For (bad) historical reasons, OPAL used to create a non-standard pair
of properties "opal-interrupts" and "opal-interrupts-names" for
representing the list of interrupts it wants Linux to request on its
behalf.
Among other issues, the opal-interrupts doesn't have a way to carry
the type of interrupts, and they were assumed to be all level
sensitive.
This is wrong on some recent systems where some of them are edge
sensitive causing warnings in the XIVE code and possible misbehaviours
if they need to be retriggered (typically the NPU2 TCE error
interrupts).
This makes Linux switch to using the standard "interrupts" and
"interrupt-names" properties instead when they are available, using
standard of_irq helpers, which can carry all the desired type
information.
Newer versions of OPAL will generate those properties in addition to
the legacy ones.
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
[mpe: Fixup prefix logic to check strlen(r->name). Reinstate setting
of start = 0 in opal_event_shutdown() to avoid double free warnings]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Using irq_work for processing OPAL event interrupts is not necessary.
irq_work is typically used to schedule work from NMI context, a
softirq may be more appropriate. However OPAL events are not
particularly performance or latency critical, so they can all be
invoked by kopald.
This patch removes the irq_work queueing, and instead wakes up
kopald when there is an event to be processed. kopald processes
interrupts individually, enabling irqs and calling cond_resched
between each one to minimise latencies.
Event handlers themselves should still use threaded handlers,
workqueues, etc. as necessary to avoid high interrupts-off latencies
within any single interrupt.
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
A kernel crash in process context that calls emergency_restart from
panic will end up calling opal_event_shutdown with interrupts disabled
but not in interrupt. This causes a sleeping function to be called
which gives the following warning with sysrq+c:
Rebooting in 10 seconds..
BUG: sleeping function called from invalid context at kernel/locking/mutex.c:238
in_atomic(): 0, irqs_disabled(): 1, pid: 7669, name: bash
CPU: 20 PID: 7669 Comm: bash Tainted: G D W 4.17.0-rc5+ #3
Call Trace:
dump_stack+0xb0/0xf4 (unreliable)
___might_sleep+0x174/0x1a0
mutex_lock+0x38/0xb0
__free_irq+0x68/0x460
free_irq+0x70/0xc0
opal_event_shutdown+0xb4/0xf0
opal_shutdown+0x24/0xa0
pnv_shutdown+0x28/0x40
machine_shutdown+0x44/0x60
machine_restart+0x28/0x80
emergency_restart+0x30/0x50
panic+0x2a0/0x328
oops_end+0x1ec/0x1f0
bad_page_fault+0xe8/0x154
handle_page_fault+0x34/0x38
--- interrupt: 300 at sysrq_handle_crash+0x44/0x60
LR = __handle_sysrq+0xfc/0x260
flag_spec.62335+0x12b844/0x1e8db4 (unreliable)
__handle_sysrq+0xfc/0x260
write_sysrq_trigger+0xa8/0xb0
proc_reg_write+0xac/0x110
__vfs_write+0x6c/0x240
vfs_write+0xd0/0x240
ksys_write+0x6c/0x110
Fixes: 9f0fd0499d ("powerpc/powernv: Add a virtual irqchip for opal events")
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
In opal_event_shutdown() we free all the IRQs hanging off the
opal_event_irqchip. However it's not safe to do so if we're called
from IRQ context, because free_irq() wants to synchronise versus IRQ
context. This can lead to warnings and a stuck system.
For example from sysrq-b:
Trying to free IRQ 17 from IRQ context!
------------[ cut here ]------------
WARNING: CPU: 0 PID: 0 at kernel/irq/manage.c:1461 __free_irq+0x398/0x8d0
...
NIP __free_irq+0x398/0x8d0
LR __free_irq+0x394/0x8d0
Call Trace:
__free_irq+0x394/0x8d0 (unreliable)
free_irq+0xa4/0x140
opal_event_shutdown+0x128/0x180
opal_shutdown+0x1c/0xb0
pnv_shutdown+0x20/0x40
machine_restart+0x38/0x90
emergency_restart+0x28/0x40
sysrq_handle_reboot+0x24/0x40
__handle_sysrq+0x198/0x590
hvc_poll+0x48c/0x8c0
hvc_handle_interrupt+0x1c/0x50
__handle_irq_event_percpu+0xe8/0x6e0
handle_irq_event_percpu+0x34/0xe0
handle_irq_event+0xc4/0x210
handle_level_irq+0x250/0x770
generic_handle_irq+0x5c/0xa0
opal_handle_events+0x11c/0x240
opal_interrupt+0x38/0x50
__handle_irq_event_percpu+0xe8/0x6e0
handle_irq_event_percpu+0x34/0xe0
handle_irq_event+0xc4/0x210
handle_fasteoi_irq+0x174/0xa10
generic_handle_irq+0x5c/0xa0
__do_irq+0xbc/0x4e0
call_do_irq+0x14/0x24
do_IRQ+0x18c/0x540
hardware_interrupt_common+0x158/0x180
We can avoid that by using disable_irq_nosync() rather than
free_irq(). Although it doesn't fully free the IRQ, it should be
sufficient when we're shutting down, particularly in an emergency.
Add an in_interrupt() check and use free_irq() when we're shutting
down normally. It's probably OK to use disable_irq_nosync() in that
case too, but for now it's safer to leave that behaviour as-is.
Fixes: 9f0fd0499d ("powerpc/powernv: Add a virtual irqchip for opal events")
Reported-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Recent versions of OPAL can provide names for the various OPAL interrupts,
so let's use them. This also modernises the code that fetches the
interrupt array to use the helpers provided by the generic code instead
of hand-parsing the property.
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
[mpe: Free irqs on error, check allocation of names, consolidate error
handling, whitespace.]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
NO_IRQ has been == 0 on powerpc for just over ten years (since commit
0ebfff1491 ("[POWERPC] Add new interrupt mapping core and change
platforms to use it")). It's also 0 on most other arches.
Although it's fairly harmless, every now and then it causes confusion
when a driver is built on powerpc and another arch which doesn't define
NO_IRQ. There's at least 6 definitions of NO_IRQ in drivers/, at least
some of which are to work around that problem.
So we'd like to remove it. This is fairly trivial in the arch code, we
just convert:
if (irq == NO_IRQ) to if (!irq)
if (irq != NO_IRQ) to if (irq)
irq = NO_IRQ; to irq = 0;
return NO_IRQ; to return 0;
And a few other odd cases as well.
At least for now we keep the #define NO_IRQ, because there is driver
code that uses NO_IRQ and the fixes to remove those will go via other
trees.
Note we also change some occurrences in PPC sound drivers, drivers/ps3,
and drivers/macintosh.
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
This sets the type of the interrupt appropriately. We set it as follow:
- If not mapped from the device-tree, we use edge. This is the case
of the virtual interrupts and PCI MSIs for example.
- If mapped from the device-tree and #interrupt-cells is 2 (PAPR
compliant), we use the second cell to set the appropriate type
- If mapped from the device-tree and #interrupt-cells is 1 (current
OPAL on P8 does that), we assume level sensitive since those are
typically going to be the PSI LSIs which are level sensitive.
Additionally, we mark the interrupts requested via the opal_interrupts
property all level. This is a bit fishy but the best we can do until we
fix OPAL to properly expose them with a complete descriptor. It is also
correct for the current HW anyway as OPAL interrupts are currently PCI
error and PSI interrupts which are level.
Finally now that edge interrupts are properly identified, we can enable
CONFIG_HARDIRQS_SW_RESEND which will make the core re-send them if
they occur while masked, which some drivers rely upon.
This fixes issues with lost interrupts on some Mellanox adapters.
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Commit 25642e1459 ("powerpc/opal-irqchip: Fix double endian
conversion") fixed an endian bug by calling opal_handle_events() in
opal_event_unmask().
However this introduced a deadlock if we find an event is active
during unmasking and call opal_handle_events() again. The bad call
sequence is:
opal_interrupt()
-> opal_handle_events()
-> generic_handle_irq()
-> handle_level_irq()
-> raw_spin_lock(&desc->lock)
handle_irq_event(desc)
unmask_irq(desc)
-> opal_event_unmask()
-> opal_handle_events()
-> generic_handle_irq()
-> handle_level_irq()
-> raw_spin_lock(&desc->lock) (BOOM)
When generating multiple opal events in quick succession this would lead
to the following stall warnings:
EEH: Fenced PHB#0 detected, location: U78C9.001.WZS09XA-P1-C32
INFO: rcu_sched detected stalls on CPUs/tasks:
12-...: (1 GPs behind) idle=68f/140000000000001/0 softirq=860/861 fqs=2065
15-...: (1 GPs behind) idle=be5/140000000000001/0 softirq=1142/1143 fqs=2065
(detected by 13, t=2102 jiffies, g=1325, c=1324, q=602)
NMI watchdog: BUG: soft lockup - CPU#18 stuck for 22s! [irqbalance:2696]
INFO: rcu_sched detected stalls on CPUs/tasks:
12-...: (1 GPs behind) idle=68f/140000000000001/0 softirq=860/861 fqs=8371
15-...: (1 GPs behind) idle=be5/140000000000001/0 softirq=1142/1143 fqs=8371
(detected by 20, t=8407 jiffies, g=1325, c=1324, q=1290)
This patch corrects the problem by queuing the work if an event is
active during unmasking, which is similar to the pre-endian fix
behaviour.
Fixes: 25642e1459 ("powerpc/opal-irqchip: Fix double endian conversion")
Signed-off-by: Alistair Popple <alistair@popple.id.au>
Reported-by: Andrew Donnellan <andrew.donnellan@au1.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
The OPAL event calls return a mask of events that are active in big
endian format. This is checked when unmasking the events in the
irqchip by comparison with a cached value. The cached value was stored
in big endian format but should've been converted to CPU endian
first.
This bug leads to OPAL event delivery being delayed or dropped on some
systems. Symptoms may include a non-functional console.
The bug is fixed by calling opal_handle_events(...) instead of
duplicating code in opal_event_unmask(...).
Fixes: 9f0fd0499d ("powerpc/powernv: Add a virtual irqchip for opal events")
Cc: stable@vger.kernel.org # v4.2+
Reported-by: Douglas L Lehr <dllehr@us.ibm.com>
Signed-off-by: Alistair Popple <alistair@popple.id.au>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
The struct irq_domain contains a "struct device_node *" field
(of_node) that is almost the only link between the irqdomain
and the device tree infrastructure.
In order to prepare for the removal of that field, convert all
users to use irq_domain_get_of_node() instead.
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Reviewed-and-tested-by: Hanjun Guo <hanjun.guo@linaro.org>
Tested-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com>
Cc: <linux-arm-kernel@lists.infradead.org>
Cc: Tomasz Nowicki <tomasz.nowicki@linaro.org>
Cc: Suravee Suthikulpanit <Suravee.Suthikulpanit@amd.com>
Cc: Graeme Gregory <graeme@xora.org.uk>
Cc: Jake Oshins <jakeo@microsoft.com>
Cc: Jiang Liu <jiang.liu@linux.intel.com>
Cc: Jason Cooper <jason@lakedaemon.net>
Cc: Rafael J. Wysocki <rjw@rjwysocki.net>
Link: http://lkml.kernel.org/r/1444737105-31573-2-git-send-email-marc.zyngier@arm.com
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
It is not uncommon (at least with the ARM stuff) to have a piece
of hardware that implements different flavours of "interrupts".
A typical example of this is the GICv3 ITS, which implements
standard PCI/MSI support, but also some form of "generic MSI".
So far, the PCI/MSI domain is registered using the ITS device_node,
so that irq_find_host can return it. On the contrary, the raw MSI
domain is not registered with an device_node, making it impossible
to be looked up by another subsystem (obviously, using the same
device_node twice would only result in confusion, as it is not
defined which one irq_find_host would return).
A solution to this is to "type" domains that may be aliasing, and
to be able to lookup an device_node that matches a given type.
For this, we introduce irq_find_matching_host() as a superset
of irq_find_host:
struct irq_domain *irq_find_matching_host(struct device_node *node,
enum irq_domain_bus_token bus_token);
where bus_token is the "type" we want to match the domain against
(so far, only DOMAIN_BUS_ANY is defined). This result in some
moderately invasive changes on the PPC side (which is the only
user of the .match method).
This has otherwise no functionnal change.
Reviewed-by: Hanjun Guo <hanjun.guo@linaro.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Cc: <linux-arm-kernel@lists.infradead.org>
Cc: Yijing Wang <wangyijing@huawei.com>
Cc: Ma Jun <majun258@huawei.com>
Cc: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com>
Cc: Duc Dang <dhdang@apm.com>
Cc: Bjorn Helgaas <bhelgaas@google.com>
Cc: Jiang Liu <jiang.liu@linux.intel.com>
Cc: Jason Cooper <jason@lakedaemon.net>
Link: http://lkml.kernel.org/r/1438091186-10244-2-git-send-email-marc.zyngier@arm.com
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
The eeh subsystem for powernv requires the opal event irqchip to be
initialised prior to initialisation or the following errors are
produced (and eeh doesn't work as expected):
irq: XICS didn't like hwirq-0x9 to VIRQ17 mapping (rc=-22)
pnv_eeh_post_init: Can't request OPAL event interrupt (0)
On powernv eeh is initialised from a subsys_initcall due to a check
for machine_is(powernv) in eeh_init(). This patch increases the
initcall priority of opal_event_init() to an arch_initcall to ensure
the opal event interface is initialised prior to any users of it.
Signed-off-by: Alistair Popple <alistair@popple.id.au>
Reported-by: Daniel Axtens <dja@axtens.net>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
All users of the old opal events notifier have been converted over to
the irq domain so remove the event notifier functions.
Signed-off-by: Alistair Popple <alistair@popple.id.au>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Whenever an interrupt is received for opal the linux kernel gets a
bitfield indicating certain events that have occurred and need handling
by the various device drivers. Currently this is handled using a
notifier interface where we call every device driver that has
registered to receive opal events.
This approach has several drawbacks. For example each driver has to do
its own checking to see if the event is relevant as well as event
masking. There is also no easy method of recording the number of times
we receive particular events.
This patch solves these issues by exposing opal events via the
standard interrupt APIs by adding a new interrupt chip and
domain. Drivers can then register for the appropriate events using
standard kernel calls such as irq_of_parse_and_map().
Signed-off-by: Alistair Popple <alistair@popple.id.au>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Benjamin Herrenschmidt