This allows platforms to hook into the initialization early to setup
things like scheduler clocks, etc.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Rather than storing each machine init hook separately, store a
pointer to the machine description record and dereference this
instead. This pointer is only available while the init sections
are present, which is not a problem as we only use it from init
code.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Per subarch interrupt handler macros V3.
This patch breaks out code from the irq_handler macro
into arch_irq_handler and arch_irq_handler_default.
The macros are put in the header file "entry-macro-multi.S"
The arch_irq_handler_default macro is designed to be
used by irq_handler in entry-armv.S while arch_irq_handler
is suitable for per-subarch use.
Signed-off-by: Magnus Damm <damm@opensource.se>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Normally different ARM platform has different way to decode the IRQ
hardware status and demultiplex to the corresponding IRQ handler.
This is highly optimized by macro irq_handler in entry-armv.S, and
each machine defines their own macro to decode the IRQ number.
However, this prevents multiple machine classes to be built into a
single kernel.
By allowing each machine to specify thier own handler, and making
function pointer 'handle_arch_irq' to point to it at run time, this
can be solved. And introduce CONFIG_MULTI_IRQ_HANDLER to allow both
solutions to work.
Comparing with the highly optimized macro of irq_handler, the new
function must be written with care not to lose too much performance.
And the IPI stuff on SMP is expected to move to the provided arch
IRQ handler as well.
The assembly code to invoke handle_arch_irq is optimized by Russell
King.
Signed-off-by: Eric Miao <eric.miao@canonical.com>
Acked-by: Nicolas Pitre <nicolas.pitre@linaro.org>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
If the irqsoff tracer is in use, stop tracing the interrupt disable
interval when returning to userspace. Tracing userspace execution time
as interrupts disabled time is not helpful for kernel performance
analysis purposes. Only do so if the irqsoff tracer is enabled, to
avoid overhead for lockdep, which doesn't care.
Signed-off-by: Todd Poynor <toddpoynor@google.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Provide common sched_clock() infrastructure for platforms to use to
create a 64-bit ns based sched_clock() implementation from a counter
running at a non-variable clock rate.
This implementation is based upon maintaining an epoch for the counter
and an epoch for the nanosecond time. When we desire a sched_clock()
time, we calculate the number of counter ticks since the last epoch
update, convert this to nanoseconds and add to the epoch nanoseconds.
We regularly refresh these epochs within the counter wrap interval.
We perform a similar calculation as above, and store the new epochs.
We read and write the epochs in such a way that sched_clock() can easily
(and locklessly) detect when an update is in progress, and repeat the
loading of these constants when they're known not to be stable. The
one caveat is that sched_clock() is not called in the middle of an
update. We achieve that by disabling IRQs.
Finally, if the clock rate is known at compile time, the counter to ns
conversion factors can be specified, allowing sched_clock() to be tightly
optimized. We ensure that these factors are correct by providing an
initialization function which performs a run-time check.
Acked-by: Peter Zijlstra <peterz@infradead.org>
Tested-by: Santosh Shilimkar <santosh.shilimkar@ti.com>
Tested-by: Will Deacon <will.deacon@arm.com>
Tested-by: Mikael Pettersson <mikpe@it.uu.se>
Tested-by: Eric Miao <eric.y.miao@gmail.com>
Tested-by: Olof Johansson <olof@lixom.net>
Tested-by: Jamie Iles <jamie@jamieiles.com>
Reviewed-by: Nicolas Pitre <nicolas.pitre@linaro.org>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
When we soft-CPU hotplug a CPU, we reset the stack pointer and
jump back to start_secondary(). This allows us to restart as if
the CPU was actually reset.
However, we weren't resetting the frame pointer, which could cause
problems with backtracing. Reset the frame pointer to zero (which
means no parent frame) just like the early assembly code also does.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
smp.c is becoming too large, so split out the TLB maintainence
broadcasting into a separate smp_tlb.c file.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
When a CPU is hot unplugged, the generic tick code cleans up the
clock event device, but fails to call down to the device's set_mode
function to actually shut the device down.
To work around this, we've historically had a local_timer_stop()
callback out of the hotplug code. However, this adds needless
complexity when we have the clock event device itself available.
Explicitly call the clock event device's set_mode function with
CLOCK_EVT_MODE_UNUSED, so that the hardware can be cleanly shutdown
without any special external callbacks. When/if the generic code
is fixed, percpu_timer_stop() can be killed off.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
We used to print a bland error message which gave no clue as to the
failure when we failed to bring up a secondary CPU. Resolve this by
separating the two failure cases.
If boot_secondary() fails, we print a message indicating the returned
error code from boot_secondary():
"CPU%u: failed to boot: %d\n", cpu, ret.
However, if boot_secondary() succeeded, but the CPU did not appear to
mark itself online within the timeout, indicate that it failed to come
online:
"CPU%u: failed to come online\n", cpu
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
* __fixup_smp_on_up has been modified with support for the
THUMB2_KERNEL case. For THUMB2_KERNEL only, fixups are split
into halfwords in case of misalignment, since we can't rely on
unaligned accesses working before turning the MMU on.
No attempt is made to optimise the aligned case, since the
number of fixups is typically small, and it seems best to keep
the code as simple as possible.
* Add a rotate in the fixup_smp code in order to support
CPU_BIG_ENDIAN, as suggested by Nicolas Pitre.
* Add an assembly-time sanity-check to ALT_UP() to ensure that
the content really is the right size (4 bytes).
(No check is done for ALT_SMP(). Possibly, this could be fixed
by splitting the two uses ot ALT_SMP() (ALT_SMP...SMP_UP versus
ALT_SMP...SMP_UP_B) into two macros. In the first case,
ALT_SMP needs to expand to >= 4 bytes, not == 4.)
* smp_mpidr.h (which implements ALT_SMP()/ALT_UP() manually due
to macro limitations) has not been modified: the affected
instruction (mov) has no 16-bit encoding, so the correct
instruction size is satisfied in this case.
* A "mode" parameter has been added to smp_dmb:
smp_dmb arm @ assumes 4-byte instructions (for ARM code, e.g. kuser)
smp_dmb @ uses W() to ensure 4-byte instructions for ALT_SMP()
This avoids assembly failures due to use of W() inside smp_dmb,
when assembling pure-ARM code in the vectors page.
There might be a better way to achieve this.
* Kconfig: make SMP_ON_UP depend on
(!THUMB2_KERNEL || !BIG_ENDIAN) i.e., THUMB2_KERNEL is now
supported, but only if !BIG_ENDIAN (The fixup code for Thumb-2
currently assumes little-endian order.)
Tested using a single generic realview kernel on:
ARM RealView PB-A8 (CONFIG_THUMB2_KERNEL={n,y})
ARM RealView PBX-A9 (SMP)
Signed-off-by: Dave Martin <dave.martin@linaro.org>
Acked-by: Nicolas Pitre <nicolas.pitre@linaro.org>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Don't call idle_task_exit() with interrupts disabled, and ensure
that we have a memory barrier after interrupts are disabled but
before signalling that this CPU has shut down.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
We always need to wait for the dying CPU to reach a safe state before
taking it down, irrespective of the requirements of the platform.
Move the completion code into the ARM SMP hotplug code rather than
having each platform re-implement this.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
All platforms call trace_hardirqs_off() in their secondary startup code,
so move this into the core SMP code - it doesn't need to be in the
per-platform code.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
There is a certain amount of smp_prepare_cpus() which doesn't belong
in the platform support code - that is, code which is invariant to the
SMP implementation. Move this code into arch/arm/kernel/smp.c, and
add a platform_ prefix to the original function.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Wait for CPUs to indicate that they've stopped, after sending the
stop IPI, rather than blindly continuing on and hoping that they've
stopped in time. Print a warning if we fail to stop the other CPUs.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Use r0,r3-r6 rather than r0,r3,r4,r6,r7, which makes it easier to
understand which registers can be modified. Also document which
registers hold values which must be preserved.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
The IPI and local timer interrupts weren't being properly accounted
for in /proc/stat. Collect them from the irq_stat structure, and
return their sum.
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
This separates out the individual IPI interrupt counts from the
total IPI count, which allows better visibility of what IPIs are
being used for.
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
As per x86, align the initial column according to how many IRQs we
have. Also, provide an english explaination for the 'LOC:' and
'IPI:' lines.
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Move the ipi_count into irq_stat, which allows the ipi_data structure
to be entirely removed.
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Provide __inc_irq_stat() and __get_irq_stat() to increment and
read the irq stat counters.
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
send_ipi_message() does nothing except call smp_cross_call(). As
this is a static function, nothing external to this file calls it,
so we can easily clean up this now unnecessary indirection.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
We should not be incrementing mm_users when we startup a secondary
CPU - doing so results in mm_users incrementing by one each time we
hotplug a CPU, which will eventually wrap, and will cause problems.
Other architectures such as x86 do not increment mm_users, but only
mm_count, so we follow that pattern.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Avoid adding nasty genirq-specific code to local timers to enable PPI
interrupts. Instead, provide a gic function to do this.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
As our SMP implementation uses MESI protocols. Grouping together data
which is mostly only read together means that we avoid unnecessary
cache line bouncing when this code shares a cache line with other data.
In other words, cache lines associated with read-mostly data are
expected to spend most of their time in shared state.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
When kexec is used to start a crash kernel the other cores
are notified. These non-crashing cores will save their state
in the crash notes and then do nothing.
Signed-off-by: Per Fransson <per.xx.fransson@stericsson.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
As we've now removed the spinlock and bitmask, we have nothing left
which requires interrupts to be disabled when sending an IPI. All
current IPI-sending implementations use the GIC, which also does not
require interrupts disabled when calling gic_raise_softirq().
Remove the now unnecessary IRQ disable.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Avoid using bitmasks and locks in the percpu area for IPIs, and instead
use individual software generated interrupts to identify the reason for
the IPI. This avoids the problems of having spinlocks in the percpu
area.
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
This allows us to use smp_cross_call() to trigger a number of different
software generated interrupts, rather than combining them all on one
SGI. Recover the SGI number via do_IPI.
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
If a section is not marked with SHF_ALLOC, it will be discarded
by the module code. Therefore, it is not correct to register
the unwind tables.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
There's no need to keep pointers to the ELF sections available while
the module is loaded - we only need the section pointers while we're
finding and registering the unwind tables, which can all be done during
the finalize stage of loading.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
The 32-bit conditional branches in Thumb-2 have a shorter range
(+/-512K) than their ARM counterparts (+/-32MB). The linker does
not currently generate trampolines to extend the range of these
Thumb-2 conditional branches, resulting in link errors when vmlinux
is sufficiently large, e.g.:
head.o:(.text+0x464): relocation truncated to fit: R_ARM_THM_JUMP19
This patch forces the longer-range, unconditional branch encoding
by use of an explicit IT instruction. The resulting branches are
triggered on the same conditions as before.
Signed-off-by: Dave Martin <dave.martin@linaro.org>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Directives such as .long and .word do not magically cause the
assembler location counter to become aligned in gas. As a result,
using these directives in code sections can result in misaligned
data words when building a Thumb-2 kernel (CONFIG_THUMB2_KERNEL).
This is a Bad Thing, since the ABI permits the compiler to assume
that fundamental types of word size or above are word- aligned when
accessing them from C. If the data is not really word-aligned,
this can cause impaired performance and stray alignment faults in
some circumstances.
In general, the following rules should be applied when using data
word declaration directives inside code sections:
* .quad and .double:
.align 3
* .long, .word, .single, .float:
.align (or .align 2)
* .short:
No explicit alignment required, since Thumb-2
instructions are always 2 or 4 bytes in size.
immediately after an instruction.
Reviewed-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Dave Martin <dave.martin@linaro.org>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Directives such as .long and .word do not magically cause the
assembler location counter to become aligned in gas. As a result,
using these directives in code sections can result in misaligned
data words when building a Thumb-2 kernel (CONFIG_THUMB2_KERNEL).
This is a Bad Thing, since the ABI permits the compiler to assume
that fundamental types of word size or above are word- aligned when
accessing them from C. If the data is not really word-aligned,
this can cause impaired performance and stray alignment faults in
some circumstances.
In general, the following rules should be applied when using data
word declaration directives inside code sections:
* .quad and .double:
.align 3
* .long, .word, .single, .float:
.align (or .align 2)
* .short:
No explicit alignment required, since Thumb-2
instructions are always 2 or 4 bytes in size.
immediately after an instruction.
Reviewed-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Dave Martin <dave.martin@linaro.org>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
swp_emulate is only used on ARMv7+, and includes ARMv7+ assembly
instructions. Allow the assembler to accept ARMv7 instructions,
but leave the compiler's code generation options alone.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
The ARM perf_event.c file contains all PMU backends and, as new PMUs
are introduced, will continue to grow.
This patch follows the example of x86 and splits the PMU implementations
into separate files which are then #included back into the main
file. Compile-time guards are added to each PMU file to avoid compiling
in code that is not relevant for the version of the architecture which
we are targetting.
Acked-by: Jean Pihet <j-pihet@ti.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
Currently, perf uses the PMU ID as an index into a string table
to look up the name of a given PMU.
This patch encodes the name of a PMU directly into the arm_pmu
structure so that PMU-specific code can be factored out into
separate files.
Acked-by: Jamie Iles <jamie@jamieiles.com>
Acked-by: Jean Pihet <j-pihet@ti.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
In preparation for separating the PMU-specific code, this patch adds
self-contained init functions to each PMU, therefore removing any
PMU-specific knowledge from the PMU-agnostic init_hw_perf_events
function.
Acked-by: Jamie Iles <jamie@jamieiles.com>
Acked-by: Jean Pihet <j-pihet@ti.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
Unlike other pmu functions, armv6pmu_pmu_stop is not declared static.
This patch adds the missing keyword.
Acked-by: Jamie Iles <jamie.iles@jamieiles.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
The functions for mapping PMU events (perf, cache and raw) are
common between all PMU types and differ only in the data on which
they operate.
This patch implements common definitions of these mapping functions
and changes the arm_pmu struct to hold pointers to the data which
they require. This is in anticipation of separating out the PMU-specific
code into separate files.
Acked-by: Jamie Iles <jamie.iles@jamieiles.com>
Acked-by: Jean Pihet <j-pihet@ti.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
