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 $j = 0; $j < 8; $j++) {
for (my $i = 0; $i < 32; $i++) {
- my $name = "dimm${dimm_id}";
- $dimm_id--;
+ 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);
+ $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 $numa_hostmap = get_numa_guest_to_host_map($conf) if $conf->{hugepages};
+ 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 $hugepages_host_topology = hugepages_host_topology();
hugepages_allocate($hugepages_topology, $hugepages_host_topology);
- eval { mon_cmd($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 { 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 = $@) {
} else {
- foreach_reverse_dimm($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);
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};
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;
});
}
}