linux.git/arch/powerpc/kernel/smp.c, branch v3.11

powerpc: Set cpu sibling mask before online cpu

2013-07-01T01:46:55+00:00

It seems following race is possible:

	cpu0					cpux
smp_init->cpu_up->_cpu_up
	__cpu_up
		kick_cpu(1)
-------------------------------------------------------------------------
		waiting online			...
		...				notify CPU_STARTING
							set cpux active
						set cpux online
-------------------------------------------------------------------------
		finish waiting online
		...
sched_init_smp
	init_sched_domains(cpu_active_mask)
		build_sched_domains
						set cpux sibling info
-------------------------------------------------------------------------

Execution of cpu0 and cpux could be concurrent between two separator
lines.

So if the cpux sibling information was set too late (normally
impossible, but could be triggered by adding some delay in
start_secondary, after setting cpu online), build_sched_domains()
running on cpu0 might see cpux active, with an empty sibling mask, then
cause some bad address accessing like following:

[    0.099855] Unable to handle kernel paging request for data at address 0xc00000038518078f
[    0.099868] Faulting instruction address: 0xc0000000000b7a64
[    0.099883] Oops: Kernel access of bad area, sig: 11 [#1]
[    0.099895] PREEMPT SMP NR_CPUS=16 DEBUG_PAGEALLOC NUMA pSeries
[    0.099922] Modules linked in:
[    0.099940] CPU: 0 PID: 1 Comm: swapper/0 Not tainted 3.10.0-rc1-00120-gb973425-dirty #16
[    0.099956] task: c0000001fed80000 ti: c0000001fed7c000 task.ti: c0000001fed7c000
[    0.099971] NIP: c0000000000b7a64 LR: c0000000000b7a40 CTR: c0000000000b4934
[    0.099985] REGS: c0000001fed7f760 TRAP: 0300   Not tainted  (3.10.0-rc1-00120-gb973425-dirty)
[    0.099997] MSR: 8000000000009032   CR: 24272828  XER: 20000003
[    0.100045] SOFTE: 1
[    0.100053] CFAR: c000000000445ee8
[    0.100064] DAR: c00000038518078f, DSISR: 40000000
[    0.100073]
GPR00: 0000000000000080 c0000001fed7f9e0 c000000000c84d48 0000000000000010
GPR04: 0000000000000010 0000000000000000 c0000001fc55e090 0000000000000000
GPR08: ffffffffffffffff c000000000b80b30 c000000000c962d8 00000003845ffc5f
GPR12: 0000000000000000 c00000000f33d000 c00000000000b9e4 0000000000000000
GPR16: 0000000000000000 0000000000000000 0000000000000001 0000000000000000
GPR20: c000000000ccf750 0000000000000000 c000000000c94d48 c0000001fc504000
GPR24: c0000001fc504000 c0000001fecef848 c000000000c94d48 c000000000ccf000
GPR28: c0000001fc522090 0000000000000010 c0000001fecef848 c0000001fed7fae0
[    0.100293] NIP [c0000000000b7a64] .get_group+0x84/0xc4
[    0.100307] LR [c0000000000b7a40] .get_group+0x60/0xc4
[    0.100318] Call Trace:
[    0.100332] [c0000001fed7f9e0] [c0000000000dbce4] .lock_is_held+0xa8/0xd0 (unreliable)
[    0.100354] [c0000001fed7fa70] [c0000000000bf62c] .build_sched_domains+0x728/0xd14
[    0.100375] [c0000001fed7fbe0] [c000000000af67bc] .sched_init_smp+0x4fc/0x654
[    0.100394] [c0000001fed7fce0] [c000000000adce24] .kernel_init_freeable+0x17c/0x30c
[    0.100413] [c0000001fed7fdb0] [c00000000000ba08] .kernel_init+0x24/0x12c
[    0.100431] [c0000001fed7fe30] [c000000000009f74] .ret_from_kernel_thread+0x5c/0x68
[    0.100445] Instruction dump:
[    0.100456] 38800010 38a00000 4838e3f5 60000000 7c6307b4 2fbf0000 419e0040 3d220001
[    0.100496] 78601f24 39491590 e93e0008 7d6a002a <7d69582a> f97f0000 7d4a002a e93e0010
[    0.100559] ---[ end trace 31fd0ba7d8756001 ]---

This patch tries to move the sibling maps updating before
notify_cpu_starting() and cpu online, and a write barrier there to make
sure sibling maps are updated before active and online mask.

Signed-off-by: Li Zhong 
Reviewed-by: Srivatsa S. Bhat 
Signed-off-by: Benjamin Herrenschmidt

powerpc: Delete __cpuinit usage from all users

2013-07-01T01:10:36+00:00

The __cpuinit type of throwaway sections might have made sense
some time ago when RAM was more constrained, but now the savings
do not offset the cost and complications.  For example, the fix in
commit 5e427ec2d0 ("x86: Fix bit corruption at CPU resume time")
is a good example of the nasty type of bugs that can be created
with improper use of the various __init prefixes.

After a discussion on LKML[1] it was decided that cpuinit should go
the way of devinit and be phased out.  Once all the users are gone,
we can then finally remove the macros themselves from linux/init.h.

This removes all the powerpc uses of the __cpuinit macros.  There
are no __CPUINIT users in assembly files in powerpc.

[1] https://lkml.org/lkml/2013/5/20/589

Cc: Benjamin Herrenschmidt 
Cc: Paul Mackerras 
Cc: Josh Boyer 
Cc: Matt Porter 
Cc: Kumar Gala 
Cc: linuxppc-dev@lists.ozlabs.org
Signed-off-by: Paul Gortmaker 
Signed-off-by: Benjamin Herrenschmidt

powerpc: Use generic idle loop

2013-04-08T15:39:27+00:00

Signed-off-by: Thomas Gleixner 
Cc: Linus Torvalds 
Cc: Rusty Russell 
Cc: Paul McKenney 
Cc: Peter Zijlstra 
Reviewed-by: Cc: Srivatsa S. Bhat 
Cc: Magnus Damm 
Cc: Benjamin Herrenschmidt 
Link: http://lkml.kernel.org/r/20130321215235.026838003@linutronix.de
Signed-off-by: Thomas Gleixner

powerpc: fix ics_rtas_init and start_secondary section mismatch

2013-02-08T03:05:48+00:00

It seems, we're fine with just annotating the two functions.
Thus, this fixes the following build warnings on ppc64:

WARNING: arch/powerpc/sysdev/xics/built-in.o(.text+0x1664):
The function .ics_rtas_init() references
the function __init .xics_register_ics().
This is often because .ics_rtas_init lacks a __init
annotation or the annotation of .xics_register_ics is wrong.

WARNING: arch/powerpc/sysdev/built-in.o(.text+0x6044):
The function .ics_rtas_init() references
the function __init .xics_register_ics().
This is often because .ics_rtas_init lacks a __init
annotation or the annotation of .xics_register_ics is wrong.

WARNING: arch/powerpc/kernel/built-in.o(.text+0x2db30):
The function .start_secondary() references
the function __cpuinit .vdso_getcpu_init().
This is often because .start_secondary lacks a __cpuinit
annotation or the annotation of .vdso_getcpu_init is wrong.

Cc: Benjamin Herrenschmidt 
Signed-off-by: Daniel Borkmann 
Signed-off-by: Benjamin Herrenschmidt

POWERPC: drivers: remove __dev* attributes.

2013-01-03T23:57:04+00:00

CONFIG_HOTPLUG is going away as an option.  As a result, the __dev*
markings need to be removed.

This change removes the use of __devinit, __devexit_p, __devinitdata,
__devinitconst, and __devexit from these drivers.

Based on patches originally written by Bill Pemberton, but redone by me
in order to handle some of the coding style issues better, by hand.

Cc: Bill Pemberton 
Cc: Benjamin Herrenschmidt 
Cc: Paul Mackerras 
Signed-off-by: Greg Kroah-Hartman

Merge commit 'origin/queue' into for-queue

2012-10-31T12:36:18+00:00

Conflicts:
	arch/powerpc/include/asm/Kbuild
	arch/powerpc/include/uapi/asm/Kbuild

KVM: PPC: Book3S HV: Allow KVM guests to stop secondary threads coming online

2012-10-30T09:54:53+00:00

When a Book3S HV KVM guest is running, we need the host to be in
single-thread mode, that is, all of the cores (or at least all of
the cores where the KVM guest could run) to be running only one
active hardware thread.  This is because of the hardware restriction
in POWER processors that all of the hardware threads in the core
must be in the same logical partition.  Complying with this restriction
is much easier if, from the host kernel's point of view, only one
hardware thread is active.

This adds two hooks in the SMP hotplug code to allow the KVM code to
make sure that secondary threads (i.e. hardware threads other than
thread 0) cannot come online while any KVM guest exists.  The KVM
code still has to check that any core where it runs a guest has the
secondary threads offline, but having done that check it can now be
sure that they will not come online while the guest is running.

Signed-off-by: Paul Mackerras 
Acked-by: Benjamin Herrenschmidt 
Signed-off-by: Alexander Graf

powerpc/smp: Do not disable IPI interrupts during suspend

2012-09-19T13:38:16+00:00

During suspend, all interrupts including IPI will be disabled. In this case,
the suspend process will hang in SMP. To prevent this, pass the flag
IRQF_NO_SUSPEND when requesting IPI irq.

Signed-off-by: Zhao Chenhui 
Signed-off-by: Li Yang 
Acked-by: Benjamin Herrenschmidt 
Signed-off-by: Kumar Gala

powerpc/smp: add generic_set_cpu_up() to set cpu_state as CPU_UP_PREPARE

2012-09-12T19:57:08+00:00

In the case of cpu hotplug, the cpu_state should be set to CPU_UP_PREPARE
when kicking cpu.  Otherwise, the cpu_state is always CPU_DEAD after
calling generic_set_cpu_dead(), which makes the delay in generic_cpu_die()
not happen.

Signed-off-by: Zhao Chenhui 
Signed-off-by: Kumar Gala

powerpc: Make sure IPI handlers see data written by IPI senders

2012-09-05T06:05:22+00:00

We have been observing hangs, both of KVM guest vcpu tasks and more
generally, where a process that is woken doesn't properly wake up and
continue to run, but instead sticks in TASK_WAKING state.  This
happens because the update of rq->wake_list in ttwu_queue_remote()
is not ordered with the update of ipi_message in
smp_muxed_ipi_message_pass(), and the reading of rq->wake_list in
scheduler_ipi() is not ordered with the reading of ipi_message in
smp_ipi_demux().  Thus it is possible for the IPI receiver not to see
the updated rq->wake_list and therefore conclude that there is nothing
for it to do.

In order to make sure that anything done before smp_send_reschedule()
is ordered before anything done in the resulting call to scheduler_ipi(),
this adds barriers in smp_muxed_message_pass() and smp_ipi_demux().
The barrier in smp_muxed_message_pass() is a full barrier to ensure that
there is a full ordering between the smp_send_reschedule() caller and
scheduler_ipi().  In smp_ipi_demux(), we use xchg() rather than
xchg_local() because xchg() includes release and acquire barriers.
Using xchg() rather than xchg_local() makes sense given that
ipi_message is not just accessed locally.

This moves the barrier between setting the message and calling the
cause_ipi() function into the individual cause_ipi implementations.
Most of them -- those that used outb, out_8 or similar -- already had
a full barrier because out_8 etc. include a sync before the MMIO
store.  This adds an explicit barrier in the two remaining cases.

These changes made no measurable difference to the speed of IPIs as
measured using a simple ping-pong latency test across two CPUs on
different cores of a POWER7 machine.

The analysis of the reason why processes were not waking up properly
is due to Milton Miller.

Cc: stable@vger.kernel.org # v3.0+
Reported-by: Milton Miller 
Signed-off-by: Paul Mackerras 
Signed-off-by: Benjamin Herrenschmidt