I stumbled upon an AMD box that had the BIOS using a hardware performance
counter. Instead of printing out a warning and continuing, it failed and
blocked further perf counter usage.
Looking through the history, I found this commit:
a5ebe0ba3d ("perf/x86: Check all MSRs before passing hw check")
which tweaked the rules for a Xen guest on an almost identical box and now
changed the behaviour.
Unfortunately the rules were tweaked incorrectly and will always lead to
MSR failures even though the MSRs are completely fine.
What happens now is in arch/x86/kernel/cpu/perf_event.c::check_hw_exists():
<snip>
for (i = 0; i < x86_pmu.num_counters; i++) {
reg = x86_pmu_config_addr(i);
ret = rdmsrl_safe(reg, &val);
if (ret)
goto msr_fail;
if (val & ARCH_PERFMON_EVENTSEL_ENABLE) {
bios_fail = 1;
val_fail = val;
reg_fail = reg;
}
}
<snip>
/*
* Read the current value, change it and read it back to see if it
* matches, this is needed to detect certain hardware emulators
* (qemu/kvm) that don't trap on the MSR access and always return 0s.
*/
reg = x86_pmu_event_addr(0);
^^^^
if the first perf counter is enabled, then this routine will always fail
because the counter is running. :-(
if (rdmsrl_safe(reg, &val))
goto msr_fail;
val ^= 0xffffUL;
ret = wrmsrl_safe(reg, val);
ret |= rdmsrl_safe(reg, &val_new);
if (ret || val != val_new)
goto msr_fail;
The above bios_fail used to be a 'goto' which is why it worked in the past.
Further, most vendors have migrated to using fixed counters to hide their
evilness hence this problem rarely shows up now days except on a few old boxes.
I fixed my problem and kept the spirit of the original Xen fix, by recording a
safe non-enable register to be used safely for the reading/writing check.
Because it is not enabled, this passes on bare metal boxes (like metal), but
should continue to throw an msr_fail on Xen guests because the register isn't
emulated yet.
Now I get a proper bios_fail error message and Xen should still see their
msr_fail message (untested).
Signed-off-by: Don Zickus <dzickus@redhat.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: george.dunlap@eu.citrix.com
Cc: konrad.wilk@oracle.com
Link: http://lkml.kernel.org/r/1431976608-56970-1-git-send-email-dzickus@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
The (SNB/IVB/HSW) HT bug only affects events that can be programmed
onto GP counters, therefore we should only limit the number of GP
counters that can be used per cpu -- iow we should not constrain the
FP counters.
Furthermore, we should only enfore such a limit when there are in fact
exclusive events being scheduled on either sibling.
Reported-by: Vince Weaver <vincent.weaver@maine.edu>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
[ Fixed build fail for the !CONFIG_CPU_SUP_INTEL case. ]
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Commit 43b4578071 ("perf/x86: Reduce stack usage of
x86_schedule_events()") violated the rule that 'fake' scheduling; as
used for event/group validation; should not change the event state.
This went mostly un-noticed because repeated calls of
x86_pmu::get_event_constraints() would give the same result. And
x86_pmu::put_event_constraints() would mostly not do anything.
Commit e979121b1b ("perf/x86/intel: Implement cross-HT corruption
bug workaround") made the situation much worse by actually setting the
event->hw.constraint value to NULL, so when validation and actual
scheduling interact we get NULL ptr derefs.
Fix it by removing the constraint pointer from the event and move it
back to an array, this time in cpuc instead of on the stack.
validate_group()
x86_schedule_events()
event->hw.constraint = c; # store
<context switch>
perf_task_event_sched_in()
...
x86_schedule_events();
event->hw.constraint = c2; # store
...
put_event_constraints(event); # assume failure to schedule
intel_put_event_constraints()
event->hw.constraint = NULL;
<context switch end>
c = event->hw.constraint; # read -> NULL
if (!test_bit(hwc->idx, c->idxmsk)) # <- *BOOM* NULL deref
This in particular is possible when the event in question is a
cpu-wide event and group-leader, where the validate_group() tries to
add an event to the group.
Reported-by: Vince Weaver <vincent.weaver@maine.edu>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Andrew Hunter <ahh@google.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Maria Dimakopoulou <maria.n.dimakopoulou@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Fixes: 43b4578071 ("perf/x86: Reduce stack usage of x86_schedule_events()")
Fixes: e979121b1b ("perf/x86/intel: Implement cross-HT corruption bug workaround")
Signed-off-by: Ingo Molnar <mingo@kernel.org>
The early_idt_handlers asm code generates an array of entry
points spaced nine bytes apart. It's not really clear from that
code or from the places that reference it what's going on, and
the code only works in the first place because GAS never
generates two-byte JMP instructions when jumping to global
labels.
Clean up the code to generate the correct array stride (member size)
explicitly. This should be considerably more robust against
screw-ups, as GAS will warn if a .fill directive has a negative
count. Using '. =' to advance would have been even more robust
(it would generate an actual error if it tried to move
backwards), but it would pad with nulls, confusing anyone who
tries to disassemble the code. The new scheme should be much
clearer to future readers.
While we're at it, improve the comments and rename the array and
common code.
Binutils may start relaxing jumps to non-weak labels. If so,
this change will fix our build, and we may need to backport this
change.
Before, on x86_64:
0000000000000000 <early_idt_handlers>:
0: 6a 00 pushq $0x0
2: 6a 00 pushq $0x0
4: e9 00 00 00 00 jmpq 9 <early_idt_handlers+0x9>
5: R_X86_64_PC32 early_idt_handler-0x4
...
48: 66 90 xchg %ax,%ax
4a: 6a 08 pushq $0x8
4c: e9 00 00 00 00 jmpq 51 <early_idt_handlers+0x51>
4d: R_X86_64_PC32 early_idt_handler-0x4
...
117: 6a 00 pushq $0x0
119: 6a 1f pushq $0x1f
11b: e9 00 00 00 00 jmpq 120 <early_idt_handler>
11c: R_X86_64_PC32 early_idt_handler-0x4
After:
0000000000000000 <early_idt_handler_array>:
0: 6a 00 pushq $0x0
2: 6a 00 pushq $0x0
4: e9 14 01 00 00 jmpq 11d <early_idt_handler_common>
...
48: 6a 08 pushq $0x8
4a: e9 d1 00 00 00 jmpq 120 <early_idt_handler_common>
4f: cc int3
50: cc int3
...
117: 6a 00 pushq $0x0
119: 6a 1f pushq $0x1f
11b: eb 03 jmp 120 <early_idt_handler_common>
11d: cc int3
11e: cc int3
11f: cc int3
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Acked-by: H. Peter Anvin <hpa@linux.intel.com>
Cc: Binutils <binutils@sourceware.org>
Cc: Borislav Petkov <bp@alien8.de>
Cc: H.J. Lu <hjl.tools@gmail.com>
Cc: Jan Beulich <JBeulich@suse.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/ac027962af343b0c599cbfcf50b945ad2ef3d7a8.1432336324.git.luto@kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Conflicts:
arch/x86/kernel/i387.c
This commit is conflicting:
e88221c50c ("x86/fpu: Disable XSAVES* support for now")
These functions changed a lot, move the quirk to arch/x86/kernel/fpu/init.c's
fpu__init_system_xstate_size_legacy().
Signed-off-by: Ingo Molnar <mingo@kernel.org>
The kernel's handling of 'compacted' xsave state layout is buggy:
http://marc.info/?l=linux-kernel&m=142967852317199
I don't have such a system, and the description there is vague, but
from extrapolation I guess that there were two kinds of bugs
observed:
- boot crashes, due to size calculations being wrong and the dynamic
allocation allocating a too small xstate area. (This is now fixed
in the new FPU code - but still present in stable kernels.)
- FPU state corruption and ABI breakage: if signal handlers try to
change the FPU state in standard format, which then the kernel
tries to restore in the compacted format.
These breakages are scary, but they only occur on a small number of
systems that have XSAVES* CPU support. Yet we have had XSAVES support
in the upstream kernel for a large number of stable kernel releases,
and the fixes are involved and unproven.
So do the safe resolution first: disable XSAVES* support and only
use the standard xstate format. This makes the code work and is
easy to backport.
On top of this we can work on enabling (and testing!) proper
compacted format support, without backporting pressure, on top of the
new, cleaned up FPU code.
Cc: <stable@vger.kernel.org>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Fenghua Yu <fenghua.yu@intel.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
This makes the functions kvm_guest_cpu_init and kvm_init_debugfs
static now due to having no external callers outside their
declarations in the file, kvm.c.
Signed-off-by: Nicholas Krause <xerofoify@gmail.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
We picked up a silent conflict in amdkfd with drm-fixes and drm-next,
backmerge v4.1-rc5 and fix the conflicts
Signed-off-by: Dave Airlie <airlied@redhat.com>
Conflicts:
drivers/gpu/drm/drm_irq.c
This reverts commit ff7bbb9c6a.
Sasha Levin is seeing odd jump in time values during boot of a KVM guest:
[...]
[ 0.000000] tsc: Detected 2260.998 MHz processor
[3376355.247558] Calibrating delay loop (skipped) preset value..
[...]
and bisected them to this commit.
Reported-by: Sasha Levin <sasha.levin@oracle.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
There are various internal FPU state debugging checks that never
trigger in practice, but which are useful for FPU code development.
Separate these out into CONFIG_X86_DEBUG_FPU=y, and also add a
couple of new ones.
The size difference is about 0.5K of code on defconfig:
text data bss filename
15028906 2578816 1638400 vmlinux
15029430 2578816 1638400 vmlinux
( Keep this enabled by default until the new FPU code is debugged. )
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Fenghua Yu <fenghua.yu@intel.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
The current xstate code in setup_xstate_features() assumes that
the first zero bit means the end of xfeatures - but that is not
so, the SDM clearly states that an arbitrary set of xfeatures
might be enabled - and it is also clear from the description
of the compaction feature that holes are possible:
"13-6 Vol. 1MANAGING STATE USING THE XSAVE FEATURE SET
[...]
Compacted format. Each state component i (i ≥ 2) is located at a byte
offset from the base address of the XSAVE area based on the XCOMP_BV
field in the XSAVE header:
— If XCOMP_BV[i] = 0, state component i is not in the XSAVE area.
— If XCOMP_BV[i] = 1, the following items apply:
• If XCOMP_BV[j] = 0 for every j, 2 ≤ j < i, state component i is
located at a byte offset 576 from the base address of the XSAVE
area. (This item applies if i is the first bit set in bits 62:2 of
the XCOMP_BV; it implies that state component i is located at the
beginning of the extended region.)
• Otherwise, let j, 2 ≤ j < i, be the greatest value such that
XCOMP_BV[j] = 1. Then state component i is located at a byte offset
X from the location of state component j, where X is the number of
bytes required for state component j as enumerated in
CPUID.(EAX=0DH,ECX=j):EAX. (This item implies that state component i
immediately follows the preceding state component whose bit is set
in XCOMP_BV.)"
So don't assume that the first zero xfeatures bit means the end of
all xfeatures - iterate through all of them.
I'm not aware of hardware that triggers this currently.
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Fenghua Yu <fenghua.yu@intel.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Improve the memory layout of 'struct fpu':
- change ->fpregs_active from 'int' to 'char' - it's just a single flag
and modern x86 CPUs can do efficient byte accesses.
- pack related fields closer to each other: often 'fpu->state' will not be
touched, while the other fields will - so pack them into a group.
Also add comments to each field, describing their purpose, and add
some background information about lazy restores.
Also fix an obsolete, lazy switching related comment in fpu_copy()'s description.
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Fenghua Yu <fenghua.yu@intel.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
So the handling of init_xstate_ctx has a layering violation: both
'struct xsave_struct' and 'union thread_xstate' have a
'struct i387_fxsave_struct' member:
xsave_struct::i387
thread_xstate::fxsave
The handling of init_xstate_ctx is generic, it is used on all
CPUs, with or without XSAVE instruction. So it's confusing how
the generic code passes around and handles an XSAVE specific
format.
What we really want is for init_xstate_ctx to be a proper
fpstate and we use its ::fxsave and ::xsave members, as
appropriate.
Since the xsave_struct::i387 and thread_xstate::fxsave aliases
each other this is not a functional problem.
So implement this, and move init_xstate_ctx to the generic FPU
code in the process.
Also, since init_xstate_ctx is not XSAVE specific anymore,
rename it to init_fpstate, and mark it __read_mostly,
because it's only modified once during bootup, and used
as a reference fpstate later on.
There's no change in functionality.
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Fenghua Yu <fenghua.yu@intel.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
With recent cleanups and fixes the fpu__reset() and fpu__clear()
functions have become almost identical in functionality: the only
difference is that fpu__reset() assumed that the fpstate
was already active in the eagerfpu case, while fpu__clear()
activated it if it was inactive.
This distinction almost never matters, the only case where such
fpstate activation happens if if the init thread (PID 1) gets exec()-ed
for the first time.
So keep fpu__clear() and change all fpu__reset() uses to
fpu__clear() to simpify the logic.
( In a later patch we'll further simplify fpu__clear() by making
sure that all contexts it is called on are already active. )
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Fenghua Yu <fenghua.yu@intel.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
The fpu__*() methods are closely related, but they are defined
in scattered places within the FPU code.
Concentrate them, and also uninline fpu__save(), fpu__drop()
and fpu__reset() to save about 5K of kernel text on 64-bit kernels:
text data bss dec filename
14113063 2575280 1634304 18322647 vmlinux.before
14108070 2575280 1634304 18317654 vmlinux.after
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Fenghua Yu <fenghua.yu@intel.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
