]> git.proxmox.com Git - qemu-server.git/blob - PVE/QemuServer/CPUConfig.pm
4351cfd3b4e81c9b05fc6c30a853c276bdaed052
[qemu-server.git] / PVE / QemuServer / CPUConfig.pm
1 package PVE::QemuServer::CPUConfig;
2
3 use strict;
4 use warnings;
5
6 use PVE::JSONSchema;
7 use PVE::Cluster qw(cfs_register_file cfs_read_file);
8 use PVE::QemuServer::Helpers qw(min_version);
9
10 use base qw(PVE::SectionConfig Exporter);
11
12 our @EXPORT_OK = qw(
13 print_cpu_device
14 get_cpu_options
15 );
16
17 # under certain race-conditions, this module might be loaded before pve-cluster
18 # has started completely, so ensure we don't prevent the FUSE mount with our dir
19 if (PVE::Cluster::check_cfs_is_mounted(1)) {
20 mkdir "/etc/pve/virtual-guest";
21 }
22
23 my $default_filename = "virtual-guest/cpu-models.conf";
24 cfs_register_file($default_filename,
25 sub { PVE::QemuServer::CPUConfig->parse_config(@_); },
26 sub { PVE::QemuServer::CPUConfig->write_config(@_); });
27
28 sub load_custom_model_conf {
29 return cfs_read_file($default_filename);
30 }
31
32 my $cpu_vendor_list = {
33 # Intel CPUs
34 486 => 'GenuineIntel',
35 pentium => 'GenuineIntel',
36 pentium2 => 'GenuineIntel',
37 pentium3 => 'GenuineIntel',
38 coreduo => 'GenuineIntel',
39 core2duo => 'GenuineIntel',
40 Conroe => 'GenuineIntel',
41 Penryn => 'GenuineIntel',
42 Nehalem => 'GenuineIntel',
43 'Nehalem-IBRS' => 'GenuineIntel',
44 Westmere => 'GenuineIntel',
45 'Westmere-IBRS' => 'GenuineIntel',
46 SandyBridge => 'GenuineIntel',
47 'SandyBridge-IBRS' => 'GenuineIntel',
48 IvyBridge => 'GenuineIntel',
49 'IvyBridge-IBRS' => 'GenuineIntel',
50 Haswell => 'GenuineIntel',
51 'Haswell-IBRS' => 'GenuineIntel',
52 'Haswell-noTSX' => 'GenuineIntel',
53 'Haswell-noTSX-IBRS' => 'GenuineIntel',
54 Broadwell => 'GenuineIntel',
55 'Broadwell-IBRS' => 'GenuineIntel',
56 'Broadwell-noTSX' => 'GenuineIntel',
57 'Broadwell-noTSX-IBRS' => 'GenuineIntel',
58 'Skylake-Client' => 'GenuineIntel',
59 'Skylake-Client-IBRS' => 'GenuineIntel',
60 'Skylake-Client-noTSX-IBRS' => 'GenuineIntel',
61 'Skylake-Server' => 'GenuineIntel',
62 'Skylake-Server-IBRS' => 'GenuineIntel',
63 'Skylake-Server-noTSX-IBRS' => 'GenuineIntel',
64 'Cascadelake-Server' => 'GenuineIntel',
65 'Cascadelake-Server-noTSX' => 'GenuineIntel',
66 KnightsMill => 'GenuineIntel',
67 'Icelake-Client' => 'GenuineIntel',
68 'Icelake-Client-noTSX' => 'GenuineIntel',
69 'Icelake-Server' => 'GenuineIntel',
70 'Icelake-Server-noTSX' => 'GenuineIntel',
71
72 # AMD CPUs
73 athlon => 'AuthenticAMD',
74 phenom => 'AuthenticAMD',
75 Opteron_G1 => 'AuthenticAMD',
76 Opteron_G2 => 'AuthenticAMD',
77 Opteron_G3 => 'AuthenticAMD',
78 Opteron_G4 => 'AuthenticAMD',
79 Opteron_G5 => 'AuthenticAMD',
80 EPYC => 'AuthenticAMD',
81 'EPYC-IBPB' => 'AuthenticAMD',
82 'EPYC-Rome' => 'AuthenticAMD',
83
84 # generic types, use vendor from host node
85 host => 'default',
86 kvm32 => 'default',
87 kvm64 => 'default',
88 qemu32 => 'default',
89 qemu64 => 'default',
90 max => 'default',
91 };
92
93 my @supported_cpu_flags = (
94 'pcid',
95 'spec-ctrl',
96 'ibpb',
97 'ssbd',
98 'virt-ssbd',
99 'amd-ssbd',
100 'amd-no-ssb',
101 'pdpe1gb',
102 'md-clear',
103 'hv-tlbflush',
104 'hv-evmcs',
105 'aes'
106 );
107 my $cpu_flag_supported_re = qr/([+-])(@{[join('|', @supported_cpu_flags)]})/;
108 my $cpu_flag_any_re = qr/([+-])([a-zA-Z0-9\-_\.]+)/;
109
110 our $qemu_cmdline_cpu_re = qr/^((?>[+-]?[\w\-\._=]+,?)+)$/;
111
112 my $cpu_fmt = {
113 cputype => {
114 description => "Emulated CPU type. Can be default or custom name (custom model names must be prefixed with 'custom-').",
115 type => 'string',
116 format_description => 'string',
117 default => 'kvm64',
118 default_key => 1,
119 optional => 1,
120 },
121 'reported-model' => {
122 description => "CPU model and vendor to report to the guest. Must be a QEMU/KVM supported model."
123 . " Only valid for custom CPU model definitions, default models will always report themselves to the guest OS.",
124 type => 'string',
125 enum => [ sort { lc("$a") cmp lc("$b") } keys %$cpu_vendor_list ],
126 default => 'kvm64',
127 optional => 1,
128 },
129 hidden => {
130 description => "Do not identify as a KVM virtual machine.",
131 type => 'boolean',
132 optional => 1,
133 default => 0
134 },
135 'hv-vendor-id' => {
136 type => 'string',
137 pattern => qr/[a-zA-Z0-9]{1,12}/,
138 format_description => 'vendor-id',
139 description => 'The Hyper-V vendor ID. Some drivers or programs inside Windows guests need a specific ID.',
140 optional => 1,
141 },
142 flags => {
143 description => "List of additional CPU flags separated by ';'."
144 . " Use '+FLAG' to enable, '-FLAG' to disable a flag."
145 . " Custom CPU models can specify any flag supported by"
146 . " QEMU/KVM, VM-specific flags must be from the following"
147 . " set for security reasons: @{[join(', ', @supported_cpu_flags)]}.",
148 format_description => '+FLAG[;-FLAG...]',
149 type => 'string',
150 pattern => qr/$cpu_flag_any_re(;$cpu_flag_any_re)*/,
151 optional => 1,
152 },
153 'phys-bits' => {
154 type => 'string',
155 format => 'pve-phys-bits',
156 format_description => '8-64|host',
157 description => "The physical memory address bits that are reported to"
158 . " the guest OS. Should be smaller or equal to the host's."
159 . " Set to 'host' to use value from host CPU, but note that"
160 . " doing so will break live migration to CPUs with other values.",
161 optional => 1,
162 },
163 };
164
165 PVE::JSONSchema::register_format('pve-phys-bits', \&parse_phys_bits);
166 sub parse_phys_bits {
167 my ($str, $noerr) = @_;
168
169 my $err_msg = "value must be an integer between 8 and 64 or 'host'\n";
170
171 if ($str !~ m/^(host|\d{1,2})$/) {
172 die $err_msg if !$noerr;
173 return undef;
174 }
175
176 if ($str =~ m/^\d+$/ && (int($str) < 8 || int($str) > 64)) {
177 die $err_msg if !$noerr;
178 return undef;
179 }
180
181 return $str;
182 }
183
184 # $cpu_fmt describes both the CPU config passed as part of a VM config, as well
185 # as the definition of a custom CPU model. There are some slight differences
186 # though, which we catch in the custom validation functions below.
187 PVE::JSONSchema::register_format('pve-cpu-conf', $cpu_fmt, \&validate_cpu_conf);
188 sub validate_cpu_conf {
189 my ($cpu) = @_;
190
191 # required, but can't be forced in schema since it's encoded in section
192 # header for custom models
193 die "CPU is missing cputype\n" if !$cpu->{cputype};
194 }
195 PVE::JSONSchema::register_format('pve-vm-cpu-conf', $cpu_fmt, \&validate_vm_cpu_conf);
196 sub validate_vm_cpu_conf {
197 my ($cpu) = @_;
198
199 validate_cpu_conf($cpu);
200
201 my $cputype = $cpu->{cputype};
202
203 # a VM-specific config is only valid if the cputype exists
204 if (is_custom_model($cputype)) {
205 # dies on unknown model
206 get_custom_model($cputype);
207 } else {
208 die "Built-in cputype '$cputype' is not defined (missing 'custom-' prefix?)\n"
209 if !defined($cpu_vendor_list->{$cputype});
210 }
211
212 # in a VM-specific config, certain properties are limited/forbidden
213
214 die "VM-specific CPU flags must be a subset of: @{[join(', ', @supported_cpu_flags)]}\n"
215 if ($cpu->{flags} && $cpu->{flags} !~ m/$cpu_flag_supported_re(;$cpu_flag_supported_re)*/);
216
217 die "Property 'reported-model' not allowed in VM-specific CPU config.\n"
218 if defined($cpu->{'reported-model'});
219
220 return $cpu;
221 }
222
223 # Section config settings
224 my $defaultData = {
225 # shallow copy, since SectionConfig modifies propertyList internally
226 propertyList => { %$cpu_fmt },
227 };
228
229 sub private {
230 return $defaultData;
231 }
232
233 sub options {
234 return { %$cpu_fmt };
235 }
236
237 sub type {
238 return 'cpu-model';
239 }
240
241 sub parse_section_header {
242 my ($class, $line) = @_;
243
244 my ($type, $sectionId, $errmsg, $config) =
245 $class->SUPER::parse_section_header($line);
246
247 return undef if !$type;
248 return ($type, $sectionId, $errmsg, {
249 # name is given by section header, and we can always prepend 'custom-'
250 # since we're reading the custom CPU file
251 cputype => "custom-$sectionId",
252 });
253 }
254
255 sub write_config {
256 my ($class, $filename, $cfg) = @_;
257
258 mkdir "/etc/pve/virtual-guest";
259
260 for my $model (keys %{$cfg->{ids}}) {
261 my $model_conf = $cfg->{ids}->{$model};
262
263 die "internal error: tried saving built-in CPU model (or missing prefix): $model_conf->{cputype}\n"
264 if !is_custom_model($model_conf->{cputype});
265
266 die "internal error: tried saving custom cpumodel with cputype (ignoring prefix: $model_conf->{cputype}) not equal to \$cfg->ids entry ($model)\n"
267 if "custom-$model" ne $model_conf->{cputype};
268
269 # saved in section header
270 delete $model_conf->{cputype};
271 }
272
273 $class->SUPER::write_config($filename, $cfg);
274 }
275
276 sub add_cpu_json_properties {
277 my ($prop) = @_;
278
279 foreach my $opt (keys %$cpu_fmt) {
280 $prop->{$opt} = $cpu_fmt->{$opt};
281 }
282
283 return $prop;
284 }
285
286 sub get_cpu_models {
287 my ($include_custom) = @_;
288
289 my $models = [];
290
291 for my $default_model (keys %{$cpu_vendor_list}) {
292 push @$models, {
293 name => $default_model,
294 custom => 0,
295 vendor => $cpu_vendor_list->{$default_model},
296 };
297 }
298
299 return $models if !$include_custom;
300
301 my $conf = load_custom_model_conf();
302 for my $custom_model (keys %{$conf->{ids}}) {
303 my $reported_model = $conf->{ids}->{$custom_model}->{'reported-model'};
304 $reported_model //= $cpu_fmt->{'reported-model'}->{default};
305 my $vendor = $cpu_vendor_list->{$reported_model};
306 push @$models, {
307 name => "custom-$custom_model",
308 custom => 1,
309 vendor => $vendor,
310 };
311 }
312
313 return $models;
314 }
315
316 sub is_custom_model {
317 my ($cputype) = @_;
318 return $cputype =~ m/^custom-/;
319 }
320
321 # Use this to get a single model in the format described by $cpu_fmt.
322 # Allows names with and without custom- prefix.
323 sub get_custom_model {
324 my ($name, $noerr) = @_;
325
326 $name =~ s/^custom-//;
327 my $conf = load_custom_model_conf();
328
329 my $entry = $conf->{ids}->{$name};
330 if (!defined($entry)) {
331 die "Custom cputype '$name' not found\n" if !$noerr;
332 return undef;
333 }
334
335 my $model = {};
336 for my $property (keys %$cpu_fmt) {
337 if (my $value = $entry->{$property}) {
338 $model->{$property} = $value;
339 }
340 }
341
342 return $model;
343 }
344
345 # Print a QEMU device node for a given VM configuration for hotplugging CPUs
346 sub print_cpu_device {
347 my ($conf, $id) = @_;
348
349 my $kvm = $conf->{kvm} // 1;
350 my $cpu = $kvm ? "kvm64" : "qemu64";
351 if (my $cputype = $conf->{cpu}) {
352 my $cpuconf = PVE::JSONSchema::parse_property_string('pve-vm-cpu-conf', $cputype)
353 or die "Cannot parse cpu description: $cputype\n";
354 $cpu = $cpuconf->{cputype};
355
356 if (is_custom_model($cpu)) {
357 my $custom_cpu = get_custom_model($cpu);
358
359 $cpu = $custom_cpu->{'reported-model'} //
360 $cpu_fmt->{'reported-model'}->{default};
361 }
362 }
363
364 my $cores = $conf->{cores} || 1;
365
366 my $current_core = ($id - 1) % $cores;
367 my $current_socket = int(($id - 1 - $current_core)/$cores);
368
369 return "$cpu-x86_64-cpu,id=cpu$id,socket-id=$current_socket,core-id=$current_core,thread-id=0";
370 }
371
372 # Resolves multiple arrays of hashes representing CPU flags with metadata to a
373 # single string in QEMU "-cpu" compatible format. Later arrays have higher
374 # priority.
375 #
376 # Hashes take the following format:
377 # {
378 # aes => {
379 # op => "+", # defaults to "" if undefined
380 # reason => "to support AES acceleration", # for override warnings
381 # value => "" # needed for kvm=off (value: off) etc...
382 # },
383 # ...
384 # }
385 sub resolve_cpu_flags {
386 my $flags = {};
387
388 for my $hash (@_) {
389 for my $flag_name (keys %$hash) {
390 my $flag = $hash->{$flag_name};
391 my $old_flag = $flags->{$flag_name};
392
393 $flag->{op} //= "";
394 $flag->{reason} //= "unknown origin";
395
396 if ($old_flag) {
397 my $value_changed = (defined($flag->{value}) != defined($old_flag->{value})) ||
398 (defined($flag->{value}) && $flag->{value} ne $old_flag->{value});
399
400 if ($old_flag->{op} eq $flag->{op} && !$value_changed) {
401 $flags->{$flag_name}->{reason} .= " & $flag->{reason}";
402 next;
403 }
404
405 my $old = print_cpuflag_hash($flag_name, $flags->{$flag_name});
406 my $new = print_cpuflag_hash($flag_name, $flag);
407 warn "warning: CPU flag/setting $new overwrites $old\n";
408 }
409
410 $flags->{$flag_name} = $flag;
411 }
412 }
413
414 my $flag_str = '';
415 # sort for command line stability
416 for my $flag_name (sort keys %$flags) {
417 $flag_str .= ',';
418 $flag_str .= $flags->{$flag_name}->{op};
419 $flag_str .= $flag_name;
420 $flag_str .= "=$flags->{$flag_name}->{value}"
421 if $flags->{$flag_name}->{value};
422 }
423
424 return $flag_str;
425 }
426
427 sub print_cpuflag_hash {
428 my ($flag_name, $flag) = @_;
429 my $formatted = "'$flag->{op}$flag_name";
430 $formatted .= "=$flag->{value}" if defined($flag->{value});
431 $formatted .= "'";
432 $formatted .= " ($flag->{reason})" if defined($flag->{reason});
433 return $formatted;
434 }
435
436 sub parse_cpuflag_list {
437 my ($re, $reason, $flaglist) = @_;
438
439 my $res = {};
440 return $res if !$flaglist;
441
442 foreach my $flag (split(";", $flaglist)) {
443 if ($flag =~ $re) {
444 $res->{$2} = { op => $1, reason => $reason };
445 }
446 }
447
448 return $res;
449 }
450
451 # Calculate QEMU's '-cpu' argument from a given VM configuration
452 sub get_cpu_options {
453 my ($conf, $arch, $kvm, $kvm_off, $machine_version, $winversion, $gpu_passthrough) = @_;
454
455 my $cputype = $kvm ? "kvm64" : "qemu64";
456 if ($arch eq 'aarch64') {
457 $cputype = 'cortex-a57';
458 }
459
460 my $cpu = {};
461 my $custom_cpu;
462 my $hv_vendor_id;
463 if (my $cpu_prop_str = $conf->{cpu}) {
464 $cpu = PVE::JSONSchema::parse_property_string('pve-vm-cpu-conf', $cpu_prop_str)
465 or die "Cannot parse cpu description: $cpu_prop_str\n";
466
467 $cputype = $cpu->{cputype};
468
469 if (is_custom_model($cputype)) {
470 $custom_cpu = get_custom_model($cputype);
471
472 $cputype = $custom_cpu->{'reported-model'} //
473 $cpu_fmt->{'reported-model'}->{default};
474 $kvm_off = $custom_cpu->{hidden}
475 if defined($custom_cpu->{hidden});
476 $hv_vendor_id = $custom_cpu->{'hv-vendor-id'};
477 }
478
479 # VM-specific settings override custom CPU config
480 $kvm_off = $cpu->{hidden}
481 if defined($cpu->{hidden});
482 $hv_vendor_id = $cpu->{'hv-vendor-id'}
483 if defined($cpu->{'hv-vendor-id'});
484 }
485
486 my $pve_flags = get_pve_cpu_flags($conf, $kvm, $cputype, $arch,
487 $machine_version);
488
489 my $hv_flags = get_hyperv_enlightenments($winversion, $machine_version,
490 $conf->{bios}, $gpu_passthrough, $hv_vendor_id) if $kvm;
491
492 my $custom_cputype_flags = parse_cpuflag_list($cpu_flag_any_re,
493 "set by custom CPU model", $custom_cpu->{flags});
494
495 my $vm_flags = parse_cpuflag_list($cpu_flag_supported_re,
496 "manually set for VM", $cpu->{flags});
497
498 my $pve_forced_flags = {};
499 $pve_forced_flags->{'enforce'} = {
500 reason => "error if requested CPU settings not available",
501 } if $cputype ne 'host' && $kvm && $arch eq 'x86_64';
502 $pve_forced_flags->{'kvm'} = {
503 value => "off",
504 reason => "hide KVM virtualization from guest",
505 } if $kvm_off;
506
507 # $cputype is the "reported-model" for custom types, so we can just look up
508 # the vendor in the default list
509 my $cpu_vendor = $cpu_vendor_list->{$cputype};
510 if ($cpu_vendor) {
511 $pve_forced_flags->{'vendor'} = {
512 value => $cpu_vendor,
513 } if $cpu_vendor ne 'default';
514 } elsif ($arch ne 'aarch64') {
515 die "internal error"; # should not happen
516 }
517
518 my $cpu_str = $cputype;
519
520 # will be resolved in parameter order
521 $cpu_str .= resolve_cpu_flags($pve_flags, $hv_flags, $custom_cputype_flags,
522 $vm_flags, $pve_forced_flags);
523
524 my $phys_bits = '';
525 foreach my $conf ($custom_cpu, $cpu) {
526 next if !defined($conf);
527 my $conf_val = $conf->{'phys-bits'};
528 next if !$conf_val;
529 if ($conf_val eq 'host') {
530 $phys_bits = ",host-phys-bits=true";
531 } else {
532 $phys_bits = ",phys-bits=$conf_val";
533 }
534 }
535 $cpu_str .= $phys_bits;
536
537 return ('-cpu', $cpu_str);
538 }
539
540 # Some hardcoded flags required by certain configurations
541 sub get_pve_cpu_flags {
542 my ($conf, $kvm, $cputype, $arch, $machine_version) = @_;
543
544 my $pve_flags = {};
545 my $pve_msg = "set by PVE;";
546
547 $pve_flags->{'lahf_lm'} = {
548 op => '+',
549 reason => "$pve_msg to support Windows 8.1+",
550 } if $cputype eq 'kvm64' && $arch eq 'x86_64';
551
552 $pve_flags->{'x2apic'} = {
553 op => '-',
554 reason => "$pve_msg incompatible with Solaris",
555 } if $conf->{ostype} && $conf->{ostype} eq 'solaris';
556
557 $pve_flags->{'sep'} = {
558 op => '+',
559 reason => "$pve_msg to support Windows 8+ and improve Windows XP+",
560 } if $cputype eq 'kvm64' || $cputype eq 'kvm32';
561
562 $pve_flags->{'rdtscp'} = {
563 op => '-',
564 reason => "$pve_msg broken on AMD Opteron",
565 } if $cputype =~ m/^Opteron/;
566
567 if (min_version($machine_version, 2, 3) && $kvm && $arch eq 'x86_64') {
568 $pve_flags->{'kvm_pv_unhalt'} = {
569 op => '+',
570 reason => "$pve_msg to improve Linux guest spinlock performance",
571 };
572 $pve_flags->{'kvm_pv_eoi'} = {
573 op => '+',
574 reason => "$pve_msg to improve Linux guest interrupt performance",
575 };
576 }
577
578 return $pve_flags;
579 }
580
581 sub get_hyperv_enlightenments {
582 my ($winversion, $machine_version, $bios, $gpu_passthrough, $hv_vendor_id) = @_;
583
584 return if $winversion < 6;
585 return if $bios && $bios eq 'ovmf' && $winversion < 8;
586
587 my $flags = {};
588 my $default_reason = "automatic Hyper-V enlightenment for Windows";
589 my $flagfn = sub {
590 my ($flag, $value, $reason) = @_;
591 $flags->{$flag} = {
592 reason => $reason // $default_reason,
593 value => $value,
594 }
595 };
596
597 my $hv_vendor_set = defined($hv_vendor_id);
598 if ($gpu_passthrough || $hv_vendor_set) {
599 $hv_vendor_id //= 'proxmox';
600 $flagfn->('hv_vendor_id', $hv_vendor_id, $hv_vendor_set ?
601 "custom hv_vendor_id set" : "NVIDIA workaround for GPU passthrough");
602 }
603
604 if (min_version($machine_version, 2, 3)) {
605 $flagfn->('hv_spinlocks', '0x1fff');
606 $flagfn->('hv_vapic');
607 $flagfn->('hv_time');
608 } else {
609 $flagfn->('hv_spinlocks', '0xffff');
610 }
611
612 if (min_version($machine_version, 2, 6)) {
613 $flagfn->('hv_reset');
614 $flagfn->('hv_vpindex');
615 $flagfn->('hv_runtime');
616 }
617
618 if ($winversion >= 7) {
619 my $win7_reason = $default_reason . " 7 and higher";
620 $flagfn->('hv_relaxed', undef, $win7_reason);
621
622 if (min_version($machine_version, 2, 12)) {
623 $flagfn->('hv_synic', undef, $win7_reason);
624 $flagfn->('hv_stimer', undef, $win7_reason);
625 }
626
627 if (min_version($machine_version, 3, 1)) {
628 $flagfn->('hv_ipi', undef, $win7_reason);
629 }
630 }
631
632 return $flags;
633 }
634
635 sub get_cpu_from_running_vm {
636 my ($pid) = @_;
637
638 my $cmdline = PVE::QemuServer::Helpers::parse_cmdline($pid);
639 die "could not read commandline of running machine\n"
640 if !$cmdline->{cpu}->{value};
641
642 # sanitize and untaint value
643 $cmdline->{cpu}->{value} =~ $qemu_cmdline_cpu_re;
644 return $1;
645 }
646
647 __PACKAGE__->register();
648 __PACKAGE__->init();
649
650 1;