GitHub/exynos8895/android_kernel_samsung_universal8895.git
12 years agoperf tool: Save cmdline from user in file header vs what is passed to record
David Ahern [Mon, 30 Jul 2012 02:53:51 +0000 (20:53 -0600)]
perf tool: Save cmdline from user in file header vs what is passed to record

A number of builtin commands process some user args and then pass the rest to
cmd_record. cmd_record then saves argc/argv that it receives into the header of
the perf data file. But this loses the arguments handled by the first command
-- ie., the real command line from the user. This patch saves the command line
as typed by the user rather than what was passed to cmd_record.

As an example consider the command:
$ perf kvm --guest --host --guestmount=/tmp/guest-mount record
    -fo /tmp/perf.data -ag -- sleep 10

Currently the command saved to the header is:
    cmdline : /tmp/p3.5/perf record -o perf.data.kvm -fo /tmp/perf.data -ag -- sleep 1

(ignore the duplicated -o -- the first would be yet another bug with perf-kvm).

With this patch the command line saved to the header is:
cmdline : /tmp/p3.5/perf kvm --guest --host --guestmount=/tmp/guest-mount
    record -fo /tmp/perf.data -ag -- sleep 1

v2: simplified to saving the command in parse_options per Stephane's suggestion

Signed-off-by: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/1343616831-6408-1-git-send-email-dsahern@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf top: Error handling for counter creation should parallel perf-record
David Ahern [Mon, 30 Jul 2012 02:53:03 +0000 (20:53 -0600)]
perf top: Error handling for counter creation should parallel perf-record

5a7ed29 fixed up perf-record but not perf-top. Similar argument holds
for it -- fallback to PMU only if it does not exist and handle invalid
attributes separately.

Signed-off-by: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Robert Richter <robert.richter@amd.com>
Link: http://lkml.kernel.org/r/1343616783-6360-1-git-send-email-dsahern@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf session: Remove no longer used synthesize_sample method
Arnaldo Carvalho de Melo [Fri, 3 Aug 2012 00:01:17 +0000 (21:01 -0300)]
perf session: Remove no longer used synthesize_sample method

Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-jd8tqbx8o8bs4t4g50vyhoc2@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf evsel: Adopt parse_sample method from perf_event
Arnaldo Carvalho de Melo [Thu, 2 Aug 2012 15:23:46 +0000 (12:23 -0300)]
perf evsel: Adopt parse_sample method from perf_event

Since we need evsel->{attr.{sample_{id_all,type}},sample_size},
reducing the number of parameters tools have to pass.

Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-wdtmgak0ihgsmw1brb54a8h4@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf evlist: Introduce perf_evlist__parse_sample
Arnaldo Carvalho de Melo [Thu, 2 Aug 2012 14:42:57 +0000 (11:42 -0300)]
perf evlist: Introduce perf_evlist__parse_sample

That is a more compact form of perf_session__parse_sample and to support
multiple evlists per perf_session is the way to go anyway.

Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-vkxx3j5qktoj11bvcwmfjj13@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf session: Use perf_evlist__id_hdr_size more extensively
Arnaldo Carvalho de Melo [Wed, 1 Aug 2012 22:31:00 +0000 (19:31 -0300)]
perf session: Use perf_evlist__id_hdr_size more extensively

Removing perf_session->id_hdr_size, as it can be obtained from the
evsel/evlist.

Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-1nwc2kslu7gsfblu98xbqbll@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf session: Use perf_evlist__sample_id_all more extensively
Arnaldo Carvalho de Melo [Wed, 1 Aug 2012 22:25:26 +0000 (19:25 -0300)]
perf session: Use perf_evlist__sample_id_all more extensively

Removing perf_session->sample_id_all, as it can be obtained from the
evsel/evlist.

Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-ok58u1mlc5ci9b6p36r52uh1@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf session: Use perf_evlist__sample_type more extensively
Arnaldo Carvalho de Melo [Wed, 1 Aug 2012 22:15:52 +0000 (19:15 -0300)]
perf session: Use perf_evlist__sample_type more extensively

Removing perf_session->sample_type, as it can be obtained from the
evsel/evlist.

Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-mnt1zwlik7sp7z6ljc9kyefg@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf evsel: Precalculate the sample size
Arnaldo Carvalho de Melo [Wed, 1 Aug 2012 21:53:11 +0000 (18:53 -0300)]
perf evsel: Precalculate the sample size

So that we don't have to store it in the perf_session instance, because
in the future perf_session instances may have multiple evlists, each
with different sample_type/sizes.

Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-ptod86fxkpgq3h62m9refkv4@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf target: Fix check on buffer size
Namhyung Kim [Thu, 26 Jul 2012 01:50:10 +0000 (10:50 +0900)]
perf target: Fix check on buffer size

It was a mistake just replace assert to BUG_ON since its condition
should be negated. Fix it.

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Cc: "Kirill A. Shutemov" <kirill@shutemov.name>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Kirill A. Shutemov <kirill@shutemov.name>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Ulrich Drepper <drepper@gmail.com>
Link: http://lkml.kernel.org/r/1343267410-7758-1-git-send-email-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf symbols: Fix array sizes for binary types arrays
Jiri Olsa [Wed, 1 Aug 2012 12:47:57 +0000 (14:47 +0200)]
perf symbols: Fix array sizes for binary types arrays

Following commit introduced wrong array boundaries, that could lead to
SIGSEGV.

  perf symbols: Factor DSO symtab types to generic binary types
  commit 44f24cb3156a1e7d2b6bb501b7f6153aed08994c
  Author: Jiri Olsa <jolsa@redhat.com>

Fixing to use proper array size.

Reported-by: Markus Trippelsdorf <markus@trippelsdorf.de>
Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Markus Trippelsdorf <markus@trippelsdorf.de>
Link: http://lkml.kernel.org/r/1343825277-10517-1-git-send-email-jolsa@redhat.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf test: Fix parse events automated tests
Jiri Olsa [Wed, 1 Aug 2012 12:48:58 +0000 (14:48 +0200)]
perf test: Fix parse events automated tests

Parse events tests got broken after following commit:

  perf tools: Fix trace events storms due to weight demux
  commit 0983cc0dbca45250cbb5984bec7c303ac265b8e5
  Author: Frederic Weisbecker <fweisbec@gmail.com>

that added PERF_SAMPLE_PERIOD sample type for tracepoints.

Updating related tests.

Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1343825338-10618-1-git-send-email-jolsa@redhat.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agouprobes: __replace_page() needs munlock_vma_page()
Oleg Nesterov [Sun, 29 Jul 2012 18:22:49 +0000 (20:22 +0200)]
uprobes: __replace_page() needs munlock_vma_page()

Like do_wp_page(), __replace_page() should do munlock_vma_page()
for the case when the old page still has other !VM_LOCKED
mappings. Unfortunately this needs mm/internal.h.

Also, move put_page() outside of ptl lock. This doesn't really
matter but looks a bit better.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Srikar Dronamraju <srikar.vnet.ibm.com>
Cc: Anton Arapov <anton@redhat.com>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Link: http://lkml.kernel.org/r/20120729182249.GA20372@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
12 years agouprobes: Rename vma_address() and make it return "unsigned long"
Oleg Nesterov [Sun, 29 Jul 2012 18:22:47 +0000 (20:22 +0200)]
uprobes: Rename vma_address() and make it return "unsigned long"

1. vma_address() returns loff_t, this looks confusing and this
   is unnecessary after the previous change. Make it return "ulong",
   all callers truncate the result anyway.

2. Its name conflicts with mm/rmap.c:vma_address(), rename it to
   offset_to_vaddr(), this matches vaddr_to_offset().

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Srikar Dronamraju <srikar.vnet.ibm.com>
Cc: Anton Arapov <anton@redhat.com>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Link: http://lkml.kernel.org/r/20120729182247.GA20365@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
12 years agouprobes: Fix register_for_each_vma()->vma_address() check
Oleg Nesterov [Sun, 29 Jul 2012 18:22:44 +0000 (20:22 +0200)]
uprobes: Fix register_for_each_vma()->vma_address() check

1. register_for_each_vma() checks that vma_address() == vaddr,
   but this is not enough. We should also ensure that
   vaddr >= vm_start, find_vma() guarantees "vaddr < vm_end" only.

2. After the prevous changes, register_for_each_vma() is the
   only reason why vma_address() has to return loff_t, all other
   users know that we have the valid mapping at this offset and
   thus the overflow is not possible.

   Change the code to use vaddr_to_offset() instead, imho this looks
   more clean/understandable and now we can change vma_address().

3. While at it, remove the unnecessary type-cast.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Srikar Dronamraju <srikar.vnet.ibm.com>
Cc: Anton Arapov <anton@redhat.com>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Link: http://lkml.kernel.org/r/20120729182244.GA20362@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
12 years agouprobes: Introduce vaddr_to_offset(vma, vaddr)
Oleg Nesterov [Sun, 29 Jul 2012 18:22:42 +0000 (20:22 +0200)]
uprobes: Introduce vaddr_to_offset(vma, vaddr)

Add the new helper, vaddr_to_offset(vma, vaddr) which returns
the offset in vma->vm_file this vaddr is mapped at.

Change build_probe_list() and find_active_uprobe() to use the
new helper, the next patch adds another user.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Srikar Dronamraju <srikar.vnet.ibm.com>
Cc: Anton Arapov <anton@redhat.com>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Link: http://lkml.kernel.org/r/20120729182242.GA20355@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
12 years agouprobes: Teach build_probe_list() to consider the range
Oleg Nesterov [Sun, 29 Jul 2012 18:22:40 +0000 (20:22 +0200)]
uprobes: Teach build_probe_list() to consider the range

Currently build_probe_list() builds the list of all uprobes
attached to the given inode, and the caller should filter out
those who don't fall into the [start,end) range, this is
sub-optimal.

This patch turns find_least_offset_node() into
find_node_in_range() which returns the first node inside the
[min,max] range, and changes build_probe_list() to use this node
as a starting point for rb_prev() and rb_next() to find all
other nodes the caller needs. The resulting list is no longer
sorted but we do not care.

This can speed up both build_probe_list() and the callers, but
there is another reason to introduce find_node_in_range(). It
can be used to figure out whether the given vma has uprobes or
not, this will be needed soon.

While at it, shift INIT_LIST_HEAD(tmp_list) into
build_probe_list().

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Srikar Dronamraju <srikar.vnet.ibm.com>
Cc: Anton Arapov <anton@redhat.com>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Link: http://lkml.kernel.org/r/20120729182240.GA20352@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
12 years agouprobes: Remove insert_vm_struct()->uprobe_mmap()
Oleg Nesterov [Sun, 29 Jul 2012 18:22:38 +0000 (20:22 +0200)]
uprobes: Remove insert_vm_struct()->uprobe_mmap()

Remove insert_vm_struct()->uprobe_mmap(). It is not needed, nobody
except arch/ia64/kernel/perfmon.c uses insert_vm_struct(vma)
with vma->vm_file != NULL.

And it is wrong. Again, get_user_pages() can not succeed before
vma_link(vma) makes is visible to find_vma(). And even if this
worked, we must not insert the new bp before this mapping is
visible to vma_prio_tree_foreach() for uprobe_unregister().

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Srikar Dronamraju <srikar.vnet.ibm.com>
Cc: Anton Arapov <anton@redhat.com>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Link: http://lkml.kernel.org/r/20120729182238.GA20349@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
12 years agouprobes: Remove copy_vma()->uprobe_mmap()
Oleg Nesterov [Sun, 29 Jul 2012 18:22:36 +0000 (20:22 +0200)]
uprobes: Remove copy_vma()->uprobe_mmap()

Remove copy_vma()->uprobe_mmap(new_vma), it is absolutely wrong.

This new_vma was just initialized to represent the new unmapped
area, [vm_start, vm_end) was returned by get_unmapped_area() in
the caller.

This means that uprobe_mmap()->get_user_pages() will fail for
sure, simply because find_vma() can never succeed. And I
verified that sys_mremap()->mremap_to() indeed always fails with
the wrong ENOMEM code if [addr, addr+old_len] is probed.

And why this uprobe_mmap() was added? I believe the intent was
wrong. Note that the caller is going to do move_page_tables(),
all registered uprobes are already faulted in, we only change
the virtual addresses.

NOTE: However, somehow we need to close the race with
uprobe_register() which relies on map_info->vaddr. This needs
another fix I'll try to do later. Probably we need uprobe_mmap()
in move_vma() but we can not do this right now, this can confuse
uprobes_state.counter (which I still hope we are going to kill).

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Srikar Dronamraju <srikar.vnet.ibm.com>
Cc: Anton Arapov <anton@redhat.com>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Link: http://lkml.kernel.org/r/20120729182236.GA20342@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
12 years agouprobes: Fix overflow in vma_address()/find_active_uprobe()
Oleg Nesterov [Sun, 29 Jul 2012 18:22:33 +0000 (20:22 +0200)]
uprobes: Fix overflow in vma_address()/find_active_uprobe()

vma->vm_pgoff is "unsigned long", it should be promoted to
loff_t before the multiplication to avoid the overflow.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Srikar Dronamraju <srikar.vnet.ibm.com>
Cc: Anton Arapov <anton@redhat.com>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Link: http://lkml.kernel.org/r/20120729182233.GA20339@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
12 years agouprobes: Suppress uprobe_munmap() from mmput()
Oleg Nesterov [Sun, 29 Jul 2012 18:22:31 +0000 (20:22 +0200)]
uprobes: Suppress uprobe_munmap() from mmput()

uprobe_munmap() does get_user_pages() and it is also called from
the final mmput()->exit_mmap() path. This slows down
exit/mmput() for no reason, and I think  it is simply
dangerous/wrong to try to fault-in a page into the dying mm. If
nothing else, this happens after the last sync_mm_rss(), afaics
handle_mm_fault() can change the task->rss_stat and make the
subsequent check_mm() unhappy.

Change uprobe_munmap() to check mm->mm_users != 0.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Srikar Dronamraju <srikar.vnet.ibm.com>
Cc: Anton Arapov <anton@redhat.com>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Link: http://lkml.kernel.org/r/20120729182231.GA20336@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
12 years agouprobes: Uprobe_mmap/munmap needs list_for_each_entry_safe()
Oleg Nesterov [Sun, 29 Jul 2012 18:22:29 +0000 (20:22 +0200)]
uprobes: Uprobe_mmap/munmap needs list_for_each_entry_safe()

The bug was introduced by me in 449d0d7c ("uprobes: Simplify the
usage of uprobe->pending_list").

Yes, we do not care about uprobe->pending_list after return and
nobody can remove the current list entry, but put_uprobe(uprobe)
can actually free it and thus we need list_for_each_safe().

Reported-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Srikar Dronamraju <srikar.vnet.ibm.com>
Cc: Anton Arapov <anton@redhat.com>
Link: http://lkml.kernel.org/r/20120729182229.GA20329@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
12 years agouprobes: Clean up and document write_opcode()->lock_page(old_page)
Oleg Nesterov [Sun, 29 Jul 2012 18:22:20 +0000 (20:22 +0200)]
uprobes: Clean up and document write_opcode()->lock_page(old_page)

The comment above write_opcode()->lock_page(old_page) tells
about the race with do_wp_page(). I don't really understand
which exactly race it means, but afaics this lock_page() was not
enough to close all races with do_wp_page().

Anyway, since:

   77fc4af1b59d uprobes: Change register_for_each_vma() to take mm->mmap_sem for writing

this code is always called with ->mmap_sem held for writing,
so we can forget about do_wp_page().

However, we can't simply remove this lock_page(), and the only
(afaics) reason is __replace_page()->try_to_free_swap().

Nothing in write_opcode() needs it, move it into
__replace_page() and fix the comment.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Srikar Dronamraju <srikar.vnet.ibm.com>
Cc: Anton Arapov <anton@redhat.com>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Link: http://lkml.kernel.org/r/20120729182220.GA20322@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
12 years agouprobes: Kill write_opcode()->lock_page(new_page)
Oleg Nesterov [Sun, 29 Jul 2012 18:22:18 +0000 (20:22 +0200)]
uprobes: Kill write_opcode()->lock_page(new_page)

write_opcode() does lock_page(new_page) for no reason. Nobody
can see this page until __replace_page() exposes it under ptl
lock, and we do nothing with this page after pte_unmap_unlock().

If nothing else, the similar code in do_wp_page() doesn't lock
the new page for page_add_new_anon_rmap/set_pte_at_notify.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Srikar Dronamraju <srikar.vnet.ibm.com>
Cc: Anton Arapov <anton@redhat.com>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Link: http://lkml.kernel.org/r/20120729182218.GA20315@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
12 years agouprobes: __replace_page() should not use page_address_in_vma()
Oleg Nesterov [Sun, 29 Jul 2012 18:22:16 +0000 (20:22 +0200)]
uprobes: __replace_page() should not use page_address_in_vma()

page_address_in_vma(old_page) in __replace_page() is ugly and
wrong. The caller already knows the correct virtual address,
this page was found by get_user_pages(vaddr).

However, page_address_in_vma() can actually fail if
page->mapping was cleared by __delete_from_page_cache() after
get_user_pages() returns. But this means the race with page
reclaim, write_opcode() should not fail, it should retry and
read this page again. Probably the race with remove_mapping() is
not possible due to page_freeze_refs() logic, but afaics at
least shmem_writepage()->shmem_delete_from_page_cache() can
clear ->mapping.

We could change __replace_page() to return -EAGAIN in this case,
but it would be better to simply use the caller's vaddr and rely
on page_check_address().

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Srikar Dronamraju <srikar.vnet.ibm.com>
Cc: Anton Arapov <anton@redhat.com>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Link: http://lkml.kernel.org/r/20120729182216.GA20311@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
12 years agouprobes: Don't recheck vma/f_mapping in write_opcode()
Oleg Nesterov [Sun, 29 Jul 2012 18:22:12 +0000 (20:22 +0200)]
uprobes: Don't recheck vma/f_mapping in write_opcode()

write_opcode() rechecks valid_vma() and ->f_mapping, this is
pointless. The caller, register_for_each_vma() or uprobe_mmap(),
has already done these checks under mmap_sem.

To clarify, uprobe_mmap() checks valid_vma() only, but we can
rely on build_probe_list(vm_file->f_mapping->host).

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Srikar Dronamraju <srikar.vnet.ibm.com>
Cc: Anton Arapov <anton@redhat.com>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Link: http://lkml.kernel.org/r/20120729182212.GA20304@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
12 years agoperf/x86: Fix missing struct before structure name
Jovi Zhang [Tue, 17 Jul 2012 02:14:41 +0000 (10:14 +0800)]
perf/x86: Fix missing struct before structure name

When CONFIG_PERF_EVENTS disabled, there will have a compiliation
error, because missing struct before structure name.

Signed-off-by: Jovi Zhang <bookjovi@gmail.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Jiri Kosina <trivial@kernel.org>
Link: http://lkml.kernel.org/r/CACV3sbKF%3DCX%2B2jWEWesfCA6rBoQ3wDM4-5ac9MuBtVbCtMRHdQ@mail.gmail.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
12 years agoperf/x86: Fix format definition of SNB-EP uncore QPI box
Yan, Zheng [Tue, 24 Jul 2012 02:44:10 +0000 (10:44 +0800)]
perf/x86: Fix format definition of SNB-EP uncore QPI box

The event control register of SNB-EP uncore QPI box has a one bit
extension at bit position 21.

Reported-by: Stephane Eranian <eranian@google.com>
Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1343097850-4348-1-git-send-email-zheng.z.yan@intel.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
12 years agoperf/x86: Make bitfield unsigned
Peter Zijlstra [Mon, 9 Jul 2012 11:50:23 +0000 (13:50 +0200)]
perf/x86: Make bitfield unsigned

Fix:

 arch/x86/kernel/cpu/perf_event.h:377:43: sparse: dubious one-bit signed bitfield

Cc: Borislav Petkov <bp@amd64.org>
Reported-by: Fengguang Wu <fengguang.wu@intel.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/n/tip-2jxkmktkppkclj1qe6qxd7ah@git.kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
12 years agoperf/x86: Fix LLC-* and node-* events on Intel SandyBridge
Yan, Zheng [Tue, 17 Jul 2012 09:27:55 +0000 (17:27 +0800)]
perf/x86: Fix LLC-* and node-* events on Intel SandyBridge

LLC-* and node-* events require using the OFFCORE_RESPONSE events
on SandyBridge, but the hw_cache_extra_regs is left uninitialized.
This patch adds the missing extra register configure table for
SandyBridge.

Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1342517275-2875-1-git-send-email-zheng.z.yan@intel.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
12 years agoperf/x86: Add Intel Nehalem-EX uncore support
Yan, Zheng [Thu, 5 Jul 2012 06:32:17 +0000 (14:32 +0800)]
perf/x86: Add Intel Nehalem-EX uncore support

The uncore subsystem in Nehalem-EX consists of 7 components
(U-Box, C-Box, B-Box, S-Box, R-Box, M-Box and W-Box). This
patch is large because the way to program these boxes is
diverse.

Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/4FF534F1.3030307@intel.com
[ Improved the code. ]
Signed-off-by: Ingo Molnar <mingo@kernel.org>
12 years agoperf/x86: Fix typo in format definition of uncore PCU filter
Yan, Zheng [Mon, 23 Jul 2012 06:23:30 +0000 (14:23 +0800)]
perf/x86: Fix typo in format definition of uncore PCU filter

The format definition of uncore PCU filter should be filter_band*
instead of filter_brand*.

Reported-by: Stephane Eranian <eranian@google.com>
Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1343024611-4692-1-git-send-email-zheng.z.yan@intel.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
12 years agosched: Deliver sched_switch events to the current task
Andrew Vagin [Thu, 12 Jul 2012 10:14:29 +0000 (14:14 +0400)]
sched: Deliver sched_switch events to the current task

Otherwise they can't be filtered for a defined task:

  perf record -e sched:sched_switch ./foo

This command doesn't report any events without this patch.

I think it isn't a security concern if someone knows who will
be executed next - this can already be observed by polling /proc
state. By default perf is disabled for non-root users in any case.

I need these events for profiling sleep times.  sched_switch is used for
getting callchains and sched_stat_* is used for getting time periods.
These events are combined in user space, then it can be analyzed by
perf tools.

Signed-off-by: Andrew Vagin <avagin@openvz.org>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Arun Sharma <asharma@fb.com>
Link: http://lkml.kernel.org/r/1342088069-1005148-1-git-send-email-avagin@openvz.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
12 years agoMerge tag 'perf-core-for-mingo' of git://git.kernel.org/pub/scm/linux/kernel/git...
Ingo Molnar [Wed, 25 Jul 2012 18:52:59 +0000 (20:52 +0200)]
Merge tag 'perf-core-for-mingo' of git://git./linux/kernel/git/acme/linux into perf/core

Pull perf/core fixes and improvements from Arnaldo Carvalho de Melo:

 * libtraceevent build fixes from Namhyung Kim

 * Prevent overflow when calculating annotation buckets size, from Cody Schafer

 * Set breakpoint events sample period to 1, just like trace events, from Jovi Zhang

 * Fix strerror_r usage, from Kirill Shutemov

 * Fix duplicate function declaration problems with bison 2.6, from Kirill Shutemov

 * Add DSO caching facility (and perf test entry) to speed up binary file
   reading, will be used by the DWARF unwind code, from Jiri Olsa

 * Fix trace event storms by setting PERF_SAMPLE_PERIOD, from Frederic Weisbecker

 * perf kvm fixes from David Ahern

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
12 years agoperf annotate: Prevent overflow in size calculation
Cody Schafer [Fri, 20 Jul 2012 03:05:25 +0000 (20:05 -0700)]
perf annotate: Prevent overflow in size calculation

A large enough symbol size causes an overflow in the size parameter to
the histogram allocation, leading to a segfault in
symbol__inc_addr_samples later on when this histogram is accessed.

In the case of being called via perf-report, this returns back and
gracefully ignores the sample, eventually ignoring the chained return
value of perf_session_deliver_event in flush_sample_queue.

Signed-off-by: Cody Schafer <cody@linux.vnet.ibm.com>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Sukadev Bhattiprolu <sukadev@linux.vnet.ibm.com>
Link: http://lkml.kernel.org/r/1342753525-4521-1-git-send-email-cody@linux.vnet.ibm.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agotools lib traceevent: Ignore TRACEEVENT-CFLAGS file
Namhyung Kim [Fri, 6 Jul 2012 07:21:33 +0000 (16:21 +0900)]
tools lib traceevent: Ignore TRACEEVENT-CFLAGS file

The TRACEEVENT-CFLAGS file is used to detect any change on compiler
flags. Just ignore it.

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Cc: David Ahern <dsahern@gmail.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1341559297-25725-3-git-send-email-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agotools lib traceevent: Detect build environment changes
Namhyung Kim [Fri, 6 Jul 2012 07:21:32 +0000 (16:21 +0900)]
tools lib traceevent: Detect build environment changes

Cross compiling perf requires setting ARCH and CROSS_COMPILE variables,
but libtraceevent couldn't detect the changes so it ends up believing no
recompiling is required. Thus the linker failed like:

     LINK perf
 ../lib/traceevent//libtraceevent.a: member ../lib/traceevent//libtraceevent.a(event-parse.o) in archive is not an object
 collect2: ld returned 1 exit status
 make: *** [perf] Error 1

This patch fixes this by adding TRACEEVENT-CFLAGS file like
PERF-CFLAGS to track those changes.

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Cc: David Ahern <dsahern@gmail.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1341559297-25725-2-git-send-email-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf tools: Fix build error with bison 2.6
Kirill A. Shutemov [Mon, 23 Jul 2012 21:04:07 +0000 (00:04 +0300)]
perf tools: Fix build error with bison 2.6

Bison 2.6 started to generate parse_events_parse() declaration in header. In
this case we have redundant redeclaration:

util/parse-events.c:29:5: error: redundant redeclaration of â€˜parse_events_parse’ [-Werror=redundant-decls]
In file included from util/parse-events.c:14:0:
util/parse-events-bison.h:99:5: note: previous declaration of â€˜parse_events_parse’ was here
cc1: all warnings being treated as errors

Let's disable -Wredundant-decls for util/parse-events.c since it includes
header we can't control.

Signed-off-by: Kirill A. Shutemov <kirill@shutemov.name>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/20120723210407.GA25186@shutemov.name
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf tools: use XSI-complaint version of strerror_r() instead of GNU-specific
Kirill A. Shutemov [Mon, 23 Jul 2012 21:06:54 +0000 (00:06 +0300)]
perf tools: use XSI-complaint version of strerror_r() instead of GNU-specific

Perf uses GNU-specific version of strerror_r(). The GNU-specific strerror_r()
returns a pointer to a string containing the error message.  This may be either
a pointer to a string that the function stores in buf, or a pointer to some
(immutable) static string (in which case buf is unused).

In glibc-2.16 GNU version was marked with attribute warn_unused_result.  It
triggers few warnings in perf:

util/target.c: In function â€˜perf_target__strerror’:
util/target.c:114:13: error: ignoring return value of â€˜strerror_r’, declared with attribute warn_unused_result [-Werror=unused-result]
ui/browsers/hists.c: In function â€˜hist_browser__dump’:
ui/browsers/hists.c:981:13: error: ignoring return value of â€˜strerror_r’, declared with attribute warn_unused_result [-Werror=unused-result]

They are bugs.

Let's fix strerror_r() usage.

Signed-off-by: Kirill A. Shutemov <kirill@shutemov.name>
Acked-by: Ulrich Drepper <drepper@gmail.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Ulrich Drepper <drepper@gmail.com>
Link: http://lkml.kernel.org/r/20120723210654.GA25248@shutemov.name
[ committer note: s/assert/BUG_ON/g ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf tools: Make the breakpoint events sample period default to 1
Jovi Zhang [Sat, 14 Jul 2012 19:03:10 +0000 (03:03 +0800)]
perf tools: Make the breakpoint events sample period default to 1

There have one problem about hw_breakpoint perf event, as watched, the
events reported to userspace is not correctly, sometime one trigger
bp_event report several events, sometime bp_event cannot go through to
user.

The root cause is attr->freq is 1 passed to kernel defaultly in bp
events, this make kernel calculate event period not as expect, make
sample period to 1 will change attr->freq to 0, to fix this problem.

This patch is similar with commit f92128 about tracepoint events:
    perf: Make the trace events sample period default to 1

Signed-off-by: Jovi Zhang <bookjovi@gmail.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/CACV3sbLF8taiCq_VYW-sgRJyupeMzg58C7ZXfMe3xZUiH_Mx6w@mail.gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf test: Add dso data caching tests
Jiri Olsa [Sun, 22 Jul 2012 12:14:40 +0000 (14:14 +0200)]
perf test: Add dso data caching tests

Adding automated test for DSO data reading. Testing raw/cached reads
from different file/cache locations.

Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Cc: Arun Sharma <asharma@fb.com>
Cc: Benjamin Redelings <benjamin.redelings@nescent.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Cyrill Gorcunov <gorcunov@openvz.org>
Cc: Frank Ch. Eigler <fche@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Robert Richter <robert.richter@amd.com>
Cc: Stephane Eranian <eranian@google.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
Cc: Ulrich Drepper <drepper@gmail.com>
Link: http://lkml.kernel.org/r/1342959280-5361-18-git-send-email-jolsa@redhat.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf symbols: Add dso data caching
Jiri Olsa [Sun, 22 Jul 2012 12:14:39 +0000 (14:14 +0200)]
perf symbols: Add dso data caching

Adding dso data caching so we don't need to open/read/close, each time
we want dso data.

The DSO data caching affects following functions:
  dso__data_read_offset
  dso__data_read_addr

Each DSO read tries to find the data (based on offset) inside the cache.
If it's not present it fills the cache from file, and returns the data.
If it is present, data are returned with no file read.

Each data read is cached by reading cache page sized/aligned amount of
DSO data. The cache page size is hardcoded to 4096.  The cache is using
RB tree with file offset as a sort key.

Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Cc: Arun Sharma <asharma@fb.com>
Cc: Benjamin Redelings <benjamin.redelings@nescent.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Cyrill Gorcunov <gorcunov@openvz.org>
Cc: Frank Ch. Eigler <fche@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Robert Richter <robert.richter@amd.com>
Cc: Stephane Eranian <eranian@google.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
Cc: Ulrich Drepper <drepper@gmail.com>
Link: http://lkml.kernel.org/r/1342959280-5361-17-git-send-email-jolsa@redhat.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf symbols: Add interface to read DSO image data
Jiri Olsa [Sun, 22 Jul 2012 12:14:33 +0000 (14:14 +0200)]
perf symbols: Add interface to read DSO image data

Adding following interface for DSO object to allow
reading of DSO image data:

  dso__data_fd
    - opens DSO and returns file descriptor
      Binary types are used to locate/open DSO in following order:
        DSO_BINARY_TYPE__BUILD_ID_CACHE
        DSO_BINARY_TYPE__SYSTEM_PATH_DSO
      In other word we first try to open DSO build-id path,
      and if that fails we try to open DSO system path.

  dso__data_read_offset
    - reads DSO data from specified offset

  dso__data_read_addr
    - reads DSO data from specified address/map.

Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Cc: Arun Sharma <asharma@fb.com>
Cc: Benjamin Redelings <benjamin.redelings@nescent.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Cyrill Gorcunov <gorcunov@openvz.org>
Cc: Frank Ch. Eigler <fche@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Robert Richter <robert.richter@amd.com>
Cc: Stephane Eranian <eranian@google.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
Cc: Ulrich Drepper <drepper@gmail.com>
Link: http://lkml.kernel.org/r/1342959280-5361-11-git-send-email-jolsa@redhat.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf symbols: Factor DSO symtab types to generic binary types
Jiri Olsa [Sun, 22 Jul 2012 12:14:32 +0000 (14:14 +0200)]
perf symbols: Factor DSO symtab types to generic binary types

Adding interface to access DSOs so it could be used
from another place.

New DSO binary type is added - making current SYMTAB__*
types more general:
   DSO_BINARY_TYPE__* = SYMTAB__*

Following function is added to return path based on the specified
binary type:
   dso__binary_type_file

Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Cc: Arun Sharma <asharma@fb.com>
Cc: Benjamin Redelings <benjamin.redelings@nescent.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Cyrill Gorcunov <gorcunov@openvz.org>
Cc: Frank Ch. Eigler <fche@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Robert Richter <robert.richter@amd.com>
Cc: Stephane Eranian <eranian@google.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
Cc: Ulrich Drepper <drepper@gmail.com>
Link: http://lkml.kernel.org/r/1342959280-5361-10-git-send-email-jolsa@redhat.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf hists: Print newline between hists callchains
Frederic Weisbecker [Wed, 18 Jul 2012 17:10:56 +0000 (19:10 +0200)]
perf hists: Print newline between hists callchains

Tiny cosmetic fix. The lack of a newline between hists callchains was
looking slightly messy.

Before:

     0.24%      swapper  [kernel.kallsyms]  [k] _raw_spin_lock_irq
                |
                --- _raw_spin_lock_irq
                    run_timer_softirq
                    __do_softirq
                    call_softirq
                    do_softirq
                    irq_exit
                    smp_apic_timer_interrupt
                    apic_timer_interrupt
                    default_idle
                    amd_e400_idle
                    cpu_idle
                    start_secondary
     0.10%         perf  [kernel.kallsyms]  [k] lock_is_held
                   |
                   --- lock_is_held
                       __might_sleep
                       mutex_lock_nested
                       perf_event_for_each_child
                       perf_ioctl
                       do_vfs_ioctl
                       sys_ioctl
                       system_call_fastpath
                       ioctl
                       cmd_record
                       run_builtin
                       main
                       __libc_start_main

After:

     0.24%      swapper  [kernel.kallsyms]  [k] _raw_spin_lock_irq
                |
                --- _raw_spin_lock_irq
                    run_timer_softirq
                    __do_softirq
                    call_softirq
                    do_softirq
                    irq_exit
                    smp_apic_timer_interrupt
                    apic_timer_interrupt
                    default_idle
                    amd_e400_idle
                    cpu_idle
                    start_secondary

     0.10%         perf  [kernel.kallsyms]  [k] lock_is_held
                   |
                   --- lock_is_held
                       __might_sleep
                       mutex_lock_nested
                       perf_event_for_each_child
                       perf_ioctl
                       do_vfs_ioctl
                       sys_ioctl
                       system_call_fastpath
                       ioctl
                       cmd_record
                       run_builtin
                       main
                       __libc_start_main

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1342631456-7233-3-git-send-email-fweisbec@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf tools: Fix trace events storms due to weight demux
Frederic Weisbecker [Wed, 18 Jul 2012 17:10:54 +0000 (19:10 +0200)]
perf tools: Fix trace events storms due to weight demux

Trace events have a period (weight) of 1 by default. This can be
overriden on events definition by using the __perf_count() macro.

For example, the sched_stat_runtime() is weighted with the runtime of
the task that fired the event.

By default, perf handles such weighted event by dividing it into
individual events carrying a weight of 1. For example if
sched_stat_runtime is fired and the task has run 5000000 nsecs, perf
divides it into 5000000 events in the buffer.

This behaviour makes weighted events unusable because they quickly
fullfill the buffers and we lose most events.

The commit 5d81e5cfb37a174e8ddc0413e2e70cdf05807ace ("events: Don't
divide events if it has field period") solves this problem by sending
only one event when PERF_SAMPLE_PERIOD flag is set. The weight is
carried in the sample itself such that we don't need to demultiplex it
anymore.

This patch provides the last missing piece to use this feature by
setting PERF_SAMPLE_PERIOD from perf tools when we deal with trace
events.

Before:
$ ./perf record -e sched:* -a sleep 1
[ perf record: Woken up 3 times to write data ]
[ perf record: Captured and wrote 1.619 MB perf.data (~70749 samples) ]
Warning:
Processed 16909 events and lost 1 chunks!

Check IO/CPU overload!

$ ./perf script
perf  1894 [003]   824.898327: sched_migrate_task: comm=perf pid=1898 prio=120 orig_cpu=2 dest_cpu=0
perf  1894 [003]   824.898335: sched_stat_sleep: comm=perf pid=1898 delay=113179500 [ns]
perf  1894 [003]   824.898336: sched_stat_sleep: comm=perf pid=1898 delay=113179500 [ns]
perf  1894 [003]   824.898337: sched_stat_sleep: comm=perf pid=1898 delay=113179500 [ns]
perf  1894 [003]   824.898338: sched_stat_sleep: comm=perf pid=1898 delay=113179500 [ns]
perf  1894 [003]   824.898339: sched_stat_sleep: comm=perf pid=1898 delay=113179500 [ns]
perf  1894 [003]   824.898340: sched_stat_sleep: comm=perf pid=1898 delay=113179500 [ns]
perf  1894 [003]   824.898341: sched_stat_sleep: comm=perf pid=1898 delay=113179500 [ns]
[...]

After:
$ ./perf record -e sched:* -a sleep 1
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.074 MB perf.data (~3228 samples) ]

$ ./perf script

perf  1461 [000]   554.286957: sched_migrate_task: comm=perf pid=1465 prio=120 orig_cpu=3 dest_cpu=1
perf  1461 [000]   554.286964: sched_stat_sleep: comm=perf pid=1465 delay=133047190 [ns]
perf  1461 [000]   554.286967: sched_wakeup: comm=perf pid=1465 prio=120 success=1 target_cpu=001
swapper     0 [001]   554.286976: sched_stat_wait: comm=perf pid=1465 delay=0 [ns]
swapper     0 [001]   554.286983: sched_switch: prev_comm=swapper/1 prev_pid=0 prev_prio=120 prev_state=R ==> next_comm=perf
[...]

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1342631456-7233-1-git-send-email-fweisbec@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf hists: Return correct number of characters printed in callchain
Frederic Weisbecker [Wed, 18 Jul 2012 17:10:55 +0000 (19:10 +0200)]
perf hists: Return correct number of characters printed in callchain

Include the omitted number of characters printed for the first entry.

Not that it really matters because nobody seem to care about the number
of printed characters for now. But just in case.

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1342631456-7233-2-git-send-email-fweisbec@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf tools: Dump exclude_{guest,host}, precise_ip header info too
David Ahern [Fri, 20 Jul 2012 23:25:52 +0000 (17:25 -0600)]
perf tools: Dump exclude_{guest,host}, precise_ip header info too

Adds the attributes to the event line in the header dump.
From:
 event : name = cycles, type = 0, config = 0x0, config1 = 0x0,
      config2 = 0x0, excl_usr = 0, excl_kern = 0, ...
to
  event : name = cycles, type = 0, config = 0x0, config1 = 0x0,
      config2 = 0x0, excl_usr = 0, excl_kern = 0, excl_host = 0,
      excl_guest = 0, precise_ip = 0, ...

Signed-off-by: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/1342826756-64663-8-git-send-email-dsahern@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf kvm: Limit repetitive guestmount message to once per directory
David Ahern [Fri, 20 Jul 2012 23:25:51 +0000 (17:25 -0600)]
perf kvm: Limit repetitive guestmount message to once per directory

After 7ed97ad use of the guestmount option without a subdir for *each*
VM generates an error message for each sample related to that VM. Once
per VM is enough.

Signed-off-by: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1342826756-64663-7-git-send-email-dsahern@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf kvm: Fix bug resolving guest kernel syms
David Ahern [Fri, 20 Jul 2012 23:25:49 +0000 (17:25 -0600)]
perf kvm: Fix bug resolving guest kernel syms

Guest kernel symbols are not resolved despite passing the information
needed to resolve them. e.g.,

perf kvm --guest --guestmount=/tmp/guest-mount record -a -- sleep 1
perf kvm --guest --guestmount=/tmp/guest-mount report --stdio

    36.55%  [guest/11399]  [unknown]         [g] 0xffffffff81600bc8
    33.19%  [guest/10474]  [unknown]         [g] 0x00000000c0116e00
    30.26%  [guest/11094]  [unknown]         [g] 0xffffffff8100a288
    43.69%  [guest/10474]  [unknown]         [g] 0x00000000c0103d90
    37.38%  [guest/11399]  [unknown]         [g] 0xffffffff81600bc8
    12.24%  [guest/11094]  [unknown]         [g] 0xffffffff810aa91d
     6.69%  [guest/11094]  [unknown]         [u] 0x00007fa784d721c3

which is just pathetic.

After a maddening 2 days sifting through perf minutia I found it --
id_hdr_size is not initialized for guest machines. This shows up on the
report side as random garbage for the cpu and timestamp, e.g.,

29816 7310572949125804849 0x1ac0 [0x50]: PERF_RECORD_MMAP ...

That messes up the sample sorting such that synthesized guest maps are
processed last.

With this patch you get a much more helpful report:

  12.11%  [guest/11399]  [guest.kernel.kallsyms.11399]  [g] irqtime_account_process_tick
  10.58%  [guest/11399]  [guest.kernel.kallsyms.11399]  [g] run_timer_softirq
   6.95%  [guest/11094]  [guest.kernel.kallsyms.11094]  [g] printk_needs_cpu
   6.50%  [guest/11094]  [guest.kernel.kallsyms.11094]  [g] do_timer
   6.45%  [guest/11399]  [guest.kernel.kallsyms.11399]  [g] idle_balance
   4.90%  [guest/11094]  [guest.kernel.kallsyms.11094]  [g] native_read_tsc
    ...

v2:
- changed rbtree walk to use rb_first per Namhyung's suggestion

Tested-by: Jiri Olsa <jolsa@redhat.com>
Signed-off-by: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1342826756-64663-5-git-send-email-dsahern@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf kvm: Guest userspace samples should not be lumped with host uspace
David Ahern [Fri, 20 Jul 2012 23:25:48 +0000 (17:25 -0600)]
perf kvm: Guest userspace samples should not be lumped with host uspace

e.g., perf kvm --host  --guest report -i perf.data --stdio -D
shows:

599127912065356 0x143b8 [0x48]: PERF_RECORD_SAMPLE(IP, 5): 5671/5676: 0x7fdf95a061c0 period: 1 addr: 0
... chain: nr:2
.....  0: ffffffffffffff80
.....  1: fffffffffffffe00
 ... thread: qemu-kvm:5671
 ...... dso: <not found>

(IP, 5) means sample in guest userspace. Those samples should not be lumped
into the VMM's host thread. i.e, the report output:

    56.86%  qemu-kvm  [unknown]         [u] 0x00007fdf95a061c0

With this patch the output emphasizes it is a guest userspace hit:

    56.86%  [guest/5671]  [unknown]         [u] 0x00007fdf95a061c0

Looking at 3 VMs (2 64-bit, 1 32-bit) with each running a CPU bound
process (openssl speed), perf report currently shows:

  93.84%  117726   qemu-kvm  [unknown]   [u] 0x00007fd7dcaea8e5

which is wrong. With this patch you get:

  31.50%   39258   [guest/18772]  [unknown]   [u] 0x00007fd7dcaea8e5
  31.50%   39236   [guest/11230]  [unknown]   [u] 0x0000000000a57340
  30.84%   39232   [guest/18395]  [unknown]   [u] 0x00007f66f641e107

Tested-by: Jiri Olsa <jolsa@redhat.com>
Signed-off-by: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1342826756-64663-4-git-send-email-dsahern@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf kvm: Set name for VM process in guest machine
David Ahern [Fri, 20 Jul 2012 23:25:47 +0000 (17:25 -0600)]
perf kvm: Set name for VM process in guest machine

COMM events are not generated in the context of a guest machine, so the
thread name is never set for the VMM process. For example, the qemu-kvm
name applies to the process in the host machine, not the guest machine.
So, samples for guest machines are currently displayed as:

    99.67%     :5671  [unknown]         [g] 0xffffffff81366b41

where 5671 is the pid of the VMM. With this patch the samples in the guest
machine are shown as:

    18.43%  [guest/5671]  [unknown]           [g] 0xffffffff810d68b7

Tested-by: Jiri Olsa <jolsa@redhat.com>
Signed-off-by: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1342826756-64663-3-git-send-email-dsahern@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoperf symbols: Add machine id to modules debug message
David Ahern [Fri, 20 Jul 2012 23:25:46 +0000 (17:25 -0600)]
perf symbols: Add machine id to modules debug message

Current debug message is:
Problems creating module maps, continuing anyway...

When running multiple VMs it would be nice to know which machine the
message is referring to:

$ perf kvm --guest --guestmount=/tmp/guest-mount record -av -- sleep 10
Problems creating module maps for guest 6613, continuing anyway...

Signed-off-by: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1342826756-64663-2-git-send-email-dsahern@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
12 years agoMerge branch 'x86-build-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Sun, 22 Jul 2012 19:46:27 +0000 (12:46 -0700)]
Merge branch 'x86-build-for-linus' of git://git./linux/kernel/git/tip/tip

Pull a x86/build change from Ingo Molnar.

This makes the default stack alignment on x86-64 be just 8, allowing for
improved code generation (it can avoid some unnecessary extra alignment
logic and use just pure push/pop sequences) and smaller stack frames.

We can't generally do SSE with 16-byte alignment issues in the kernel anyway.

* 'x86-build-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  x86-64, gcc: Use -mpreferred-stack-boundary=3 if supported

12 years agoMerge branch 'x86-uv-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Sun, 22 Jul 2012 19:37:15 +0000 (12:37 -0700)]
Merge branch 'x86-uv-for-linus' of git://git./linux/kernel/git/tip/tip

Pull x86/uv changes from Ingo Molnar:
 "UV2 BAU productization fixes.

  The BAU (Broadcast Assist Unit) is SGI's fancy out of line way on UV
  hardware to do TLB flushes, instead of the normal APIC IPI methods.
  The commits here fix / work around hangs in their latest hardware
  iteration (UV2).

  My understanding is that the main purpose of the out of line
  signalling channel is to improve scalability: the UV APIC hardware
  glue does not handle broadcasting to many CPUs very well, and this
  matters most for TLB shootdowns.

  [ I don't agree with all aspects of the current approach: in hindsight
    it would have been better to link the BAU at the IPI/APIC driver
    level instead of the TLB shootdown level, where TLB flushes are
    really just one of the uses of broadcast SMP messages.  Doing that
    would improve scalability in some other ways and it would also
    remove a few uglies from the TLB path.  It would also be nice to
    push more is_uv_system() tests into proper x86_init or x86_platform
    callbacks.  Cliff? ]"

* 'x86-uv-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  x86/uv: Work around UV2 BAU hangs
  x86/uv: Implement UV BAU runtime enable and disable control via /proc/sgi_uv/
  x86/uv: Fix the UV BAU destination timeout period

12 years agoMerge branch 'x86-reboot-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 22 Jul 2012 19:25:47 +0000 (12:25 -0700)]
Merge branch 'x86-reboot-for-linus' of git://git./linux/kernel/git/tip/tip

Pull x86/reboot changes from Ingo Molnar:
 "Now that the revampted x86 real-mode trampoline code is upstream and
  seems to be working well, we can extend the 64-bit reboot code to be
  as capable as the 32-bit one."

* 'x86-reboot-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  x86-64, reboot: Be more paranoid in 64-bit reboot=bios
  x86, reboot: Drop redundant write of reboot_mode
  x86-64, reboot: Allow reboot=bios and reboot-cpu override on x86-64

12 years agoMerge branch 'x86-platform-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 22 Jul 2012 19:19:36 +0000 (12:19 -0700)]
Merge branch 'x86-platform-for-linus' of git://git./linux/kernel/git/tip/tip

Pull x86 platform changes from Ingo Molnar:
 "This tree mostly involves various APIC driver cleanups/robustization,
  and vSMP motivated platform callback improvements/cleanups"

Fix up trivial conflict due to printk cleanup right next to return value
change.

* 'x86-platform-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (29 commits)
  Revert "x86/early_printk: Replace obsolete simple_strtoul() usage with kstrtoint()"
  x86/apic/x2apic: Use multiple cluster members for the irq destination only with the explicit affinity
  x86/apic/x2apic: Limit the vector reservation to the user specified mask
  x86/apic: Optimize cpu traversal in __assign_irq_vector() using domain membership
  x86/vsmp: Fix vector_allocation_domain's return value
  irq/apic: Use config_enabled(CONFIG_SMP) checks to clean up irq_set_affinity() for UP
  x86/vsmp: Fix linker error when CONFIG_PROC_FS is not set
  x86/apic/es7000: Make apicid of a cluster (not CPU) from a cpumask
  x86/apic/es7000+summit: Always make valid apicid from a cpumask
  x86/apic/es7000+summit: Fix compile warning in cpu_mask_to_apicid()
  x86/apic: Fix ugly casting and branching in cpu_mask_to_apicid_and()
  x86/apic: Eliminate cpu_mask_to_apicid() operation
  x86/x2apic/cluster: Vector_allocation_domain() should return a value
  x86/apic/irq_remap: Silence a bogus pr_err()
  x86/vsmp: Ignore IOAPIC IRQ affinity if possible
  x86/apic: Make cpu_mask_to_apicid() operations check cpu_online_mask
  x86/apic: Make cpu_mask_to_apicid() operations return error code
  x86/apic: Avoid useless scanning thru a cpumask in assign_irq_vector()
  x86/apic: Try to spread IRQ vectors to different priority levels
  x86/apic: Factor out default vector_allocation_domain() operation
  ...

12 years agoMerge branch 'x86-debug-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Sun, 22 Jul 2012 19:04:44 +0000 (12:04 -0700)]
Merge branch 'x86-debug-for-linus' of git://git./linux/kernel/git/tip/tip

Pull debug-for-linus git tree from Ingo Molnar.

Fix up trivial conflict in arch/x86/kernel/cpu/perf_event_intel.c due to
a printk() having changed to a pr_info() differently in the two branches.

* 'x86-debug-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  x86: Move call to print_modules() out of show_regs()
  x86/mm: Mark free_initrd_mem() as __init
  x86/microcode: Mark microcode_id[] as __initconst
  x86/nmi: Clean up register_nmi_handler() usage
  x86: Save cr2 in NMI in case NMIs take a page fault (for i386)
  x86: Remove cmpxchg from i386 NMI nesting code
  x86: Save cr2 in NMI in case NMIs take a page fault
  x86/debug: Add KERN_<LEVEL> to bare printks, convert printks to pr_<level>

12 years agoMerge branch 'x86-asm-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Sun, 22 Jul 2012 18:42:28 +0000 (11:42 -0700)]
Merge branch 'x86-asm-for-linus' of git://git./linux/kernel/git/tip/tip

Pull x86/asm changes from Ingo Molnar:
 "Assorted single-commit improvements, as usual"

* 'x86-asm-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  x86/mm/mtrr: Slightly simplify print_mtrr_state()
  x86/mm/mtrr: Fix alignment determination in range_to_mtrr()
  x86/copy_user_generic: Optimize copy_user_generic with CPU erms feature
  x86/alternatives: Use atomic_xchg() instead atomic_dec_and_test() for stop_machine_text_poke()

12 years agoMerge branch 'timers-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 22 Jul 2012 18:35:46 +0000 (11:35 -0700)]
Merge branch 'timers-core-for-linus' of git://git./linux/kernel/git/tip/tip

Pull timer core changes from Ingo Molnar:
 "Continued cleanups of the core time and NTP code, plus more nohz work
  preparing for tick-less userspace execution."

* 'timers-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  time: Rework timekeeping functions to take timekeeper ptr as argument
  time: Move xtime_nsec adjustment underflow handling timekeeping_adjust
  time: Move arch_gettimeoffset() usage into timekeeping_get_ns()
  time: Refactor accumulation of nsecs to secs
  time: Condense timekeeper.xtime into xtime_sec
  time: Explicitly use u32 instead of int for shift values
  time: Whitespace cleanups per Ingo%27s requests
  nohz: Move next idle expiry time record into idle logic area
  nohz: Move ts->idle_calls incrementation into strict idle logic
  nohz: Rename ts->idle_tick to ts->last_tick
  nohz: Make nohz API agnostic against idle ticks cputime accounting
  nohz: Separate idle sleeping time accounting from nohz logic
  timers: Improve get_next_timer_interrupt()
  timers: Add accounting of non deferrable timers
  timers: Consolidate base->next_timer update
  timers: Create detach_if_pending() and use it

12 years agoMerge branch 'smp-hotplug-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 22 Jul 2012 18:22:15 +0000 (11:22 -0700)]
Merge branch 'smp-hotplug-for-linus' of git://git./linux/kernel/git/tip/tip

Pull smp/hotplug changes from Ingo Molnar:
 "Various cleanups to the SMP hotplug code - a continuing effort of
  Thomas et al"

* 'smp-hotplug-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  smpboot: Remove leftover declaration
  smp: Remove num_booting_cpus()
  smp: Remove ipi_call_lock[_irq]()/ipi_call_unlock[_irq]()
  POWERPC: Smp: remove call to ipi_call_lock()/ipi_call_unlock()
  SPARC: SMP: Remove call to ipi_call_lock_irq()/ipi_call_unlock_irq()
  ia64: SMP: Remove call to ipi_call_lock_irq()/ipi_call_unlock_irq()
  x86-smp-remove-call-to-ipi_call_lock-ipi_call_unlock
  tile: SMP: Remove call to ipi_call_lock()/ipi_call_unlock()
  S390: Smp: remove call to ipi_call_lock()/ipi_call_unlock()
  parisc: Smp: remove call to ipi_call_lock()/ipi_call_unlock()
  mn10300: SMP: Remove call to ipi_call_lock()/ipi_call_unlock()
  hexagon: SMP: Remove call to ipi_call_lock()/ipi_call_unlock()

12 years agoMerge branch 'perf-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Sun, 22 Jul 2012 18:10:36 +0000 (11:10 -0700)]
Merge branch 'perf-core-for-linus' of git://git./linux/kernel/git/tip/tip

Pull perf events changes from Ingo Molnar:

 "- kernel side:

   - Intel uncore PMU support for Nehalem and Sandy Bridge CPUs, we
     support both the events available via the MSR and via the PCI
     access space.

   - various uprobes cleanups and restructurings

   - PMU driver quirks by microcode version and required x86 microcode
     loader cleanups/robustization

   - various tracing robustness updates

   - static keys: remove obsolete static_branch()

  - tooling side:

   - GTK browser improvements

   - perf report browser: support screenshots to file

   - more automated tests

   - perf kvm improvements

   - perf bench refinements

   - build environment improvements

   - pipe mode improvements

   - libtraceevent updates, we have now hopefully merged most bits with
     the out of tree forked code base

  ... and many other goodies."

* 'perf-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (138 commits)
  tracing: Check for allocation failure in __tracing_open()
  perf/x86: Fix intel_perfmon_event_mapformatting
  jump label: Remove static_branch()
  tracepoint: Use static_key_false(), since static_branch() is deprecated
  perf/x86: Uncore filter support for SandyBridge-EP
  perf/x86: Detect number of instances of uncore CBox
  perf/x86: Fix event constraint for SandyBridge-EP C-Box
  perf/x86: Use 0xff as pseudo code for fixed uncore event
  perf/x86: Save a few bytes in 'struct x86_pmu'
  perf/x86: Add a microcode revision check for SNB-PEBS
  perf/x86: Improve debug output in check_hw_exists()
  perf/x86/amd: Unify AMD's generic and family 15h pmus
  perf/x86: Move Intel specific code to intel_pmu_init()
  perf/x86: Rename Intel specific macros
  perf/x86: Fix USER/KERNEL tagging of samples
  perf tools: Split event symbols arrays to hw and sw parts
  perf tools: Split out PE_VALUE_SYM parsing token to SW and HW tokens
  perf tools: Add empty rule for new line in event syntax parsing
  perf test: Use ARRAY_SIZE in parse events tests
  tools lib traceevent: Cleanup realloc use
  ...

12 years agoMerge branch 'core-rcu-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Sun, 22 Jul 2012 17:45:05 +0000 (10:45 -0700)]
Merge branch 'core-rcu-for-linus' of git://git./linux/kernel/git/tip/tip

Pull RCU changes from Ingo Molnar:
 "Quoting from Paul, the major features of this series are:

  1. Preventing latency spikes of more than 200 microseconds for
     kernels built with NR_CPUS=4096, which is reportedly becoming the
     default for some distros.  This is a first step, as it does not
     help with systems that actually -have- 4096 CPUs (work on this case
     is in progress, but is not yet ready for mainline).

     This category also includes improving concurrency of rcu_barrier(),
     placed here due to conflicts.  Posted to LKML at:

      https://lkml.org/lkml/2012/6/22/381

     Note that patches 18-22 of that series have been defered to 3.7, as
     they have not yet proven themselves to be mainline-ready (and yes,
     these are the ones intended to get rid of RCU's latency spikes for
     systems that actually have 4096 CPUs).

  2. Updates to documentation and rcutorture fixes, the latter category
     including improvements to rcu_barrier() testing.  Posted to LKML at

      http://lkml.indiana.edu/hypermail/linux/kernel/1206.1/04094.html.

  3. Miscellaneous fixes posted to LKML at:

      https://lkml.org/lkml/2012/6/22/500

     with the exception of the last commit, which was posted here:

      http://www.gossamer-threads.com/lists/linux/kernel/1561830

  4. RCU_FAST_NO_HZ fixes and improvements.  Posted to LKML at:

      http://lkml.indiana.edu/hypermail/linux/kernel/1206.1/00006.html
      http://www.gossamer-threads.com/lists/linux/kernel/1561833

     The first four patches of the first series went into 3.5 to fix a
     regression.

  5. Code-style fixes.  These were posted to LKML at

      http://lkml.indiana.edu/hypermail/linux/kernel/1205.2/01180.html
      http://lkml.indiana.edu/hypermail/linux/kernel/1205.2/01181.html"

* 'core-rcu-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (48 commits)
  rcu: Fix broken strings in RCU's source code.
  rcu: Fix code-style issues involving "else"
  rcu: Introduce check for callback list/count mismatch
  rcu: Make RCU_FAST_NO_HZ respect nohz= boot parameter
  rcu: Fix qlen_lazy breakage
  rcu: Round FAST_NO_HZ lazy timeout to nearest second
  rcu: The rcu_needs_cpu() function is not a quiescent state
  rcu: Dump only the current CPU's buffers for idle-entry/exit warnings
  rcu: Add check for CPUs going offline with callbacks queued
  rcu: Disable preemption in rcu_blocking_is_gp()
  rcu: Prevent uninitialized string in RCU CPU stall info
  rcu: Fix rcu_is_cpu_idle() #ifdef in TINY_RCU
  rcu: Split RCU core processing out of __call_rcu()
  rcu: Prevent __call_rcu() from invoking RCU core on offline CPUs
  rcu: Make __call_rcu() handle invocation from idle
  rcu: Remove function versions of __kfree_rcu and __is_kfree_rcu_offset
  rcu: Consolidate tree/tiny __rcu_read_{,un}lock() implementations
  rcu: Remove return value from rcu_assign_pointer()
  key: Remove extraneous parentheses from rcu_assign_keypointer()
  rcu: Remove return value from RCU_INIT_POINTER()
  ...

12 years agoMerge branch 'core-iommu-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 22 Jul 2012 17:39:32 +0000 (10:39 -0700)]
Merge branch 'core-iommu-for-linus' of git://git./linux/kernel/git/tip/tip

Pull core/iommu changes from Ingo Molnar.

* 'core-iommu-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  iommu/dmar: Use pr_format() instead of PREFIX to tidy up pr_*() calls
  iommu/dmar: Reserve mmio space used by the IOMMU, if the BIOS forgets to
  iommu/dmar: Replace printks with appropriate pr_*()

12 years agoRevert "x86/early_printk: Replace obsolete simple_strtoul() usage with kstrtoint()"
Ingo Molnar [Fri, 22 Jun 2012 14:25:19 +0000 (16:25 +0200)]
Revert "x86/early_printk: Replace obsolete simple_strtoul() usage with kstrtoint()"

This reverts commit fbd24153c48b8425b09c161a020483cd77da870e.

This commit is subtly buggy: kstrto*int() can return an error but
it's not checked in every path. simple_strtoul() on the other hand
could not fail, so this patch subtly intruduces new failure modes.

Signed-off-by: Shuah Khan <shuahkhan@gmail.com>
Link: http://lkml.kernel.org/r/1338424803.3569.5.camel@lorien2
Signed-off-by: Ingo Molnar <mingo@kernel.org>
12 years agoLinux 3.5
Linus Torvalds [Sat, 21 Jul 2012 20:58:29 +0000 (13:58 -0700)]
Linux 3.5

12 years agoRemove SYSTEM_SUSPEND_DISK system state
Rafael J. Wysocki [Sat, 21 Jul 2012 18:24:52 +0000 (20:24 +0200)]
Remove SYSTEM_SUSPEND_DISK system state

The SYSTEM_SUSPEND_DISK system state is never used, so drop it.

Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agoMerge branch 'anton-kgdb' (kgdb dmesg fixups)
Linus Torvalds [Sat, 21 Jul 2012 17:34:13 +0000 (10:34 -0700)]
Merge branch 'anton-kgdb' (kgdb dmesg fixups)

Merge emailed kgdb dmesg fixups patches from Anton Vorontsov:
 "The dmesg command appears to be broken after the printk rework.  The
  old logic in the kdb code makes no sense in terms of current
  printk/logging storage format, and KDB simply hangs forever upon
  entering 'dmesg' command.

  The first patch revives the command by switching to kmsg_dumper
  iterator.  As a side-effect, the code is now much more simpler.

  A few changes were needed in the printk.c: we needed unlocked variant
  of the kmsg_dumper iterator, but these can surely wait for 3.6.

  It's probably too late even for the first patch to go to 3.5, but I'll
  try to convince otherwise.  :-) Here we go:

   - The current code is broken for sure, and has no hope to work at
     all.  It is a regression
   - The new code works for me, and probably works for everyone else;
   - If it compiles (and I urge everyone to compile-test it on your
     setup), it hardly can make things worse."

* Merge emailed patches from Anton Vorontsov: (4 commits)
  kdb: Switch to nolock variants of kmsg_dump functions
  printk: Implement some unlocked kmsg_dump functions
  printk: Remove kdb_syslog_data
  kdb: Revive dmesg command

12 years agokdb: Switch to nolock variants of kmsg_dump functions
Anton Vorontsov [Sat, 21 Jul 2012 00:28:25 +0000 (17:28 -0700)]
kdb: Switch to nolock variants of kmsg_dump functions

The locked variants are prone to deadlocks (suppose we got to the
debugger w/ the logbuf lock held), so let's switch to nolock variants.

Signed-off-by: Anton Vorontsov <anton.vorontsov@linaro.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agoprintk: Implement some unlocked kmsg_dump functions
Anton Vorontsov [Sat, 21 Jul 2012 00:28:07 +0000 (17:28 -0700)]
printk: Implement some unlocked kmsg_dump functions

If used from KDB, the locked variants are prone to deadlocks (suppose we
got to the debugger w/ the logbuf lock held).

So, we have to implement a few routines that grab no logbuf lock.

Yet we don't need these functions in modules, so we don't export them.

Signed-off-by: Anton Vorontsov <anton.vorontsov@linaro.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agoprintk: Remove kdb_syslog_data
Anton Vorontsov [Sat, 21 Jul 2012 00:27:54 +0000 (17:27 -0700)]
printk: Remove kdb_syslog_data

The function is no longer needed, so remove it.

Signed-off-by: Anton Vorontsov <anton.vorontsov@linaro.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agokdb: Revive dmesg command
Anton Vorontsov [Sat, 21 Jul 2012 00:27:37 +0000 (17:27 -0700)]
kdb: Revive dmesg command

The kgdb dmesg command is broken after the printk rework.  The old logic
in kdb code makes no sense in terms of current printk/logging storage
format, and KDB simply hangs forever.

This patch revives the command by switching to kmsg_dumper iterator.

The code is now much more simpler and shorter.

Signed-off-by: Anton Vorontsov <anton.vorontsov@linaro.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agoMerge branch 'upstream' of git://git.linux-mips.org/pub/scm/ralf/upstream-linus
Linus Torvalds [Fri, 20 Jul 2012 19:02:02 +0000 (12:02 -0700)]
Merge branch 'upstream' of git://git.linux-mips.org/ralf/upstream-linus

Pull late MIPS fixes from Ralf Baechle:
 "This fixes a number of lose ends in the MIPS code and various bug
  fixes.

  Aside of dropping some patch that should not be in this pull request
  everything has sat in -next for quite a while and there are no known
  issues.

  The biggest patch in this patch set moves the allocation of an array
  that is aliased to a function (for runtime generated code) to
  assembler code.  This avoids an issue with certain toolchains when
  building for microMIPS."

* 'upstream' of git://git.linux-mips.org/pub/scm/ralf/upstream-linus: (35 commits)
  MIPS: PCI: Move fixups from __init to __devinit.
  MIPS: Fix bug.h MIPS build regression
  MIPS: sync-r4k: remove redundant irq operation
  MIPS: smp: Warn on too early irq enable
  MIPS: call set_cpu_online() on cpu being brought up with irq disabled
  MIPS: call ->smp_finish() a little late
  MIPS: Yosemite: delay irq enable to ->smp_finish()
  MIPS: SMTC: delay irq enable to ->smp_finish()
  MIPS: BMIPS: delay irq enable to ->smp_finish()
  MIPS: Octeon: delay enable irq to ->smp_finish()
  MIPS: Oprofile: Fix build as a module.
  MIPS: BCM63XX: Fix BCM6368 IPSec clock bit
  MIPS: perf: Fix build error caused by unused counters_per_cpu_to_total()
  MIPS: Fix Magic SysRq L kernel crash.
  MIPS: BMIPS: Fix duplicate header inclusion.
  mips: mark const init data with __initconst instead of __initdata
  MIPS: cmpxchg.h: Add missing include
  MIPS: Malta may also be equipped with MIPS64 R2 processors.
  MIPS: Fix typo multipy -> multiply
  MIPS: Cavium: Fix duplicate ARCH_SPARSEMEM_ENABLE in kconfig.
  ...

12 years agoMerge tag 'dm-3.5-fixes-2' of git://git.kernel.org/pub/scm/linux/kernel/git/agk/linux-dm
Linus Torvalds [Fri, 20 Jul 2012 18:51:22 +0000 (11:51 -0700)]
Merge tag 'dm-3.5-fixes-2' of git://git./linux/kernel/git/agk/linux-dm

Pull device-mapper discard fixes from Alasdair G Kergon:
  - avoid a crash in dm-raid1 when discards coincide with mirror
    recovery;
  - avoid discarding shared data that's still needed in dm-thin;
  - don't guarantee that discarded blocks will be wiped in dm-raid1.

* tag 'dm-3.5-fixes-2' of git://git.kernel.org/pub/scm/linux/kernel/git/agk/linux-dm:
  dm raid1: set discard_zeroes_data_unsupported
  dm thin: do not send discards to shared blocks
  dm raid1: fix crash with mirror recovery and discard

12 years agoMerge branch 'for-linus' of git://git.open-osd.org/linux-open-osd
Linus Torvalds [Fri, 20 Jul 2012 18:43:53 +0000 (11:43 -0700)]
Merge branch 'for-linus' of git://git.open-osd.org/linux-open-osd

Pull pnfs/ore fixes from Boaz Harrosh:
 "These are catastrophic fixes to the pnfs objects-layout that were just
  discovered.  They are also destined for @stable.

  I have found these and worked on them at around RC1 time but
  unfortunately went to the hospital for kidney stones and had a very
  slow recovery.  I refrained from sending them as is, before proper
  testing, and surly I have found a bug just yesterday.

  So now they are all well tested, and have my sign-off.  Other then
  fixing the problem at hand, and assuming there are no bugs at the new
  code, there is low risk to any surrounding code.  And in anyway they
  affect only these paths that are now broken.  That is RAID5 in pnfs
  objects-layout code.  It does also affect exofs (which was not broken)
  but I have tested exofs and it is lower priority then objects-layout
  because no one is using exofs, but objects-layout has lots of users."

* 'for-linus' of git://git.open-osd.org/linux-open-osd:
  pnfs-obj: Fix __r4w_get_page when offset is beyond i_size
  pnfs-obj: don't leak objio_state if ore_write/read fails
  ore: Unlock r4w pages in exact reverse order of locking
  ore: Remove support of partial IO request (NFS crash)
  ore: Fix NFS crash by supporting any unaligned RAID IO

12 years agoMerge tag 'upstream-3.5-rc8' of git://git.infradead.org/linux-ubifs
Linus Torvalds [Fri, 20 Jul 2012 18:42:30 +0000 (11:42 -0700)]
Merge tag 'upstream-3.5-rc8' of git://git.infradead.org/linux-ubifs

Pull UBIFS free space fix-up bugfix from Artem Bityutskiy:
 "It's been reported already twice recently:

    http://lists.infradead.org/pipermail/linux-mtd/2012-May/041408.html
    http://lists.infradead.org/pipermail/linux-mtd/2012-June/042422.html

  and we finally have the fix.  I am quite confident the fix is correct
  because I could reproduce the problem with nandsim and verify the fix.
  It was also verified by Iwo (the reporter).

  I am also confident that this is OK to merge the fix so late because
  this patch affects only the fixup functionality, which is not used by
  most users."

* tag 'upstream-3.5-rc8' of git://git.infradead.org/linux-ubifs:
  UBIFS: fix a bug in empty space fix-up

12 years agodm raid1: set discard_zeroes_data_unsupported
Mikulas Patocka [Fri, 20 Jul 2012 13:25:07 +0000 (14:25 +0100)]
dm raid1: set discard_zeroes_data_unsupported

We can't guarantee that REQ_DISCARD on dm-mirror zeroes the data even if
the underlying disks support zero on discard.  So this patch sets
ti->discard_zeroes_data_unsupported.

For example, if the mirror is in the process of resynchronizing, it may
happen that kcopyd reads a piece of data, then discard is sent on the
same area and then kcopyd writes the piece of data to another leg.
Consequently, the data is not zeroed.

The flag was made available by commit 983c7db347db8ce2d8453fd1d89b7a4bb6920d56
(dm crypt: always disable discard_zeroes_data).

Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Cc: stable@kernel.org
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
12 years agodm thin: do not send discards to shared blocks
Mikulas Patocka [Fri, 20 Jul 2012 13:25:05 +0000 (14:25 +0100)]
dm thin: do not send discards to shared blocks

When process_discard receives a partial discard that doesn't cover a
full block, it sends this discard down to that block. Unfortunately, the
block can be shared and the discard would corrupt the other snapshots
sharing this block.

This patch detects block sharing and ends the discard with success when
sending it to the shared block.

The above change means that if the device supports discard it can't be
guaranteed that a discard request zeroes data. Therefore, we set
ti->discard_zeroes_data_unsupported.

Thin target discard support with this bug arrived in commit
104655fd4dcebd50068ef30253a001da72e3a081 (dm thin: support discards).

Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Cc: stable@kernel.org
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
12 years agodm raid1: fix crash with mirror recovery and discard
Mikulas Patocka [Fri, 20 Jul 2012 13:25:03 +0000 (14:25 +0100)]
dm raid1: fix crash with mirror recovery and discard

This patch fixes a crash when a discard request is sent during mirror
recovery.

Firstly, some background.  Generally, the following sequence happens during
mirror synchronization:
- function do_recovery is called
- do_recovery calls dm_rh_recovery_prepare
- dm_rh_recovery_prepare uses a semaphore to limit the number
  simultaneously recovered regions (by default the semaphore value is 1,
  so only one region at a time is recovered)
- dm_rh_recovery_prepare calls __rh_recovery_prepare,
  __rh_recovery_prepare asks the log driver for the next region to
  recover. Then, it sets the region state to DM_RH_RECOVERING. If there
  are no pending I/Os on this region, the region is added to
  quiesced_regions list. If there are pending I/Os, the region is not
  added to any list. It is added to the quiesced_regions list later (by
  dm_rh_dec function) when all I/Os finish.
- when the region is on quiesced_regions list, there are no I/Os in
  flight on this region. The region is popped from the list in
  dm_rh_recovery_start function. Then, a kcopyd job is started in the
  recover function.
- when the kcopyd job finishes, recovery_complete is called. It calls
  dm_rh_recovery_end. dm_rh_recovery_end adds the region to
  recovered_regions or failed_recovered_regions list (depending on
  whether the copy operation was successful or not).

The above mechanism assumes that if the region is in DM_RH_RECOVERING
state, no new I/Os are started on this region. When I/O is started,
dm_rh_inc_pending is called, which increases reg->pending count. When
I/O is finished, dm_rh_dec is called. It decreases reg->pending count.
If the count is zero and the region was in DM_RH_RECOVERING state,
dm_rh_dec adds it to the quiesced_regions list.

Consequently, if we call dm_rh_inc_pending/dm_rh_dec while the region is
in DM_RH_RECOVERING state, it could be added to quiesced_regions list
multiple times or it could be added to this list when kcopyd is copying
data (it is assumed that the region is not on any list while kcopyd does
its jobs). This results in memory corruption and crash.

There already exist bypasses for REQ_FLUSH requests: REQ_FLUSH requests
do not belong to any region, so they are always added to the sync list
in do_writes. dm_rh_inc_pending does not increase count for REQ_FLUSH
requests. In mirror_end_io, dm_rh_dec is never called for REQ_FLUSH
requests. These bypasses avoid the crash possibility described above.

These bypasses were improperly implemented for REQ_DISCARD when
the mirror target gained discard support in commit
5fc2ffeabb9ee0fc0e71ff16b49f34f0ed3d05b4 (dm raid1: support discard).

In do_writes, REQ_DISCARD requests is always added to the sync queue and
immediately dispatched (even if the region is in DM_RH_RECOVERING).  However,
dm_rh_inc and dm_rh_dec is called for REQ_DISCARD resusts.  So it violates the
rule that no I/Os are started on DM_RH_RECOVERING regions, and causes the list
corruption described above.

This patch changes it so that REQ_DISCARD requests follow the same path
as REQ_FLUSH. This avoids the crash.

Reference: https://bugzilla.redhat.com/837607

Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Cc: stable@kernel.org
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
12 years agopnfs-obj: Fix __r4w_get_page when offset is beyond i_size
Boaz Harrosh [Thu, 7 Jun 2012 23:02:30 +0000 (02:02 +0300)]
pnfs-obj: Fix __r4w_get_page when offset is beyond i_size

It is very common for the end of the file to be unaligned on
stripe size. But since we know it's beyond file's end then
the XOR should be preformed with all zeros.

Old code used to just read zeros out of the OSD devices, which is a great
waist. But what scares me more about this situation is that, we now have
pages attached to the file's mapping that are beyond i_size. I don't
like the kind of bugs this calls for.

Fix both birds, by returning a global zero_page, if offset is beyond
i_size.

TODO:
Change the API to ->__r4w_get_page() so a NULL can be
returned without being considered as error, since XOR API
treats NULL entries as zero_pages.

[Bug since 3.2. Should apply the same way to all Kernels since]
CC: Stable Tree <stable@kernel.org>
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
12 years agopnfs-obj: don't leak objio_state if ore_write/read fails
Boaz Harrosh [Fri, 8 Jun 2012 02:29:40 +0000 (05:29 +0300)]
pnfs-obj: don't leak objio_state if ore_write/read fails

[Bug since 3.2 Kernel]
CC: Stable Tree <stable@kernel.org>
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
12 years agoore: Unlock r4w pages in exact reverse order of locking
Boaz Harrosh [Wed, 11 Jul 2012 12:27:13 +0000 (15:27 +0300)]
ore: Unlock r4w pages in exact reverse order of locking

The read-4-write pages are locked in address ascending order.
But where unlocked in a way easiest for coding. Fix that,
locks should be released in opposite order of locking, .i.e
descending address order.

I have not hit this dead-lock. It was found by inspecting the
dbug print-outs. I suspect there is an higher lock at caller that
protects us, but fix it regardless.

Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
12 years agoore: Remove support of partial IO request (NFS crash)
Boaz Harrosh [Fri, 8 Jun 2012 01:30:40 +0000 (04:30 +0300)]
ore: Remove support of partial IO request (NFS crash)

Do to OOM situations the ore might fail to allocate all resources
needed for IO of the full request. If some progress was possible
it would proceed with a partial/short request, for the sake of
forward progress.

Since this crashes NFS-core and exofs is just fine without it just
remove this contraption, and fail.

TODO:
Support real forward progress with some reserved allocations
of resources, such as mem pools and/or bio_sets

[Bug since 3.2 Kernel]
CC: Stable Tree <stable@kernel.org>
CC: Benny Halevy <bhalevy@tonian.com>
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
12 years agoore: Fix NFS crash by supporting any unaligned RAID IO
Boaz Harrosh [Thu, 7 Jun 2012 22:19:07 +0000 (01:19 +0300)]
ore: Fix NFS crash by supporting any unaligned RAID IO

In RAID_5/6 We used to not permit an IO that it's end
byte is not stripe_size aligned and spans more than one stripe.
.i.e the caller must check if after submission the actual
transferred bytes is shorter, and would need to resubmit
a new IO with the remainder.

Exofs supports this, and NFS was supposed to support this
as well with it's short write mechanism. But late testing has
exposed a CRASH when this is used with none-RPC layout-drivers.

The change at NFS is deep and risky, in it's place the fix
at ORE to lift the limitation is actually clean and simple.
So here it is below.

The principal here is that in the case of unaligned IO on
both ends, beginning and end, we will send two read requests
one like old code, before the calculation of the first stripe,
and also a new site, before the calculation of the last stripe.
If any "boundary" is aligned or the complete IO is within a single
stripe. we do a single read like before.

The code is clean and simple by splitting the old _read_4_write
into 3 even parts:
1._read_4_write_first_stripe
2. _read_4_write_last_stripe
3. _read_4_write_execute

And calling 1+3 at the same place as before. 2+3 before last
stripe, and in the case of all in a single stripe then 1+2+3
is preformed additively.

Why did I not think of it before. Well I had a strike of
genius because I have stared at this code for 2 years, and did
not find this simple solution, til today. Not that I did not try.

This solution is much better for NFS than the previous supposedly
solution because the short write was dealt  with out-of-band after
IO_done, which would cause for a seeky IO pattern where as in here
we execute in order. At both solutions we do 2 separate reads, only
here we do it within a single IO request. (And actually combine two
writes into a single submission)

NFS/exofs code need not change since the ORE API communicates the new
shorter length on return, what will happen is that this case would not
occur anymore.

hurray!!

[Stable this is an NFS bug since 3.2 Kernel should apply cleanly]
CC: Stable Tree <stable@kernel.org>
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
12 years agoUBIFS: fix a bug in empty space fix-up
Artem Bityutskiy [Sat, 14 Jul 2012 11:33:09 +0000 (14:33 +0300)]
UBIFS: fix a bug in empty space fix-up

UBIFS has a feature called "empty space fix-up" which is a quirk to work-around
limitations of dumb flasher programs. Namely, of those flashers that are unable
to skip NAND pages full of 0xFFs while flashing, resulting in empty space at
the end of half-filled eraseblocks to be unusable for UBIFS. This feature is
relatively new (introduced in v3.0).

The fix-up routine (fixup_free_space()) is executed only once at the very first
mount if the superblock has the 'space_fixup' flag set (can be done with -F
option of mkfs.ubifs). It basically reads all the UBIFS data and metadata and
writes it back to the same LEB. The routine assumes the image is pristine and
does not have anything in the journal.

There was a bug in 'fixup_free_space()' where it fixed up the log incorrectly.
All but one LEB of the log of a pristine file-system are empty. And one
contains just a commit start node. And 'fixup_free_space()' just unmapped this
LEB, which resulted in wiping the commit start node. As a result, some users
were unable to mount the file-system next time with the following symptom:

UBIFS error (pid 1): replay_log_leb: first log node at LEB 3:0 is not CS node
UBIFS error (pid 1): replay_log_leb: log error detected while replaying the log at LEB 3:0

The root-cause of this bug was that 'fixup_free_space()' wrongly assumed
that the beginning of empty space in the log head (c->lhead_offs) was known
on mount. However, it is not the case - it was always 0. UBIFS does not store
in it the master node and finds out by scanning the log on every mount.

The fix is simple - just pass commit start node size instead of 0 to
'fixup_leb()'.

Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@linux.intel.com>
Cc: stable@vger.kernel.org [v3.0+]
Reported-by: Iwo Mergler <Iwo.Mergler@netcommwireless.com>
Tested-by: Iwo Mergler <Iwo.Mergler@netcommwireless.com>
Reported-by: James Nute <newten82@gmail.com>
12 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/sage/ceph...
Linus Torvalds [Thu, 19 Jul 2012 23:11:28 +0000 (16:11 -0700)]
Merge branch 'for-linus' of git://git./linux/kernel/git/sage/ceph-client

Pull last minute Ceph fixes from Sage Weil:
 "The important one fixes a bug in the socket failure handling behavior
  that was turned up in some recent failure injection testing.  The
  other two are minor bug fixes."

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/sage/ceph-client:
  rbd: endian bug in rbd_req_cb()
  rbd: Fix ceph_snap_context size calculation
  libceph: fix messenger retry

12 years agoMerge tag 'md-3.5-fixes' of git://neil.brown.name/md
Linus Torvalds [Thu, 19 Jul 2012 15:27:13 +0000 (08:27 -0700)]
Merge tag 'md-3.5-fixes' of git://neil.brown.name/md

Pull three md bugfixes from NeilBrown:
 "One of the bugs was introduced in 3.5-rc1.  Others have been there for
  longer."

* tag 'md-3.5-fixes' of git://neil.brown.name/md:
  md/raid1: close some possible races on write errors during resync
  md: avoid crash when stopping md array races with closing other open fds.
  md: fix bug in handling of new_data_offset

12 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
Linus Torvalds [Thu, 19 Jul 2012 15:21:13 +0000 (08:21 -0700)]
Merge git://git./linux/kernel/git/davem/net

Pull networking changes from David Miller:
 "Ok, we should be good to go now"

1) We have to statically initialize the init_net device list head rather
   than do so in an initcall, otherwise netprio_cgroup crashes if it's
   built statically rather than modular (Mark D.  Rustad)

2) Fix SKB null oopser in CIPSO ipv4 option processing (Paul Moore)

3) Qlogic maintainers update (Anirban Chakraborty)

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net:
  net: Statically initialize init_net.dev_base_head
  MAINTAINERS: Changes in qlcnic and qlge maintainers list
  cipso: don't follow a NULL pointer when setsockopt() is called

12 years agoMerge branch 'upstream-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jikos/hid
Linus Torvalds [Thu, 19 Jul 2012 15:15:55 +0000 (08:15 -0700)]
Merge branch 'upstream-fixes' of git://git./linux/kernel/git/jikos/hid

Pull HID update from Jiri Kosina:
 "A final round of changes for HID for 3.5: just device ID additions."

* 'upstream-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jikos/hid:
  HID: hid-multitouch: add support for Zytronic panels
  HID: add Sennheiser BTD500USB device support
  HID: add battery quirk for Apple Wireless ANSI

12 years agocx25821: Remove bad strcpy to read-only char*
Ezequiel Garcia [Wed, 18 Jul 2012 13:05:26 +0000 (10:05 -0300)]
cx25821: Remove bad strcpy to read-only char*

The strcpy was being used to set the name of the board.  Since the
destination char* was read-only and the name is set statically at
compile time; this was both wrong and redundant.

The type of char* is changed to const char* to prevent future errors.

Reported-by: Radek Masin <radek@masin.eu>
Signed-off-by: Ezequiel Garcia <elezegarcia@gmail.com>
[ Taking directly due to vacations   - Linus ]
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agoHID: hid-multitouch: add support for Zytronic panels
Benjamin Tissoires [Tue, 19 Jun 2012 12:39:54 +0000 (14:39 +0200)]
HID: hid-multitouch: add support for Zytronic panels

Signed-off-by: Benjamin Tissoires <benjamin.tissoires@enac.fr>
Signed-off-by: Jiri Kosina <jkosina@suse.cz>
12 years agoMIPS: PCI: Move fixups from __init to __devinit.
Sebastian Andrzej Siewior [Thu, 19 Jul 2012 05:13:54 +0000 (07:13 +0200)]
MIPS: PCI: Move fixups from __init to __devinit.

Fixups are executed once the pci-device is found which is during boot
process so __init seems fine as long as the platform does not support
hotplug.
However it is possible to remove the PCI bus at run time and have it
rediscovered again via "echo 1 > /sys/bus/pci/rescan" and this will call
the fixups again.

[ralf@linux-mips.org: Made piixirqmap[] in malta_piix_func0_fixup()
__initdata.]

Signed-off-by: Sebastian Andrzej Siewior <sebastian@breakpoint.cc>
Cc: linux-mips@linux-mips.org
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
12 years agoMIPS: Fix bug.h MIPS build regression
Yoichi Yuasa [Thu, 19 Jul 2012 05:13:54 +0000 (07:13 +0200)]
MIPS: Fix bug.h MIPS build regression

Commit: 3777808873b0c49c5cf27e44c948dfb02675d578 [bug.h: need linux/kernel.h
for TAINT_WARN.] breaks all MIPS builds.

  CC      arch/mips/kernel/machine_kexec.o
In file included from include/linux/kernel.h:20:0,
                 from include/asm-generic/bug.h:35,
                 from /home/yuasa/src/linux/kernel/git/linux-2.6/arch/mips/include/asm/bug.h:41,
                 from /home/yuasa/src/linux/kernel/git/linux-2.6/arch/mips/include/asm/bitops.h:20,
                 from include/linux/bitops.h:22,
                 from include/linux/signal.h:38,
                 from include/linux/elfcore.h:5,
                 from include/linux/kexec.h:60,
                 from arch/mips/kernel/machine_kexec.c:9:
include/linux/log2.h: In function '__ilog2_u32':
include/linux/log2.h:34:2: error: implicit declaration of function 'fls' [-Werror=implicit-function-declaration]
include/linux/log2.h: In function '__ilog2_u64':
include/linux/log2.h:42:2: error: implicit declaration of function 'fls64' [-Werror=implicit-function-declaration]
include/linux/log2.h: In function '__roundup_pow_of_two':
include/linux/log2.h:63:2: error: implicit declaration of function 'fls_long' [-Werror=implicit-function-declaration]
In file included from include/linux/bitops.h:22:0,
                 from include/linux/signal.h:38,
                 from include/linux/elfcore.h:5,
                 from include/linux/kexec.h:60,
                 from arch/mips/kernel/machine_kexec.c:9:
/home/yuasa/src/linux/kernel/git/linux-2.6/arch/mips/include/asm/bitops.h: At top level:
/home/yuasa/src/linux/kernel/git/linux-2.6/arch/mips/include/asm/bitops.h:615:19: error: static declaration of 'fls' follows non-static declaration
include/linux/log2.h:34:9: note: previous implicit declaration of 'fls' was here
In file included from /home/yuasa/src/linux/kernel/git/linux-2.6/arch/mips/include/asm/bitops.h:651:0,
                 from include/linux/bitops.h:22,
                 from include/linux/signal.h:38,
                 from include/linux/elfcore.h:5,
                 from include/linux/kexec.h:60,
                 from arch/mips/kernel/machine_kexec.c:9:
include/asm-generic/bitops/fls64.h:18:28: error: static declaration of 'fls64' follows non-static declaration
include/linux/log2.h:42:9: note: previous implicit declaration of 'fls64' was here
In file included from include/linux/signal.h:38:0,
                 from include/linux/elfcore.h:5,
                 from include/linux/kexec.h:60,
                 from arch/mips/kernel/machine_kexec.c:9:
include/linux/bitops.h:160:24: error: conflicting types for 'fls_long'
include/linux/log2.h:63:16: note: previous implicit declaration of 'fls_long' was here
cc1: all warnings being treated as errors

make[2]: *** [arch/mips/kernel/machine_kexec.o] Error 1

Signed-off-by: Yoichi Yuasa <yuasa@linux-mips.org>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Paul Mundt <lethal@linux-sh.org>
Cc: yuasa@linux-mips.org
Cc: linux-kernel@vger.kernel.org
Cc: Linuxppc-dev <linuxppc-dev@ozlabs.org>
Cc: Linux MIPS Mailing List <linux-mips@linux-mips.org>
Cc: Linux-sh list <linux-sh@vger.kernel.org>
Cc: Chris Zankel <chris@zankel.net>
Patchwork: https://patchwork.linux-mips.org/patch/4000/
Tested-by: John Crispin <blogic@openwrt.org>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
12 years agoMIPS: sync-r4k: remove redundant irq operation
Yong Zhang [Thu, 19 Jul 2012 07:13:54 +0000 (09:13 +0200)]
MIPS: sync-r4k: remove redundant irq operation

Since we have delayed irq enabling to ->smp_finish()

Signed-off-by: Yong Zhang <yong.zhang0@gmail.com>
Cc: Sergei Shtylyov <sshtylyov@mvista.com>
Cc: David Daney <david.daney@cavium.com>
Acked-by: David Daney <david.daney@cavium.com>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
12 years agoMIPS: smp: Warn on too early irq enable
Yong Zhang [Thu, 19 Jul 2012 07:13:53 +0000 (09:13 +0200)]
MIPS: smp: Warn on too early irq enable

Just to catch a potential issue.

Signed-off-by: Yong Zhang <yong.zhang0@gmail.com>
Cc: Sergei Shtylyov <sshtylyov@mvista.com>
Cc: David Daney <david.daney@cavium.com>
Acked-by: David Daney <david.daney@cavium.com>
Patchwork: https://patchwork.linux-mips.org/patch/3852/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
12 years agoMIPS: call set_cpu_online() on cpu being brought up with irq disabled
Yong Zhang [Thu, 19 Jul 2012 07:13:53 +0000 (09:13 +0200)]
MIPS: call set_cpu_online() on cpu being brought up with irq disabled

To prevent a problem as commit 5fbd036b [sched: Cleanup cpu_active madness]
and commit 2baab4e9 [sched: Fix select_fallback_rq() vs cpu_active/cpu_online]
try to resolve, move set_cpu_online() to the brought up CPU and with irq
disabled.

Signed-off-by: Yong Zhang <yong.zhang0@gmail.com>
Cc: Sergei Shtylyov <sshtylyov@mvista.com>
Cc: David Daney <david.daney@cavium.com>
Acked-by: David Daney <david.daney@cavium.com>
Patchwork: https://patchwork.linux-mips.org/patch/3851/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
12 years agoMIPS: call ->smp_finish() a little late
Yong Zhang [Thu, 19 Jul 2012 07:13:53 +0000 (09:13 +0200)]
MIPS: call ->smp_finish() a little late

We have move irq enable to ->smp_finish. Place ->smp_finish() a little
late to prepare for move set_cpu_online() into start_secondary.
And it's not necessary to call cpu_set(cpu, cpu_callin_map) and
synchronise_count_slave() with irq enabled.

Signed-off-by: Yong Zhang <yong.zhang0@gmail.com>
Cc: Sergei Shtylyov <sshtylyov@mvista.com>
Cc: David Daney <david.daney@cavium.com>
Acked-by: David Daney <david.daney@cavium.com>
Patchwork: https://patchwork.linux-mips.org/patch/3850/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
12 years agoMIPS: Yosemite: delay irq enable to ->smp_finish()
Yong Zhang [Thu, 19 Jul 2012 07:13:53 +0000 (09:13 +0200)]
MIPS: Yosemite: delay irq enable to ->smp_finish()

To prepare for smoothing set_cpu_[active|online]() mess up

Signed-off-by: Yong Zhang <yong.zhang0@gmail.com>
Cc: Sergei Shtylyov <sshtylyov@mvista.com>
Cc: David Daney <david.daney@cavium.com>
Acked-by: David Daney <david.daney@cavium.com>
Patchwork: https://patchwork.linux-mips.org/patch/3848/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
12 years agoMIPS: SMTC: delay irq enable to ->smp_finish()
Yong Zhang [Thu, 19 Jul 2012 07:13:53 +0000 (09:13 +0200)]
MIPS: SMTC: delay irq enable to ->smp_finish()

To prepare for smoothing set_cpu_[active|online]() mess up

Signed-off-by: Yong Zhang <yong.zhang0@gmail.com>
Cc: Sergei Shtylyov <sshtylyov@mvista.com>
Cc: David Daney <david.daney@cavium.com>
Acked-by: David Daney <david.daney@cavium.com>
Patchwork: https://patchwork.linux-mips.org/patch/3847/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
12 years agoMIPS: BMIPS: delay irq enable to ->smp_finish()
Yong Zhang [Thu, 19 Jul 2012 07:13:53 +0000 (09:13 +0200)]
MIPS: BMIPS: delay irq enable to ->smp_finish()

To prepare for smoothing set_cpu_[active|online]() mess up

Signed-off-by: Yong Zhang <yong.zhang0@gmail.com>
Cc: Sergei Shtylyov <sshtylyov@mvista.com>
Cc: David Daney <david.daney@cavium.com>
Acked-by: David Daney <david.daney@cavium.com>
Patchwork: https://patchwork.linux-mips.org/patch/3846/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>