commit 753649dbc4 upstream.
Network folks reported that directing all MSI-X vectors of their multi
queue NICs to a single core can cause interrupt stack overflows when
enough interrupts fire at the same time.
This is caused by the fact that we run interrupt handlers by default
with interrupts enabled unless the driver reuqests the interrupt with
the IRQF_DISABLED set. The NIC handlers do not set this flag, so
simultaneous interrupts can nest unlimited and cause the stack
overflow.
The only safe counter measure is to run the interrupt handlers with
interrupts disabled. We can't switch to this mode in general right
now, but it is safe to do so for MSI interrupts.
Force IRQF_DISABLED for MSI interrupt handlers.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: Linus Torvalds <torvalds@osdl.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Alan Cox <alan@lxorguk.ukuu.org.uk>
Cc: David Miller <davem@davemloft.net>
Cc: Greg Kroah-Hartman <gregkh@suse.de>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
* 'irq-threaded-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
genirq: Do not mask oneshot edge type interrupts
genirq: Support nested threaded irq handling
genirq: Add buslock support
genirq: Add oneshot support
The wake_up_process() of the new irq thread in __setup_irq() is too
early as the irqaction is not yet fully initialized especially
action->irq is not yet set. The interrupt thread might dereference the
wrong irq descriptor.
Move the wakeup after the action is installed and action->irq has been
set.
Reported-by: Michael Buesch <mb@bu3sch.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Michael Buesch <mb@bu3sch.de>
Interrupt chips which are behind a slow bus (i2c, spi ...) and
demultiplex other interrupt sources need to run their interrupt
handler in a thread.
The demultiplexed interrupt handlers need to run in thread context as
well and need to finish before the demux handler thread can reenable
the interrupt line. So the easiest way is to run the sub device
handlers in the context of the demultiplexing handler thread.
To avoid that a separate thread is created for the subdevices the
function set_nested_irq_thread() is provided which sets the
IRQ_NESTED_THREAD flag in the interrupt descriptor.
A driver which calls request_threaded_irq() must not be aware of the
fact that the threaded handler is called in the context of the
demultiplexing handler thread. The setup code checks the
IRQ_NESTED_THREAD flag which was set from the irq chip setup code and
does not setup a separate thread for the interrupt. The primary
function which is provided by the device driver is replaced by an
internal dummy function which warns when it is called.
For the demultiplexing handler a helper function handle_nested_irq()
is provided which calls the demux interrupt thread function in the
context of the caller and does the proper interrupt accounting and
takes the interrupt disabled status of the demultiplexed subdevice
into account.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Mark Brown <broonie@opensource.wolfsonmicro.com>
Cc: Dmitry Torokhov <dmitry.torokhov@gmail.com>
Cc: Trilok Soni <soni.trilok@gmail.com>
Cc: Pavel Machek <pavel@ucw.cz>
Cc: Brian Swetland <swetland@google.com>
Cc: Joonyoung Shim <jy0922.shim@samsung.com>
Cc: m.szyprowski@samsung.com
Cc: t.fujak@samsung.com
Cc: kyungmin.park@samsung.com,
Cc: David Brownell <david-b@pacbell.net>
Cc: Daniel Ribeiro <drwyrm@gmail.com>
Cc: arve@android.com
Cc: Barry Song <21cnbao@gmail.com>
Some interrupt chips are connected to a "slow" bus (i2c, spi ...). The
bus access needs to sleep and therefor cannot be called in atomic
contexts.
Some of the generic interrupt management functions like disable_irq(),
enable_irq() ... call interrupt chip functions with the irq_desc->lock
held and interrupts disabled. This does not work for such devices.
Provide a separate synchronization mechanism for such interrupt
chips. The irq_chip structure is extended by two optional functions
(bus_lock and bus_sync_and_unlock).
The idea is to serialize the bus access for those operations in the
core code so that drivers which are behind that bus operated interrupt
controller do not have to worry about it and just can use the normal
interfaces. To achieve this we add two function pointers to the
irq_chip: bus_lock and bus_sync_unlock.
bus_lock() is called to serialize access to the interrupt controller
bus.
Now the core code can issue chip->mask/unmask ... commands without
changing the fast path code at all. The chip implementation merily
stores that information in a chip private data structure and
returns. No bus interaction as these functions are called from atomic
context.
After that bus_sync_unlock() is called outside the atomic context. Now
the chip implementation issues the bus commands, waits for completion
and unlocks the interrupt controller bus.
The irq_chip implementation as pseudo code:
struct irq_chip_data {
struct mutex mutex;
unsigned int irq_offset;
unsigned long mask;
unsigned long mask_status;
}
static void bus_lock(unsigned int irq)
{
struct irq_chip_data *data = get_irq_desc_chip_data(irq);
mutex_lock(&data->mutex);
}
static void mask(unsigned int irq)
{
struct irq_chip_data *data = get_irq_desc_chip_data(irq);
irq -= data->irq_offset;
data->mask |= (1 << irq);
}
static void unmask(unsigned int irq)
{
struct irq_chip_data *data = get_irq_desc_chip_data(irq);
irq -= data->irq_offset;
data->mask &= ~(1 << irq);
}
static void bus_sync_unlock(unsigned int irq)
{
struct irq_chip_data *data = get_irq_desc_chip_data(irq);
if (data->mask != data->mask_status) {
do_bus_magic_to_set_mask(data->mask);
data->mask_status = data->mask;
}
mutex_unlock(&data->mutex);
}
The device drivers can use request_threaded_irq, free_irq, disable_irq
and enable_irq as usual with the only restriction that the calls need
to come from non atomic context.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Mark Brown <broonie@opensource.wolfsonmicro.com>
Cc: Dmitry Torokhov <dmitry.torokhov@gmail.com>
Cc: Trilok Soni <soni.trilok@gmail.com>
Cc: Pavel Machek <pavel@ucw.cz>
Cc: Brian Swetland <swetland@google.com>
Cc: Joonyoung Shim <jy0922.shim@samsung.com>
Cc: m.szyprowski@samsung.com
Cc: t.fujak@samsung.com
Cc: kyungmin.park@samsung.com,
Cc: David Brownell <david-b@pacbell.net>
Cc: Daniel Ribeiro <drwyrm@gmail.com>
Cc: arve@android.com
Cc: Barry Song <21cnbao@gmail.com>
For threaded interrupt handlers we expect the hard interrupt handler
part to mask the interrupt on the originating device. The interrupt
line itself is reenabled after the hard interrupt handler has
executed.
This requires access to the originating device from hard interrupt
context which is not always possible. There are devices which can only
be accessed via a bus (i2c, spi, ...). The bus access requires thread
context. For such devices we need to keep the interrupt line masked
until the threaded handler has executed.
Add a new flag IRQF_ONESHOT which allows drivers to request that the
interrupt is not unmasked after the hard interrupt context handler has
been executed and the thread has been woken. The interrupt line is
unmasked after the thread handler function has been executed.
Note that for now IRQF_ONESHOT cannot be used with IRQF_SHARED to
avoid complex accounting mechanisms.
For oneshot interrupts the primary handler simply returns
IRQ_WAKE_THREAD and does nothing else. A generic implementation
irq_default_primary_handler() is provided to avoid useless copies all
over the place. It is automatically installed when
request_threaded_irq() is called with handler=NULL and
thread_fn!=NULL.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Mark Brown <broonie@opensource.wolfsonmicro.com>
Cc: Dmitry Torokhov <dmitry.torokhov@gmail.com>
Cc: Trilok Soni <soni.trilok@gmail.com>
Cc: Pavel Machek <pavel@ucw.cz>
Cc: Brian Swetland <swetland@google.com>
Cc: Joonyoung Shim <jy0922.shim@samsung.com>
Cc: m.szyprowski@samsung.com
Cc: t.fujak@samsung.com
Cc: kyungmin.park@samsung.com,
Cc: David Brownell <david-b@pacbell.net>
Cc: Daniel Ribeiro <drwyrm@gmail.com>
Cc: arve@android.com
Cc: Barry Song <21cnbao@gmail.com>
free_irq() can remove an irqaction while the corresponding interrupt
is in progress, but free_irq() sets action->thread to NULL
unconditionally, which might lead to a NULL pointer dereference in
handle_IRQ_event() when the hard interrupt context tries to wake up
the handler thread.
Prevent this by moving the thread stop after synchronize_irq(). No
need to set action->thread to NULL either as action is going to be
freed anyway.
This fixes a boot crash reported against preempt-rt which uses the
mainline irq threads code to implement full irq threading.
[ tglx: removed local irqthread variable ]
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Since genirq: Delegate irq affinity setting to the irq thread
(591d2fb02e) compilation with
CONFIG_SMP=n fails with following error:
/usr/src/linux-2.6/kernel/irq/manage.c:
In function 'irq_thread_check_affinity':
/usr/src/linux-2.6/kernel/irq/manage.c:475:
error: 'struct irq_desc' has no member named 'affinity'
make[4]: *** [kernel/irq/manage.o] Error 1
That commit adds a new function irq_thread_check_affinity() which
uses struct irq_desc.affinity which is only available for CONFIG_SMP=y.
Move that function under #ifdef CONFIG_SMP.
[ tglx@brownpaperbag: compile and boot tested on UP and SMP ]
Signed-off-by: Bruno Premont <bonbons@linux-vserver.org>
LKML-Reference: <20090722222232.2eb3e1c4@neptune.home>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
irq_set_thread_affinity() calls set_cpus_allowed_ptr() which might
sleep, but irq_set_thread_affinity() is called with desc->lock held
and can be called from hard interrupt context as well. The code has
another bug as it does not hold a ref on the task struct as required
by set_cpus_allowed_ptr().
Just set the IRQTF_AFFINITY bit in action->thread_flags. The next time
the thread runs it migrates itself. Solves all of the above problems
nicely.
Add kerneldoc to irq_set_thread_affinity() while at it.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
LKML-Reference: <new-submission>
* 'irq-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
genirq, irq.h: Fix kernel-doc warnings
genirq: fix comment to say IRQ_WAKE_THREAD
Trying to implement a driver to use threaded irqs, I was confused when the
return value to use that was described in the comment above
request_threaded_irq was not defined.
Turns out that the enum is IRQ_WAKE_THREAD where as the comment said
IRQ_THREAD_WAKE.
[Impact: do not confuse developers with wrong comments ]
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
LKML-Reference: <alpine.DEB.2.00.0905121431020.13338@gandalf.stny.rr.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
irq_set_affinity() and move_masked_irq() try to assign affinity
before calling chip set_affinity(). Some archs are assigning it
in ->set_affinity() again.
We do something like:
cpumask_cpy(desc->affinity, mask);
desc->chip->set_affinity(mask);
But in the failure path, affinity should not be touched - otherwise
we'll end up with a different affinity mask despite the failure to
migrate the IRQ.
So try to update the afffinity only if set_affinity returns with 0.
Also call irq_set_thread_affinity accordingly.
v2: update after "irq, x86: Remove IRQ_DISABLED check in process context IRQ move"
v3: according to Ingo, change set_affinity() in irq_chip should return int.
v4: update comments by removing moving irq_desc code.
[ Impact: fix /proc/irq/*/smp_affinity setting corner case bug ]
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Suresh Siddha <suresh.b.siddha@intel.com>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Rusty Russell <rusty@rustcorp.com.au>
LKML-Reference: <49F65509.60307@kernel.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
As discussed in the thread here:
http://marc.info/?l=linux-kernel&m=123964468521142&w=2
Eric W. Biederman observed:
> It looks like some additional bugs have slipped in since last I looked.
>
> set_irq_affinity does this:
> ifdef CONFIG_GENERIC_PENDING_IRQ
> if (desc->status & IRQ_MOVE_PCNTXT || desc->status & IRQ_DISABLED) {
> cpumask_copy(desc->affinity, cpumask);
> desc->chip->set_affinity(irq, cpumask);
> } else {
> desc->status |= IRQ_MOVE_PENDING;
> cpumask_copy(desc->pending_mask, cpumask);
> }
> #else
>
> That IRQ_DISABLED case is a software state and as such it has nothing to
> do with how safe it is to move an irq in process context.
[...]
>
> The only reason we migrate MSIs in interrupt context today is that there
> wasn't infrastructure for support migration both in interrupt context
> and outside of it.
Yes. The idea here was to force the MSI migration to happen in process
context. One of the patches in the series did
disable_irq(dev->irq);
irq_set_affinity(dev->irq, cpumask_of(dev->cpu));
enable_irq(dev->irq);
with the above patch adding irq/manage code check for interrupt disabled
and moving the interrupt in process context.
IIRC, there was no IRQ_MOVE_PCNTXT when we were developing this HPET
code and we ended up having this ugly hack. IRQ_MOVE_PCNTXT was there
when we eventually submitted the patch upstream. But, looks like I did a
blind rebasing instead of using IRQ_MOVE_PCNTXT in hpet MSI code.
Below patch fixes this. i.e., revert commit 932775a4ab
and add PCNTXT to HPET MSI setup. Also removes copying of desc->affinity
in generic code as set_affinity routines are doing it internally.
Reported-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
Acked-by: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: "Li Shaohua" <shaohua.li@intel.com>
Cc: Gary Hade <garyhade@us.ibm.com>
Cc: "lcm@us.ibm.com" <lcm@us.ibm.com>
Cc: suresh.b.siddha@intel.com
LKML-Reference: <20090413222058.GB8211@linux-os.sc.intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Introduce helper functions allowing us to prevent device drivers from
getting any interrupts (without disabling interrupts on the CPU)
during suspend (or hibernation) and to make them start to receive
interrupts again during the subsequent resume. These functions make it
possible to keep timer interrupts enabled while the "late" suspend and
"early" resume callbacks provided by device drivers are being
executed. In turn, this allows device drivers' "late" suspend and
"early" resume callbacks to sleep, execute ACPI callbacks etc.
The functions introduced here will be used to rework the handling of
interrupts during suspend (hibernation) and resume. Namely,
interrupts will only be disabled on the CPU right before suspending
sysdevs, while device drivers will be prevented from receiving
interrupts, with the help of the new helper function, before their
"late" suspend callbacks run (and analogously during resume).
Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Acked-by: Ingo Molnar <mingo@elte.hu>
Delta patch to address the review comments.
- Implement warning when IRQ_WAKE_THREAD is requested and no
thread handler installed
- coding style fixes
Pointed-out-by: Christoph Hellwig <hch@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Add support for threaded interrupt handlers:
A device driver can request that its main interrupt handler runs in a
thread. To achive this the device driver requests the interrupt with
request_threaded_irq() and provides additionally to the handler a
thread function. The handler function is called in hard interrupt
context and needs to check whether the interrupt originated from the
device. If the interrupt originated from the device then the handler
can either return IRQ_HANDLED or IRQ_WAKE_THREAD. IRQ_HANDLED is
returned when no further action is required. IRQ_WAKE_THREAD causes
the genirq code to invoke the threaded (main) handler. When
IRQ_WAKE_THREAD is returned handler must have disabled the interrupt
on the device level. This is mandatory for shared interrupt handlers,
but we need to do it as well for obscure x86 hardware where disabling
an interrupt on the IO_APIC level redirects the interrupt to the
legacy PIC interrupt lines.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Ingo Molnar <mingo@elte.hu>
Impact: cleanup
The code is only compiled if CONFIG_GENERIC_HARDIRQS=y so another
check for this define in the code is redundant. Remove it.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Export the setup_irq() and remove_irq() symbols.
I'd like to export these functions since I have timer
code that needs to use setup_irq() early on (too early
for request_irq()), and the same code can also be
compiled as a module.
Signed-off-by: Magnus Damm <damm@igel.co.jp>
LKML-Reference: <20090312120559.2926.82371.sendpatchset@rx1.opensource.se>
[ changed to _GPL as these are special APIs deep inside the irq layer. ]
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: add new API
This patch adds a remove_irq() function for releasing
interrupts requested with setup_irq().
Without this patch we have no way of releasing such
interrupts since free_irq() today tries to kfree()
the irqaction passed with setup_irq().
Signed-off-by: Magnus Damm <damm@igel.co.jp>
LKML-Reference: <20090312120542.2926.56609.sendpatchset@rx1.opensource.se>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
