]> git.proxmox.com Git - qemu-server.git/blob - PVE/QemuServer/CPUConfig.pm
62505914d39af2764046ed48072729f21c493bbe
[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 description => "The physical memory address bits that are reported to"
157 . " the guest OS. Should be smaller or equal to the host's."
158 . " Set to 'host' to use value from host CPU, but note that"
159 . " doing so will break live migration to CPUs with other values.",
160 optional => 1,
161 },
162 };
163
164 PVE::JSONSchema::register_format('pve-phys-bits', \&parse_phys_bits);
165 sub parse_phys_bits {
166 my ($str, $noerr) = @_;
167
168 my $err_msg = "value must be an integer between 8 and 64 or 'host'\n";
169
170 if ($str !~ m/^(host|\d{1,2})$/) {
171 die $err_msg if !$noerr;
172 return undef;
173 }
174
175 if ($str =~ m/^\d+$/ && (int($str) < 8 || int($str) > 64)) {
176 die $err_msg if !$noerr;
177 return undef;
178 }
179
180 return $str;
181 }
182
183 # $cpu_fmt describes both the CPU config passed as part of a VM config, as well
184 # as the definition of a custom CPU model. There are some slight differences
185 # though, which we catch in the custom verification function below.
186 PVE::JSONSchema::register_format('pve-cpu-conf', \&parse_cpu_conf_basic);
187 sub parse_cpu_conf_basic {
188 my ($cpu_str, $noerr) = @_;
189
190 my $cpu = eval { PVE::JSONSchema::parse_property_string($cpu_fmt, $cpu_str) };
191 if ($@) {
192 die $@ if !$noerr;
193 return undef;
194 }
195
196 # required, but can't be forced in schema since it's encoded in section
197 # header for custom models
198 if (!$cpu->{cputype}) {
199 die "CPU is missing cputype\n" if !$noerr;
200 return undef;
201 }
202
203 return $cpu;
204 }
205
206 PVE::JSONSchema::register_format('pve-vm-cpu-conf', \&parse_vm_cpu_conf);
207 sub parse_vm_cpu_conf {
208 my ($cpu_str, $noerr) = @_;
209
210 my $cpu = parse_cpu_conf_basic($cpu_str, $noerr);
211 return undef if !$cpu;
212
213 my $cputype = $cpu->{cputype};
214
215 # a VM-specific config is only valid if the cputype exists
216 if (is_custom_model($cputype)) {
217 eval { get_custom_model($cputype); };
218 if ($@) {
219 die $@ if !$noerr;
220 return undef;
221 }
222 } else {
223 if (!defined($cpu_vendor_list->{$cputype})) {
224 die "Built-in cputype '$cputype' is not defined (missing 'custom-' prefix?)\n" if !$noerr;
225 return undef;
226 }
227 }
228
229 # in a VM-specific config, certain properties are limited/forbidden
230
231 if ($cpu->{flags} && $cpu->{flags} !~ m/$cpu_flag_supported_re(;$cpu_flag_supported_re)*/) {
232 die "VM-specific CPU flags must be a subset of: @{[join(', ', @supported_cpu_flags)]}\n"
233 if !$noerr;
234 return undef;
235 }
236
237 die "Property 'reported-model' not allowed in VM-specific CPU config.\n"
238 if defined($cpu->{'reported-model'});
239
240 return $cpu;
241 }
242
243 # Section config settings
244 my $defaultData = {
245 # shallow copy, since SectionConfig modifies propertyList internally
246 propertyList => { %$cpu_fmt },
247 };
248
249 sub private {
250 return $defaultData;
251 }
252
253 sub options {
254 return { %$cpu_fmt };
255 }
256
257 sub type {
258 return 'cpu-model';
259 }
260
261 sub parse_section_header {
262 my ($class, $line) = @_;
263
264 my ($type, $sectionId, $errmsg, $config) =
265 $class->SUPER::parse_section_header($line);
266
267 return undef if !$type;
268 return ($type, $sectionId, $errmsg, {
269 # name is given by section header, and we can always prepend 'custom-'
270 # since we're reading the custom CPU file
271 cputype => "custom-$sectionId",
272 });
273 }
274
275 sub write_config {
276 my ($class, $filename, $cfg) = @_;
277
278 mkdir "/etc/pve/virtual-guest";
279
280 for my $model (keys %{$cfg->{ids}}) {
281 my $model_conf = $cfg->{ids}->{$model};
282
283 die "internal error: tried saving built-in CPU model (or missing prefix): $model_conf->{cputype}\n"
284 if !is_custom_model($model_conf->{cputype});
285
286 die "internal error: tried saving custom cpumodel with cputype (ignoring prefix: $model_conf->{cputype}) not equal to \$cfg->ids entry ($model)\n"
287 if "custom-$model" ne $model_conf->{cputype};
288
289 # saved in section header
290 delete $model_conf->{cputype};
291 }
292
293 $class->SUPER::write_config($filename, $cfg);
294 }
295
296 sub add_cpu_json_properties {
297 my ($prop) = @_;
298
299 foreach my $opt (keys %$cpu_fmt) {
300 $prop->{$opt} = $cpu_fmt->{$opt};
301 }
302
303 return $prop;
304 }
305
306 sub get_cpu_models {
307 my ($include_custom) = @_;
308
309 my $models = [];
310
311 for my $default_model (keys %{$cpu_vendor_list}) {
312 push @$models, {
313 name => $default_model,
314 custom => 0,
315 vendor => $cpu_vendor_list->{$default_model},
316 };
317 }
318
319 return $models if !$include_custom;
320
321 my $conf = load_custom_model_conf();
322 for my $custom_model (keys %{$conf->{ids}}) {
323 my $reported_model = $conf->{ids}->{$custom_model}->{'reported-model'};
324 $reported_model //= $cpu_fmt->{'reported-model'}->{default};
325 my $vendor = $cpu_vendor_list->{$reported_model};
326 push @$models, {
327 name => "custom-$custom_model",
328 custom => 1,
329 vendor => $vendor,
330 };
331 }
332
333 return $models;
334 }
335
336 sub is_custom_model {
337 my ($cputype) = @_;
338 return $cputype =~ m/^custom-/;
339 }
340
341 # Use this to get a single model in the format described by $cpu_fmt.
342 # Allows names with and without custom- prefix.
343 sub get_custom_model {
344 my ($name, $noerr) = @_;
345
346 $name =~ s/^custom-//;
347 my $conf = load_custom_model_conf();
348
349 my $entry = $conf->{ids}->{$name};
350 if (!defined($entry)) {
351 die "Custom cputype '$name' not found\n" if !$noerr;
352 return undef;
353 }
354
355 my $model = {};
356 for my $property (keys %$cpu_fmt) {
357 if (my $value = $entry->{$property}) {
358 $model->{$property} = $value;
359 }
360 }
361
362 return $model;
363 }
364
365 # Print a QEMU device node for a given VM configuration for hotplugging CPUs
366 sub print_cpu_device {
367 my ($conf, $id) = @_;
368
369 my $kvm = $conf->{kvm} // 1;
370 my $cpu = $kvm ? "kvm64" : "qemu64";
371 if (my $cputype = $conf->{cpu}) {
372 my $cpuconf = parse_cpu_conf_basic($cputype)
373 or die "Cannot parse cpu description: $cputype\n";
374 $cpu = $cpuconf->{cputype};
375
376 if (is_custom_model($cpu)) {
377 my $custom_cpu = get_custom_model($cpu);
378
379 $cpu = $custom_cpu->{'reported-model'} //
380 $cpu_fmt->{'reported-model'}->{default};
381 }
382 }
383
384 my $cores = $conf->{cores} || 1;
385
386 my $current_core = ($id - 1) % $cores;
387 my $current_socket = int(($id - 1 - $current_core)/$cores);
388
389 return "$cpu-x86_64-cpu,id=cpu$id,socket-id=$current_socket,core-id=$current_core,thread-id=0";
390 }
391
392 # Resolves multiple arrays of hashes representing CPU flags with metadata to a
393 # single string in QEMU "-cpu" compatible format. Later arrays have higher
394 # priority.
395 #
396 # Hashes take the following format:
397 # {
398 # aes => {
399 # op => "+", # defaults to "" if undefined
400 # reason => "to support AES acceleration", # for override warnings
401 # value => "" # needed for kvm=off (value: off) etc...
402 # },
403 # ...
404 # }
405 sub resolve_cpu_flags {
406 my $flags = {};
407
408 for my $hash (@_) {
409 for my $flag_name (keys %$hash) {
410 my $flag = $hash->{$flag_name};
411 my $old_flag = $flags->{$flag_name};
412
413 $flag->{op} //= "";
414 $flag->{reason} //= "unknown origin";
415
416 if ($old_flag) {
417 my $value_changed = (defined($flag->{value}) != defined($old_flag->{value})) ||
418 (defined($flag->{value}) && $flag->{value} ne $old_flag->{value});
419
420 if ($old_flag->{op} eq $flag->{op} && !$value_changed) {
421 $flags->{$flag_name}->{reason} .= " & $flag->{reason}";
422 next;
423 }
424
425 my $old = print_cpuflag_hash($flag_name, $flags->{$flag_name});
426 my $new = print_cpuflag_hash($flag_name, $flag);
427 warn "warning: CPU flag/setting $new overwrites $old\n";
428 }
429
430 $flags->{$flag_name} = $flag;
431 }
432 }
433
434 my $flag_str = '';
435 # sort for command line stability
436 for my $flag_name (sort keys %$flags) {
437 $flag_str .= ',';
438 $flag_str .= $flags->{$flag_name}->{op};
439 $flag_str .= $flag_name;
440 $flag_str .= "=$flags->{$flag_name}->{value}"
441 if $flags->{$flag_name}->{value};
442 }
443
444 return $flag_str;
445 }
446
447 sub print_cpuflag_hash {
448 my ($flag_name, $flag) = @_;
449 my $formatted = "'$flag->{op}$flag_name";
450 $formatted .= "=$flag->{value}" if defined($flag->{value});
451 $formatted .= "'";
452 $formatted .= " ($flag->{reason})" if defined($flag->{reason});
453 return $formatted;
454 }
455
456 sub parse_cpuflag_list {
457 my ($re, $reason, $flaglist) = @_;
458
459 my $res = {};
460 return $res if !$flaglist;
461
462 foreach my $flag (split(";", $flaglist)) {
463 if ($flag =~ $re) {
464 $res->{$2} = { op => $1, reason => $reason };
465 }
466 }
467
468 return $res;
469 }
470
471 # Calculate QEMU's '-cpu' argument from a given VM configuration
472 sub get_cpu_options {
473 my ($conf, $arch, $kvm, $kvm_off, $machine_version, $winversion, $gpu_passthrough) = @_;
474
475 my $cputype = $kvm ? "kvm64" : "qemu64";
476 if ($arch eq 'aarch64') {
477 $cputype = 'cortex-a57';
478 }
479
480 my $cpu = {};
481 my $custom_cpu;
482 my $hv_vendor_id;
483 if (my $cpu_prop_str = $conf->{cpu}) {
484 $cpu = parse_vm_cpu_conf($cpu_prop_str)
485 or die "Cannot parse cpu description: $cpu_prop_str\n";
486
487 $cputype = $cpu->{cputype};
488
489 if (is_custom_model($cputype)) {
490 $custom_cpu = get_custom_model($cputype);
491
492 $cputype = $custom_cpu->{'reported-model'} //
493 $cpu_fmt->{'reported-model'}->{default};
494 $kvm_off = $custom_cpu->{hidden}
495 if defined($custom_cpu->{hidden});
496 $hv_vendor_id = $custom_cpu->{'hv-vendor-id'};
497 }
498
499 # VM-specific settings override custom CPU config
500 $kvm_off = $cpu->{hidden}
501 if defined($cpu->{hidden});
502 $hv_vendor_id = $cpu->{'hv-vendor-id'}
503 if defined($cpu->{'hv-vendor-id'});
504 }
505
506 my $pve_flags = get_pve_cpu_flags($conf, $kvm, $cputype, $arch,
507 $machine_version);
508
509 my $hv_flags = get_hyperv_enlightenments($winversion, $machine_version,
510 $conf->{bios}, $gpu_passthrough, $hv_vendor_id) if $kvm;
511
512 my $custom_cputype_flags = parse_cpuflag_list($cpu_flag_any_re,
513 "set by custom CPU model", $custom_cpu->{flags});
514
515 my $vm_flags = parse_cpuflag_list($cpu_flag_supported_re,
516 "manually set for VM", $cpu->{flags});
517
518 my $pve_forced_flags = {};
519 $pve_forced_flags->{'enforce'} = {
520 reason => "error if requested CPU settings not available",
521 } if $cputype ne 'host' && $kvm && $arch eq 'x86_64';
522 $pve_forced_flags->{'kvm'} = {
523 value => "off",
524 reason => "hide KVM virtualization from guest",
525 } if $kvm_off;
526
527 # $cputype is the "reported-model" for custom types, so we can just look up
528 # the vendor in the default list
529 my $cpu_vendor = $cpu_vendor_list->{$cputype};
530 if ($cpu_vendor) {
531 $pve_forced_flags->{'vendor'} = {
532 value => $cpu_vendor,
533 } if $cpu_vendor ne 'default';
534 } elsif ($arch ne 'aarch64') {
535 die "internal error"; # should not happen
536 }
537
538 my $cpu_str = $cputype;
539
540 # will be resolved in parameter order
541 $cpu_str .= resolve_cpu_flags($pve_flags, $hv_flags, $custom_cputype_flags,
542 $vm_flags, $pve_forced_flags);
543
544 my $phys_bits = '';
545 foreach my $conf ($custom_cpu, $cpu) {
546 next if !defined($conf);
547 my $conf_val = $conf->{'phys-bits'};
548 next if !$conf_val;
549 if ($conf_val eq 'host') {
550 $phys_bits = ",host-phys-bits=true";
551 } else {
552 $phys_bits = ",phys-bits=$conf_val";
553 }
554 }
555 $cpu_str .= $phys_bits;
556
557 return ('-cpu', $cpu_str);
558 }
559
560 # Some hardcoded flags required by certain configurations
561 sub get_pve_cpu_flags {
562 my ($conf, $kvm, $cputype, $arch, $machine_version) = @_;
563
564 my $pve_flags = {};
565 my $pve_msg = "set by PVE;";
566
567 $pve_flags->{'lahf_lm'} = {
568 op => '+',
569 reason => "$pve_msg to support Windows 8.1+",
570 } if $cputype eq 'kvm64' && $arch eq 'x86_64';
571
572 $pve_flags->{'x2apic'} = {
573 op => '-',
574 reason => "$pve_msg incompatible with Solaris",
575 } if $conf->{ostype} && $conf->{ostype} eq 'solaris';
576
577 $pve_flags->{'sep'} = {
578 op => '+',
579 reason => "$pve_msg to support Windows 8+ and improve Windows XP+",
580 } if $cputype eq 'kvm64' || $cputype eq 'kvm32';
581
582 $pve_flags->{'rdtscp'} = {
583 op => '-',
584 reason => "$pve_msg broken on AMD Opteron",
585 } if $cputype =~ m/^Opteron/;
586
587 if (min_version($machine_version, 2, 3) && $kvm && $arch eq 'x86_64') {
588 $pve_flags->{'kvm_pv_unhalt'} = {
589 op => '+',
590 reason => "$pve_msg to improve Linux guest spinlock performance",
591 };
592 $pve_flags->{'kvm_pv_eoi'} = {
593 op => '+',
594 reason => "$pve_msg to improve Linux guest interrupt performance",
595 };
596 }
597
598 return $pve_flags;
599 }
600
601 sub get_hyperv_enlightenments {
602 my ($winversion, $machine_version, $bios, $gpu_passthrough, $hv_vendor_id) = @_;
603
604 return if $winversion < 6;
605 return if $bios && $bios eq 'ovmf' && $winversion < 8;
606
607 my $flags = {};
608 my $default_reason = "automatic Hyper-V enlightenment for Windows";
609 my $flagfn = sub {
610 my ($flag, $value, $reason) = @_;
611 $flags->{$flag} = {
612 reason => $reason // $default_reason,
613 value => $value,
614 }
615 };
616
617 my $hv_vendor_set = defined($hv_vendor_id);
618 if ($gpu_passthrough || $hv_vendor_set) {
619 $hv_vendor_id //= 'proxmox';
620 $flagfn->('hv_vendor_id', $hv_vendor_id, $hv_vendor_set ?
621 "custom hv_vendor_id set" : "NVIDIA workaround for GPU passthrough");
622 }
623
624 if (min_version($machine_version, 2, 3)) {
625 $flagfn->('hv_spinlocks', '0x1fff');
626 $flagfn->('hv_vapic');
627 $flagfn->('hv_time');
628 } else {
629 $flagfn->('hv_spinlocks', '0xffff');
630 }
631
632 if (min_version($machine_version, 2, 6)) {
633 $flagfn->('hv_reset');
634 $flagfn->('hv_vpindex');
635 $flagfn->('hv_runtime');
636 }
637
638 if ($winversion >= 7) {
639 my $win7_reason = $default_reason . " 7 and higher";
640 $flagfn->('hv_relaxed', undef, $win7_reason);
641
642 if (min_version($machine_version, 2, 12)) {
643 $flagfn->('hv_synic', undef, $win7_reason);
644 $flagfn->('hv_stimer', undef, $win7_reason);
645 }
646
647 if (min_version($machine_version, 3, 1)) {
648 $flagfn->('hv_ipi', undef, $win7_reason);
649 }
650 }
651
652 return $flags;
653 }
654
655 sub get_cpu_from_running_vm {
656 my ($pid) = @_;
657
658 my $cmdline = PVE::QemuServer::Helpers::parse_cmdline($pid);
659 die "could not read commandline of running machine\n"
660 if !$cmdline->{cpu}->{value};
661
662 # sanitize and untaint value
663 $cmdline->{cpu}->{value} =~ $qemu_cmdline_cpu_re;
664 return $1;
665 }
666
667 __PACKAGE__->register();
668 __PACKAGE__->init();
669
670 1;