A number of small optimisations to FPU handling, in particular:
- move the task USEDFPU flag from the thread_info flags field (which
is accessed asynchronously to the thread) to a new status field,
which is only accessed by the thread itself. This allows locking to
be removed in most cases, or can be reduced to a preempt_lock().
This mimics the i386 behaviour.
- move the modification of regs->sr and thread_info->status flags out
of save_fpu() to __unlazy_fpu(). This gives the compiler a better
chance to optimise things, as well as making save_fpu() symmetrical
with restore_fpu() and init_fpu().
- implement prepare_to_copy(), so that when creating a thread, we can
unlazy the FPU prior to copying the thread data structures.
Also make sure that the FPU is disabled while in the kernel, in
particular while booting, and for newly created kernel threads,
In a very artificial benchmark, the execution time for 2500000
context switches was reduced from 50 to 45 seconds.
Signed-off-by: Stuart Menefy <stuart.menefy@st.com>
Signed-off-by: Paul Mundt <lethal@linux-sh.org>
sh port of the sLeAZY-fpu feature currently implemented for some architectures
such us i386.
Right now the SH kernel has a 100% lazy fpu behaviour.
This is of course great for applications that have very sporadic or no FPU use.
However for very frequent FPU users... you take an extra trap every context
switch.
The patch below adds a simple heuristic to this code: after 5 consecutive
context switches of FPU use, the lazy behavior is disabled and the context
gets restored every context switch.
After 256 switches, this is reset and the 100% lazy behavior is returned.
Tests with LMbench showed no regression.
I saw a little improvement due to the prefetching (~2%).
The tests below also show that, with this sLeazy patch, indeed,
the number of FPU exceptions is reduced.
To test this. I hacked the lat_ctx LMBench to use the FPU a little more.
sLeasy implementation
===========================================
switch_to calls | 79326
sleasy calls | 42577
do_fpu_state_restore calls| 59232
restore_fpu calls | 59032
Exceptions: 0x800 (FPU disabled ): 16604
100% Leazy (default implementation)
===========================================
switch_to calls | 79690
do_fpu_state_restore calls | 53299
restore_fpu calls | 53101
Exceptions: 0x800 (FPU disabled ): 53273
Signed-off-by: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Signed-off-by: Stuart Menefy <stuart.menefy@st.com>
Signed-off-by: Paul Mundt <lethal@linux-sh.org>
The typename member of struct irq_chip was kept for migration purposes
and is obsolete since more than 2 years. Fix up the leftovers.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: linux-sh@vger.kernel.org
Signed-off-by: Paul Mundt <lethal@linux-sh.org>
sh64 doesn't use GENERIC_BUG, which presently causes the handle_BUG()
code to blow up. Fix up the dependencies and get it all building again.
Signed-off-by: Paul Mundt <lethal@linux-sh.org>
This fixes up the build and behaviour for various configurations. Namely
the CONFIG_32BIT cases where legacy mappings do not exist, as well as the
sh64 build.
Signed-off-by: Paul Mundt <lethal@linux-sh.org>
The dwarf unwinder presently attempts to provide a sane PC value if none
is provided, however the logic is broken and cases where a previous valid
dwarf frame exists along with a bogus PC value can still proceed. This
fixes up the test and prevents the unwinder from blowing up.
Signed-off-by: Paul Mundt <lethal@linux-sh.org>
This implements preliminary support for perf callchains (at the moment
only the kernel side is implemented). The actual implementation itself is
just a simple wrapper around the unwinder API, which allows for callchain
generation with or without the dwarf unwinder.
Signed-off-by: Paul Mundt <lethal@linux-sh.org>
Wanted by the SPU2 UIO driver, which really ought to be handling this
itself. Default enable it for now, until the driver gets a bit more
intelligent.
Signed-off-by: Paul Mundt <lethal@linux-sh.org>
'config' can be unintialized, and although it's not really an error, it
still manages to trigger the -Werror with certain toolchains. Initialize
it early to shut up gcc.
Signed-off-by: Paul Mundt <lethal@linux-sh.org>
Add sh7724 code to save and restore RWDT state during
R-standby. Without this patch the watchdog will generate
a reset shortly after resuming from R-standby.
Signed-off-by: Magnus Damm <damm@opensource.se>
Signed-off-by: Paul Mundt <lethal@linux-sh.org>
Different CPUs will have different starting vectors, with varying
amounts of reserved or unusable vector space prior to the first slot.
This introduces a legacy vector reservation system that inserts itself in
between the CPU vector map registration and the platform specific IRQ
setup. This works fine in practice as the only new vectors that boards
need to establish on their own should be dynamically allocated rather
than arbitrarily assigned. As a plus, this also makes all of the
converted platforms sparseirq ready.
Signed-off-by: Paul Mundt <lethal@linux-sh.org>
Add R-standby specific bits to the SuperH Mobile sleep code.
Signed-off-by: Magnus Damm <damm@opensource.se>
Signed-off-by: Paul Mundt <lethal@linux-sh.org>
Use RSMEM instead of ILMEM for sleep mode code storage on SH7724.
This allows us to use R-standby mode on SH7724.
Signed-off-by: Magnus Damm <damm@opensource.se>
Signed-off-by: Paul Mundt <lethal@linux-sh.org>
Add MMU and cache handling functionality to the SuperH Mobile
sleep code. The MMU and cache registers are saved and restored.
The MMU is disabled and the cache is flushed and disabled before
entering sleep modes if the SUSP_SH_MMU flag is set. This flag
should be set in the case of R-standby and most likely for future
U-standby support as well.
Signed-off-by: Magnus Damm <damm@opensource.se>
Signed-off-by: Paul Mundt <lethal@linux-sh.org>
Add code to keep track of supported sleep modes. This to
only export cpuidle modes that are backed by board support
code. Also, do not allow suspend-to-ram if sdram board code
is missing.
Signed-off-by: Magnus Damm <damm@opensource.se>
Signed-off-by: Paul Mundt <lethal@linux-sh.org>
Rework the SuperH Mobile sleep code from including
board specific code to allowing each board to provide
pre/post code snippets. These snippets should contain
sdram management code to enter and leave self-refresh.
Signed-off-by: Magnus Damm <damm@opensource.se>
Signed-off-by: Paul Mundt <lethal@linux-sh.org>
Add code to allow boards registering self-contained
functions for going to/from self-refresh. At this
point the board code is unused. When all supported
boards have been converted then the new sleep code
will make use of these functions.
Signed-off-by: Magnus Damm <damm@opensource.se>
Signed-off-by: Paul Mundt <lethal@linux-sh.org>
Make use of the recently added notifier chains for sh7724
r-standby register save/restore handling. At this point
only the BSC and INTC are handled.
Signed-off-by: Magnus Damm <damm@opensource.se>
Signed-off-by: Paul Mundt <lethal@linux-sh.org>
This patch adds atomic notifier chains for pre/post
sleep events. Useful for cpu code and boards that
need to save and restore register state before and
after entering a sleep mode.
Signed-off-by: Magnus Damm <damm@opensource.se>
Signed-off-by: Paul Mundt <lethal@linux-sh.org>
num_events should be compared > MAX_HWEVENTS and not >=. The latter was
used as a debugging test which accidentally slipped in.
Signed-off-by: Paul Mundt <lethal@linux-sh.org>
This adds in preliminary support for the SH-4A performance counters.
Presently only the first 2 counters are supported, as these are the ones
of the most interest to the perf tool and end users. Counter chaining is
not presently handled, so these are simply implemented as 32-bit
counters.
This also establishes a perf event support framework for other hardware
counters, which the existing SH-4 oprofile code will migrate over to as
the SH-4A support evolves.
Signed-off-by: Paul Mundt <lethal@linux-sh.org>
Many of these symbols went away completely, or we just never cared about
them in the first place. Trim the exports down to the essential set.
Signed-off-by: Paul Mundt <lethal@linux-sh.org>
These were previously hidden in sh_ksyms_32, despite also being needed
for sh64 now that the cache.c code is shared.
Signed-off-by: Paul Mundt <lethal@linux-sh.org>
Leaving this configurable caused more trouble than it was ever worth, so
just make it explicit. Boards that are verified one way or the other can
fix up their selects accordingly. We presently default to non-coherent
for most platforms.
Signed-off-by: Paul Mundt <lethal@linux-sh.org>
Now that SH's irqflags functions are out of line it becomes necessary to
mark them as "notrace" so that we don't try to trace them.
[ Do the same for irq_64.c -- PFM. ]
Signed-off-by: Matt Fleming <matt@console-pimps.org>
Signed-off-by: Paul Mundt <lethal@linux-sh.org>
This unbreaks kexec support. Without this fix all
cases of kexec fails since __pa() does not behave
like PHYSADDR(). The downside is that we also kill
the code blocking users running old kexec-tools.
Signed-off-by: Magnus Damm <damm@opensource.se>
Signed-off-by: Paul Mundt <lethal@linux-sh.org>
When CONFIG_FUNCTION_GRAPH_TRACER is enabled the function graph tracer
may patch return addresses on the stack with the address of
return_to_handler(). This really confuses the DWARF unwinder because it
will try find the caller of return_to_handler(), not the caller of the
real return address.
So teach the DWARF unwinder how to find the real return address whenever
it encounters return_to_handler().
This patch does not cope very well when multiple return addresses on the
stack have been patched. To make it work properly it would require state
to track how many return_to_handler()'s have been seen so that we'd know
where to look in current->curr_ret_stack[]. So for now, instead of
trying to handle this, just moan if more than one return address on the
stack has been patched.
Signed-off-by: Matt Fleming <matt@console-pimps.org>
Signed-off-by: Paul Mundt <lethal@linux-sh.org>
This adds an __irq_entry annotation for do_IRQ() so that the IRQ
annotation in the function graph tracer works as advertized. We already
have the IRQENTRY section wired up, so this is just a trivial addition
to actually make use of it.
Signed-off-by: Paul Mundt <lethal@linux-sh.org>
This moves the current dma_alloc/free_coherent() calls to a generic
variant and plugs them in for the nommu default. Other variants can
override the defaults in the dma mapping ops directly.
Signed-off-by: Paul Mundt <lethal@linux-sh.org>
SCIF2 and the FPU exceptions happen to share vector numbers, one in
EXPEVT and the other in INTEVT. This is a violation of the interface and
should have never made it in to silicon. On top of that, the demux hack
that was added for special dispatch is rather error prone, and introduces
more problems than it solves. Kill all of it off, and just refuse to deal
with SCIF2 outright.
Signed-off-by: Paul Mundt <lethal@linux-sh.org>
This simplifies the irqflags support by switching over to the asm-generic
version. The necessary support functions are brought out-of-line for both
SHcompact and SHmedia instruction sets.
Signed-off-by: Paul Mundt <lethal@linux-sh.org>
This code was added for some ancient SH-4 solution engines with peculiar
boot ROMs that did silly things to the UBC MSTP bits. None of these have
been in the wild for years, and these days the clock framework wraps up
the MSTP bits, meaning that the UBC code is one of the few interfaces
that is stomping MSTP bits underneath the clock framework. At this point
the risks far outweigh any benefit this code provided, so just kill it
off.
Signed-off-by: Paul Mundt <lethal@linux-sh.org>
This enables SCHED_MC support for SH-X3 multi-cores. Presently this is
just a simple wrapper around the possible map, but this allows for
tying in support for some of the more exotic NUMA clusters where we can
actually do something with the topology.
Signed-off-by: Paul Mundt <lethal@linux-sh.org>
In the case where need_resched() is set in between the cpu_idle() and
pm_idle() calls we were missing an else case for just re-enabling local
IRQs and bailing out. This was noticed by the irqs_disabled() warning,
even though IRQs were being re-enabled elsewhere.
Signed-off-by: Paul Mundt <lethal@linux-sh.org>
This follows the x86 change and moves check_pgt_cache() up under the
!need_resched() tight loop, rather than simply calling in to it when
exiting idle.
Signed-off-by: Paul Mundt <lethal@linux-sh.org>
This does a bit of chainsawing of the idle loop code to get light sleep
working on SMP. Previously this was forcing secondary CPUs in to sleep
mode with them not coming back if they didn't have their own local
timers. Given that we use clockevents broadcasting by default, the CPU
managing the clockevents can't have IRQs disabled before entering its
sleep state.
This unfortunately leaves us with the age-old need_resched() race in
between local_irq_enable() and cpu_sleep(), but at present this is
unavoidable. After some more experimentation it may be possible to layer
on SR.BL bit manipulation over top of this scheme to inhibit the race
condition, but given the current potential for missing wakeups, this is
left as a future exercise.
Signed-off-by: Paul Mundt <lethal@linux-sh.org>
Paul Mundt