HD-audio has quite a few asymmetrical ways of accessing verbs, and one
of typical ones is GET/SET_POWER_STATE verbs. While it takes only the
power state for setting, it returns a combination of states for
getting. For making the state handling simpler, this patch adds a
code to translate the value returned from GET_POWER_STATE to return
only the actual state or -1 for error. In that way, the driver can
simplify the power state management.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
HD-audio spec is inconvenient regarding the handling of stereo volume
controls. It can set and get only single channel at once (although
there is a special option to set the same value to both channels).
This patch provides a fake pseudo-register via the regmap access so
that the stereo channels can be read and written by a single call.
It'd be useful, for example, for implementing DAPM widgets.
A stereo amp pseudo register consists of the encoding like the normal
amp verbs but it has both SET_LEFT (bit 13) and SET_RIGHT (bit 12)
bits set. The regmap reads and writes a 16bit value for this pseudo
register where the upper 8bit is for the right chanel and the lower
8bit for the left channel.
Note that the driver doesn't recognize conflicts when both stereo and
mono channel registers are mixed. Mixing them would certainly confuse
the operation. So, use carefully.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Like the previous patches, this patch converts also to the regmap, at
this time, the cached verb writes are the target. But this conversion
needs a bit more caution than before.
- In the old code, we just record any verbs as is, and restore them at
resume. For the regmap scheme, this doesn't work, since a few verbs
like AMP or DIGI_CONVERT are asymmetrical. Such verbs are converted
either to the dedicated function (snd_hda_regmap_xxx_amp()) or
changed to the unified verb.
- Some verbs have to be declared as vendor-specific ones before
accessing via regmap.
Also, the minor optimization with codec->cached_write flag is dropped
in a few places, as this would confuse the operation. Further
optimizations will be brought in the later patches, if any.
This conversion ends up with a drop of significant amount of codes,
mostly the helper codes that are no longer used.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Codecs may have own vendor-specific verbs, and we need to allow each
driver to give such verbs for cached accesses. Here a verb can be put
into a single array and looked through it at readable and writeable
callbacks.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
The amp hash table was used for recording the cached reads of some
capability values like pin caps or amp caps. Now all these are moved
to regmap as well.
One addition to the regmap helper is codec->caps_overwriting flag.
This is set in snd_hdac_override_parm(), and the regmap helper accepts
any register while this flag is set, so that it can overwrite even the
read-only verb like AC_VERB_PARAMETERS. The flag is cleared
immediately in snd_hdac_override_parm(), as it's a once-off flag.
Along with these changes, the no longer needed amp hash and relevant
fields are removed from hda_codec struct now.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
This patch converts the amp access functions to the regmap helpers.
The amp values were formerly cached in the own hash table. Now it's
dropped by the regmap's cache.
The only tricky conversion is snd_hda_codec_amp_init(). This function
shouldn't do anything if the amp was already initialized. For
achieving this behavior, a value is read once at first temporarily in
the cache-only mode. Only if it returns an error, i.e. the item
still doesn't exist in the cache, it proceeds to the update.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Sometimes we need the uncached reads, e.g. for refreshing the tree.
This patch provides the helper function for that and uses it for
refreshing widgets, reading subtrees and the whole proc reads.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Let's start converting the access functions to regmap.
The first one is the simplest, just converting the codec parameter
read helper function snd_hda_param_read().
Signed-off-by: Takashi Iwai <tiwai@suse.de>
This patch adds an infrastructure to support regmap-based verb
accesses. Because o the asymmetric nature of HD-audio verbs,
especially the amp verbs, we need to translate the verbs as a sort of
pseudo registers to be mapped uniquely in regmap.
In this patch, a pseudo register is built from the NID, the
AC_VERB_GET_* and 8bit parameters, i.e. almost in the form to be sent
to HD-audio bus but without codec address field. OTOH, for writing,
the same pseudo register is translated to AC_VERB_SET_* automatically.
The AC_VERB_SET_AMP_* verb is re-encoded from the corresponding
AC_VERB_GET_AMP_* verb and parameter at writing.
Some verbs has a single command for read but multiple for writes. A
write for such a verb is split automatically to multiple verbs.
The patch provides also a few handy helper functions. They are
designed to be accessible even without regmap. When no regmap is set
up (e.g. before the codec device instantiation), the direct hardware
access is used. Also, it tries to avoid the unnecessary power-up.
The power up/down sequence is performed only on demand.
The codec driver needs to call snd_hdac_regmap_exit() and
snd_hdac_regmap_exit() at probe and remove if it wants the regmap
access.
There is one flag added to hdac_device. When the flag lazy_cache is
set, regmap helper ignores a write for a suspended device and returns
as if it was actually written. It reduces the hardware access pretty
much, e.g. when adjusting the mixer volume while in idle. This
assumes that the driver will sync the cache later at resume properly,
so use it carefully.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Now let's take the basic tracepoints back to the HD-audio driver.
The three bus tracepoints, hda_send_cmd, hda_get_response and
hda_unsol_event are revived but in a slightly different form.
Since we don't assign the card number there, print the bus device name
instead.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
When the driver is unloaded before the codec is bound, it still keeps
the runtime PM refcount up, and results in the unbalance. This patch
covers these cases by introducing a flag indicating the runtime PM
initialization and handling the codec registration procedure more
properly. It also fixes the missing input beep device as a gratis,
too.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Add an overriding exec_verb op to struct hdac_device so that the call
via snd_hdac_exec_verb() can switch to a different route depending on
the setup. The codec driver sets this field so that it can handle the
errors or applying quirks appropriately. Furthermore, this mechanism
will be used for smooth transition for the regmap support in later
patches.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
This patch changes the sysfs files assigned to the codec device on the
bus which were formerly identical with hwdep sysfs files. Now it
shows only a few core parameter, vendor_id, subsystem_id, revision_id,
afg, mfg, vendor_name and chip_name.
In addition, now a widget tree is added to the bus device sysfs
directory for showing the widget topology and attributes. It's just a
flat tree consisting of subdirectories named as the widget NID
including various attributes like widget capability bits. The AFG
(usually NID 0x01) is always found there, and it contains always
amp_in_caps, amp_out_caps and power_caps files. Each of these
attributes show a single value. The rest are the widget nodes
belonging to that AFG. Note that the child node might not start from
0x02 but from another value like 0x0a.
Each child node may contain caps, pin_caps, amp_in_caps, amp_out_caps,
power_caps and connections files. The caps (representing the widget
capability bits) always contain a value. The rest may contain
value(s) if the attribute exists on the node. Only connections file
show multiple values while other attributes have zero or one single
value.
An example of ls -R output is like below:
% ls -R /sys/bus/hdaudio/devices/hdaudioC0D0/
/sys/bus/hdaudio/devices/hdaudioC0D0/widgets/:
01/ 04/ 07/ 0a/ 0d/ 10/ 13/ 16/ 19/ 1c/ 1f/ 22/
02/ 05/ 08/ 0b/ 0e/ 11/ 14/ 17/ 1a/ 1d/ 20/ 23/
03/ 06/ 09/ 0c/ 0f/ 12/ 15/ 18/ 1b/ 1e/ 21/
/sys/bus/hdaudio/devices/hdaudioC0D0/widgets/01:
amp_in_caps amp_out_caps power_caps
/sys/bus/hdaudio/devices/hdaudioC0D0/widgets/02:
amp_in_caps amp_out_caps caps connections pin_caps pin_cfg
power_caps
/sys/bus/hdaudio/devices/hdaudioC0D0/widgets/03:
.....
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Now some codes and functionalities of hda_codec struct are moved to
hdac_device struct. A few basic attributes like the codec address,
vendor ID number, FG numbers, etc are moved to hdac_device, and they
are accessed like codec->core.addr. The basic verb exec functions are
moved, too.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
A few basic codes for communicating over HD-audio bus are moved to
struct hdac_bus now. It has only command and get_response ops in
addition to the unsolicited event handling.
Note that the codec-side tracing support is disabled temporarily
during this transition due to the code shuffling. It will be
re-enabled later once when all pieces are settled down.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Define the common hd-audio driver and device types to bind over
snd_hda_bus_type publicly. This allows to implement other type of
device and driver code over hd-audio bus.
Now both struct hda_codec and struct hda_codec_driver inherit these
new struct hdac_device and struct hdac_driver, respectively.
The bus registration is done in subsys_initcall() to assure it
before any other driver registrations.
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Use %pS for actual addresses, otherwise you'll get bad output
on arches like ppc64 where %pF expects a function descriptor.
Signed-off-by: Scott Wood <scottwood@freescale.com>
Signed-off-by: Lee Jones <lee.jones@linaro.org>
Even if bus is not hot-pluggable, the devices can be unbound from the
driver via sysfs, so we should not be using __exit annotations on
remove() methods. The only exception is drivers registered with
platform_driver_probe() which specifically disables sysfs bind/unbind
attributes.
Signed-off-by: Dmitry Torokhov <dmitry.torokhov@gmail.com>
Acked-by: Tony Lindgren <tony@atomide.com>
Signed-off-by: Lee Jones <lee.jones@linaro.org>
This patch modifies the HYP init code so it can deal with system
RAM residing at an offset which exceeds the reach of VA_BITS.
Like for EL1, this involves configuring an additional level of
translation for the ID map. However, in case of EL2, this implies
that all translations use the extra level, as we cannot seamlessly
switch between translation tables with different numbers of
translation levels.
So add an extra translation table at the root level. Since the
ID map and the runtime HYP map are guaranteed not to overlap, they
can share this root level, and we can essentially merge these two
tables into one.
Tested-by: Marc Zyngier <marc.zyngier@arm.com>
Reviewed-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Will Deacon <will.deacon@arm.com>
The page size and the number of translation levels, and hence the supported
virtual address range, are build-time configurables on arm64 whose optimal
values are use case dependent. However, in the current implementation, if
the system's RAM is located at a very high offset, the virtual address range
needs to reflect that merely because the identity mapping, which is only used
to enable or disable the MMU, requires the extended virtual range to map the
physical memory at an equal virtual offset.
This patch relaxes that requirement, by increasing the number of translation
levels for the identity mapping only, and only when actually needed, i.e.,
when system RAM's offset is found to be out of reach at runtime.
Tested-by: Laura Abbott <lauraa@codeaurora.org>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Tested-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Will Deacon <will.deacon@arm.com>
Rework of the KVM HYP bounce page from Ard Biesheuvel. Subsequent arm64
idmap rework depends on this, so merge it here with Marc Zyngier's
blessing (kvm-arm co-maintainer).
Make the code which sets up the pmd depend on PT_NLEVELS == 3, not on
CONFIG_64BIT. The reason is, that a 64bit kernel with a page size
greater than 4k doesn't need the pmd and thus has PT_NLEVELS = 2.
Signed-off-by: Helge Deller <deller@gmx.de>
The patch dc6c9a35b6 that counts pmds
allocated for a process introduced a bug on 64-bit PA-RISC kernels.
The PA-RISC architecture preallocates one pmd with each pgd. This
preallocated pmd can never be freed - pmd_free does nothing when it is
called with this pmd. When the kernel attempts to free this preallocated
pmd, it decreases the count of allocated pmds. The result is that the
counter underflows and this error is reported.
This patch fixes the bug by artifically incrementing the counter in
pmd_free when the kernel tries to free the preallocated pmd.
Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Helge Deller <deller@gmx.de>
Commit 06f75a1f62 ("ARM, arm64: kvm: get rid of the bounce
page") uses ld's builtin function LOG2CEIL() to align the
KVM init code to a log2 upper bound of its size. However,
this function turns out to be a fairly recent addition to
binutils, which breaks the build for older toolchains.
So instead, implement a replacement LOG2_ROUNDUP() using
the C preprocessor.
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Will Deacon <will.deacon@arm.com>
"hdev->req->nbytes" is an unsigned int so we so we lose the upper 3 bits
to the shift wrap bug.
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
GCC complains about that %u is the wrong format string for size_t and
also that "ret" is unused.
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Convert pr_info() and pr_err() log messages to dev_info() and dev_err(),
respectively, where able. This adds the module name and PCI B:D:F to
indicate which QAT device generated the log message. The "QAT:" is removed
from these log messages as that is now unnecessary. A few of these log
messages have additional spelling/contextual fixes.
Signed-off-by: Bruce Allan <bruce.w.allan@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
This define is a duplicate of the one in ./include/linux/pci_ids.h
Signed-off-by: Bruce Allan <bruce.w.allan@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The Kconfig entry for CRYPTO_DEV_IMGTEC_HASH incorrectly selects
CRYPTO_SHA224, which does not exist (and is covered by CRYPTO_SHA256
which covers both 224 and 256). Remove it.
Also correct typo CRYPTO_ALG_API to be CRYPTO_ALGPI.
Reported-by: Valentin Rothberg <valentinrothberg@gmail.com>
Signed-off-by: James Hartley <james.hartley@imgtec.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
This allows us to remove some unnecessary ifdefs. There should
be no change to the generated code.
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Brad Spengler <spender@grsecurity.net>
Cc: Denys Vlasenko <dvlasenk@redhat.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/f7e00f0d668e253abf0bd8bf36491ac47bd761ff.1426728647.git.luto@kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
user_mode_vm() and user_mode() are now the same. Change all callers
of user_mode_vm() to user_mode().
The next patch will remove the definition of user_mode_vm.
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Brad Spengler <spender@grsecurity.net>
Cc: Denys Vlasenko <dvlasenk@redhat.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/43b1f57f3df70df5a08b0925897c660725015554.1426728647.git.luto@kernel.org
[ Merged to a more recent kernel. ]
Signed-off-by: Ingo Molnar <mingo@kernel.org>
user_mode() is now identical to user_mode_vm(). Subsequent patches
will change all callers of user_mode_vm() to user_mode() and then
delete user_mode_vm().
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Brad Spengler <spender@grsecurity.net>
Cc: Denys Vlasenko <dvlasenk@redhat.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/0dd03eacb5f0a2b5ba0240de25347a31b493c289.1426728647.git.luto@kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
A few of the user_mode() checks in traps.c are immediately after
explicit checks for vm86 mode. Change them to user_mode_ignore_vm86().
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Brad Spengler <spender@grsecurity.net>
Cc: Denys Vlasenko <dvlasenk@redhat.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/0b324d5b75c3402be07f8d3c6245ed7f4995029e.1426728647.git.luto@kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
There's no point in checking the VM bit on 64-bit, and, since
we're explicitly checking it, we can use user_mode_ignore_vm86()
after the check.
While we're at it, rearrange the #ifdef slightly to make the code
flow a bit clearer.
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Brad Spengler <spender@grsecurity.net>
Cc: Denys Vlasenko <dvlasenk@redhat.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/dc1457a734feccd03a19bb3538a7648582f57cdd.1426728647.git.luto@kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
user_mode() is dangerous and user_mode_vm() has a confusing name.
Add user_mode_ignore_vm86() (equivalent to current user_mode()).
We'll change the small number of legitimate users of user_mode()
to user_mode_ignore_vm86().
Inspired by grsec, although this works rather differently.
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Brad Spengler <spender@grsecurity.net>
Cc: Denys Vlasenko <dvlasenk@redhat.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/202c56ca63823c338af8e2e54948dbe222da6343.1426728647.git.luto@kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
When debugging the latencies on a 40 core box, where we hit 300 to
500 microsecond latencies, I found there was a huge contention on the
runqueue locks.
Investigating it further, running ftrace, I found that it was due to
the pulling of RT tasks.
The test that was run was the following:
cyclictest --numa -p95 -m -d0 -i100
This created a thread on each CPU, that would set its wakeup in iterations
of 100 microseconds. The -d0 means that all the threads had the same
interval (100us). Each thread sleeps for 100us and wakes up and measures
its latencies.
cyclictest is maintained at:
git://git.kernel.org/pub/scm/linux/kernel/git/clrkwllms/rt-tests.git
What happened was another RT task would be scheduled on one of the CPUs
that was running our test, when the other CPU tests went to sleep and
scheduled idle. This caused the "pull" operation to execute on all
these CPUs. Each one of these saw the RT task that was overloaded on
the CPU of the test that was still running, and each one tried
to grab that task in a thundering herd way.
To grab the task, each thread would do a double rq lock grab, grabbing
its own lock as well as the rq of the overloaded CPU. As the sched
domains on this box was rather flat for its size, I saw up to 12 CPUs
block on this lock at once. This caused a ripple affect with the
rq locks especially since the taking was done via a double rq lock, which
means that several of the CPUs had their own rq locks held while trying
to take this rq lock. As these locks were blocked, any wakeups or load
balanceing on these CPUs would also block on these locks, and the wait
time escalated.
I've tried various methods to lessen the load, but things like an
atomic counter to only let one CPU grab the task wont work, because
the task may have a limited affinity, and we may pick the wrong
CPU to take that lock and do the pull, to only find out that the
CPU we picked isn't in the task's affinity.
Instead of doing the PULL, I now have the CPUs that want the pull to
send over an IPI to the overloaded CPU, and let that CPU pick what
CPU to push the task to. No more need to grab the rq lock, and the
push/pull algorithm still works fine.
With this patch, the latency dropped to just 150us over a 20 hour run.
Without the patch, the huge latencies would trigger in seconds.
I've created a new sched feature called RT_PUSH_IPI, which is enabled
by default.
When RT_PUSH_IPI is not enabled, the old method of grabbing the rq locks
and having the pulling CPU do the work is implemented. When RT_PUSH_IPI
is enabled, the IPI is sent to the overloaded CPU to do a push.
To enabled or disable this at run time:
# mount -t debugfs nodev /sys/kernel/debug
# echo RT_PUSH_IPI > /sys/kernel/debug/sched_features
or
# echo NO_RT_PUSH_IPI > /sys/kernel/debug/sched_features
Update: This original patch would send an IPI to all CPUs in the RT overload
list. But that could theoretically cause the reverse issue. That is, there
could be lots of overloaded RT queues and one CPU lowers its priority. It would
then send an IPI to all the overloaded RT queues and they could then all try
to grab the rq lock of the CPU lowering its priority, and then we have the
same problem.
The latest design sends out only one IPI to the first overloaded CPU. It tries to
push any tasks that it can, and then looks for the next overloaded CPU that can
push to the source CPU. The IPIs stop when all overloaded CPUs that have pushable
tasks that have priorities greater than the source CPU are covered. In case the
source CPU lowers its priority again, a flag is set to tell the IPI traversal to
restart with the first RT overloaded CPU after the source CPU.
Parts-suggested-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Joern Engel <joern@purestorage.com>
Cc: Clark Williams <williams@redhat.com>
Cc: Mike Galbraith <umgwanakikbuti@gmail.com>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/20150318144946.2f3cc982@gandalf.local.home
Signed-off-by: Ingo Molnar <mingo@kernel.org>
When CONFIG_IRQ_WORK is not defined (difficult to do, as it also
requires CONFIG_PRINTK not to be defined), we get a build failure:
kernel/built-in.o: In function `flush_smp_call_function_queue':
kernel/smp.c:263: undefined reference to `irq_work_run'
kernel/smp.c:263: undefined reference to `irq_work_run'
Makefile:933: recipe for target 'vmlinux' failed
Simplest thing to do is to make irq_work_run() a nop when not set.
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Link: http://lkml.kernel.org/r/20150319101851.4d224d9b@gandalf.local.home
Signed-off-by: Ingo Molnar <mingo@kernel.org>
The hrtimer mode of broadcast queues hrtimers in the idle entry
path so as to wakeup cpus in deep idle states. The associated
call graph is :
cpuidle_idle_call()
|____ clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ENTER, ....))
|_____tick_broadcast_set_event()
|____clockevents_program_event()
|____bc_set_next()
The hrtimer_{start/cancel} functions call into tracing which uses RCU.
But it is not legal to call into RCU in cpuidle because it is one of the
quiescent states. Hence protect this region with RCU_NONIDLE which informs
RCU that the cpu is momentarily non-idle.
As an aside it is helpful to point out that the clock event device that is
programmed here is not a per-cpu clock device; it is a
pseudo clock device, used by the broadcast framework alone.
The per-cpu clock device programming never goes through bc_set_next().
Signed-off-by: Preeti U Murthy <preeti@linux.vnet.ibm.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: linuxppc-dev@ozlabs.org
Cc: mpe@ellerman.id.au
Cc: tglx@linutronix.de
Link: http://lkml.kernel.org/r/20150318104705.17763.56668.stgit@preeti.in.ibm.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Module unload calls lockdep_free_key_range(), which removes entries
from the data structures. Most of the lockdep code OTOH assumes the
data structures are append only; in specific see the comments in
add_lock_to_list() and look_up_lock_class().
Clearly this has only worked by accident; make it work proper. The
actual scenario to make it go boom would involve the memory freed by
the module unlock being re-allocated and re-used for a lock inside of
a rcu-sched grace period. This is a very unlikely scenario, still
better plug the hole.
Use RCU list iteration in all places and ammend the comments.
Change lockdep_free_key_range() to issue a sync_sched() between
removal from the lists and returning -- which results in the memory
being freed. Further ensure the callers are placed correctly and
comment the requirements.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Andrey Tsyvarev <tsyvarev@ispras.ru>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
When non-realtime tasks get priority-inheritance boosted to a realtime
scheduling class, RLIMIT_RTTIME starts to apply to them. However, the
counter used for checking this (the same one used for SCHED_RR
timeslices) was not getting reset. This meant that tasks running with a
non-realtime scheduling class which are repeatedly boosted to a realtime
one, but never block while they are running realtime, eventually hit the
timeout without ever running for a time over the limit. This patch
resets the realtime timeslice counter when un-PI-boosting from an RT to
a non-RT scheduling class.
I have some test code with two threads and a shared PTHREAD_PRIO_INHERIT
mutex which induces priority boosting and spins while boosted that gets
killed by a SIGXCPU on non-fixed kernels but doesn't with this patch
applied. It happens much faster with a CONFIG_PREEMPT_RT kernel, and
does happen eventually with PREEMPT_VOLUNTARY kernels.
Signed-off-by: Brian Silverman <brian@peloton-tech.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: austin@peloton-tech.com
Cc: <stable@vger.kernel.org>
Link: http://lkml.kernel.org/r/1424305436-6716-1-git-send-email-brian@peloton-tech.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>