There are many use cases where people will want to monitor more tasks
than there exist RMIDs in the hardware, meaning that we have to perform
some kind of multiplexing.
We do this by "rotating" the RMIDs in a workqueue, and assigning an RMID
to a waiting event when the RMID becomes unused.
This scheme reserves one RMID at all times for rotation. When we need to
schedule a new event we give it the reserved RMID, pick a victim event
from the front of the global CQM list and wait for the victim's RMID to
drop to zero occupancy, before it becomes the new reserved RMID.
We put the victim's RMID onto the limbo list, where it resides for a
"minimum queue time", which is intended to save ourselves an expensive
smp IPI when the RMID is unlikely to have a occupancy value below
__intel_cqm_threshold.
If we fail to recycle an RMID, even after waiting the minimum queue time
then we need to increment __intel_cqm_threshold. There is an upper bound
on this threshold, __intel_cqm_max_threshold, which is programmable from
userland as /sys/devices/intel_cqm/max_recycling_threshold.
The comments above __intel_cqm_rmid_rotate() have more details.
Signed-off-by: Matt Fleming <matt.fleming@intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Arnaldo Carvalho de Melo <acme@kernel.org>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Kanaka Juvva <kanaka.d.juvva@intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Vikas Shivappa <vikas.shivappa@linux.intel.com>
Link: http://lkml.kernel.org/r/1422038748-21397-9-git-send-email-matt@codeblueprint.co.uk
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Add support for task events as well as system-wide events. This change
has a big impact on the way that we gather LLC occupancy values in
intel_cqm_event_read().
Currently, for system-wide (per-cpu) events we defer processing to
userspace which knows how to discard all but one cpu result per package.
Things aren't so simple for task events because we need to do the value
aggregation ourselves. To do this, we defer updating the LLC occupancy
value in event->count from intel_cqm_event_read() and do an SMP
cross-call to read values for all packages in intel_cqm_event_count().
We need to ensure that we only do this for one task event per cache
group, otherwise we'll report duplicate values.
If we're a system-wide event we want to fallback to the default
perf_event_count() implementation. Refactor this into a common function
so that we don't duplicate the code.
Also, introduce PERF_TYPE_INTEL_CQM, since we need a way to track an
event's task (if the event isn't per-cpu) inside of the Intel CQM PMU
driver. This task information is only availble in the upper layers of
the perf infrastructure.
Other perf backends stash the target task in event->hw.*target so we
need to do something similar. The task is used to determine whether
events should share a cache group and an RMID.
Signed-off-by: Matt Fleming <matt.fleming@intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Arnaldo Carvalho de Melo <acme@kernel.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Kanaka Juvva <kanaka.d.juvva@intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Vikas Shivappa <vikas.shivappa@linux.intel.com>
Cc: linux-api@vger.kernel.org
Link: http://lkml.kernel.org/r/1422038748-21397-8-git-send-email-matt@codeblueprint.co.uk
Signed-off-by: Ingo Molnar <mingo@kernel.org>
This is based on a patch originally by hpa.
With the current improvements to the alternatives, we can simply use %P1
as a mem8 operand constraint and rely on the toolchain to generate the
proper instruction sizes. For example, on 32-bit, where we use an empty
old instruction we get:
apply_alternatives: feat: 6*32+8, old: (c104648b, len: 4), repl: (c195566c, len: 4)
c104648b: alt_insn: 90 90 90 90
c195566c: rpl_insn: 0f 0d 4b 5c
...
apply_alternatives: feat: 6*32+8, old: (c18e09b4, len: 3), repl: (c1955948, len: 3)
c18e09b4: alt_insn: 90 90 90
c1955948: rpl_insn: 0f 0d 08
...
apply_alternatives: feat: 6*32+8, old: (c1190cf9, len: 7), repl: (c1955a79, len: 7)
c1190cf9: alt_insn: 90 90 90 90 90 90 90
c1955a79: rpl_insn: 0f 0d 0d a0 d4 85 c1
all with the proper padding done depending on the size of the
replacement instruction the compiler generates.
Signed-off-by: Borislav Petkov <bp@suse.de>
Cc: H. Peter Anvin <hpa@linux.intel.com>
Pull misc x86 fixes from Ingo Molnar:
"This contains:
- EFI fixes
- a boot printout fix
- ASLR/kASLR fixes
- intel microcode driver fixes
- other misc fixes
Most of the linecount comes from an EFI revert"
* 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
x86/mm/ASLR: Avoid PAGE_SIZE redefinition for UML subarch
x86/microcode/intel: Handle truncated microcode images more robustly
x86/microcode/intel: Guard against stack overflow in the loader
x86, mm/ASLR: Fix stack randomization on 64-bit systems
x86/mm/init: Fix incorrect page size in init_memory_mapping() printks
x86/mm/ASLR: Propagate base load address calculation
Documentation/x86: Fix path in zero-page.txt
x86/apic: Fix the devicetree build in certain configs
Revert "efi/libstub: Call get_memory_map() to obtain map and desc sizes"
x86/efi: Avoid triple faults during EFI mixed mode calls
Pull microcode fixes from Borislav Petkov:
- Two fixes hardening microcode data handling. (Quentin Casasnovas)
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Pull RAS updates from Borislav Petkov:
"- Enable AMD thresholding IRQ by default if supported. (Aravind Gopalakrishnan)
- Unify mce_panic() message pattern. (Derek Che)
- A bit more involved simplification of the CMCI logic after yet another
report about race condition with the adaptive logic. (Borislav Petkov)
- ACPI APEI EINJ fleshing out of the user documentation. (Borislav Petkov)
- Minor cleanup. (Jan Beulich.)"
Signed-off-by: Ingo Molnar <mingo@kernel.org>
We setup APIC vectors for threshold errors if interrupt_capable.
However, we don't set interrupt_enable by default. Rework
threshold_restart_bank() so that when we set up lvt_offset, we also set
IntType to APIC and also enable thresholding interrupts for banks which
support it by default.
User is still allowed to disable interrupts through sysfs.
While at it, check if status is valid before we proceed to log error
using mce_log. This is because, in multi-node platforms, only the NBC
(Node Base Core, i.e. the first core in the node) has valid status info
in its MCA registers. So, the decoding of status values on the non-NBC
leads to noise on kernel logs like so:
EDAC DEBUG: amd64_inject_write_store: section=0x80000000 word_bits=0x10020001
[Hardware Error]: Corrected error, no action required.
[Hardware Error]: CPU:25 (15:2:0) MC4_STATUS[-|CE|-|-|-
[Hardware Error]: Corrected error, no action required.
[Hardware Error]: CPU:26 (15:2:0) MC4_STATUS[-|CE|-|-|-
<...>
WARNING: CPU: 25 PID: 0 at drivers/edac/amd64_edac.c:2147 decode_bus_error+0x1ba/0x2a0()
WARNING: CPU: 26 PID: 0 at drivers/edac/amd64_edac.c:2147 decode_bus_error+0x1ba/0x2a0()
Something is rotten in the state of Denmark.
Suggested-by: Borislav Petkov <bp@suse.de>
Signed-off-by: Aravind Gopalakrishnan <aravind.gopalakrishnan@amd.com>
Link: http://lkml.kernel.org/r/1422896561-7695-1-git-send-email-aravind.gopalakrishnan@amd.com
[ Massage commit message. ]
Signed-off-by: Borislav Petkov <bp@suse.de>
There is another mce_panic call with "Fatal machine check on current CPU" in
the same mce.c file, why not keep them all in same pattern
mce_panic("Fatal machine check on current CPU", &m, msg);
Signed-off-by: Derek Che <drc@yahoo-inc.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
Signed-off-by: Borislav Petkov <bp@suse.de>
Initially, this started with the yet another report about a race
condition in the CMCI storm adaptive period length thing. Yes, we have
to admit, it is fragile and error prone. So let's simplify it.
The simpler logic is: now, after we enter storm mode, we go straight to
polling with CMCI_STORM_INTERVAL, i.e. once a second. We remain in storm
mode as long as we see errors being logged while polling.
Theoretically, if we see an uninterrupted error stream, we will remain
in storm mode indefinitely and keep polling the MSRs.
However, when the storm is actually a burst of errors, once we have
logged them all, we back out of it after ~5 mins of polling and no more
errors logged.
If we encounter an error during those 5 minutes, we reset the polling
interval to 5 mins.
Making machine_check_poll() return a bool and denoting whether it has
seen an error or not lets us simplify a bunch of code and move the storm
handling private to mce_intel.c.
Some minor cleanups while at it.
Reported-by: Calvin Owens <calvinowens@fb.com>
Tested-by: Tony Luck <tony.luck@intel.com>
Link: http://lkml.kernel.org/r/1417746575-23299-1-git-send-email-calvinowens@fb.com
Signed-off-by: Borislav Petkov <bp@suse.de>
Haswell has a new feature that utilizes the existing LBR facility to
record call chains. To enable this feature, bits (JCC, NEAR_IND_JMP,
NEAR_REL_JMP, FAR_BRANCH, EN_CALLSTACK) in LBR_SELECT must be set to 1,
bits (NEAR_REL_CALL, NEAR-IND_CALL, NEAR_RET) must be cleared. Due to
a hardware bug of Haswell, this feature doesn't work well with
FREEZE_LBRS_ON_PMI.
When the call stack feature is enabled, the LBR stack will capture
unfiltered call data normally, but as return instructions are executed,
the last captured branch record is flushed from the on-chip registers
in a last-in first-out (LIFO) manner. Thus, branch information relative
to leaf functions will not be captured, while preserving the call stack
information of the main line execution path.
This patch defines a separate lbr_sel map for Haswell. The map contains
a new entry for the call stack feature.
Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com>
Signed-off-by: Kan Liang <kan.liang@intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Arnaldo Carvalho de Melo <acme@kernel.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: eranian@google.com
Cc: jolsa@redhat.com
Link: http://lkml.kernel.org/r/1415156173-10035-5-git-send-email-kan.liang@intel.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Pull mcelog regression fix from Tony Luck:
"Fix regression - functions on the mce notifier chain should not be
able to decide that an event should not be logged"
* tag 'please-pull-fixmcelog' of git://git.kernel.org/pub/scm/linux/kernel/git/ras/ras:
x86/mce: Fix regression. All error records should report via /dev/mcelog
Pull x86 perf updates from Ingo Molnar:
"This series tightens up RDPMC permissions: currently even highly
sandboxed x86 execution environments (such as seccomp) have permission
to execute RDPMC, which may leak various perf events / PMU state such
as timing information and other CPU execution details.
This 'all is allowed' RDPMC mode is still preserved as the
(non-default) /sys/devices/cpu/rdpmc=2 setting. The new default is
that RDPMC access is only allowed if a perf event is mmap-ed (which is
needed to correctly interpret RDPMC counter values in any case).
As a side effect of these changes CR4 handling is cleaned up in the
x86 code and a shadow copy of the CR4 value is added.
The extra CR4 manipulation adds ~ <50ns to the context switch cost
between rdpmc-capable and rdpmc-non-capable mms"
* 'perf-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
perf/x86: Add /sys/devices/cpu/rdpmc=2 to allow rdpmc for all tasks
perf/x86: Only allow rdpmc if a perf_event is mapped
perf: Pass the event to arch_perf_update_userpage()
perf: Add pmu callbacks to track event mapping and unmapping
x86: Add a comment clarifying LDT context switching
x86: Store a per-cpu shadow copy of CR4
x86: Clean up cr4 manipulation
printk and friends can now format bitmaps using '%*pb[l]'. cpumask
and nodemask also provide cpumask_pr_args() and nodemask_pr_args()
respectively which can be used to generate the two printf arguments
necessary to format the specified cpu/nodemask.
* Unnecessary buffer size calculation and condition on the lenght
removed from intel_cacheinfo.c::show_shared_cpu_map_func().
* uv_nmi_nr_cpus_pr() got overly smart and implemented "..."
abbreviation if the output stretched over the predefined 1024 byte
buffer. Replaced with plain printk.
Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Mike Travis <travis@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Pull x86 asm changes from Ingo Molnar:
"The main changes in this cycle were the x86/entry and sysret
enhancements from Andy Lutomirski, see merge commits 772a9aca12 and
b57c0b5175 for details"
[ Exectutive summary: IST exceptions that interrupt user space will run
on the regular kernel stack instead of the IST stack. Which
simplifies things particularly on return to user space.
The sysret cleanup ends up simplifying the logic on when we can use
sysret vs when we have to use iret. - Linus ]
* 'x86-asm-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
x86_64, entry: Remove the syscall exit audit and schedule optimizations
x86_64, entry: Use sysret to return to userspace when possible
x86, traps: Fix ist_enter from userspace
x86, vdso: teach 'make clean' remove vdso64 binaries
x86_64 entry: Fix RCX for ptraced syscalls
x86: entry_64.S: fold SAVE_ARGS_IRQ macro into its sole user
x86: ia32entry.S: fix wrong symbolic constant usage: R11->ARGOFFSET
x86: entry_64.S: delete unused code
x86, mce: Get rid of TIF_MCE_NOTIFY and associated mce tricks
x86, traps: Add ist_begin_non_atomic and ist_end_non_atomic
x86: Clean up current_stack_pointer
x86, traps: Track entry into and exit from IST context
x86, entry: Switch stacks on a paranoid entry from userspace
Pull x86 APIC updates from Ingo Molnar:
"Continued fallout of the conversion of the x86 IRQ code to the
hierarchical irqdomain framework: more cleanups, simplifications,
memory allocation behavior enhancements, mainly in the interrupt
remapping and APIC code"
* 'x86-apic-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (44 commits)
x86, init: Fix UP boot regression on x86_64
iommu/amd: Fix irq remapping detection logic
x86/acpi: Make acpi_[un]register_gsi_ioapic() depend on CONFIG_X86_LOCAL_APIC
x86: Consolidate boot cpu timer setup
x86/apic: Reuse apic_bsp_setup() for UP APIC setup
x86/smpboot: Sanitize uniprocessor init
x86/smpboot: Move apic init code to apic.c
init: Get rid of x86isms
x86/apic: Move apic_init_uniprocessor code
x86/smpboot: Cleanup ioapic handling
x86/apic: Sanitize ioapic handling
x86/ioapic: Add proper checks to setp/enable_IO_APIC()
x86/ioapic: Provide stub functions for IOAPIC%3Dn
x86/smpboot: Move smpboot inlines to code
x86/x2apic: Use state information for disable
x86/x2apic: Split enable and setup function
x86/x2apic: Disable x2apic from nox2apic setup
x86/x2apic: Add proper state tracking
x86/x2apic: Clarify remapping mode for x2apic enablement
x86/x2apic: Move code in conditional region
...
Pull perf updates from Ingo Molnar:
"Kernel side changes:
- AMD range breakpoints support:
Extend breakpoint tools and core to support address range through
perf event with initial backend support for AMD extended
breakpoints.
The syntax is:
perf record -e mem:addr/len:type
For example set write breakpoint from 0x1000 to 0x1200 (0x1000 + 512)
perf record -e mem:0x1000/512:w
- event throttling/rotating fixes
- various event group handling fixes, cleanups and general paranoia
code to be more robust against bugs in the future.
- kernel stack overhead fixes
User-visible tooling side changes:
- Show precise number of samples in at the end of a 'record' session,
if processing build ids, since we will then traverse the whole
perf.data file and see all the PERF_RECORD_SAMPLE records,
otherwise stop showing the previous off-base heuristicly counted
number of "samples" (Namhyung Kim).
- Support to read compressed module from build-id cache (Namhyung
Kim)
- Enable sampling loads and stores simultaneously in 'perf mem'
(Stephane Eranian)
- 'perf diff' output improvements (Namhyung Kim)
- Fix error reporting for evsel pgfault constructor (Arnaldo Carvalho
de Melo)
Tooling side infrastructure changes:
- Cache eh/debug frame offset for dwarf unwind (Namhyung Kim)
- Support parsing parameterized events (Cody P Schafer)
- Add support for IP address formats in libtraceevent (David Ahern)
Plus other misc fixes"
* 'perf-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (48 commits)
perf: Decouple unthrottling and rotating
perf: Drop module reference on event init failure
perf: Use POLLIN instead of POLL_IN for perf poll data in flag
perf: Fix put_event() ctx lock
perf: Fix move_group() order
perf: Fix event->ctx locking
perf: Add a bit of paranoia
perf symbols: Convert lseek + read to pread
perf tools: Use perf_data_file__fd() consistently
perf symbols: Support to read compressed module from build-id cache
perf evsel: Set attr.task bit for a tracking event
perf header: Set header version correctly
perf record: Show precise number of samples
perf tools: Do not use __perf_session__process_events() directly
perf callchain: Cache eh/debug frame offset for dwarf unwind
perf tools: Provide stub for missing pthread_attr_setaffinity_np
perf evsel: Don't rely on malloc working for sz 0
tools lib traceevent: Add support for IP address formats
perf ui/tui: Show fatal error message only if exists
perf tests: Fix typo in sample-parsing.c
...
I'm getting complaints from validation teams that have updated their
Linux kernels from ancient versions to current. They don't see the
error logs they expect. I tell the to unload any EDAC drivers[1], and
things start working again. The problem is that we short-circuit
the logging process if any function on the decoder chain claims to
have dealt with the problem:
ret = atomic_notifier_call_chain(&x86_mce_decoder_chain, 0, m);
if (ret == NOTIFY_STOP)
return;
The logic we used when we added this code was that we did not want
to confuse users with double reports of the same error.
But it turns out users are not confused - they are upset that they
don't see a log where their tools used to find a log.
I could also get into a long description of how the consumer of this
log does more than just decode model specific details of the error.
It keeps counts, tracks thresholds, takes actions and runs scripts
that can alert administrators to problems.
[1] We've recently compounded the problem because the acpi_extlog
driver also registers for this notifier and also returns NOTIFY_STOP.
Signed-off-by: Tony Luck <tony.luck@intel.com>
Pull timer and x86 fix from Ingo Molnar:
"A CLOCK_TAI early expiry fix and an x86 microcode driver oops fix"
* 'timers-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
hrtimer: Fix incorrect tai offset calculation for non high-res timer systems
* 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
x86, microcode: Return error from driver init code when loader is disabled
We currently allow any process to use rdpmc. This significantly
weakens the protection offered by PR_TSC_DISABLED, and it could be
helpful to users attempting to exploit timing attacks.
Since we can't enable access to individual counters, use a very
coarse heuristic to limit access to rdpmc: allow access only when
a perf_event is mmapped. This protects seccomp sandboxes.
There is plenty of room to further tighen these restrictions. For
example, this allows rdpmc for any x86_pmu event, but it's only
useful for self-monitoring tasks.
As a side effect, cap_user_rdpmc will now be false for AMD uncore
events. This isn't a real regression, since .event_idx is disabled
for these events anyway for the time being. Whenever that gets
re-added, the cap_user_rdpmc code can be adjusted or refactored
accordingly.
Signed-off-by: Andy Lutomirski <luto@amacapital.net>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Arnaldo Carvalho de Melo <acme@kernel.org>
Cc: Kees Cook <keescook@chromium.org>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Vince Weaver <vince@deater.net>
Cc: "hillf.zj" <hillf.zj@alibaba-inc.com>
Cc: Valdis Kletnieks <Valdis.Kletnieks@vt.edu>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Link: http://lkml.kernel.org/r/a2bdb3cf3a1d70c26980d7c6dddfbaa69f3182bf.1414190806.git.luto@amacapital.net
Signed-off-by: Ingo Molnar <mingo@kernel.org>
The new hw_breakpoint bits are now ready for v3.20, merge them
into the main branch, to avoid conflicts.
Conflicts:
tools/perf/Documentation/perf-record.txt
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Pull x86/entry enhancements from Andy Lutomirski:
" This is my accumulated x86 entry work, part 1, for 3.20. The meat
of this is an IST rework. When an IST exception interrupts user
space, we will handle it on the per-thread kernel stack instead of
on the IST stack. This sounds messy, but it actually simplifies the
IST entry/exit code, because it eliminates some ugly games we used
to play in order to handle rescheduling, signal delivery, etc on the
way out of an IST exception.
The IST rework introduces proper context tracking to IST exception
handlers. I haven't seen any bug reports, but the old code could
have incorrectly treated an IST exception handler as an RCU extended
quiescent state.
The memory failure change (included in this pull request with
Borislav and Tony's permission) eliminates a bunch of code that
is no longer needed now that user memory failure handlers are
called in process context.
Finally, this includes a few on Denys' uncontroversial and Obviously
Correct (tm) cleanups.
The IST and memory failure changes have been in -next for a while.
LKML references:
IST rework:
http://lkml.kernel.org/r/cover.1416604491.git.luto@amacapital.net
Memory failure change:
http://lkml.kernel.org/r/54ab2ffa301102cd6e@agluck-desk.sc.intel.com
Denys' cleanups:
http://lkml.kernel.org/r/1420927210-19738-1-git-send-email-dvlasenk@redhat.com
"
This tree semantically depends on and is based on the following RCU commit:
734d168013 ("rcu: Make rcu_nmi_enter() handle nesting")
... and for that reason won't be pushed upstream before the RCU bits hit Linus's tree.
Signed-off-by: Ingo Molnar <mingo@kernel.org>
