Make it possible to re-evaluate cpu_features.

Add cpuctl(4) ioctl CPUCTL_EVAL_CPU_FEATURES which forces re-read of
cpu_features, cpu_features2, cpu_stdext_features, and
std_stdext_features2.

The intent is to allow the kernel to see the changes in the CPU
features after micocode update.  Of course, the update is not atomic
across variables and not synchronized with readers.  See the man page
warning as well.

Reviewed by:	imp (previous version), jilles
Sponsored by:	The FreeBSD Foundation
MFC after:	1 week
Differential revision:	https://reviews.freebsd.org/D13770
This commit is contained in:
Konstantin Belousov 2018-01-05 21:06:19 +00:00
parent b4f55763ce
commit 0530a9360f
Notes: svn2git 2020-12-20 02:59:44 +00:00
svn path=/head/; revision=327597
7 changed files with 128 additions and 47 deletions

View file

@ -1535,7 +1535,7 @@ hammer_time(u_int64_t modulep, u_int64_t physfree)
kmdp = init_ops.parse_preload_data(modulep);
identify_cpu();
identify_cpu1();
identify_hypervisor();
/* Init basic tunables, hz etc */

View file

@ -73,6 +73,7 @@ static int cpuctl_do_cpuid(int cpu, cpuctl_cpuid_args_t *data,
struct thread *td);
static int cpuctl_do_cpuid_count(int cpu, cpuctl_cpuid_count_args_t *data,
struct thread *td);
static int cpuctl_do_eval_cpu_features(int cpu, struct thread *td);
static int cpuctl_do_update(int cpu, cpuctl_update_args_t *data,
struct thread *td);
static int update_intel(int cpu, cpuctl_update_args_t *args,
@ -159,7 +160,8 @@ cpuctl_ioctl(struct cdev *dev, u_long cmd, caddr_t data,
}
/* Require write flag for "write" requests. */
if ((cmd == CPUCTL_MSRCBIT || cmd == CPUCTL_MSRSBIT ||
cmd == CPUCTL_UPDATE || cmd == CPUCTL_WRMSR) &&
cmd == CPUCTL_UPDATE || cmd == CPUCTL_WRMSR ||
cmd == CPUCTL_EVAL_CPU_FEATURES) &&
(flags & FWRITE) == 0)
return (EPERM);
switch (cmd) {
@ -187,6 +189,9 @@ cpuctl_ioctl(struct cdev *dev, u_long cmd, caddr_t data,
ret = cpuctl_do_cpuid_count(cpu,
(cpuctl_cpuid_count_args_t *)data, td);
break;
case CPUCTL_EVAL_CPU_FEATURES:
ret = cpuctl_do_eval_cpu_features(cpu, td);
break;
default:
ret = EINVAL;
break;
@ -504,6 +509,29 @@ update_via(int cpu, cpuctl_update_args_t *args, struct thread *td)
return (ret);
}
static int
cpuctl_do_eval_cpu_features(int cpu, struct thread *td)
{
int is_bound = 0;
int oldcpu;
KASSERT(cpu >= 0 && cpu <= mp_maxid,
("[cpuctl,%d]: bad cpu number %d", __LINE__, cpu));
#ifdef __i386__
if (cpu_id == 0)
return (ENODEV);
#endif
oldcpu = td->td_oncpu;
is_bound = cpu_sched_is_bound(td);
set_cpu(cpu, td);
identify_cpu1();
identify_cpu2();
restore_cpu(oldcpu, is_bound, td);
return (0);
}
int
cpuctl_open(struct cdev *dev, int flags, int fmt __unused, struct thread *td)
{

View file

@ -59,5 +59,6 @@ typedef struct {
#define CPUCTL_MSRSBIT _IOWR('c', 5, cpuctl_msr_args_t)
#define CPUCTL_MSRCBIT _IOWR('c', 6, cpuctl_msr_args_t)
#define CPUCTL_CPUID_COUNT _IOWR('c', 7, cpuctl_cpuid_count_args_t)
#define CPUCTL_EVAL_CPU_FEATURES _IO('c', 8)
#endif /* _CPUCTL_H_ */

View file

@ -119,7 +119,8 @@ void cpu_setregs(void);
void dump_add_page(vm_paddr_t);
void dump_drop_page(vm_paddr_t);
void finishidentcpu(void);
void identify_cpu(void);
void identify_cpu1(void);
void identify_cpu2(void);
void identify_hypervisor(void);
void initializecpu(void);
void initializecpucache(void);

View file

@ -1385,9 +1385,8 @@ fix_cpuid(void)
return (false);
}
#ifdef __amd64__
void
identify_cpu(void)
identify_cpu1(void)
{
u_int regs[4];
@ -1404,7 +1403,29 @@ identify_cpu(void)
cpu_feature = regs[3];
cpu_feature2 = regs[2];
}
#endif
void
identify_cpu2(void)
{
u_int regs[4], cpu_stdext_disable;
if (cpu_high >= 7) {
cpuid_count(7, 0, regs);
cpu_stdext_feature = regs[1];
/*
* Some hypervisors failed to filter out unsupported
* extended features. Allow to disable the
* extensions, activation of which requires setting a
* bit in CR4, and which VM monitors do not support.
*/
cpu_stdext_disable = 0;
TUNABLE_INT_FETCH("hw.cpu_stdext_disable", &cpu_stdext_disable);
cpu_stdext_feature &= ~cpu_stdext_disable;
cpu_stdext_feature2 = regs[2];
}
}
/*
* Final stage of CPU identification.
@ -1412,7 +1433,7 @@ identify_cpu(void)
void
finishidentcpu(void)
{
u_int regs[4], cpu_stdext_disable;
u_int regs[4];
#ifdef __i386__
u_char ccr3;
#endif
@ -1431,22 +1452,7 @@ finishidentcpu(void)
cpu_mon_max_size = regs[1] & CPUID5_MON_MAX_SIZE;
}
if (cpu_high >= 7) {
cpuid_count(7, 0, regs);
cpu_stdext_feature = regs[1];
/*
* Some hypervisors failed to filter out unsupported
* extended features. Allow to disable the
* extensions, activation of which requires setting a
* bit in CR4, and which VM monitors do not support.
*/
cpu_stdext_disable = 0;
TUNABLE_INT_FETCH("hw.cpu_stdext_disable", &cpu_stdext_disable);
cpu_stdext_feature &= ~cpu_stdext_disable;
cpu_stdext_feature2 = regs[2];
}
identify_cpu2();
#ifdef __i386__
if (cpu_high > 0 &&

View file

@ -24,7 +24,7 @@
.\"
.\" $FreeBSD$
.\"
.Dd September 30, 2017
.Dd January 5, 2018
.Dt CPUCONTROL 8
.Os
.Sh NAME
@ -34,46 +34,51 @@
device
.Sh SYNOPSIS
.Nm
.Bk
.Op Fl v
.Fl m Ar msr
.Bk
.Ar device
.Ek
.Bk
.Nm
.Op Fl v
.Fl m Ar msr Ns = Ns Ar value
.Bk
.Ar device
.Ek
.Bk
.Nm
.Op Fl v
.Fl m Ar msr Ns &= Ns Ar mask
.Bk
.Ar device
.Ek
.Bk
.Nm
.Op Fl v
.Fl m Ar msr Ns |= Ns Ar mask
.Bk
.Ar device
.Ek
.Bk
.Nm
.Op Fl v
.Fl i Ar level
.Bk
.Ar device
.Ek
.Bk
.Nm
.Op Fl v
.Fl i Ar level,level_type
.Bk
.Ar device
.Ek
.Bk
.Nm
.Op Fl vn
.Op Fl d Ar datadir
.Fl u
.Ar device
.Ek
.Bk
.Nm
.Fl e
.Ar device
.Ek
.Sh DESCRIPTION
@ -136,6 +141,20 @@ The
.Nm
utility will walk through the configured data directories
and apply all firmware updates available for this CPU.
.It Fl e
Re-evaluate the kernel flags indicating the present CPU features.
This command is typically executed after a firmware update was applied
which changes information reported by the
.Dv CPUID
instruction.
.Pp
.Bf -symbolic
Only execute the
.Fl e
command after the microcode update was applied to all CPUs in the system.
The kernel does not operate correctly if the features of processors are
not identical.
.Ef
.It Fl v
Increase the verbosity level.
.It Fl h

View file

@ -63,6 +63,7 @@ int verbosity_level = 0;
#define FLAG_M 0x02
#define FLAG_U 0x04
#define FLAG_N 0x08
#define FLAG_E 0x10
#define OP_INVAL 0x00
#define OP_READ 0x01
@ -117,7 +118,7 @@ usage(void)
if (name == NULL)
name = "cpuctl";
fprintf(stderr, "Usage: %s [-vh] [-d datadir] [-m msr[=value] | "
"-i level | -i level,level_type | -u] device\n", name);
"-i level | -i level,level_type | -e | -u] device\n", name);
exit(EX_USAGE);
}
@ -340,6 +341,25 @@ do_msr(const char *cmdarg, const char *dev)
return (0);
}
static int
do_eval_cpu_features(const char *dev)
{
int fd, error;
assert(dev != NULL);
fd = open(dev, O_RDWR);
if (fd < 0) {
WARN(0, "error opening %s for writing", dev);
return (1);
}
error = ioctl(fd, CPUCTL_EVAL_CPU_FEATURES, NULL);
if (error < 0)
WARN(0, "ioctl(%s, CPUCTL_EVAL_CPU_FEATURES)", dev);
close(fd);
return (error);
}
static int
do_update(const char *dev)
{
@ -430,11 +450,14 @@ main(int argc, char *argv[])
error = 0;
cmdarg = ""; /* To keep gcc3 happy. */
while ((c = getopt(argc, argv, "d:hi:m:nuv")) != -1) {
while ((c = getopt(argc, argv, "d:ehi:m:nuv")) != -1) {
switch (c) {
case 'd':
datadir_add(optarg);
break;
case 'e':
flags |= FLAG_E;
break;
case 'i':
flags |= FLAG_I;
cmdarg = optarg;
@ -468,22 +491,25 @@ main(int argc, char *argv[])
if ((flags & FLAG_N) == 0)
datadir_add(DEFAULT_DATADIR);
dev = argv[0];
c = flags & (FLAG_I | FLAG_M | FLAG_U);
c = flags & (FLAG_E | FLAG_I | FLAG_M | FLAG_U);
switch (c) {
case FLAG_I:
if (strstr(cmdarg, ",") != NULL)
error = do_cpuid_count(cmdarg, dev);
else
error = do_cpuid(cmdarg, dev);
break;
case FLAG_M:
error = do_msr(cmdarg, dev);
break;
case FLAG_U:
error = do_update(dev);
break;
default:
usage(); /* Only one command can be selected. */
case FLAG_I:
if (strstr(cmdarg, ",") != NULL)
error = do_cpuid_count(cmdarg, dev);
else
error = do_cpuid(cmdarg, dev);
break;
case FLAG_M:
error = do_msr(cmdarg, dev);
break;
case FLAG_U:
error = do_update(dev);
break;
case FLAG_E:
error = do_eval_cpu_features(dev);
break;
default:
usage(); /* Only one command can be selected. */
}
SLIST_FREE(&datadirs, next, free);
return (error == 0 ? 0 : 1);