This is the gereric part of R4000/R4400 errata workarounds. They include
compiler and assembler support as well as some source code modifications
to address the problems with some combinations of multiply/divide+shift
instructions as well as the daddi and daddiu instructions.
Changes included are as follows:
1. New Kconfig options to select workarounds by platforms as necessary.
2. Arch top-level Makefile to pass necessary options to the compiler; also
incompatible configurations are detected (-mno-sym32 unsupported as
horribly intrusive for little gain).
3. Bug detection updated and shuffled -- the multiply/divide+shift problem
is lethal enough that if not worked around it makes the kernel crash in
time_init() because of a division by zero; the daddiu erratum might
also trigger early potentially, though I have not observed it. On the
other hand the daddi detection code requires the exception subsystem to
have been initialised (and is there mainly for information).
4. r4k_daddiu_bug() added so that the existence of the erratum can be
queried by code at the run time as necessary; useful for generated code
like TLB fault and copy/clear page handlers.
5. __udelay() updated as it uses multiplication in inline assembly.
Note that -mdaddi requires modified toolchain (which has been maintained
by myself and available from my site for ~4years now -- versions covered
are GCC 2.95.4 - 4.1.2 and binutils from 2.13 onwards). The -mfix-r4000
and -mfix-r4400 have been standard for a while though.
Signed-off-by: Maciej W. Rozycki <macro@linux-mips.org>
Signed-off-by: Ralf Baechle <ralf@linux-mips.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>
All kobjects require a dynamically allocated name now. We no longer
need to keep track if the name is statically assigned, we can just
unconditionally free() all kobject names on cleanup.
Signed-off-by: Kay Sievers <kay.sievers@vrfy.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
This seems as reasonable assumption and gets some SNI machines to work
which currently must rely on the cp0 counter as clocksource.
Signed-off-by: Thomas Bogendoerfer <tsbogend@alpha.franken.de>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Frank Rowand <frank.rowand@am.sony.com> reports:
> In linux-2.6.24-rc4 the Toshiba RBTX4927 hangs on boot.
>
> The cause is that plat_time_init() from arch/mips/tx4927/common/
> tx4927_setup.c does not override the __weak plat_time_init() from
> arch/mips/kernel/time.c. This is due to a compiler bug in gcc 4.1.1. The
> bug is reported to not exist in earlier versions of gcc, and to be fixed in
> 4.1.2. The problem is that the __weak plat_time_init() is empty and thus
> gets optimized out of existence (thus the linker is never given the option
> to replace the __weak function).
[ He meant the call to plat_time_init() from time_init() gets optimized away ]
> For more info on the gcc bug see
>
> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=27781
>
> The attached patch is one workaround. Another possible workaround
[ His patch adds -fno-unit-at-a-time for time.c ]
> would be to change the __weak plat_time_init() to be a non-empty
> function.
The __weak definition of plat_time_init was only ever meant to be a
migration helper to keep platforms that don't have a plat_time_init
compiling. A few greps says that all platforms now supply their own
plat_time_init() so the weak definition is no longer needed. So I
instead delete it.
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Sibyte SOCs only have 32-bit PCI. Due to the sparse use of the address
space only the first 1GB of memory is mapped at physical addresses
below 1GB. If a system has more than 1GB of memory 32-bit DMA will
not be able to reach all of it.
For now this patch is good enough to keep Sibyte users happy but it seems
eventually something like swiotlb will be needed for Sibyte.
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
In particular as-is it's not suited for multicore and mutiprocessors
systems where there is on guarantee that the counter are synchronized
or running from the same clock at all. This broke Sibyte and probably
others since the "[MIPS] Handle R4000/R4400 mfc0 from count register."
commit.
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
The R4000 and R4400 have an errata where if the cp0 count register is read
in the exact moment when it matches the compare register no interrupt will
be generated.
This bug may be triggered if the cp0 count register is being used as
clocksource and the compare interrupt as clockevent. So a simple
workaround is to avoid using the compare for both facilities on the
affected CPUs.
This is different from the workaround suggested in the old errata documents;
at some opportunity probably the official version should be implemented
and tested. Another thing to find out is which processor versions
exactly are affected. I only have errata documents upto R4400 V3.0
available so for the moment the code treats all R4000 and R4400 as broken.
This is potencially a problem for some machines that have no other decent
clocksource available; this workaround will cause them to fall back to
another clocksource, worst case the "jiffies" source.
This matters to any sort of device that is wired to one of the CPU
interrupt pins on an SMP system. Typically the scenario is most easily
triggered with the count/compare timer interrupt where the same interrupt
number and thus irq_desc is used on each processor.
CPU A CPU B
do_IRQ()
generic_handle_irq()
handle_level_irq()
spin_lock(desc_lock)
set IRQ_INPROGRESS
spin_unlock(desc_lock)
do_IRQ()
generic_handle_irq()
handle_level_irq()
spin_lock(desc_lock)
IRQ_INPROGRESS set => bail out
spin_lock(desc_lock)
clear IRQ_INPROGRESS
spin_unlock(desc_lock)
In case of the cp0 compare interrupt this means the interrupt will be
acked and not handled or re-armed on CPU b, so there won't be any timer
interrupt until the count register wraps around.
With kernels 2.6.20 ... 2.6.23 we usually were lucky that things were just
working right on VSMP because the count registers are synchronized on
bootup so it takes something that disables interrupts for a long time on
one processor to trigger this one.
For scenarios where an interrupt is multicasted or broadcasted over several
CPUs the existing code was safe and the fix will break it. There is no
way to know in the interrupt controller code because it is abstracted from
the platform code. I think we do not have such a setup currently, so this
should be ok.
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Adds a JR.HB after halting a TC, to ensure that the TC has really halted.
only modifies the TCSTATUS register when the TC is safely halted.
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
We have no guarantee by the generic time code that the timer is stopped
when the ->next_event method is called. Modifying the Timer Initial Count
register while the timer is enabled has UNPREDICTABLE effect according to
the BCM1250/BCM1125/BCM1125H User Manual. So stop the timer before
reprogramming.
This is a paranoia fix; no ill effects have been observed previously.
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
For the old minimum of a single tick a value of zero would be programmed
into the init value register which in the BCM1250/BCM1125/BCM1125H User
Manual in the Timer Special Cases section is documented to have
UNPREDICTABLE effect.
Observable sympthoms of this bug were hangs of several seconds on the
console during bootup and later if both dyntick and highres timer options
were activated.
In theory contiguous mode of the timers is also affected but in an act of
hopeless lack of realism I'll assume nobody will ever configure a KERNEL
for HZ > 500kHz but if so I leave that to evolution to sort out.
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Shadow register support would not possibly have worked on multicore
systems. The support code for it was also depending not on MIPS R2 but
VSMP or SMTC kernels even though it makes perfect sense with UP kernels.
SR sets are a scarce resource and the expected usage pattern is that
users actually hardcode the register set numbers in their code. So fix
the allocator by ditching it. Move the remaining CPU probe bits into
the generic CPU probe.
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
The Jazz machines have to use the PIT timer for dyntick and highresolution
kernels. This may break because currently just like i386 used to do MIPS
uses two separate spinlocks in the actual PIT code and the PC speaker
code. So switch to do it the same that x86 currently does PIT locking.
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
* Do not include unnecessary headers.
* Do not mention time.README.
* Do not mention mips_timer_ack.
* Make clocksource_mips static. It is now dedicated to c0_timer.
* Initialize clocksource_mips.read statically.
* Remove null_hpt_read.
* Remove an argument of plat_timer_setup. It is just a placeholder.
Signed-off-by: Atsushi Nemoto <anemo@mba.ocn.ne.jp>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
A comment on ptrace_getregs() states "Registers are sign extended to
fill the available space." but it is not true. Fix code to match the
comment. Also fix casts on each caller to get rid of some warnings.
Signed-off-by: Atsushi Nemoto <anemo@mba.ocn.ne.jp>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Mmap with MAP_FIXED was not validating the addr and len parameters. This
leads to the failure of GCC's gcc.c-torture/execute/loop-2[fg].c testcases
when using the o32 ABI on a 64 bit kernel.
These testcases try to mmap 65536 bytes at 0x7fff8000 and then access all
the memory. In 2.6.18 and 2.6.23.1 (and likely other versions as well)
the kernel maps the requested memory, but since half of it is above
0x80000000 a SIGBUS is generated when it is accessed.
This patch moves the len validation above the MAP_FIXED processing so that
it is always validated. It also adds validation to the addr parameter for
MAP_FIXED mappings.
Signed-off-by: David Daney <ddaney@avtrex.com>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Try increasingly longer time periods starting of at 0x10 cycles. This
should be fast on hardware and work nicely with emulators.
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
The expression "(long)(read_c0_count() - cnt)" can never be a negative
value on 64-bit kernel. Cast to "int" before comparison.
Signed-off-by: Atsushi Nemoto <anemo@mba.ocn.ne.jp>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Convert jmr3927_clock_event_device to more generic
txx9tmr_clock_event_device which supports one-shot mode. The
txx9tmr_clock_event_device can be used for TX49 too if the cp0 timer
interrupt was not available.
Convert jmr3927_hpt_read to txx9_clocksource driver which does not
depends jiffies anymore. The txx9_clocksource itself can be used for
TX49, but normally TX49 uses higher precision clocksource_mips.
Signed-off-by: Atsushi Nemoto <anemo@mba.ocn.ne.jp>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
clockevent_delta2ns() use the shift and mult value, so
clockevent_set_clock() should be called first.
Pointed out by Atsushi Nemoto.
Signed-off-by: Yoichi Yuasa <yoichi_yuasa@tripeaks.co.jp>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
set_next_event() and set_mode() are always called with interrupt disabled.
irqsave and irqrestore are not necessary for spinlock.
Pointed out by Atsushi Nemoto.
Signed-off-by: Yoichi Yuasa <yoichi_yuasa@tripeaks.co.jp>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Modify the SMTC initialization code to allow boot-time specification not
only of how many VPEs and TCs to use, but also how many TCs out of the
allowed pool are to be bound to VPE 0. The new boot option is "vpe0tcs=N",
where N is an integer. Using it in combination with the existing options
allows arbitrary assignments across the 2 VPEs of a 34K. e.g. "maxtcs=3
vpe0tcs=1" forces VPE0 to have 1 TC, while VPE1 has 2, and "maxtcs=4
vpe0tcs=3" forces VPE0 to have 3 TCs, while VPE1 gets 1. If no vpe0tcs
option is specified, the traditional algorithm of evenly dividing TCs
between available VPEs, with the odd "slop" going to VPE0, is retained.
The reason for doing this is to allow a finer balancing of TCs which can
handle I/O interrupts on Malta (those on VPE 0) and those which cannot.
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
And make use of it for Cobalt. A few others such as the Malta could make
use of it as well.
Signed-off-by: Yoichi Yuasa <yoichi_yuasa@tripeaks.co.jp>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Fix the various misspellings of "system", controller", "interrupt" and
"[un]necessary".
Signed-off-by: Robert P. J. Day <rpjday@mindspring.com>
Signed-off-by: Adrian Bunk <bunk@kernel.org>
* 'upstream' of git://ftp.linux-mips.org/pub/scm/upstream-linus:
[MIPS] Delete totally outdated Documentation/mips/time.README
[MIPS] Kill duplicated setup_irq() for cp0 timer
[MIPS] Sibyte: Finish conversion to modern time APIs.
[MIPS] time: Helpers to compute clocksource/event shift and mult values.
[MIPS] SMTC: Build fix.
[MIPS] time: Delete dead code.
[MIPS] MIPSsim: Strip defconfig file to the bones.
One of the easiest things to isolate is the pid printed in kernel log.
There was a patch, that made this for arch-independent code, this one makes
so for arch/xxx files.
It took some time to cross-compile it, but hopefully these are all the
printks in arch code.
Signed-off-by: Alexey Dobriyan <adobriyan@openvz.org>
Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Cc: <linux-arch@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
