabilities are read, can/should we then check if the
> rx/tx values are set and cause a failure at startup time? Probably
> reduces some complexity in qemuBuildNicDevStr too.
Well, I don't know. I mean, currently this patch is written in
fault-tolerant fashion - if qemu doesn't support these two settings,
domain can start just fine. If it does support them, they are applied.
On one hand, this is very user friendly. On the other it doesn't follow
how we treat other settings. For instance, the aforementioned
chardev_tls - if qemu doesn't support it starting a domain fails. So
maybe after all I need to change this behaviour I've implemented. Either
don't enable it in the config or install newer qemu. I'll send v2 shortly.
Michal
More information about the libvir-list
mailing list