When a CPU is offlined on POWER currently, we call rtas_stop_self() and hand
the CPU back to the resource pool. This path is used for DLPAR which will
cause a change in the LPAR configuration which will be visible outside.
This patch changes the default state a CPU is put into when it is offlined.
On platforms which support ceding the processor to the hypervisor with
latency hint specifier value, during a cpu offline operation,
instead of calling rtas_stop_self(), we cede the vCPU to the hypervisor
while passing a latency hint specifier value. The Hypervisor can use this hint
to provide better energy savings. Also, during the offline
operation, the control of the vCPU remains with the LPAR as oppposed to
returning it to the resource pool.
The patch achieves this by creating an infrastructure to set the
preferred_offline_state() which can be either
- CPU_STATE_OFFLINE: which is the current behaviour of calling
rtas_stop_self()
- CPU_STATE_INACTIVE: which cedes the vCPU to the hypervisor with the latency
hint specifier.
The codepath which wants to perform a DLPAR operation can set the
preferred_offline_state() of a CPU to CPU_STATE_OFFLINE before invoking
cpu_down().
The patch also provides a boot-time command line argument to disable/enable
CPU_STATE_INACTIVE.
Signed-off-by: Gautham R Shenoy <ego@in.ibm.com>
Signed-off-by: Nathan Fontenot <nfont@austin.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Use the accessors rather than frobbing bits directly (the new versions
are const).
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Mike Travis <travis@sgi.com>
Currently, pseries_cpu_die() calls msleep() while polling RTAS for
the status of the dying cpu.
However, if the cpu that is going down also happens to be the one
doing the tick then we're hosed as the tick_do_timer_cpu 'baton' is
only passed later on in tick_shutdown() when _cpu_down() does the
CPU_DEAD notification. Therefore jiffies won't be updated anymore.
This replaces that msleep() with a cpu_relax() to make sure we're not
going to schedule at that point.
With this patch my test box survives a 100k iterations hotplug stress
test on _all_ cpus, whereas without it, it quickly dies after ~50
iterations.
Signed-off-by: Sebastien Dugue <sebastien.dugue@bull.net>
Cc: Michael Ellerman <michael@ellerman.id.au>
Signed-off-by: Paul Mackerras <paulus@samba.org>
This splits off the kexec path bits of the xics_teardown_cpu() routine
into its own xics_kexec_teardown_cpu() routine. With the previous
combined routine the CPPR for a cpu that is being removed may have its
CPPR reset in the plpar_eoi() call (which explicitly sets the CPPR to
a non-zero value). Splitting of the kexec bits of the code prevents
this from happening in the cpu remove path.
Once again, this does not cause the cpu remove from the kernel to
fail, but it does cause cpu dlpar operations to not be able to return
the cpu to the hypervisor.
Signed-off-by: Nathan Fontenot <nfont@austin.ibm.com>
Signed-off-by: Paul Mackerras <paulus@samba.org>
Replace all lock_cpu_hotplug/unlock_cpu_hotplug from the kernel and use
get_online_cpus and put_online_cpus instead as it highlights the
refcount semantics in these operations.
The new API guarantees protection against the cpu-hotplug operation, but
it doesn't guarantee serialized access to any of the local data
structures. Hence the changes needs to be reviewed.
In case of pseries_add_processor/pseries_remove_processor, use
cpu_maps_update_begin()/cpu_maps_update_done() as we're modifying the
cpu_present_map there.
Signed-off-by: Gautham R Shenoy <ego@in.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Before we register the SLB shadow buffer, we need to invalidate the
entries in the buffer, otherwise we can end up stale entries from when
we previously offlined the CPU.
This does this invalidate as well as unregistering the buffer with
PHYP before we offline the cpu. Tested and fixes crashes seen on
970MP (thanks to tonyb) and POWER5.
Signed-off-by: Michael Neuling <mikey@neuling.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>
Don't allow cpu hotplug on systems lacking XICS interrupt controller
(i.e. with an MPIC interrupt controller), since the current pSeries
platform code is hardcoded for XICS.
This works around the bug reported by Paul Mackerras where the
disable_nonboot_cpus() call recently added to the shutdown path will
cause an oops on older pSeries machines.
Signed-off-by: Olof Johansson <olof@lixom.net>
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
Acked-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>
Purely cosmetic. Change pSeries to pseries inline with other parts of the
kernel, and fix an overly long line.
Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Signed-off-by: Paul Mackerras <paulus@samba.org>
To support cpu hotplug on pseries we require two RTAS tokens. The cpu
hotplug machinery should only be wired up if these tokens are found in
the device tree.
Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Signed-off-by: Paul Mackerras <paulus@samba.org>
Move the rest of the hotplug cpu code from platforms/pseries/smp.c into
platforms/pseries/hotplug-cpu.c.
Wire up the smp_ops callbacks and the notifier in the hotplug cpu initcall,
rather than in smp_init_pseries(). No change in behaviour.
Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Signed-off-by: Paul Mackerras <paulus@samba.org>
Move pSeries_mach_cpu_die() into platforms/pseries/hotplug-cpu.c,
this allows rtas_stop_self() to be static so remove the prototype.
Wire up pSeries_mach_cpu_die() in the initcall, rather than statically
in setup.c, the initcall will still run prior to the cpu hotplug code
being callable, so there should be no change in behaviour.
Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Signed-off-by: Paul Mackerras <paulus@samba.org>
As the first step in consolidating the pseries hotplug cpu code,
create platforms/pseries/hotplug-cpu.c and move rtas_stop_self()
into it. Do the rtas token initialisation in a new initcall, rather
than rtas_initialize().
Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Acked-by: Linas Vepstas <linas@austin.ibm.com>
Signed-off-by: Paul Mackerras <paulus@samba.org>
