use strict;
use warnings;
-use PVE::QemuServer;
+
use PVE::Tools qw(run_command lock_file lock_file_full file_read_firstline dir_glob_foreach);
use PVE::Exception qw(raise raise_param_exc);
+use PVE::QemuServer;
+use PVE::QemuServer::Monitor qw(mon_cmd);
+
my $MAX_NUMA = 8;
-my $MAX_MEM = 4194304;
my $STATICMEM = 1024;
+my $_host_bits;
+my sub get_host_phys_address_bits {
+ return $_host_bits if defined($_host_bits);
+
+ my $fh = IO::File->new ('/proc/cpuinfo', "r") or return;
+ while (defined(my $line = <$fh>)) {
+ # hopefully we never need to care about mixed (big.LITTLE) archs
+ if ($line =~ m/^address sizes\s*:\s*(\d+)\s*bits physical/i) {
+ $_host_bits = int($1);
+ $fh->close();
+ return $_host_bits;
+ }
+ }
+ $fh->close();
+ return; # undef, cannot really do anything..
+}
+
+my sub get_max_mem {
+ my ($conf) = @_;
+
+ my $cpu = {};
+ if (my $cpu_prop_str = $conf->{cpu}) {
+ $cpu = PVE::JSONSchema::parse_property_string('pve-vm-cpu-conf', $cpu_prop_str)
+ or die "Cannot parse cpu description: $cpu_prop_str\n";
+ }
+ my $bits;
+ if (my $phys_bits = $cpu->{'phys-bits'}) {
+ if ($phys_bits eq 'host') {
+ $bits = get_host_phys_address_bits();
+ } elsif ($phys_bits =~ /^(\d+)$/) {
+ $bits = int($phys_bits);
+ }
+ }
+
+ if (!defined($bits)) {
+ my $host_bits = get_host_phys_address_bits() // 36; # fixme: what fallback?
+ if ($cpu->{cputype} && $cpu->{cputype} =~ /^(host|max)$/) {
+ $bits = $host_bits;
+ } else {
+ $bits = $host_bits > 40 ? 40 : $host_bits; # take the smaller one
+ }
+ }
+
+ # remove 20 bits to get MB and half that as QEMU needs some overhead
+ my $bits_to_max_mem = int(1 << ($bits - 21));
+
+ return $bits_to_max_mem > 4*1024*1024 ? 4*1024*1024 : $bits_to_max_mem;
+}
+
+sub get_numa_node_list {
+ my ($conf) = @_;
+ my @numa_map;
+ for (my $i = 0; $i < $MAX_NUMA; $i++) {
+ my $entry = $conf->{"numa$i"} or next;
+ my $numa = PVE::QemuServer::parse_numa($entry) or next;
+ push @numa_map, $i;
+ }
+ return @numa_map if @numa_map;
+ my $sockets = $conf->{sockets} || 1;
+ return (0..($sockets-1));
+}
+
+# only valid when numa nodes map to a single host node
+sub get_numa_guest_to_host_map {
+ my ($conf) = @_;
+ my $map = {};
+ for (my $i = 0; $i < $MAX_NUMA; $i++) {
+ my $entry = $conf->{"numa$i"} or next;
+ my $numa = PVE::QemuServer::parse_numa($entry) or next;
+ $map->{$i} = print_numa_hostnodes($numa->{hostnodes});
+ }
+ return $map if %$map;
+ my $sockets = $conf->{sockets} || 1;
+ return {map { $_ => $_ } (0..($sockets-1))};
+}
+
sub foreach_dimm{
my ($conf, $vmid, $memory, $sockets, $func) = @_;
return if $current_size == $memory;
+ my @numa_map = get_numa_node_list($conf);
+
for (my $j = 0; $j < 8; $j++) {
for (my $i = 0; $i < 32; $i++) {
my $name = "dimm${dimm_id}";
$dimm_id++;
- my $numanode = $i % $sockets;
+ my $numanode = $numa_map[$i % @numa_map];
$current_size += $dimm_size;
&$func($conf, $vmid, $name, $dimm_size, $numanode, $current_size, $memory);
return $current_size if $current_size >= $memory;
return if $current_size == $memory;
+ my @numa_map = get_numa_node_list($conf);
+
for (my $j = 0; $j < 8; $j++) {
for (my $i = 0; $i < 32; $i++) {
- my $name = "dimm${dimm_id}";
- $dimm_id--;
- my $numanode = $i % $sockets;
- $current_size -= $dimm_size;
- &$func($conf, $vmid, $name, $dimm_size, $numanode, $current_size, $memory);
+ my $name = "dimm${dimm_id}";
+ $dimm_id--;
+ my $numanode = $numa_map[(31-$i) % @numa_map];
+ $current_size -= $dimm_size;
+ &$func($conf, $vmid, $name, $dimm_size, $numanode, $current_size, $memory);
return $current_size if $current_size <= $memory;
}
$dimm_size /= 2;
$static_memory = $static_memory * $sockets if ($conf->{hugepages} && $conf->{hugepages} == 1024);
die "memory can't be lower than $static_memory MB" if $value < $static_memory;
- die "you cannot add more memory than $MAX_MEM MB!\n" if $memory > $MAX_MEM;
+ my $MAX_MEM = get_max_mem($conf);
+ die "you cannot add more memory than max mem $MAX_MEM MB!\n" if $memory > $MAX_MEM;
+
+ if ($value > $memory) {
- if($value > $memory) {
+ my $numa_hostmap;
- foreach_dimm($conf, $vmid, $value, $sockets, sub {
+ foreach_dimm($conf, $vmid, $value, $sockets, sub {
my ($conf, $vmid, $name, $dimm_size, $numanode, $current_size, $memory) = @_;
return if $current_size <= $conf->{memory};
if ($conf->{hugepages}) {
+ $numa_hostmap = get_numa_guest_to_host_map($conf) if !$numa_hostmap;
my $hugepages_size = hugepages_size($conf, $dimm_size);
my $path = hugepages_mount_path($hugepages_size);
- my $hugepages_topology->{$hugepages_size}->{$numanode} = hugepages_nr($dimm_size, $hugepages_size);
+ my $host_numanode = $numa_hostmap->{$numanode};
+ my $hugepages_topology->{$hugepages_size}->{$host_numanode} = hugepages_nr($dimm_size, $hugepages_size);
my $code = sub {
my $hugepages_host_topology = hugepages_host_topology();
hugepages_allocate($hugepages_topology, $hugepages_host_topology);
- eval { PVE::QemuServer::vm_mon_cmd($vmid, "object-add", 'qom-type' => "memory-backend-file", id => "mem-$name", props => {
- size => int($dimm_size*1024*1024), 'mem-path' => $path, share => JSON::true, prealloc => JSON::true } ); };
+ eval { mon_cmd($vmid, "object-add", 'qom-type' => "memory-backend-file", id => "mem-$name", size => int($dimm_size*1024*1024), 'mem-path' => $path, share => JSON::true, prealloc => JSON::true ) };
if (my $err = $@) {
hugepages_reset($hugepages_host_topology);
die $err;
eval { hugepages_update_locked($code); };
} else {
- eval { PVE::QemuServer::vm_mon_cmd($vmid, "object-add", 'qom-type' => "memory-backend-ram", id => "mem-$name", props => { size => int($dimm_size*1024*1024) } ) };
+ eval { mon_cmd($vmid, "object-add", 'qom-type' => "memory-backend-ram", id => "mem-$name", size => int($dimm_size*1024*1024) ) };
}
if (my $err = $@) {
die $err;
}
- eval { PVE::QemuServer::vm_mon_cmd($vmid, "device_add", driver => "pc-dimm", id => "$name", memdev => "mem-$name", node => $numanode) };
+ eval { mon_cmd($vmid, "device_add", driver => "pc-dimm", id => "$name", memdev => "mem-$name", node => $numanode) };
if (my $err = $@) {
eval { PVE::QemuServer::qemu_objectdel($vmid, "mem-$name"); };
die $err;
} else {
- foreach_reverse_dimm($conf, $vmid, $value, $sockets, sub {
+ foreach_reverse_dimm($conf, $vmid, $value, $sockets, sub {
my ($conf, $vmid, $name, $dimm_size, $numanode, $current_size, $memory) = @_;
return if $current_size >= $conf->{memory};
print "try to unplug memory dimm $name\n";
my $retry = 0;
- while (1) {
+ while (1) {
eval { PVE::QemuServer::qemu_devicedel($vmid, $name) };
sleep 3;
my $dimm_list = qemu_dimm_list($vmid);
sub qemu_dimm_list {
my ($vmid) = @_;
- my $dimmarray = PVE::QemuServer::vm_mon_cmd_nocheck($vmid, "query-memory-devices");
+ my $dimmarray = mon_cmd($vmid, "query-memory-devices");
my $dimms = {};
foreach my $dimm (@$dimmarray) {
sub config {
my ($conf, $vmid, $sockets, $cores, $defaults, $hotplug_features, $cmd) = @_;
-
+
my $memory = $conf->{memory} || $defaults->{memory};
my $static_memory = 0;
if ($hotplug_features->{memory}) {
die "NUMA needs to be enabled for memory hotplug\n" if !$conf->{numa};
+ my $MAX_MEM = get_max_mem($conf);
die "Total memory is bigger than ${MAX_MEM}MB\n" if $memory > $MAX_MEM;
+
+ for (my $i = 0; $i < $MAX_NUMA; $i++) {
+ die "cannot enable memory hotplugging with custom NUMA topology\n"
+ if $conf->{"numa$i"};
+ }
+
my $sockets = 1;
$sockets = $conf->{sockets} if $conf->{sockets};
# cpus
my $cpulists = $numa->{cpus};
die "missing NUMA node$i cpus\n" if !defined($cpulists);
- my $cpus = join(',', map {
+ my $cpus = join(',cpus=', map {
my ($start, $end) = @$_;
defined($end) ? "$start-$end" : $start
} @$cpulists);
if $numa_totalmemory && $numa_totalmemory != $static_memory;
#if no custom tology, we split memory and cores across numa nodes
- if(!$numa_totalmemory) {
-
+ if (!$numa_totalmemory) {
my $numa_memory = ($static_memory / $sockets);
for (my $i = 0; $i < $sockets; $i++) {
die "host NUMA node$i doesn't exist\n" if ! -d "/sys/devices/system/node/node$i/" && $conf->{hugepages};
- my $cpustart = ($cores * $i);
- my $cpuend = ($cpustart + $cores - 1) if $cores && $cores > 1;
- my $cpus = $cpustart;
- $cpus .= "-$cpuend" if $cpuend;
-
my $mem_object = print_mem_object($conf, "ram-node$i", $numa_memory);
-
push @$cmd, '-object', $mem_object;
+
+ my $cpus = ($cores * $i);
+ $cpus .= "-" . ($cpus + $cores - 1) if $cores > 1;
+
push @$cmd, '-numa', "node,nodeid=$i,cpus=$cpus,memdev=ram-node$i";
}
}
push @$cmd, "-object" , $mem_object;
push @$cmd, "-device", "pc-dimm,id=$name,memdev=mem-$name,node=$numanode";
- #if dimm_memory is not aligned to dimm map
- if($current_size > $memory) {
- $conf->{memory} = $current_size;
- PVE::QemuConfig->write_config($vmid, $conf);
- }
+ die "memory size ($memory) must be aligned to $dimm_size for hotplugging\n"
+ if $current_size > $memory;
});
}
}
my $mountdata = PVE::ProcFSTools::parse_proc_mounts();
foreach my $size (qw(2048 1048576)) {
- return if (! -d "/sys/kernel/mm/hugepages/hugepages-${size}kB");
+ next if (! -d "/sys/kernel/mm/hugepages/hugepages-${size}kB");
my $path = "/run/hugepages/kvm/${size}kB";
}
sub hugepages_size {
- my ($conf, $size) = @_;
+ my ($conf, $size) = @_;
+ die "hugepages option is not enabled" if !$conf->{hugepages};
+ die "memory size '$size' is not a positive even integer; cannot use for hugepages\n"
+ if $size <= 0 || $size & 1;
- die "hugepages option is not enabled" if !$conf->{hugepages};
+ my $page_chunk = sub { -d "/sys/kernel/mm/hugepages/hugepages-". ($_[0] * 1024) ."kB" };
+ die "your system doesn't support hugepages\n" if !$page_chunk->(2) && !$page_chunk->(1024);
- if ($conf->{hugepages} eq 'any') {
+ if ($conf->{hugepages} eq 'any') {
- #try to use 1GB if available && memory size is matching
- if (-d "/sys/kernel/mm/hugepages/hugepages-1048576kB" && ($size % 1024 == 0)) {
+ # try to use 1GB if available && memory size is matching
+ if ($page_chunk->(1024) && ($size & 1023) == 0) {
return 1024;
- } else {
+ } elsif ($page_chunk->(2)) {
return 2;
+ } else {
+ die "host only supports 1024 GB hugepages, but requested size '$size' is not a multiple of 1024 MB\n"
}
+ } else {
- } else {
-
- my $hugepagesize = $conf->{hugepages} * 1024 . "kB";
+ my $hugepagesize = $conf->{hugepages};
- if (! -d "/sys/kernel/mm/hugepages/hugepages-$hugepagesize") {
- die "your system doesn't support hugepages of $hugepagesize";
+ if (!$page_chunk->($hugepagesize)) {
+ die "your system doesn't support hugepages of $hugepagesize MB\n";
+ } elsif (($size % $hugepagesize) != 0) {
+ die "Memory size $size is not a multiple of the requested hugepages size $hugepagesize\n";
}
- die "Memory size $size is not a multiple of the requested hugepages size $hugepagesize" if ($size % $conf->{hugepages}) != 0;
- return $conf->{hugepages};
- }
+ return $hugepagesize
+ }
}
sub hugepages_topology {
}
if ($hotplug_features->{memory}) {
+ my $numa_hostmap = get_numa_guest_to_host_map($conf);
+
foreach_dimm($conf, undef, $memory, $sockets, sub {
my ($conf, undef, $name, $dimm_size, $numanode, $current_size, $memory) = @_;
+ $numanode = $numa_hostmap->{$numanode};
+
my $hugepages_size = hugepages_size($conf, $dimm_size);
$hugepages_topology->{$hugepages_size}->{$numanode} += hugepages_nr($dimm_size, $hugepages_size);
});
}
+sub hugepages_default_nr_hugepages {
+ my ($size) = @_;
+
+ my $cmdline = PVE::Tools::file_read_firstline("/proc/cmdline");
+ my $args = PVE::Tools::split_args($cmdline);
+
+ my $parsed_size = 2; # default is 2M
+
+ foreach my $arg (@$args) {
+ if ($arg eq "hugepagesz=2M") {
+ $parsed_size = 2;
+ } elsif ($arg eq "hugepagesz=1G") {
+ $parsed_size = 1024;
+ } elsif ($arg =~ m/^hugepages=(\d+)?$/) {
+ if ($parsed_size == $size) {
+ return $1;
+ }
+ }
+ }
+
+ return 0;
+}
+
sub hugepages_pre_deallocate {
my ($hugepages_topology) = @_;
my $hugepages_size = $size * 1024;
my $path = "/sys/kernel/mm/hugepages/hugepages-${hugepages_size}kB/";
- my $hugepages_nr = PVE::Tools::file_read_firstline($path."nr_hugepages");
- PVE::ProcFSTools::write_proc_entry($path."nr_hugepages", 0);
+ my $hugepages_nr = hugepages_default_nr_hugepages($size);
+ PVE::ProcFSTools::write_proc_entry($path."nr_hugepages", $hugepages_nr);
}
}