Jack Ren and Eric Miao tracked down the following long standing
problem in the NOHZ code:
scheduler switch to idle task
enable interrupts
Window starts here
----> interrupt happens (does not set NEED_RESCHED)
irq_exit() stops the tick
----> interrupt happens (does set NEED_RESCHED)
return from schedule()
cpu_idle(): preempt_disable();
Window ends here
The interrupts can happen at any point inside the race window. The
first interrupt stops the tick, the second one causes the scheduler to
rerun and switch away from idle again and we end up with the tick
disabled.
The fact that it needs two interrupts where the first one does not set
NEED_RESCHED and the second one does made the bug obscure and extremly
hard to reproduce and analyse. Kudos to Jack and Eric.
Solution: Limit the NOHZ functionality to the idle loop to make sure
that we can not run into such a situation ever again.
cpu_idle()
{
preempt_disable();
while(1) {
tick_nohz_stop_sched_tick(1); <- tell NOHZ code that we
are in the idle loop
while (!need_resched())
halt();
tick_nohz_restart_sched_tick(); <- disables NOHZ mode
preempt_enable_no_resched();
schedule();
preempt_disable();
}
}
In hindsight we should have done this forever, but ...
/me grabs a large brown paperbag.
Debugged-by: Jack Ren <jack.ren@marvell.com>,
Debugged-by: eric miao <eric.y.miao@gmail.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
When we fully commit to returning back to kernel mode from
a trap, zero out the regs->magic value to prevent false
positives during stack backtraces.
Signed-off-by: David S. Miller <davem@davemloft.net>
The offset to the pt_regs area was wrong, so we weren't
looking at the right location for the magic cookie.
A trap frame is composed of a "struct sparc_stackf" then
a "struct pt_regs", the code was using "struct reg_window"
instead of "struct sparc_stackf".
Signed-off-by: David S. Miller <davem@davemloft.net>
Because of the silly way I set up the initial stack for
new kernel threads, there is a loop at the top of the
stack.
To fix this, properly add another stack frame that is copied
from the parent and terminate it in the child by setting
the frame pointer in that frame to zero.
Signed-off-by: David S. Miller <davem@davemloft.net>
When a cpu really is stuck in the kernel, it can be often
impossible to figure out which cpu is stuck where. The
worst case is when the stuck cpu has interrupts disabled.
Therefore, implement a global cpu state capture that uses
SMP message interrupts which are not disabled by the
normal IRQ enable/disable APIs of the kernel.
As long as we can get a sysrq 'y' to the kernel, we can
get a dump. Even if the console interrupt cpu is wedged,
we can trigger it from userspace using /proc/sysrq-trigger
The output is made compact so that this facility is more
useful on high cpu count systems, which is where this
facility will likely find itself the most useful :)
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch removes the CVS keywords that weren't updated for a long time
from comments.
Signed-off-by: Adrian Bunk <bunk@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Just like mmap, we need to validate address ranges regardless
of MAP_FIXED.
sparc{,64}_mmap_check()'s flag argument is unused, remove.
Based upon a report and preliminary patch by
Jan Lieskovsky <jlieskov@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
So, forever, we've had this ptrace_signal_deliver implementation
which tries to handle all of the nasties that can occur when the
debugger looks at a process about to take a signal. It's meant
to address all of these issues inside of the kernel so that the
debugger need not be mindful of such things.
Problem is, this doesn't work.
The idea was that we should do the syscall restart business first, so
that the debugger captures that state. Otherwise, if the debugger for
example saves the child's state, makes the child execute something
else, then restores the saved state, we won't handle the syscall
restart properly because we lose the "we're in a syscall" state.
The code here worked for most cases, but if the debugger actually
passes the signal through to the child unaltered, it's possible that
we would do a syscall restart when we shouldn't have.
In particular this breaks the case of debugging a process under a gdb
which is being debugged by yet another gdb. gdb uses sigsuspend
to wait for SIGCHLD of the inferior, but if gdb itself is being
debugged by a top-level gdb we get a ptrace_stop(). The top-level gdb
does a PTRACE_CONT with SIGCHLD to let the inferior gdb see the
signal. But ptrace_signal_deliver() assumed the debugger would cancel
out the signal and therefore did a syscall restart, because the return
error was ERESTARTNOHAND.
Fix this by simply making ptrace_signal_deliver() a nop, and providing
a way for the debugger to control system call restarting properly:
1) Report a "in syscall" software bit in regs->{tstate,psr}.
It is set early on in trap entry to a system call and is fully
visible to the debugger via ptrace() and regsets.
2) Test this bit right before doing a syscall restart. We have
to do a final recheck right after get_signal_to_deliver() in
case the debugger cleared the bit during ptrace_stop().
3) Clear the bit in trap return so we don't accidently try to set
that bit in the real register.
As a result we also get a ptrace_{is,clear}_syscall() for sparc32 just
like sparc64 has.
M68K has this same exact bug, and is now the only other user of the
ptrace_signal_deliver hook. It needs to be fixed in the same exact
way as sparc.
Signed-off-by: David S. Miller <davem@davemloft.net>
Forever we had a PTRACE_SUNOS_DETACH which was unconditionally
recognized, regardless of the personality of the process.
Unfortunately, this value is what ended up in the GLIBC sys/ptrace.h
header file on sparc as PTRACE_DETACH and PT_DETACH.
So continue to recognize this old value. Luckily, it doesn't conflict
with anything we actually care about.
Signed-off-by: David S. Miller <davem@davemloft.net>
We need to be more liberal about the alignment of the buffer given to
us by sigaltstack(). The user should not need to be mindful of all of
the alignment constraints we have for the stack frame.
This mirrors how we handle this situation in clone() as well.
Also, we align the stack even in non-SA_ONSTACK cases so that signals
due to bad stack alignment can be delivered properly. This makes such
errors easier to debug and recover from.
Finally, add the sanity check x86 has to make sure we won't overflow
the signal stack.
This fixes glibc testcases nptl/tst-cancel20.c and
nptl/tst-cancelx20.c
Signed-off-by: David S. Miller <davem@davemloft.net>
We clobber %i1 as well as %i0 for these system calls,
because they give two return values.
Therefore, on error, we have to restore %i1 properly
or else the restart explodes since it uses the wrong
arguments.
This fixes glibc's nptl/tst-eintr1.c testcase.
Signed-off-by: David S. Miller <davem@davemloft.net>
This reverts commit 2664ef44cf.
Ingo moved around where the softlockup dependency sits
so this change is no longer necessary.
Signed-off-by: David S. Miller <davem@davemloft.net>
The change I put into copy_thread() just papered over the real
problem.
When we are looking to see if we should do a syscall restart, when
deliverying a signal, we should only interpret the syscall return
value as an error if the carry condition code(s) are set.
Otherwise it's a success return.
Also, sigreturn paths should do a pt_regs_clear_trap_type().
It turns out that doing a syscall restart when returning from a fork()
does and should happen, from time to time. Even if copy_thread()
returns success, copy_process() can still unwind and signal
-ERESTARTNOINTR in the parent.
Signed-off-by: David S. Miller <davem@davemloft.net>
It just creates confusion, errors, and bugs.
For one thing, this can cause dup sysfs or procfs nodes to get
created:
[ 1.198015] proc_dir_entry '00.0' already registered
[ 1.198036] Call Trace:
[ 1.198052] [00000000004f2534] create_proc_entry+0x7c/0x98
[ 1.198092] [00000000005719e4] pci_proc_attach_device+0xa4/0xd4
[ 1.198126] [00000000007d991c] pci_proc_init+0x64/0x88
[ 1.198158] [00000000007c62a4] kernel_init+0x190/0x330
[ 1.198183] [0000000000426cf8] kernel_thread+0x38/0x48
[ 1.198210] [00000000006a0d90] rest_init+0x18/0x5c
Signed-off-by: David S. Miller <davem@davemloft.net>
Remove dulicated include file <asm/timer.h> in arch/sparc64/kernel/smp.c.
Signed-off-by: Huang Weiyi <hwy@cn.fujitsu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Current limitations:
1) On SMP single stepping has some fundamental issues,
shared with other sw single-step architectures such
as mips and arm.
2) On 32-bit sparc we don't support SMP kgdb yet. That
requires some reworking of the IPI mechanisms and
infrastructure on that platform.
Signed-off-by: David S. Miller <davem@davemloft.net>
entry.S was a hodge-podge of several totally unrelated
sets of assembler routines, ranging from FPU trap handlers
to hypervisor call functions.
Split it up into topic-sized pieces.
Signed-off-by: David S. Miller <davem@davemloft.net>
This fixes a regression added by
238468b2ac ("[SPARC64]: Use trap type
stored in pt_regs to handle syscall restart.")
Because we now encode the "returning from syscall" status in the
pt_regs area, we have to be mindful to zap it out in the child
of a fork.
During a parallel kernel build I saw an accidental -EINTR return
from vfork() in 'make' because of this bug.
Signed-off-by: David S. Miller <davem@davemloft.net>
If we use this from more than one place, it's better to
have helpers instead of twiddling magic constants all
over.
Add pt_regs_trap_type(), pt_regs_clear_trap_type(), and
pt_regs_is_syscall().
Use them in do_signal().
Signed-off-by: David S. Miller <davem@davemloft.net>
Back around the same time we were bootstrapping the first 32-bit sparc
Linux kernel with a SunOS userland, we made the signal frame match
that of SunOS.
By the time we even started putting together a native Linux userland
for 32-bit Sparc we realized this layout wasn't sufficient for Linux's
needs.
Therefore we changed the layout, yet kept support for the old style
signal frame layout in there. The detection mechanism is that we had
sys_sigaction() start passing in a negative signal number to indicate
"new style signal frames please".
Anyways, no binaries exist in the world that use the old stuff. In
fact, I bet Jakub Jelinek and myself are the only two people who ever
had such binaries to be honest.
So let's get rid of this stuff.
I added an assertion using WARN_ON_ONCE() that makes sure 32-bit
applications are passing in that negative signal number still.
Signed-off-by: David S. Miller <davem@davemloft.net>
I must have disabled this due to other bugs which were fixed over
time. And this is needed in order for child devices of "pmu"
to get proper resource values.
Signed-off-by: David S. Miller <davem@davemloft.net>
It's completely superfluous, CONFIG_COMPAT is sufficient.
What this used to be is an umbrella for enabling code shared
by all 32-bit compat binary support types. But with the
removal of SunOS and Solaris support, the only one left is
Linux 32-bit ELF.
Update defconfig.
Signed-off-by: David S. Miller <davem@davemloft.net>
Kernel bugzilla 10273
As reported by Jos van der Ende, ever since commit
5a606b72a4 ("[SPARC64]: Do not ACK an
INO if it is disabled or inprogress.") sun4u interrupts
can get stuck.
What this changset did was add the following conditional to
the various IRQ chip ->enable() handlers on sparc64:
if (unlikely(desc->status & (IRQ_DISABLED|IRQ_INPROGRESS)))
return;
which is correct, however it means that special care is needed
in the ->enable() method.
Specifically we must put the interrupt into IDLE state during
an enable, or else it might never be sent out again.
Setting the INO interrupt state to IDLE resets the state machine,
the interrupt input to the INO is retested by the hardware, and
if an interrupt is being signalled by the device, the INO
moves back into TRANSMIT state, and an interrupt vector is sent
to the cpu.
The two sun4v IRQ chip handlers were already doing this properly,
only sun4u got it wrong.
Signed-off-by: David S. Miller <davem@davemloft.net>
Otherwise all sorts of bad things can happen, including
spurious softlockup reports.
Other platforms have this same bug, in one form or
another, just don't see the issue because they
don't sleep as long as sparc64 can in NOHZ.
Thanks to some brilliant debugging by Peter Zijlstra.
Signed-off-by: David S. Miller <davem@davemloft.net>
Now that we have a magic cookie in the pt_regs, we can
properly detect trap frames in stack bactraces.
Signed-off-by: David S. Miller <davem@davemloft.net>
Now that we indicate the "restart system call" in the
trap type field of pt_regs->magic, we don't need to
set the %l6 boolean in all of the trap return paths.
And we therefore don't need to pass it to do_notify_resume().
Signed-off-by: David S. Miller <davem@davemloft.net>
Now that we can check the trap type directly, we don't need the
funny restart_syscall indication from the trap return paths.
Signed-off-by: David S. Miller <davem@davemloft.net>
This sets us up for several simplifications and facilities:
1) The magic cookie lets us identify trap frames more
accurately in stack backtraces.
2) The trap type lets us simplify all of the "are we in
a syscall" state management and checks.
3) We can now see if a task off the cpu is sleeping in
a system call or not. In fact, we can see what
trap it is sleeping in whatever the type. The utrace
guys will use this.
Based upon some discussions with Roland McGrath.
Signed-off-by: David S. Miller <davem@davemloft.net>
The following cleanups are now possible:
- arch/sparc64/kernel/entry.S:ret_sys_call no longer has to be global
- arch/sparc64/kernel/sparc64_ksyms.c:
remove no longer used prototypes
Signed-off-by: Adrian Bunk <bunk@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Currently there is only code to parse NUMA attributes on
sun4v/niagara systems, but later on we will add such parsing
for older systems.
Signed-off-by: David S. Miller <davem@davemloft.net>
We have to do it like this before we can move the PROM and MDESC device
tree code over to using lmb_alloc().
Signed-off-by: David S. Miller <davem@davemloft.net>
None of these files use any of the functionality promised by
asm/semaphore.h. It's possible that they rely on it dragging in some
unrelated header file, but I can't build all these files, so we'll have
fix any build failures as they come up.
Signed-off-by: Matthew Wilcox <willy@linux.intel.com>
Semaphores are no longer performance-critical, so a generic C
implementation is better for maintainability, debuggability and
extensibility. Thanks to Peter Zijlstra for fixing the lockdep
warning. Thanks to Harvey Harrison for pointing out that the
unlikely() was unnecessary.
Signed-off-by: Matthew Wilcox <willy@linux.intel.com>
Acked-by: Ingo Molnar <mingo@elte.hu>
1) ptrace should pass 'current' to task_user_regset_view()
2) When fetching general registers using a 64-bit view, and
the target is 32-bit, we have to convert.
3) Skip the whole register window get/set code block if
the user isn't asking to access anything in there.
Otherwise we have problems if the user doesn't have
an address space setup. Fetching ptrace register is
still valid at such a time, and ptrace does not try
to access the register window area of the regset.
Signed-off-by: David S. Miller <davem@davemloft.net>
