linux-stable.git/arch/arc, branch v3.16.35

ARC: dw2 unwind: Ignore CIE version !=1 gracefully instead of bailing

2016-01-25T10:43:41+00:00

commit 323f41f9e7d0cb5b1d1586aded6682855f1e646d upstream.

ARC dwarf unwinder only supports CIE version == 1
The boot time dwarf sanitizer (part of binary lookup table constructor)
would simply bail if it saw CIE version == 3, rendering unwinder with a
NULL lookup table.

It seems libgcc linked with kernel does have such entries.

With fallback linear search removed, and a NULL binary lookup table,
unwinder fails to generate any stack trace.

So allow graceful ignoring of unsupported CIE entries.

This problem was initially seen in Alexey's setup (and not mine) as he
was using buildroot built toolchain (libgcc) which doesn't get built with
CFLAGS_FOR_TARGET="-gdwarf-2 which is my default

Fixes STAR 9000985048: "kernel unwinder broken with stock tools"

Fixes: 2e22502c080f ARC: dw2 unwind: Remove falllback linear search thru FDE entries
Reported-by Alexey Brodkin 
Signed-off-by: Vineet Gupta 
Signed-off-by: Luis Henriques

ARC: dw2 unwind: Reinstante unwinding out of modules

2016-01-25T10:43:40+00:00

commit bc79c9a7216562a2035d2f64f73626613c1300d0 upstream.

The fix which removed linear searching of dwarf (because binary lookup
data always exists) missed out on the fact that modules don't get the
binary lookup tables info. This caused unwinding out of modules to stop
working.

So add binary lookup header setup (equivalent of eh_frame_hdr setup) to
modules as well.

While at it, confine the header setup to within unwinder code,
reducing one API exposed out of unwinder code.

Fixes: 2e22502c080f ARC: dw2 unwind: Remove falllback linear search thru FDE entries
Signed-off-by: Vineet Gupta 
Signed-off-by: Luis Henriques

ARC: dw2 unwind: Remove falllback linear search thru FDE entries

2015-12-18T16:06:32+00:00

commit 2e22502c080f27afeab5e6f11e618fb7bc7aea53 upstream.

Fixes STAR 9000953410: "perf callgraph profiling causing RCU stalls"

| perf record -g -c 15000 -e cycles /sbin/hackbench
|
| INFO: rcu_preempt self-detected stall on CPU
| 1: (1 GPs behind) idle=609/140000000000002/0 softirq=2914/2915 fqs=603
| Task dump for CPU 1:

in-kernel dwarf unwinder has a fast binary lookup and a fallback linear
search (which iterates thru each of ~11K entries) thus takes 2 orders of
magnitude longer (~3 million cycles vs. 2000). Routines written in hand
assembler lack dwarf info (as we don't support assembler CFI pseudo-ops
yet) fail the unwinder binary lookup, hit linear search, failing
nevertheless in the end.

However the linear search is pointless as binary lookup tables are created
from it in first place. It is impossible to have binary lookup fail while
succeed the linear search. It is pure waste of cycles thus removed by
this patch.

This manifested as RCU stalls / NMI watchdog splat when running
hackbench under perf with callgraph profiling. The triggering condition
was perf counter overflowing in routine lacking dwarf info (like memset)
leading to patheic 3 million cycle unwinder slow path and by the time it
returned new interrupts were already pending (Timer, IPI) and taken
rightaway. The original memset didn't make forward progress, system kept
accruing more interrupts and more unwinder delayes in a vicious feedback
loop, ultimately triggering the NMI diagnostic.

Signed-off-by: Vineet Gupta 
Signed-off-by: Luis Henriques

ARC: make sure instruction_pointer() returns unsigned value

2015-08-10T08:54:19+00:00

commit f51e2f1911122879eefefa4c592dea8bf794b39c upstream.

Currently instruction_pointer() returns pt_regs->ret and so return value
is of type "long", which implicitly stands for "signed long".

While that's perfectly fine when dealing with 32-bit values if return
value of instruction_pointer() gets assigned to 64-bit variable sign
extension may happen.

And at least in one real use-case it happens already.
In perf_prepare_sample() return value of perf_instruction_pointer()
(which is an alias to instruction_pointer() in case of ARC) is assigned
to (struct perf_sample_data)->ip (which type is "u64").

And what we see if instuction pointer points to user-space application
that in case of ARC lays below 0x8000_0000 "ip" gets set properly with
leading 32 zeros. But if instruction pointer points to kernel address
space that starts from 0x8000_0000 then "ip" is set with 32 leadig
"f"-s. I.e. id instruction_pointer() returns 0x8100_0000, "ip" will be
assigned with 0xffff_ffff__8100_0000. Which is obviously wrong.

In particular that issuse broke output of perf, because perf was unable
to associate addresses like 0xffff_ffff__8100_0000 with anything from
/proc/kallsyms.

That's what we used to see:
 ----------->8----------
  6.27%  ls       [unknown]                [k] 0xffffffff8046c5cc
  2.96%  ls       libuClibc-0.9.34-git.so  [.] memcpy
  2.25%  ls       libuClibc-0.9.34-git.so  [.] memset
  1.66%  ls       [unknown]                [k] 0xffffffff80666536
  1.54%  ls       libuClibc-0.9.34-git.so  [.] 0x000224d6
  1.18%  ls       libuClibc-0.9.34-git.so  [.] 0x00022472
 ----------->8----------

With that change perf output looks much better now:
 ----------->8----------
  8.21%  ls       [kernel.kallsyms]        [k] memset
  3.52%  ls       libuClibc-0.9.34-git.so  [.] memcpy
  2.11%  ls       libuClibc-0.9.34-git.so  [.] malloc
  1.88%  ls       libuClibc-0.9.34-git.so  [.] memset
  1.64%  ls       [kernel.kallsyms]        [k] _raw_spin_unlock_irqrestore
  1.41%  ls       [kernel.kallsyms]        [k] __d_lookup_rcu
 ----------->8----------

Signed-off-by: Alexey Brodkin 
Cc: arc-linux-dev@synopsys.com
Cc: linux-kernel@vger.kernel.org
Signed-off-by: Vineet Gupta 
Signed-off-by: Luis Henriques

ARC: Override toplevel default -O2 with -O3

2015-08-10T08:54:10+00:00

commit 97709069214eb75312c14946803b9da4d3814203 upstream.

ARC kernels have historically been built with -O3, despite top level
Makefile defaulting to -O2. This was facilitated by implicitly ordering
of arch makefile include AFTER top level assigned -O2.

An upstream fix to top level a1c48bb160f ("Makefile: Fix unrecognized
cross-compiler command line options") changed the ordering, making ARC
-O3 defunct.

Fix that by NOT relying on any ordering whatsoever and use the proper
arch override facility now present in kbuild (ARCH_*FLAGS)

Depends-on: ("kbuild: Allow arch Makefiles to override {cpp,ld,c}flags")
Suggested-by: Michal Marek 
Cc: Geert Uytterhoeven 
Signed-off-by: Vineet Gupta 
Signed-off-by: Luis Henriques

ARC: add smp barriers around atomics per Documentation/atomic_ops.txt

2015-07-15T09:00:53+00:00

commit 2576c28e3f623ed401db7e6197241865328620ef upstream.

 - arch_spin_lock/unlock were lacking the ACQUIRE/RELEASE barriers
   Since ARCv2 only provides load/load, store/store and all/all, we need
   the full barrier

 - LLOCK/SCOND based atomics, bitops, cmpxchg, which return modified
   values were lacking the explicit smp barriers.

 - Non LLOCK/SCOND varaints don't need the explicit barriers since that
   is implicity provided by the spin locks used to implement the
   critical section (the spin lock barriers in turn are also fixed in
   this commit as explained above

Cc: Paul E. McKenney 
Acked-by: Peter Zijlstra (Intel) 
Signed-off-by: Vineet Gupta 
Signed-off-by: Luis Henriques

locking,arch,arc: Fold atomic_ops

2015-07-15T09:00:53+00:00

commit f7d11e93ee97a37da1947b7c4e1794705a6f360c upstream.

Many of the atomic op implementations are the same except for one
instruction; fold the lot into a few CPP macros and reduce LoC.

This also prepares for easy addition of new ops.

Signed-off-by: Peter Zijlstra 
Cc: Linus Torvalds 
Cc: Paul E. McKenney 
Cc: Vineet Gupta 
Link: http://lkml.kernel.org/r/20140508135851.886055622@infradead.org
Signed-off-by: Ingo Molnar 
[ luis: 3.16 prereq for:
  2576c28e3f62 "ARC: add smp barriers around atomics per Documentation/atomic_ops.txt"
Signed-off-by: Luis Henriques

ARC: add compiler barrier to LLSC based cmpxchg

2015-07-15T09:00:52+00:00

commit d57f727264f1425a94689bafc7e99e502cb135b5 upstream.

When auditing cmpxchg call sites, Chuck noted that gcc was optimizing
away some of the desired LDs.

|	do {
|		new = old = *ipi_data_ptr;
|		new |= 1U << msg;
|	} while (cmpxchg(ipi_data_ptr, old, new) != old);

was generating to below

| 8015cef8:	ld         r2,[r4,0]  <-- First LD
| 8015cefc:	bset       r1,r2,r1
|
| 8015cf00:	llock      r3,[r4]  <-- atomic op
| 8015cf04:	brne       r3,r2,8015cf10
| 8015cf08:	scond      r1,[r4]
| 8015cf0c:	bnz        8015cf00
|
| 8015cf10:	brne       r3,r2,8015cf00  <-- Branch doesn't go to orig LD

Although this was fixed by adding a ACCESS_ONCE in this call site, it
seems safer (for now at least) to add compiler barrier to LLSC based
cmpxchg

Reported-by: Chuck Jordan 
Acked-by: Peter Zijlstra (Intel) 
Signed-off-by: Vineet Gupta 
Signed-off-by: Luis Henriques

ARC: signal handling robustify

2015-05-12T08:36:34+00:00

commit e4140819dadc3624accac8294881bca8a3cba4ed upstream.

A malicious signal handler / restorer can DOS the system by fudging the
user regs saved on stack, causing weird things such as sigreturn returning
to user mode PC but cpu state still being kernel mode....

Ensure that in sigreturn path status32 always has U bit; any other bogosity
(gargbage PC etc) will be taken care of by normal user mode exceptions mechanisms.

Reproducer signal handler:

    void handle_sig(int signo, siginfo_t *info, void *context)
    {
	ucontext_t *uc = context;
	struct user_regs_struct *regs = &(uc->uc_mcontext.regs);

	regs->scratch.status32 = 0;
    }

Before the fix, kernel would go off to weeds like below:

    --------->8-----------
    [ARCLinux]$ ./signal-test
    Path: /signal-test
    CPU: 0 PID: 61 Comm: signal-test Not tainted 4.0.0-rc5+ #65
    task: 8f177880 ti: 5ffe6000 task.ti: 8f15c000

    [ECR   ]: 0x00220200 => Invalid Write @ 0x00000010 by insn @ 0x00010698
    [EFA   ]: 0x00000010
    [BLINK ]: 0x2007c1ee
    [ERET  ]: 0x10698
    [STAT32]: 0x00000000 :                                   <--------
    BTA: 0x00010680	 SP: 0x5ffe7e48	 FP: 0x00000000
    LPS: 0x20003c6c	LPE: 0x20003c70	LPC: 0x00000000
    ...
    --------->8-----------

Reported-by: Alexey Brodkin 
Signed-off-by: Vineet Gupta 
[ luis: backported to 3.16: used Vineet's backport to 3.14 ]
Signed-off-by: Luis Henriques

ARC: SA_SIGINFO ucontext regs off-by-one

2015-04-10T09:04:07+00:00

commit 6914e1e3f63caa829431160f0f7093292daef2d5 upstream.

The regfile provided to SA_SIGINFO signal handler as ucontext was off by
one due to pt_regs gutter cleanups in 2013.

Before handling signal, user pt_regs are copied onto user_regs_struct and copied
back later. Both structs are binary compatible. This was all fine until
commit 2fa919045b72 (ARC: pt_regs update #2) which removed the empty stack slot
at top of pt_regs (corresponding to first pad) and made the corresponding
fixup in struct user_regs_struct (the pad in there was moved out of
@scratch - not removed altogether as it is part of ptrace ABI)

 struct user_regs_struct {
+       long pad;
        struct {
-               long pad;
                long bta, lp_start, lp_end,....
        } scratch;
 ...
 }

This meant that now user_regs_struct was off by 1 reg w.r.t pt_regs and
signal code needs to user_regs_struct.scratch to reflect it as pt_regs,
which is what this commit does.

This problem was hidden for 2 years, because both save/restore, despite
using wrong location, were using the same location. Only an interim
inspection (reproducer below) exposed the issue.

     void handle_segv(int signo, siginfo_t *info, void *context)
     {
 	ucontext_t *uc = context;
	struct user_regs_struct *regs = &(uc->uc_mcontext.regs);

	printf("regs %x %x\n",               <=== prints 7 8 (vs. 8 9)
               regs->scratch.r8, regs->scratch.r9);
     }

     int main()
     {
	struct sigaction sa;

	sa.sa_sigaction = handle_segv;
	sa.sa_flags = SA_SIGINFO;
	sigemptyset(&sa.sa_mask);
	sigaction(SIGSEGV, &sa, NULL);

	asm volatile(
	"mov	r7, 7	\n"
	"mov	r8, 8	\n"
	"mov	r9, 9	\n"
	"mov	r10, 10	\n"
	:::"r7","r8","r9","r10");

	*((unsigned int*)0x10) = 0;
     }

Fixes: 2fa919045b72ec892e "ARC: pt_regs update #2: Remove unused gutter at start of pt_regs"
Signed-off-by: Vineet Gupta 
Signed-off-by: Luis Henriques