The setup/remove_state/instance() functions in the hotplug core code are
serialized against concurrent CPU hotplug, but unfortunately not serialized
against themself.
As a consequence a concurrent invocation of these function results in
corruption of the callback machinery because two instances try to invoke
callbacks on remote cpus at the same time. This results in missing callback
invocations and initiator threads waiting forever on the completion.
The obvious solution to replace get_cpu_online() with cpu_hotplug_begin()
is not possible because at least one callsite calls into these functions
from a get_online_cpu() locked region.
Extend the protection scope of the cpuhp_state_mutex from solely protecting
the state arrays to cover the callback invocation machinery as well.
Fixes: 5b7aa87e04 ("cpu/hotplug: Implement setup/removal interface")
Reported-and-tested-by: Bart Van Assche <[email protected]>
Signed-off-by: Sebastian Andrzej Siewior <[email protected]>
Cc: [email protected]
Cc: [email protected]
Cc: [email protected]
Cc: [email protected]
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Thomas Gleixner <[email protected]>
Since the change in commit:
fd7a4bed18 ("sched, rt: Convert switched_{from, to}_rt() / prio_changed_rt() to balance callbacks")
... we don't reschedule a task under certain circumstances:
Lets say task-A, SCHED_OTHER, is running on CPU0 (and it may run only on
CPU0) and holds a PI lock. This task is removed from the CPU because it
used up its time slice and another SCHED_OTHER task is running. Task-B on
CPU1 runs at RT priority and asks for the lock owned by task-A. This
results in a priority boost for task-A. Task-B goes to sleep until the
lock has been made available. Task-A is already runnable (but not active),
so it receives no wake up.
The reality now is that task-A gets on the CPU once the scheduler decides
to remove the current task despite the fact that a high priority task is
enqueued and waiting. This may take a long time.
The desired behaviour is that CPU0 immediately reschedules after the
priority boost which made task-A the task with the lowest priority.
Suggested-by: Peter Zijlstra <[email protected]>
Signed-off-by: Sebastian Andrzej Siewior <[email protected]>
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Cc: Linus Torvalds <[email protected]>
Cc: Mike Galbraith <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Fixes: fd7a4bed18 ("sched, rt: Convert switched_{from, to}_rt() prio_changed_rt() to balance callbacks")
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Ingo Molnar <[email protected]>
RCU_EXPEDITE_BOOT should speed up the boot process by enforcing
synchronize_rcu_expedited() instead of synchronize_rcu() during the boot
process. There should be no reason why one does not want this and there
is no need worry about real time latency at this point.
Therefore make it default.
Note that users wishing to avoid expediting entirely, for example when
bringing up new hardware possibly having flaky IPIs, can use the
rcu_normal boot parameter to override boot-time expediting.
Signed-off-by: Sebastian Andrzej Siewior <[email protected]>
[ paulmck: Reworded commit log. ]
Signed-off-by: Paul E. McKenney <[email protected]>
Reviewed-by: Josh Triplett <[email protected]>
Before commit b32614c034 ("tracing/rb: Convert to hotplug state
machine") the allocated cpumask was initialized to the mask of ONLINE or
POSSIBLE CPUs. After the CPU hotplug changes the buffer initialisation
moved to trace_rb_cpu_prepare() but I forgot to initially set the
cpumask to zero. This is done now.
Link: http://lkml.kernel.org/r/[email protected]
Fixes: b32614c034 ("tracing/rb: Convert to hotplug state machine")
Reported-by: Tetsuo Handa <[email protected]>
Tested-by: Tetsuo Handa <[email protected]>
Signed-off-by: Sebastian Andrzej Siewior <[email protected]>
Signed-off-by: Steven Rostedt <[email protected]>
Install the callbacks via the state machine and let the core invoke the
callbacks on the already online CPUs. When the hotplug state is
unregistered the cleanup function is called for each cpu. So both cpu loops
in init() and exit() are not longer required.
Cc: Jean Delvare <[email protected]>
Cc: Guenter Roeck <[email protected]>
Cc: [email protected]
Signed-off-by: Sebastian Andrzej Siewior <[email protected]>
Signed-off-by: Guenter Roeck <[email protected]>
Before commit b32614c034 ("tracing/rb: Convert to hotplug state machine")
the allocated cpumask was initialized to the mask of online or possible
CPUs. After the CPU hotplug changes the buffer initialization moved to
trace_rb_cpu_prepare() but the cpumask is allocated with alloc_cpumask()
and therefor has random content. As a consequence the cpu buffers are not
initialized and a later access dereferences a NULL pointer.
Use zalloc_cpumask() instead so trace_rb_cpu_prepare() initializes the
buffers properly.
Fixes: b32614c034 ("tracing/rb: Convert to hotplug state machine")
Reported-by: Tetsuo Handa <[email protected]>
Signed-off-by: Sebastian Andrzej Siewior <[email protected]>
Cc: [email protected]
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Thomas Gleixner <[email protected]>
Install the callbacks via the state machine. The notifier in struct
ring_buffer is replaced by the multi instance interface. Upon
__ring_buffer_alloc() invocation, cpuhp_state_add_instance() will invoke
the trace_rb_cpu_prepare() on each CPU.
This callback may now fail. This means __ring_buffer_alloc() will fail and
cleanup (like previously) and during a CPU up event this failure will not
allow the CPU to come up.
Signed-off-by: Sebastian Andrzej Siewior <[email protected]>
Cc: Steven Rostedt <[email protected]>
Cc: [email protected]
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Thomas Gleixner <[email protected]>
Install the callbacks via the state machine and let the core invoke the
callbacks on the already online CPUs.
The two smp_call_function_single() invocations in intel_idle_cpu_init() have
been removed because intel_idle_cpu_init() is now invoked via the hotplug
callback which runs on the target CPU. The IRQ-off calling convention for
auto_demotion_disable() and c1e_promotion_disable() has not been preserved
because only those two modify the MSR during CPU intialization.
Signed-off-by: Sebastian Andrzej Siewior <[email protected]>
Acked-by: Jacob Pan <[email protected]>
Signed-off-by: Rafael J. Wysocki <[email protected]>
Install the callbacks via the state machine and let the core invoke
the callbacks on the already online CPUs.
Replace the wrmsr/rdmrs_on_cpu() calls in the hotplug callbacks as they are
guaranteed to be invoked on the incoming/outgoing cpu.
Signed-off-by: Sebastian Andrzej Siewior <[email protected]>
Signed-off-by: Thomas Gleixner <[email protected]>
Tested-by: Srinivas Pandruvada <[email protected]>
Signed-off-by: Zhang Rui <[email protected]>
This is a conversation to the new hotplug state machine with
the difference that CPU_DEAD becomes CPU_PREDOWN.
At the same time it makes the handling of the two states symmetrical.
stop_power_clamp_worker() is called unconditionally and the controversial
error message is removed.
Finally, the hotplug state callbacks are removed after the powerclamping
is stopped to avoid a potential race.
Signed-off-by: Sebastian Andrzej Siewior <[email protected]>
[[email protected]: Fixed the possible race in powerclamp_exit()]
Signed-off-by: Petr Mladek <[email protected]>
Signed-off-by: Jacob Pan <[email protected]>
Signed-off-by: Rafael J. Wysocki <[email protected]>
The online / pre_down callback is invoked on the target CPU since commit
1cf4f629d9 ("cpu/hotplug: Move online calls to hotplugged cpu") which means
for the hotplug callback we can use rmdsrl() instead of rdmsr_on_cpus().
This leaves us with set_boost() as the only user which still needs to
read/write the MSR on different CPUs. There is no point in doing that
update on all cpus with the read modify write magic via per cpu data. We
simply can issue a function call on all online CPUs which also means that we
need half that many IPIs.
Signed-off-by: Sebastian Andrzej Siewior <[email protected]>
Acked-by: Viresh Kumar <[email protected]>
Signed-off-by: Rafael J. Wysocki <[email protected]>
Install the callbacks via the state machine as a first step. The init/exit
code is a duplicate of the hotplug code. This is cleaned up in a
consecutive patch.
Signed-off-by: Sebastian Andrzej Siewior <[email protected]>
Signed-off-by: Thomas Gleixner <[email protected]>
Tested-by: Jacob Pan <[email protected]>
Signed-off-by: Rafael J. Wysocki <[email protected]>
Install the callbacks via the state machine and let the core invoke
the callbacks on the already online CPUs.
The smp_call_function_single() is dropped because the ONLINE callback is
invoked on the target CPU since commit 1cf4f629d9 ("cpu/hotplug: Move
online calls to hotplugged cpu"). smp_call_function_single() invokes the
invoked function with interrupts disabled, but this calling convention is
not preserved as the MSR is not modified by anything else than this code.
Signed-off-by: Sebastian Andrzej Siewior <[email protected]>
Cc: Bjorn Helgaas <[email protected]>
Cc: [email protected]
Cc: [email protected]
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Thomas Gleixner <[email protected]>
Install the callbacks via the state machine and let the core invoke
the callbacks on the already online CPUs.
The previous convention of keeping the files around until the CPU is dead
has not been preserved as there is no point to keep them available when the
cpu is going down. This makes the hotplug call symmetric.
Signed-off-by: Sebastian Andrzej Siewior <[email protected]>
Acked-by: "David S. Miller" <[email protected]>
Cc: [email protected]
Cc: [email protected]
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Thomas Gleixner <[email protected]>
Install the callbacks via the state machine and let the core invoke
the callbacks on the already online CPUs.
The previous convention of keeping the files around until the CPU is dead
has not been preserved as there is no point to keep them available when the
cpu is going down. This makes the hotplug call symmetric.
Signed-off-by: Sebastian Andrzej Siewior <[email protected]>
Cc: Benjamin Herrenschmidt <[email protected]>
Cc: [email protected]
Cc: Paul Mackerras <[email protected]>
Cc: Michael Ellerman <[email protected]>
Cc: [email protected]
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Thomas Gleixner <[email protected]>
Install the callbacks via the state machine and let the core invoke
the callbacks on the already online CPUs.
Move the callbacks to online/offline as there is no point in having the
files around before the cpu is online and until its completely gone.
[ tglx: Move the callbacks to online/offline ]
Signed-off-by: Sebastian Andrzej Siewior <[email protected]>
Cc: [email protected]
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Thomas Gleixner <[email protected]>
The CPU_ONLINE and CPU_DOWN_PREPARE look fully symmetrical and could be move
to the hotplug state machine.
On a failure during registration we have the tear down callback invoked
(mce_cpu_pre_down()) so there should be no timer around and so no need to need
keep notifier installed (this was the reason according to the comment why the
notifier was registered despite of errors).
Signed-off-by: Sebastian Andrzej Siewior <[email protected]>
Acked-by: Borislav Petkov <[email protected]>
Cc: Tony Luck <[email protected]>
Cc: [email protected]
Cc: [email protected]
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Thomas Gleixner <[email protected]>
Initially I wanted to remove mcheck_cpu_init() from identify_cpu() and let it
become an independent early hotplug callback. The main problem here was that
the init on the boot CPU may happen too late
(device_initcall_sync(mcheck_init_device)) and nobody wanted to risk receiving
and MCE event at boot time leading to a shutdown (if the MCE feature is not yet
enabled).
Here is attempt two: the timming stays as-is but the ordering of the functions
is changed:
- mcheck_cpu_init() (which is run from identify_cpu()) will setup the timer
struct but won't fire the timer. This is moved to CPU_ONLINE since its
cleanup part is in CPU_DOWN_PREPARE. So if it is okay to stop the timer early
in the shutdown phase, it should be okay to start it late in the bring up phase.
- CPU_DOWN_PREPARE disables the MCE feature flags for !INTEL CPUs in
mce_disable_cpu(). If a failure occures it would be re-enabled on all vendor
CPUs (including Intel where it was not disabled during shutdown). To keep this
working I am moving it to CPU_ONLINE. smp_call_function_single() is dropped
beause the notifier runs nowdays on the target CPU.
- CPU_ONLINE is invoking mce_device_create() + mce_threshold_create_device()
but its cleanup part is in CPU_DEAD (mce_threshold_remove_device() and
mce_device_remove()). In order to keep this symmetrical I am moving the clean
up from CPU_DEAD to CPU_DOWN_PREPARE.
Signed-off-by: Sebastian Andrzej Siewior <[email protected]>
Acked-by: Borislav Petkov <[email protected]>
Cc: Tony Luck <[email protected]>
Cc: [email protected]
Cc: [email protected]
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Thomas Gleixner <[email protected]>
The threshold_cpu_callback callbacks looks like one of the notifier and
its arguments are almost the same. Split this out and have one ONLINE
and one DEAD callback. This will come handy later once the main code
gets changed to use the callback mechanism.
Also, handle threshold_cpu_callback_online() return value so we don't
continue if the function fails.
Boris Petkov removed the callback pointer and replaced it with proper
functions.
Signed-off-by: Sebastian Andrzej Siewior <[email protected]>
Acked-by: Borislav Petkov <[email protected]>
Cc: Tony Luck <[email protected]>
Cc: [email protected]
Cc: [email protected]
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Thomas Gleixner <[email protected]>
If we try a CPU down and fail in the middle then we roll back to the
online state. This means we would perform CPU_ONLINE / mce_device_create()
without invoking CPU_DEAD / mce_device_remove() for the cleanup of what was
allocated in CPU_ONLINE.
Be prepared for this and don't allocate the struct if we have it
already.
Signed-off-by: Sebastian Andrzej Siewior <[email protected]>
Acked-by: Borislav Petkov <[email protected]>
Cc: Tony Luck <[email protected]>
Cc: [email protected]
Cc: [email protected]
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Thomas Gleixner <[email protected]>
cpuhp_setup_state() invokes the startup callback on all online cpus with
the proper protection, so we can remove the cpu hotplug protection from the
init function and the creation of the per cpu files for online cpus in
smp_add_present_cpu(). smp_add_present_cpu() is called also called from
__smp_rescan_cpus(), but this callpath never adds an online cpu, it merily
adds newly present cpus, so the creation of the cpu files is not required.
Signed-off-by: Sebastian Andrzej Siewior <[email protected]>
Signed-off-by: Thomas Gleixner <[email protected]>
Acked-by: Heiko Carstens <[email protected]>
Cc: Martin Schwidefsky <[email protected]>
Cc: [email protected]
Cc: [email protected]
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Thomas Gleixner <[email protected]>
If the gcc is configured to do -fPIE by default then the build aborts
later with:
| Unsupported relocation type: unknown type rel type name (29)
Tagging it stable so it is possible to compile recent stable kernels as
well.
Cc: [email protected]
Signed-off-by: Sebastian Andrzej Siewior <[email protected]>
Signed-off-by: Michal Marek <[email protected]>
Adding -no-PIE to the fstack protector check. -no-PIE was introduced
before -fstack-protector so there is no need for a runtime check.
Without it the build stops:
|Cannot use CONFIG_CC_STACKPROTECTOR_STRONG: -fstack-protector-strong available but compiler is broken
due to -mcmodel=kernel + -fPIE if -fPIE is enabled by default.
Tagging it stable so it is possible to compile recent stable kernels as
well.
Cc: [email protected]
Signed-off-by: Sebastian Andrzej Siewior <[email protected]>
Signed-off-by: Michal Marek <[email protected]>
Debian started to build the gcc with -fPIE by default so the kernel
build ends before it starts properly with:
|kernel/bounds.c:1:0: error: code model kernel does not support PIC mode
Also add to KBUILD_AFLAGS due to:
|gcc -Wp,-MD,arch/x86/entry/vdso/vdso32/.note.o.d … -mfentry -DCC_USING_FENTRY … vdso/vdso32/note.S
|arch/x86/entry/vdso/vdso32/note.S:1:0: sorry, unimplemented: -mfentry isn’t supported for 32-bit in combination with -fpic
Tagging it stable so it is possible to compile recent stable kernels as
well.
Cc: [email protected]
Signed-off-by: Sebastian Andrzej Siewior <[email protected]>
Signed-off-by: Michal Marek <[email protected]>
It popped up in perf testing that the worker consumes some amount of
CPU. It boils down to the increment of `ops` which causes cache line
bouncing between the individual threads.
This patch aligns the struct by 256 bytes to ensure that not a cache
line is shared among CPUs. 128 byte is the x86 worst case and grep says
that L1_CACHE_SHIFT is set to 8 on s390.
Signed-off-by: Sebastian Andrzej Siewior <[email protected]>
Cc: Davidlohr Bueso <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Arnaldo Carvalho de Melo <[email protected]>
kbuild test robot reported this against the -RT tree:
|>> arch/x86/kernel/acpi/boot.c:90:21: warning: 'acpi_ioapic_lock' defined but not used [-Wunused-variable]
| static DEFINE_MUTEX(acpi_ioapic_lock);
| ^
| include/linux/mutex_rt.h:27:15: note: in definition of macro 'DEFINE_MUTEX'
| struct mutex mutexname = __MUTEX_INITIALIZER(mutexname)
^~~~~~~~~
which is also true (as in non-used) for !RT but the compiler does not
emit a warning.
Reported-by: kbuild test robot <[email protected]>
Signed-off-by: Sebastian Andrzej Siewior <[email protected]>
Cc: Linus Torvalds <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Ingo Molnar <[email protected]>
This DMA driver is used by 8250-omap on DRA7-evm. There is one
requirement that is to pause a transfer. This is currently used on the RX
side. It is possible that the UART HW aborted the RX (UART's RX-timeout)
but the DMA controller starts the transfer shortly after.
Before we can manually purge the FIFO we need to pause the transfer,
check how many bytes it already received and terminate the transfer
without it making any progress.
From testing on the TX side it seems that it is possible that we invoke
pause once the transfer has completed which is indicated by the missing
CCR_ENABLE bit but before the interrupt has been noticed. In that case the
interrupt will come even after disabling it.
The AM572x manual says that we have to wait for the CCR_RD_ACTIVE &
CCR_WR_ACTIVE bits to be gone before programming it again here is the
drain loop. Also it looks like without the drain the TX-transfer makes
sometimes progress.
One note: The pause + resume combo is broken because after resume the
the complete transfer will be programmed again. That means the already
transferred bytes (until the pause event) will be sent again. This is
currently not important for my UART user because it does only pause +
terminate.
v3…v4:
- update subject line.
v2…v3:
- rephrase the comment based on Russell's information / feedback.
v1…v2:
- move the drain loop into omap_dma_drain_chan() instead of having it
twice.
- allow pause only for DMA_DEV_TO_MEM transfers if non-cyclic. Add a
comment why DMA_MEM_TO_DEV not allowed.
- clear pause on terminate_all. Otherwise pause() + terminate_all()
will keep the pause bit set and we can't pause the following
transfer.
Signed-off-by: Sebastian Andrzej Siewior <[email protected]>
[[email protected]: drain channel only when buffering is on, rebase to v4.8]
Signed-off-by: Vignesh R <[email protected]>
Acked-by: Peter Ujfalusi <[email protected]>
Signed-off-by: Vinod Koul <[email protected]>
This patch moves the wakeup_process() invocation so it is not done under
the ipc global lock by making use of a lockless wake_q. With this change,
the waiter is woken up once the message has been assigned and it does not
need to loop on SMP if the message points to NULL. In the signal case we
still need to check the pointer under the lock to verify the state.
This change should also avoid the introduction of preempt_disable() in -RT
which avoids a busy-loop which pools for the NULL -> !NULL change if the
waiter has a higher priority compared to the waker.
By making use of wake_qs, the logic of sysv msg queues is greatly
simplified (and very well suited as we can batch lockless wakeups),
particularly around the lockless receive algorithm.
This has been tested with Manred's pmsg-shared tool on a "AMD A10-7800
Radeon R7, 12 Compute Cores 4C+8G":
test | before | after | diff
-----------------|------------|------------|----------
pmsg-shared 8 60 | 19,347,422 | 30,442,191 | + ~57.34 %
pmsg-shared 4 60 | 21,367,197 | 35,743,458 | + ~67.28 %
pmsg-shared 2 60 | 22,884,224 | 24,278,200 | + ~6.09 %
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Sebastian Andrzej Siewior <[email protected]>
Signed-off-by: Davidlohr Bueso <[email protected]>
Acked-by: Peter Zijlstra (Intel) <[email protected]>
Cc: Manfred Spraul <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Install the callbacks via the state machine.
Signed-off-by: Sebastian Andrzej Siewior <[email protected]>
Signed-off-by: Boris Ostrovsky <[email protected]>
Signed-off-by: David Vrabel <[email protected]>
The "blk_mq_queue_reinit_dead()" just cleared the cpumask instead doing
a copy. Since we might never had an online callback we could end up with
a ZERO mask which in turn leads to crash as test robot demonstarted.
Fixes: 65d5291eee ("blk-mq: Convert to new hotplug state machine")
Signed-off-by: Sebastian Andrzej Siewior <[email protected]>
Signed-off-by: Jens Axboe <[email protected]>
The function cpufreq_register_driver() returns zero on success and since
commit 27622b061e ("cpufreq: Convert to hotplug state machine")
erroneously a positive number. Due to the "if (x) assume_error" construct
all callers assumed an error and as a consequence the cpu freq kworker
crashes with a NULL pointer dereference.
Reset the return value back to zero in the success case.
Fixes: 27622b061e ("cpufreq: Convert to hotplug state machine")
Reported-by: Borislav Petkov <[email protected]>
Reported-and-tested-by: Ingo Molnar <[email protected]>
Signed-off-by: Sebastian Andrzej Siewior <[email protected]>
Cc: [email protected]
Cc: [email protected]
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Thomas Gleixner <[email protected]>
Since commit 4d4c474124 ("perf/x86/intel/bts: Fix BTS PMI detection")
my box goes boom on boot:
| .... node #0, CPUs: #1#2#3#4#5#6#7
| BUG: unable to handle kernel NULL pointer dereference at 0000000000000018
| IP: [<ffffffff8100c463>] intel_bts_interrupt+0x43/0x130
| Call Trace:
| <NMI> d [<ffffffff8100b341>] intel_pmu_handle_irq+0x51/0x4b0
| [<ffffffff81004d47>] perf_event_nmi_handler+0x27/0x40
This happens because the code introduced in this commit dereferences the
debug store pointer unconditionally. The debug store is not guaranteed to
be available, so a NULL pointer check as on other places is required.
Fixes: 4d4c474124 ("perf/x86/intel/bts: Fix BTS PMI detection")
Signed-off-by: Sebastian Andrzej Siewior <[email protected]>
Reviewed-by: Alexander Shishkin <[email protected]>
Cc: Arnaldo Carvalho de Melo <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: [email protected]
Cc: [email protected]
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Thomas Gleixner <[email protected]>
Install the callbacks via the state machine. It uses the multi instance
infrastructure of the hotplug code to handle each interface.
virtscsi_set_affinity() is removed from virtscsi_init() because
virtscsi_cpu_notif_add() (the function which registers the instance) is invoked
right after it and the cpuhp_state_add_instance() functions invokes the startup
callback on all online CPUs.
The same thing can not be applied virtscsi_cpu_notif_remove() because
virtscsi_remove_vqs() invokes virtscsi_set_affinity() with affinity = false as
argument but the old CPU_DEAD state invoked the function with affinity = true
(which does not match the DEAD callback).
Signed-off-by: Sebastian Andrzej Siewior <[email protected]>
Cc: "James E.J. Bottomley" <[email protected]>
Cc: [email protected]
Cc: "Martin K. Petersen" <[email protected]>
Cc: "Michael S. Tsirkin" <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: [email protected]
Cc: [email protected]
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Thomas Gleixner <[email protected]>
Install the callbacks via the state machine and let the core invoke
the callbacks on the already online CPUs.
Since the online target runs always on the target CPU we can drop
smp_call_function_single(). The functions is invoked with interrupts off to
keep the old calling convention. If the maintainer things that this function
can be called with interrupts enabled then it can be removed :)
Signed-off-by: Sebastian Andrzej Siewior <[email protected]>
Cc: Robert Richter <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: [email protected]
Cc: [email protected]
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Thomas Gleixner <[email protected]>
Install the callbacks via the state machine.
CPU_UP_CANCELED_FROZEN() is not preserved: It is only there to free memory in an
error case because it is assumed if the CPU does show up on resume it won't be
seen ever again. As per Borislav:
|IOW, you don't need mc_cpu_dead().
Signed-off-by: Sebastian Andrzej Siewior <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Borislav Petkov <[email protected]>
Cc: [email protected]
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Thomas Gleixner <[email protected]>
The resent conversion of the cpu hotplug support in the uncore driver
introduced a regression due to the way the callbacks are invoked at
initialization time.
The old code called the prepare/starting/online function on each online cpu
as a block. The new code registers the hotplug callbacks in the core for
each state. The core invokes the callbacks at each registration on all
online cpus.
The code implicitely relied on the prepare/starting/online callbacks being
called as combo on a particular cpu, which was not obvious and completely
undocumented.
The resulting subtle wreckage happens due to the way how the uncore code
manages shared data structures for cpus which share an uncore resource in
hardware. The sharing is determined in the cpu starting callback, but the
prepare callback allocates per cpu data for the upcoming cpu because
potential sharing is unknown at this point. If the starting callback finds
a online cpu which shares the hardware resource it takes a refcount on the
percpu data of that cpu and puts the own data structure into a
'free_at_online' pointer of that shared data structure. The online callback
frees that.
With the old model this worked because in a starting callback only one non
unused structure (the one of the starting cpu) was available. The new code
allocates the data structures for all cpus when the prepare callback is
registered.
Now the starting function iterates through all online cpus and looks for a
data structure (skipping its own) which has a matching hardware id. The id
member of the data structure is initialized to 0, but the hardware id can
be 0 as well. The resulting wreckage is:
CPU0 finds a matching id on CPU1, takes a refcount on CPU1 data and puts
its own data structure into CPU1s data structure to be freed.
CPU1 skips CPU0 because the data structure is its allegedly unsued own.
It finds a matching id on CPU2, takes a refcount on CPU1 data and puts
its own data structure into CPU2s data structure to be freed.
....
Now the online callbacks are invoked.
CPU0 has a pointer to CPU1s data and frees the original CPU0 data. So
far so good.
CPU1 has a pointer to CPU2s data and frees the original CPU1 data, which
is still referenced by CPU0 ---> Booom
So there are two issues to be solved here:
1) The id field must be initialized at allocation time to a value which
cannot be a valid hardware id, i.e. -1
This prevents the above scenario, but now CPU1 and CPU2 both stick their
own data structure into the free_at_online pointer of CPU0. So we leak
CPU1s data structure.
2) Fix the memory leak described in #1
Instead of having a single pointer, use a hlist to enqueue the
superflous data structures which are then freed by the first cpu
invoking the online callback.
Ideally we should know the sharing _before_ invoking the prepare callback,
but that's way beyond the scope of this bug fix.
[ tglx: Rewrote changelog ]
Fixes: 96b2bd3866 ("perf/x86/amd/uncore: Convert to hotplug state machine")
Reported-and-tested-by: Eric Sandeen <[email protected]>
Signed-off-by: Sebastian Andrzej Siewior <[email protected]>
Cc: Borislav Petkov <[email protected]>
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Thomas Gleixner <[email protected]>
I have here a FPGA behind PCIe which exports SRAM which I use for
pstore. Now it seems that the FPGA no longer supports cmpxchg based
updates and writes back 0xff…ff and returns the same. This leads to
crash during crash rendering pstore useless.
Since I doubt that there is much benefit from using cmpxchg() here, I am
dropping this atomic access and use the spinlock based version.
Cc: Anton Vorontsov <[email protected]>
Cc: Colin Cross <[email protected]>
Cc: Kees Cook <[email protected]>
Cc: Tony Luck <[email protected]>
Cc: Rabin Vincent <[email protected]>
Tested-by: Rabin Vincent <[email protected]>
Signed-off-by: Sebastian Andrzej Siewior <[email protected]>
Reviewed-by: Guenter Roeck <[email protected]>
[kees: remove "_locked" suffix since it's the only option now]
Signed-off-by: Kees Cook <[email protected]>
Cc: [email protected]
A basic rmmod ramoops segfaults. Let's see why.
Since commit 34f0ec82e0 ("pstore: Correct the max_dump_cnt clearing of
ramoops") sets ->max_dump_cnt to zero before looping over ->przs but we
didn't use it before that either.
And since commit ee1d267423 ("pstore: add pstore unregister") we free
that memory on rmmod.
But even then, we looped until a NULL pointer or ERR. I don't see where
it is ensured that the last member is NULL. Let's try this instead:
simply error recovery and free. Clean up in error case where resources
were allocated. And then, in the free path, rely on ->max_dump_cnt in
the free path.
Cc: Anton Vorontsov <[email protected]>
Cc: Colin Cross <[email protected]>
Cc: Kees Cook <[email protected]>
Cc: Tony Luck <[email protected]>
Cc: Namhyung Kim <[email protected]>
Acked-by: Namhyung Kim <[email protected]>
Signed-off-by: Sebastian Andrzej Siewior <[email protected]>
Signed-off-by: Kees Cook <[email protected]>
Cc: [email protected] # 4.4.x-
