The following change fixes the x86 implementation of
trigger_all_cpu_backtrace(), which was previously (accidentally,
as far as I can tell) disabled to always return false as on
architectures that do not implement this function.
trigger_all_cpu_backtrace(), as defined in include/linux/nmi.h,
should call arch_trigger_all_cpu_backtrace() if available, or
return false if the underlying arch doesn't implement this
function.
x86 did provide a suitable arch_trigger_all_cpu_backtrace()
implementation, but it wasn't actually being used because it was
declared in asm/nmi.h, which linux/nmi.h doesn't include. Also,
linux/nmi.h couldn't easily be fixed by including asm/nmi.h,
because that file is not available on all architectures.
I am proposing to fix this by moving the x86 definition of
arch_trigger_all_cpu_backtrace() to asm/irq.h.
Tested via: echo l > /proc/sysrq-trigger
Before the change, this uses a fallback implementation which
shows backtraces on active CPUs (using
smp_call_function_interrupt() )
After the change, this shows NMI backtraces on all CPUs
Signed-off-by: Michel Lespinasse <walken@google.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Link: http://lkml.kernel.org/r/1370518875-1346-1-git-send-email-walken@google.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
arch/x86/kernel/cpu/intel_cacheinfo.c: In function ‘init_intel_cacheinfo’:
arch/x86/kernel/cpu/intel_cacheinfo.c:642:28: warning: ‘this_leaf.size’ may be used uninitialized in this function [-Wmaybe-uninitialized] arch/x86/kernel/cpu/intel_cacheinfo.c:643:29: warning: ‘this_leaf.eax.split.num_threads_sharing’ may be used uninitialized in this function [-Wmaybe-uninitialized]
This keeps on happening during randbuilds and the compiler is
wrong here:
In the case where cpuid4_cache_lookup_regs() returns 0, both
this_leaf.size and this_leaf.eax get initialized. In the case
where the CPUID leaf doesn't contain valid cache info, we error
out which init_intel_cacheinfo() handles correctly without
touching the abovementioned fields.
So shut up the warning by clearing out the struct which we hand
down.
While at it, reverse error handling and gain one indentation
level.
Signed-off-by: Borislav Petkov <bp@suse.de>
Link: http://lkml.kernel.org/r/1370710095-20547-1-git-send-email-bp@alien8.de
Signed-off-by: Ingo Molnar <mingo@kernel.org>
mem-loads is basically the same as Sandy Bridge,
but we use a separate string for changes later.
Haswell doesn't support the full precise store mode,
so we emulate it using the "DataLA" facility.
This allows to do everything, but for data sources we
can only detect L1 hit or not.
There is no explicit enable bit anymore, so we have
to tie it to a perf internal only flag.
The address is supported for all memory related PEBS
events with DataLA. Instead of only logging for the
load and store events we allow logging it for all
(it will be simply 0 if the current event does not
support it)
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Cc: Andi Kleen <ak@linux.jf.intel.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Link: http://lkml.kernel.org/r/1371515812-9646-7-git-send-email-andi@firstfloor.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Add support for the Haswell extended (fmt2) PEBS format.
It has a superset of the nhm (fmt1) PEBS fields, but has a
longer record so we need to adjust the code paths.
The main advantage is the new "EventingRip" support which
directly gives the instruction, not off-by-one instruction. So
with precise == 2 we use that directly and don't try to use LBRs
and walking basic blocks. This lowers the overhead of using
precise significantly.
Some other features are added in later patches.
Reviewed-by: Stephane Eranian <eranian@google.com>
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Cc: Andi Kleen <ak@linux.jf.intel.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Link: http://lkml.kernel.org/r/1371515812-9646-2-git-send-email-andi@firstfloor.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Implement a perf PMU to handle IOMMU performance counters and events.
The PMU only supports counting mode (e.g. perf stat). Since the counters
are shared across all cores, the PMU is implemented as "system-wide" mode.
To invoke the AMD IOMMU PMU, issue a perf tool command such as:
./perf stat -a -e amd_iommu/<events>/ <command>
or:
./perf stat -a -e amd_iommu/config=<config-data>,config1=<config1-data>/ <command>
For example:
./perf stat -a -e amd_iommu/mem_trans_total/ <command>
The resulting count will be how many IOMMU total peripheral memory
operations were performed during the command execution window.
Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Signed-off-by: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1370466709-3212-3-git-send-email-suravee.suthikulpanit@amd.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
intel_pmu_handle_irq() has a warning in it if it does too many
loops. It is a WARN_ONCE(), but the perf_event_print_debug()
call beneath it is unconditional. For the first warning, you get
a nice backtrace and message, but subsequent ones just dump the
PMU state with no leading messages. I doubt this is what was
intended.
This patch will only print the PMU state when paired with the
WARN_ON() text. It effectively open-codes WARN_ONCE()'s
one-time-only logic.
My suspicion is that the code really just wants to make sure we
do not sit in the loop and spit out a warning for every loop
iteration after the 100th. From what I've seen, this is very
unlikely to happen since we also clear the PMU state.
After this patch, instead of seeing the PMU state dumped each
time, you will just see:
[57494.894540] perf_event_intel: clearing PMU state on CPU#129
[57579.539668] perf_event_intel: clearing PMU state on CPU#10
[57587.137762] perf_event_intel: clearing PMU state on CPU#134
[57623.039912] perf_event_intel: clearing PMU state on CPU#114
[57644.559943] perf_event_intel: clearing PMU state on CPU#118
...
Signed-off-by: Dave Hansen <dave.hansen@linux.intel.com>
Signed-off-by: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20130530174559.0DB049F4@viggo.jf.intel.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
x86_schedule_events() caches event constraints on the stack during
scheduling. Given the number of possible events, this is 512 bytes of
stack; since it can be invoked under schedule() under god-knows-what,
this is causing stack blowouts.
Trade some space usage for stack safety: add a place to cache the
constraint pointer to struct perf_event. For 8 bytes per event (1% of
its size) we can save the giant stack frame.
This shouldn't change any aspect of scheduling whatsoever and while in
theory the locality's a tiny bit worse, I doubt we'll see any
performance impact either.
Tested: `perf stat whatever` does not blow up and produces
results that aren't hugely obviously wrong. I'm not sure how to run
particularly good tests of perf code, but this should not produce any
functional change whatsoever.
Signed-off-by: Andrew Hunter <ahh@google.com>
Reviewed-by: Stephane Eranian <eranian@google.com>
Signed-off-by: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1369332423-4400-1-git-send-email-ahh@google.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
This patch fixes broken support of PEBS-LL on SNB-EP/IVB-EP.
For some reason, the LDLAT extra reg definition for snb_ep
showed up as duplicate in the snb table.
This patch moves the definition of LDLAT back into the
snb_ep table.
Thanks to Don Zickus for tracking this one down.
Signed-off-by: Stephane Eranian <eranian@google.com>
Signed-off-by: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20130607212210.GA11849@quad
Signed-off-by: Ingo Molnar <mingo@kernel.org>
kernel might hung in pvclock_clocksource_read() due to
uninitialized memory might contain odd version value in
following cycle:
do {
version = __pvclock_read_cycles(src, &ret, &flags);
} while ((src->version & 1) || version != src->version);
if secondary kvmclock is accessed before it's registered with kvm.
Clear garbage in pvclock shared memory area right after it's
allocated to avoid this issue.
Ref: https://bugzilla.kernel.org/show_bug.cgi?id=59521
Signed-off-by: Igor Mammedov <imammedo@redhat.com>
[See BZ for analysis. We may want a different fix for 3.11, but
this is the safest for now - Paolo]
Cc: <stable@vger.kernel.org> # 3.8
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Joshua reported: Commit cd7b304dfaf1 (x86, range: fix missing merge
during add range) broke mtrr cleanup on his setup in 3.9.5.
corresponding commit in upstream is fbe06b7bae.
*BAD*gran_size: 64K chunk_size: 16M num_reg: 6 lose cover RAM: -0G
https://bugzilla.kernel.org/show_bug.cgi?id=59491
So it rejects new var mtrr layout.
It turns out we have some problem with initial mtrr range retrieval.
The current sequence is:
x86_get_mtrr_mem_range
==> bunchs of add_range_with_merge
==> bunchs of subract_range
==> clean_sort_range
add_range_with_merge for [0,1M)
sort_range()
add_range_with_merge could have blank slots, so we can not just
sort only, that will have final result have extra blank slot in head.
So move that calling add_range_with_merge for [0,1M), with that we
could avoid extra clean_sort_range calling.
Reported-by: Joshua Covington <joshuacov@googlemail.com>
Tested-by: Joshua Covington <joshuacov@googlemail.com>
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Link: http://lkml.kernel.org/r/1371154622-8929-2-git-send-email-yinghai@kernel.org
Cc: <stable@vger.kernel.org> v3.9
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
The package power limit notification interrupt is primarily for
system diagnosis, and should not be blindly enabled on every
system by default -- particuarly since Linux does nothing in the
handler except count how many times it has been called...
Add a new kernel cmdline parameter "int_pln_enable" for situations where
users want to oberve these events via existing system counters:
$ grep TRM /proc/interrupts
$ grep . /sys/devices/system/cpu/cpu*/thermal_throttle/*
https://bugzilla.kernel.org/show_bug.cgi?id=36182
Signed-off-by: Fenghua Yu <fenghua.yu@intel.com>
Signed-off-by: Len Brown <len.brown@intel.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
Package power limits are common on some systems under some conditions --
so printing console messages when limits are reached
causes unnecessary customer concern and support calls.
Note that even with these console messages gone,
the events can still be observed via system counters:
$ grep TRM /proc/interrupts
Shows total thermal interrupts, which includes both power
limit notifications and thermal throttling interrupts.
$ grep . /sys/devices/system/cpu/cpu*/thermal_throttle/*
Will show what caused those interrupts, core and package
throttling and power limit notifications.
https://bugzilla.kernel.org/show_bug.cgi?id=36182
Signed-off-by: Fenghua Yu <fenghua.yu@intel.com>
Signed-off-by: Len Brown <len.brown@intel.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
Added callback registration for package threshold reports. Also added
a callback to check the rate control implemented in callback or not.
If there is no rate control implemented, then there is a default rate
control similar to core threshold notification by delaying for
CHECK_INTERVAL (5 minutes) between reports.
Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
Signed-off-by: Zhang Rui <rui.zhang@intel.com>
Moving x86 to the generic idle implementation (commit 7d1a9417 "x86:
Use generic idle loop") wreckaged the stack protector.
I stupidly missed that boot_init_stack_canary() must be inlined from a
function which never returns, but I put that call into
arch_cpu_idle_prepare() which of course returns.
I pondered to play tricks with arch_cpu_idle_prepare() first, but then
I noticed, that the other archs which have implemented the
stackprotector (ARM and SH) do not initialize the canary for the
non-boot cpus.
So I decided to move the boot_init_stack_canary() call into
cpu_startup_entry() ifdeffed with an CONFIG_X86 for now. This #ifdef
is just a temporary measure as I don't want to inflict the
boot_init_stack_canary() call on ARM and SH that late in the cycle.
I'll queue a patch for 3.11 which removes the #ifdef if the ARM/SH
maintainers have no objection.
Reported-by: Wouter van Kesteren <woutershep@gmail.com>
Cc: x86@kernel.org
Cc: Russell King <linux@arm.linux.org.uk>
Cc: Paul Mundt <lethal@linux-sh.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Commit 316ad24830 ("sched/x86: Rewrite
set_cpu_sibling_map()") broke the construction of sibling maps,
which also broke the booted_cores accounting.
Before the rewrite, if smt was present, then each map was
updated for each smt sibling. After the rewrite only
cpu_sibling_mask gets updated, as the llc and core maps depend
on 'has_mc = x86_max_cores > 1' instead. This leads to problems
with topologies like the following
(qemu -smp sockets=2,cores=1,threads=2)
processor : 0
physical id : 0
siblings : 1 <= should be 2
core id : 0
cpu cores : 1
processor : 1
physical id : 0
siblings : 1 <= should be 2
core id : 0
cpu cores : 0 <= should be 1
processor : 2
physical id : 1
siblings : 1 <= should be 2
core id : 0
cpu cores : 1
processor : 3
physical id : 1
siblings : 1 <= should be 2
core id : 0
cpu cores : 0 <= should be 1
This patch restores the former construction by defining has_mc
as (has_smt || x86_max_cores > 1). This should be fine as there
were no (has_smt && !has_mc) conditions in the context.
Aso rename has_mc to has_mp now that it's not just for cores.
Signed-off-by: Andrew Jones <drjones@redhat.com>
Acked-by: Peter Zijlstra <peterz@infradead.org>
Cc: a.p.zijlstra@chello.nl
Cc: fenghua.yu@intel.com
Link: http://lkml.kernel.org/r/1369831695-11970-1-git-send-email-drjones@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Several drivers currently use mtrr_add through various #ifdef guards
and/or drm wrappers. The vast majority of them want to add WC MTRRs
on x86 systems and don't actually need the MTRR if PAT (i.e.
ioremap_wc, etc) are working.
arch_phys_wc_add and arch_phys_wc_del are new functions, available
on all architectures and configurations, that add WC MTRRs on x86 if
needed (and handle errors) and do nothing at all otherwise. They're
also easier to use than mtrr_add and mtrr_del, so the call sites can
be simplified.
As an added benefit, this will avoid wasting MTRRs and possibly
warning pointlessly on PAT-supporting systems.
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Andy Lutomirski <luto@amacapital.net>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Pull x86 fixes from Peter Anvin:
- Three EFI-related fixes
- Two early memory initialization fixes
- build fix for older binutils
- fix for an eager FPU performance regression -- currently we don't
allow the use of the FPU at interrupt time *at all* in eager mode,
which is clearly wrong.
* 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
x86: Allow FPU to be used at interrupt time even with eagerfpu
x86, crc32-pclmul: Fix build with older binutils
x86-64, init: Fix a possible wraparound bug in switchover in head_64.S
x86, range: fix missing merge during add range
x86, efi: initial the local variable of DataSize to zero
efivar: fix oops in efivar_update_sysfs_entries() caused by memory reuse
efivarfs: Never return ENOENT from firmware again
With the addition of eagerfpu the irq_fpu_usable() now returns false
negatives especially in the case of ksoftirqd and interrupted idle task,
two common cases for FPU use for example in networking/crypto. With
eagerfpu=off FPU use is possible in those contexts. This is because of
the eagerfpu check in interrupted_kernel_fpu_idle():
...
* For now, with eagerfpu we will return interrupted kernel FPU
* state as not-idle. TBD: Ideally we can change the return value
* to something like __thread_has_fpu(current). But we need to
* be careful of doing __thread_clear_has_fpu() before saving
* the FPU etc for supporting nested uses etc. For now, take
* the simple route!
...
if (use_eager_fpu())
return 0;
As eagerfpu is automatically "on" on those CPUs that also have the
features like AES-NI this patch changes the eagerfpu check to return 1 in
case the kernel_fpu_begin() has not been said yet. Once it has been the
__thread_has_fpu() will start returning 0.
Notice that with eagerfpu the __thread_has_fpu is always true initially.
FPU use is thus always possible no matter what task is under us, unless
the state has already been saved with kernel_fpu_begin().
[ hpa: this is a performance regression, not a correctness regression,
but since it can be quite serious on CPUs which need encryption at
interrupt time I am marking this for urgent/stable. ]
Signed-off-by: Pekka Riikonen <priikone@iki.fi>
Link: http://lkml.kernel.org/r/alpine.GSO.2.00.1305131356320.18@git.silcnet.org
Cc: <stable@vger.kernel.org> v3.7+
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
In head_64.S, a switchover has been used to handle kernel crossing
1G, 512G boundaries.
And commit 8170e6bed4
x86, 64bit: Use a #PF handler to materialize early mappings on demand
said:
During the switchover in head_64.S, before #PF handler is available,
we use three pages to handle kernel crossing 1G, 512G boundaries with
sharing page by playing games with page aliasing: the same page is
mapped twice in the higher-level tables with appropriate wraparound.
But from the switchover code, when we set up the PUD table:
114 addq $4096, %rdx
115 movq %rdi, %rax
116 shrq $PUD_SHIFT, %rax
117 andl $(PTRS_PER_PUD-1), %eax
118 movq %rdx, (4096+0)(%rbx,%rax,8)
119 movq %rdx, (4096+8)(%rbx,%rax,8)
It seems line 119 has a potential bug there. For example,
if the kernel is loaded at physical address 511G+1008M, that is
000000000 111111111 111111000 000000000000000000000
and the kernel _end is 512G+2M, that is
000000001 000000000 000000001 000000000000000000000
So in this example, when using the 2nd page to setup PUD (line 114~119),
rax is 511.
In line 118, we put rdx which is the address of the PMD page (the 3rd page)
into entry 511 of the PUD table. But in line 119, the entry we calculate from
(4096+8)(%rbx,%rax,8) has exceeded the PUD page. IMO, the entry in line
119 should be wraparound into entry 0 of the PUD table.
The patch fixes the bug.
Signed-off-by: Zhang Yanfei <zhangyanfei@cn.fujitsu.com>
Link: http://lkml.kernel.org/r/5191DE5A.3020302@cn.fujitsu.com
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Cc: <stable@vger.kernel.org> v3.9
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
All the virtualized platforms (KVM, lguest and Xen) have persistent
wallclocks that have more than one second of precision.
read_persistent_wallclock() and update_persistent_wallclock() allow
for nanosecond precision but their implementation on x86 with
x86_platform.get/set_wallclock() only allows for one second precision.
This means guests may see a wallclock time that is off by up to 1
second.
Make set_wallclock() and get_wallclock() take a struct timespec
parameter (which allows for nanosecond precision) so KVM and Xen
guests may start with a more accurate wallclock time and a Xen dom0
can maintain a more accurate wallclock for guests.
Signed-off-by: David Vrabel <david.vrabel@citrix.com>
Signed-off-by: John Stultz <john.stultz@linaro.org>
We're trying to use 64 bit masks but the shifts wrap so we can't use the
high 32 bits. I've fixed this by changing several types to unsigned
long long.
This is a static checker fix. The one change which is clearly needed is
"mask = 0xff << (idx * 8);" where the author obviously intended to use
all 64 bits. The other changes are mostly to silence my static checker.
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Stephane Eranian <eranian@google.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Link: http://lkml.kernel.org/r/20130518183452.GA14587@elgon.mountain
Signed-off-by: Ingo Molnar <mingo@kernel.org>
In commit 78d77df715 ("x86-64, init: Do not set NX bits on non-NX
capable hardware") we added the early_pmd_flags that gets the NX bit set
when a CPU supports NX. However, the new variable was marked __initdata,
because the main _use_ of this is in an __init routine.
However, the bit setting happens from secondary_startup_64(), which is
called not only at bootup, but on every secondary CPU start. Including
resuming from STR and at CPU hotplug time. So the value cannot be
__initdata.
Reported-bisected-and-tested-by: Michal Hocko <mhocko@suse.cz>
Cc: stable@vger.kernel.org # v3.9
Acked-by: Peter Anvin <hpa@linux.intel.com>
Cc: Fernando Luis Vázquez Cao <fernando@oss.ntt.co.jp>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
