Additional testing uncovered a situation where the MCA recovery code could
hang due to a race condition.
According to the SAL spec, SAL sends a rendezvous interrupt to all but the first
CPU that goes into MCA. This includes other CPUs that go into MCA at the same
time. Those other CPUs will go into the linux MCA handler (rather than the
slave loop) with the rendezvous interrupt pending. When all the CPUs have
completed MCA processing and the last monarch completes, freeing all the CPUs,
the CPUs with the pended rendezvous interrupt then go into the
ia64_mca_rendez_int_handler(). In ia64_mca_rendez_int_handler() the CPUs
get marked as rendezvoused, but then leave the handler (due to no MCA).
That leaves the CPUs marked as rendezvoused _before_ the next MCA event.
When the next MCA hits, the monarch will mistakenly believe that all the CPUs
are rendezvoused when they are not, opening up a window where a CPU can get
stuck in the slave loop.
This patch avoids leaving CPUs marked as rendezvoused when they are not.
Signed-off-by: Russ Anderson <rja@sgi.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
While testing the MCA recovery code, noticed that some machines would have a
five second delay rendezvousing cpus. What was happening is that
ia64_wait_for_slaves() would check to see if all the slave CPUs had
rendezvoused. If any had not, it would wait 1 millisecond then check again.
If any CPUs had still not rendezvoused, it would wait 5 seconds before
checking again.
On some configs the rendezvous takes more than 1 millisecond, causing the code
to wait the full 5 seconds, even though the last CPU rendezvoused after only
a few milliseconds.
The fix is to check every 1 millisecond to see if all the cpus have
rendezvoused. After 5 seconds the code concludes the CPUs will never
rendezvous (same as before).
The MCA code is, by definition, not performance critical, but a needless
delay of 5 seconds is senseless. The 5 seconds also adds up quickly
when running the error injection code in a loop.
This patch both simplifies the code and removes the needless delay.
Signed-off-by: Russ Anderson <rja@sgi.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
Because it is dead code and not referenced by anybody else (that file cannot
be built modular).
Signed-off-by: Satyam Sharma <satyam@infradead.org>
Cc: "Luck, Tony" <tony.luck@intel.com>
Cc: Stephane Eranian <eranian@hpl.hp.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Tony Luck <tony.luck@intel.com>
* palinfo.c:
palinfo_cpu_notifier is a CPU hotplug notifier_block, and can be
marked __cpuinitdata, and the callback function palinfo_cpu_callback()
itself can be marked __cpuinit. create_palinfo_proc_entries() is only
called from __cpuinit callback or general __init code, therefore a
candidate for __cpuinit itself. remove_palinfo_proc_entries() is only
called from __cpuinit callback or general __exit code, therefore a
candidate for __cpuexit.
* salinfo.c:
The CPU hotplug notifier_block can be __cpuinitdata. The callback
salinfo_cpu_callback() is incorrectly marked __devinit -- it must
be __cpuinit instead.
* topology.c:
cache_sysfs_init() is only called at device_initcall() time so marking
it as __cpuinit is wrong and wasteful. It should be unconditionally
__init. Also cleanup reference to hotplug notifier callback function
from this function and replace with cache_add_dev(), which could also
enable us to use other tricks to replace __cpuinit{data} annotations,
as recently discussed on this list.
cache_shared_cpu_map_setup() is only ever called from __cpuinit-marked
functions hence both its definitions (SMP or !SMP) are candidates for
__cpuinit itself. Also all_cpu_cache_info can be __cpuinitdata because
only referenced from __cpuinit code.
Signed-off-by: Satyam Sharma <satyam@infradead.org>
Cc: "Luck, Tony" <tony.luck@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Tony Luck <tony.luck@intel.com>
When PTRACE_SYSCALL was used and then PTRACE_DETACH is used, the
TIF_SYSCALL_TRACE flag is left set on the formerly-traced task. This
means that when a new tracer comes along and does PTRACE_ATTACH, it's
possible he gets a syscall tracing stop even though he's never used
PTRACE_SYSCALL. This happens if the task was in the middle of a system
call when the second PTRACE_ATTACH was done. The symptom is an
unexpected SIGTRAP when the tracer thinks that only SIGSTOP should have
been provoked by his ptrace calls so far.
A few machines already fixed this in ptrace_disable (i386, ia64, m68k).
But all other machines do not, and still have this bug. On x86_64, this
constitutes a regression in IA32 compatibility support.
Since all machines now use TIF_SYSCALL_TRACE for this, I put the
clearing of TIF_SYSCALL_TRACE in the generic ptrace_detach code rather
than adding it to every other machine's ptrace_disable.
Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This patch cleans up the `enable early console for SKI' patch
(471e7a4484), and
1. potentially allows the gensparse_defconfig to work again.
(there are other problems running a generic kernel on Ski)
2. fixes the `console registered twice' problem.
3. Cleans up the code by moving the `extern hpsim_cons' declaration to
a new asm/hpsim.h file.
Thanks to Jes for comments.
Signed-off-by: Peter Chubb <peterc@gelato.unsw.edu.au>
Signed-off-by: Tony Luck <tony.luck@intel.com>
Add additional support for CPU disable on SN platforms.
Correctly setup the smp_affinity mask for I/O error IRQs.
Restrict the use of the feature to Altix 4000 and 450 systems
running with a CPU disable capable PROM, and do not allow disabling
of CPU 0.
Signed-off-by: John Keller <jpk@sgi.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
The pending interrupts can be remaining at boot up time on some
platform. This will cause spurious interrupts when interrupt is
enabled for the first time. This patch clears IVR at the CPU
initialization to eliminate such spurious interrupts.
Signed-off-by: Kenji Kaneshige <kaneshige.kenji@jp.fujitsu.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
Fix handling for spurious interrupts not being mapped to any IRQs.
Currently, spurious interrupts that are not mapped to any IRQs are
handled as IRQ 15 (== IA64_SPURIOUS_VECTOR). But it is not proper
because vector != irq. We need special handlings for such spurious
interrupts not being mapped to any IRQs.
Signed-off-by: Kenji Kaneshige <kaneshige.kenji@jp.fujitsu.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
When using Ski to debug early startup, it's a bit of a pain not to
have printk.
This patch enables the simulated console very early.
It may be worth conditionalising on the command line... but this is
enough for now.
Signed-off-by: Peter Chubb <peterc@gelato.unsw.edu.au>
Signed-off-by: Tony Luck <tony.luck@intel.com>
The "ri" field in the processor status register only has defined
values of 0, 1, 2. Do not let ptrace set this to 3. As with
other reserved fields in registers we silently discard the value.
Signed-off-by: Shaohua Li <shaohua.li@intel.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
The core cpufreq code doesn't appear to understand returning -EAGAIN
for the get() function of the cpufreq_driver. If PAL_GET_PSTATE returns
-1, such as when running on Xen, scaling_cur_freq is happy to return
4294967285 kHz (ie. (unsigned)-11). The other drivers appear to return
0 for a failure, and doing so gives me the max frequency from
scaling_cur_frequency and "<unknown>" from cpuinfo_cur_frequency. I
believe that's the desired behavior.
Signed-off-by: Alex Williamson <alex.williamson@hp.com>
Acked-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
Explicitly put the unwind section into its own program-header. This
used to be unnecessary (probably because binutils did it for us), but
with current binutils (e.g., v2.17.50.20070804) we won't get
the PT_IA_64_UNWIND header without this patch which will break
unwinding in a debugger and simulators such as Ski.
Signed-off-by: David Mosberger-Tang <dmosberger@gmail.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
Add NOTES to linker script such that the kernel can be built with
recent versions of binutils. Without this patch, final link fails
with this error:
ld: .tmp_vmlinux1: section `.text' can't be allocated in segment 0
ld: final link failed: Bad value
This error is due to the fact that the --build-id option is used
with newer linkers to include a .notes section on the kernel, but
without the NOTES macro, that section won't be included in the kernel
which then leads to the above error message.
Signed-off-by: David Mosberger-Tang <dmosberger@gmail.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
Use local_vector_to_irq() instead of looping through all NR_IRQS.
This avoids registering the CPE handler on multiple irqs. Only
register if the irq is valid. If no valid irq is found, print an
error message and set up polling.
Signed-off-by: Russ Anderson <rja@sgi.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
Add base support for implementing platform_irq_to_vector(), and
then use it on SN2.
Signed-off-by: Kenji Kaneshige <kaneshige.kenji@jp.fujitsu.com>
Acked-by: John Keller <jpk@sgi.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
While sending interrupts to a cpu to repeatedly wake a thread, on occasion
that thread will take a full timer tick cycle (4002 usec in my case)
to wakeup.
The problem concerns a race condition in the code around the safe_halt()
call in the default_idle() routine. Setting 'nohalt' on the kernel
command line causes the long wakeups to disappear.
void
default_idle (void)
{
local_irq_enable();
while (!need_resched()) {
--> if (can_do_pal_halt)
--> safe_halt();
else
A timer tick could arrive between the check for !need_resched and the
actual call to safe_halt() (which does a pal call to PAL_HALT_LIGHT).
By the time the timer tick completes, a thread that might now need to run
could get held up for as long as a timer tick waiting for the halted cpu.
I'm proposing that we disable irq's and check need_resched again before
calling safe_halt(). Does anyone see any problem with this approach?
Signed-off-by: Dimitri Sivanich <sivanich@sgi.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
* 'release' of git://git.kernel.org/pub/scm/linux/kernel/git/aegl/linux-2.6:
[IA64] ITC: Reduce rating for ITC clock if ITCs are drifty
[IA64] SN2: Fix up sn2_rtc clock
[IA64] Fix wrong access to irq_desc[] in iosapic_register_intr().
[IA64] Fix possible race in destroy_and_reserve_irq()
[IA64] Fix registered interrupt check
[IA64] Remove a few duplicate includes
[IA64] Allow smp_call_function_single() to current cpu
[IA64] fix a few section mismatch warnings
Make sure to reduce the rating of the ITC clock if ITCs are drifty. If they
are drifting then we have not synchronized the ITC values, nor are we doing
the jitter compensation (useless since drift may increase the differentials
arbitrarily).
Without this patch it is possible that the ITC clock becomes selected as
the system clock on systems with drifty ITCs which will result in
nanosleep hanging.
One can still select the itc clock manually on such systems via
clocksource=itc
(Produces nice hangs on SGI Altix.)
Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
In error path we must unlock irq_desc[irq].lock before we change
'irq'.
Signed-off-by: Kenji Kaneshige <kaneshige.kenji@jp.fujitsu.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
Remove unused TIF_NOTIFY_RESUME flag for all processor architectures. The
flag was not used excecpt on IA-64 where the patch replaces it with
TIF_PERFMON_WORK.
Signed-off-by: stephane eranian <eranian@hpl.hp.com>
Cc: <linux-arch@vger.kernel.org>
Cc: "Luck, Tony" <tony.luck@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Currently, destroy_and_reserve_irq() sets irq_status[irq] UNUSED using
clear_irq_vector() and sets irq_status[irq] RSVD using reserve_irq().
But there is a race window because vector_lock is once released between
them. This patch fixes this race window.
Signed-off-by: Kenji Kaneshige <kaneshige.kenji@jp.fujitsu.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
Fix the problem that interrupts are not initialized correctly at PCI
hotplug or driver reloading time.
By vector domain change, the iosapic_rte_info structure was changed to
be on the iosapic_intr_info[irq].rtes list even after the interrupts
are unregistered. So iosapic_intr_info[irq].rtes list must not be
checked to see if there are registered interrupts (RTEs) on the
irq. We must check iosapic_intr_info[irq].count counter instead.
Signed-off-by: Kenji Kaneshige <kaneshige.kenji@jp.fujitsu.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
This patch removes a few duplicate includes from arch/ia64/
Acked-by: Jes Sorensen <jes@sgi.com>
Signed-off-by: Jesper Juhl <jesper.juhl@gmail.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
This removes the requirement for callers to get_cpu() to check in simple
cases. i386 and x86_64 already received a similar treatment.
Signed-off-by: Avi Kivity <avi@qumranet.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
Fix the following section mismatch warnings:
WARNING: vmlinux.o(.text+0x41902): Section mismatch: reference to .init.text:__alloc_bootmem (between 'ia64_mca_cpu_init' and 'ia64_do_tlb_purge')
WARNING: vmlinux.o(.text+0x49222): Section mismatch: reference to .init.text:__alloc_bootmem (between 'register_intr' and 'iosapic_register_intr')
WARNING: vmlinux.o(.text+0x62beb2): Section mismatch: reference to .init.text:__alloc_bootmem_node (between 'hubdev_init_node' and 'cnodeid_get_geoid')
Signed-off-by: Sam Ravnborg <sam@ravnborg.org>
Signed-off-by: Tony Luck <tony.luck@intel.com>
Fix wrong return value in parse_vector_domain().
Signed-off-by: Kenji Kaneshige <kaneshige.kenji@jp.fujitsu.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
The ia64's acpi_gsi_to_irq() function assumes irq == vector. But in
fact irq can be different from vector. This patch fix this wrong
assumption.
Signed-off-by: Kenji Kaneshige <kaneshige.kenji@jp.fujitsu.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
Add some sanity checks into __bind_irq_vector().
Signed-off-by: Kenji Kaneshige <kaneshige.kenji@jp.fujitsu.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
u32* volatile cyclone_timer means volatile auto pointer to u32,
which is clearly not what had been intended (we never even take
the address of that variable, let alone pass it to something that
could change it behind our back). u32 volatile * is what the
authors apparently wanted to say, but in reality we don't need that
qualifier there at all - it's (properly) only passed to iomem helpers
which takes care of that stuff just fine.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
* 'release' of git://git.kernel.org/pub/scm/linux/kernel/git/aegl/linux-2.6:
[IA64] Nail two more simple section mismatch errors
[IA64] fix section mismatch warnings
[IA64] rename partial_page
[IA64] Ensure that machvec is set up takes place before serial console
[IA64] vector-domain - fix vector_table
[IA64] vector-domain - handle assign_irq_vector(AUTO_ASSIGN)
pcibios_setup (between 'pci_setup' and 'quirk_mellanox_tavor')
setup_profiling_timer (between 'write_profile' and 'delayed_put_task_struct')
Signed-off-by: Tony Luck <tony.luck@intel.com>
In 741f98fe29 Sam added full
checking across the entire vmlinux image. This flushed out
a dozen new section mismatch warnings. Start the whack-a-mole
game again to stomp them out.
Signed-off-by: Tony Luck <tony.luck@intel.com>
Jens has added a partial_page thing in splice whcih conflicts with the ia64
one. Rename ia64 out of the way. (ia64 chose poorly).
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Tony Luck <tony.luck@intel.com>
Parse the machvec command line option outside of the early_param()
so that ia64_mv is set before any console intialisation that
may result from early_param parsing.
Signed-off-by: Simon Horman <horms@verge.net.au>
Signed-off-by: Tony Luck <tony.luck@intel.com>
This change fixes a panic when assign_irq_vector(irq) is called with
irq = AUTO_ASSIGN.
Signed-off-by: Yasuaki Ishimatsu <isimatu.yasuaki@jp.fujitsu.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
As it was a synonym for (CONFIG_ACPI && CONFIG_X86),
the ifdefs for it were more clutter than they were worth.
For ia64, just add a few stubs in anticipation of future
S3 or S4 support.
Signed-off-by: Len Brown <len.brown@intel.com>
* 'release' of git://git.kernel.org/pub/scm/linux/kernel/git/aegl/linux-2.6:
[IA64] Prevent people from directly including <asm/rwsem.h>.
[IA64] remove time interpolator
[IA64] Convert to generic timekeeping/clocksource
[IA64] refresh some config files for 64K pagesize
[IA64] Delete iosapic_free_rte()
[IA64] fallocate system call
[IA64] Enable percpu vector domain for IA64_DIG
[IA64] Enable percpu vector domain for IA64_GENERIC
[IA64] Support irq migration across domain
[IA64] Add support for vector domain
[IA64] Add mapping table between irq and vector
[IA64] Check if irq is sharable
[IA64] Fix invalid irq vector assumption for iosapic
[IA64] Use dynamic irq for iosapic interrupts
[IA64] Use per iosapic lock for indirect iosapic register access
[IA64] Cleanup lock order in iosapic_register_intr
[IA64] Remove duplicated members in iosapic_rte_info
[IA64] Remove block structure for locking in iosapic.c
This is a merge of Peter Keilty's initial patch (which was
revived by Bob Picco) for this with Hidetoshi Seto's fixes
and scaling improvements.
Acked-by: Bob Picco <bob.picco@hp.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
> arch/ia64/kernel/iosapic.c:597: warning: 'iosapic_free_rte' defined but not used
>
> This isn't spurious, the only call to iosapic_free_rte() has been removed, but there
> is still a call to iosapic_alloc_rte() ... which means we must have a memory leak.
I did it on purpose (and gave the warning a miss...) and I consider
iosapic_free_rte() is no longer needed.
I decided to remain iosapic_rte_info to keep gsi-to-irq binding
after device disable. Indeed it needs some extra memory, but it
is only "sizeof(iosapic_rte_info) * <the number of removed devices>"
bytes and has no memory leak becasue re-enabled devices use the
iosapic_rte_info which they used before disabling.
Signed-off-by: Yasuaki Ishimatsu <isimatu.yasuaki@jp.fujitsu.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
sys_fallocate for ia64. This uses an empty slot #1303 erroneously
marked as reserved for move_pages (which had already been allocated
as syscall #1276)
Signed-Off-By: Dave Chinner <dgc@sgi.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
Since Ingo's recent scheduler rewrite which was merged as commit
0437e109e1 sched_cacheflush is unused.
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Currently most of the per cpu data, which is accessed by different cpus,
has a ____cacheline_aligned_in_smp attribute. Move all this data to the
new per cpu shared data section: .data.percpu.shared_aligned.
This will seperate the percpu data which is referenced frequently by other
cpus from the local only percpu data.
Signed-off-by: Fenghua Yu <fenghua.yu@intel.com>
Acked-by: Suresh Siddha <suresh.b.siddha@intel.com>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Christoph Lameter <clameter@sgi.com>
Cc: "Luck, Tony" <tony.luck@intel.com>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
per cpu data section contains two types of data. One set which is
exclusively accessed by the local cpu and the other set which is per cpu,
but also shared by remote cpus. In the current kernel, these two sets are
not clearely separated out. This can potentially cause the same data
cacheline shared between the two sets of data, which will result in
unnecessary bouncing of the cacheline between cpus.
One way to fix the problem is to cacheline align the remotely accessed per
cpu data, both at the beginning and at the end. Because of the padding at
both ends, this will likely cause some memory wastage and also the
interface to achieve this is not clean.
This patch:
Moves the remotely accessed per cpu data (which is currently marked
as ____cacheline_aligned_in_smp) into a different section, where all the data
elements are cacheline aligned. And as such, this differentiates the local
only data and remotely accessed data cleanly.
Signed-off-by: Fenghua Yu <fenghua.yu@intel.com>
Acked-by: Suresh Siddha <suresh.b.siddha@intel.com>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Christoph Lameter <clameter@sgi.com>
Cc: <linux-arch@vger.kernel.org>
Cc: "Luck, Tony" <tony.luck@intel.com>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
