[valgrind] [Bug 402351] mips64 libvexmultiarch_test fails on s390x

2019-03-14 Thread Petar Jovanovic 
https://bugs.kde.org/show_bug.cgi?id=402351

Petar Jovanovic  changed:

   What|Removed |Added

 Status|ASSIGNED|RESOLVED
 Resolution|--- |FIXED

--- Comment #7 from Petar Jovanovic  ---
Fixed with
https://sourceware.org/git/?p=valgrind.git;a=commit;h=cd20c8ca5815f0abf9d1bdb47888753bdb5c5738

Sorry for a delay in taking a look at this.

-- 
You are receiving this mail because:
You are watching all bug changes.

[valgrind] [Bug 402351] mips64 libvexmultiarch_test fails on s390x

2019-03-14 Thread Petar Jovanovic 
https://bugs.kde.org/show_bug.cgi?id=402351

Petar Jovanovic  changed:

   What|Removed |Added

   Assignee|jsew...@acm.org |mips3...@gmail.com
 Status|CONFIRMED   |ASSIGNED

--- Comment #6 from Petar Jovanovic  ---
This problem is triggered when VEX is configured for MIPS64 w/ 32-bit FPUs.
Since this can happen only in artificial cases, I will go with adjusting the
test itself.

-- 
You are receiving this mail because:
You are watching all bug changes.

[valgrind] [Bug 402351] mips64 libvexmultiarch_test fails on s390x

2019-03-10 Thread Julian Seward 
https://bugs.kde.org/show_bug.cgi?id=402351

--- Comment #5 from Julian Seward  ---
> t9 = ReinterpI64asF64(GET:I64(664))
> vex: priv/host_mips_isel.c:4915 (iselInt64Expr): Assertion `!env->mode64' 
> failed.

Either (1) there is some 32-vs-64-bit guest-vs-host confusion here, or (2)
there is no confusion, and instead this is a mips32 instruction selector
problem.

For (2) the test requires iselInt64Expr to select 32-bit instructions for
GET:I64(664), which should be two 32-bit loads into two registers.

-- 
You are receiving this mail because:
You are watching all bug changes.

[valgrind] [Bug 402351] mips64 libvexmultiarch_test fails on s390x

2019-01-10 Thread Mark Wielaard 
https://bugs.kde.org/show_bug.cgi?id=402351

--- Comment #4 from Mark Wielaard  ---
For ppc64be this is the output:

 doing translating guest PPC64(1030) BigEndian 64bits to host MIPS64(1033)
BigEndian 64bits

 Front end 

0x103C34C8:-- IMark(0x103C34C8, 0, 0) --
  PUT(1296) = 0x103C34C8:I64
  PUT(1296) = GET:I64(1296); exit-NoDecode

GuestBytes 103C34C8 0   


 After pre-instr IR optimisation


IRSB {
   t0:I64   

   -- IMark(0x103C34C8, 0, 0) --
   PUT(1296) = 0x103C34C8:I64; exit-NoDecode
}


 After instrumentation 

IRSB {
   t0:I64   

   -- IMark(0x103C34C8, 0, 0) --
   PUT(1296) = 0x103C34C8:I64; exit-NoDecode
}


 After post-instr IR optimisation


IRSB {
   t0:I64   

   -- IMark(0x103C34C8, 0, 0) --
   PUT(1296) = 0x103C34C8:I64; exit-NoDecode
}


  After tree-building 

IRSB {
   t0:I64   

   -- IMark(0x103C34C8, 0, 0) --
   PUT(1296) = 0x103C34C8:I64; exit-NoDecode
}


 Instruction selection 
(evCheck) lw $9, 8($23); addiu $9, $9, -1; sw $9, 8($23); bgez $t9, nofail;
jalr *0($23); nofail:

-- -- IMark(0x103C34C8, 0, 0) --

-- PUT(1296) = 0x103C34C8:I64; exit-NoDecode
li %vR1,0x103C34C8
(xAssisted) if (guest_COND.AL) { sw %vR1, 1296($23); move $9,
$IRJumpKind_to_TRCVAL(6664); move $9, $disp_assisted; jalr $9; nop; }

  0   (evCheck) lw $9, 8($23); addiu $9, $9, -1; sw $9, 8($23); bgez $t9,
nofail; jalr *0($23); nofail:
  1   li %vR1,0x103C34C8
  2   (xAssisted) if (guest_COND.AL) { sw %vR1, 1296($23); move $9,
$IRJumpKind_to_TRCVAL(6664); move $9, $disp_assisted; jalr $9; nop; }


 Register-allocated code 

  0   (evCheck) lw $9, 8($23); addiu $9, $9, -1; sw $9, 8($23); bgez $t9,
nofail; jalr *0($23); nofail:
  1   li $24,0x103C34C8
  2   (xAssisted) if (guest_COND.AL) { sw $24, 1296($23); move $9,
$IRJumpKind_to_TRCVAL(6664); move $9, $disp_assisted; jalr $9; nop; }


 Assembly 

(evCheck) lw $9, 8($23); addiu $9, $9, -1; sw $9, 8($23); bgez $t9, nofail;
jalr *0($23); nofail:
8E E9 00 08 25 29 FF FF AE E9 00 08 05 21 00 03 DE E9 00 00 01 20 F8 09 00 00
00 00 

li $24,0x103C34C8
3C 18 10 3C 37 18 34 C8 

(xAssisted) if (guest_COND.AL) { sw $24, 1296($23); move $9,
$IRJumpKind_to_TRCVAL(6664); move $9, $disp_assisted; jalr $9; nop; }
FE F8 05 10 3C 17 00 00 36 F7 00 00 00 17 BC 38 36 F7 00 00 00 17 BC 38 36 F7
00 45 3C 09 00 00 35 29 00 00 00 09 4C 38 35 29 10 3C 00 09 4C 38 35 29 34 98
01 20 F8 09 00 00 00 00 

VexExpansionRatio 0 96   960 :10

For s390x it is:

 doing translating guest S390X(1031) BigEndian 64bits to host MIPS64(1033)
BigEndian 64bits

 Front end 

ldgr %f2,%r11
  -- IMark(0x1000920, 4, 0) --
  PUT(96) = ReinterpI64asF64(GET:I64(664))
  PUT(720) = 0x1000924:I64

ldgr %f0,%r15
  -- IMark(0x1000924, 4, 0) --
  PUT(64) = ReinterpI64asF64(GET:I64(696))
  PUT(720) = 0x1000928:I64

lay  %r15,-168(%r15)
  -- IMark(0x1000928, 6, 0) --
  t1 = 0xFF58:I64
  t0 = Add64(Add64(t1,GET:I64(696)),0x0:I64)
  PUT(696) = t0
  PUT(720) = 0x100092E:I64

lgr  %r11,%r15
  -- IMark(0x100092E, 4, 0) --
  PUT(664) = GET:I64(696)
  PUT(720) = 0x1000932:I64

stg  %r2,160(%r11)
  -- IMark(0x1000932, 6, 0) --
  t3 = 0xA0:I64
  t2 = Add64(Add64(t3,GET:I64(664)),0x0:I64)
  STbe(t2) = GET:I64(592)
  PUT(720) = 0x1000938:I64

lg   %r1,160(%r11)
  -- IMark(0x1000938, 6, 0) --
  t5 = 0xA0:I64
  t4 = Add64(Add64(t5,GET:I64(664)),0x0:I64)
  PUT(584) = LDbe:I64(t4)
  PUT(720) = 0x100093E:I64

mvhi 0(%r1),1031
  -- IMark(0x100093E, 6, 0) --
  t6 = Add64(0x0:I64,GET:I64(584))
  STbe(t6) = 0x407:I32
  PUT(720) = 0x1000944:I64

nopr
  -- IMark(0x1000944, 2, 0) --
  PUT(720) = 0x1000946:I64

lgdr %r11,%f2
  -- IMark(0x1000946, 4, 0) --
  PUT(664) = ReinterpF64asI64(GET:F64(96))
  PUT(720) = 0x100094A:I64

lgdr %r15,%f0
  -- IMark(0x100094A, 4, 0) --
  PUT(696) = ReinterpF64asI64(GET:F64(64))
  PUT(720) = 0x100094E:I64

br   %r14
  -- IMark(0x100094E, 2, 0) --

[valgrind] [Bug 402351] mips64 libvexmultiarch_test fails on s390x

2019-01-10 Thread Mark Wielaard 
https://bugs.kde.org/show_bug.cgi?id=402351

--- Comment #3 from Mark Wielaard  ---
(In reply to Petar Jovanovic from comment #2)
> I will take a look at it. Is there a way to reproduce it myself?
> I am not seeing any s390 boards on GCC farm unfortunately.

I had assumed it would also trigger on ppc64be (s390x is also big endian), but
it seems to work fine there. Which might indicate that my patch/logic is wrong.

I'll see if I can get you access to the s390x server that I replicated this on.

-- 
You are receiving this mail because:
You are watching all bug changes.

[valgrind] [Bug 402351] mips64 libvexmultiarch_test fails on s390x

2019-01-10 Thread Petar Jovanovic 
https://bugs.kde.org/show_bug.cgi?id=402351

--- Comment #2 from Petar Jovanovic  ---
I will take a look at it. Is there a way to reproduce it myself?
I am not seeing any s390 boards on GCC farm unfortunately.

-- 
You are receiving this mail because:
You are watching all bug changes.

[valgrind] [Bug 402351] mips64 libvexmultiarch_test fails on s390x

2019-01-06 Thread Mark Wielaard 
https://bugs.kde.org/show_bug.cgi?id=402351

Mark Wielaard  changed:

   What|Removed |Added

 Status|REPORTED|CONFIRMED
 CC||mips3...@gmail.com
 Ever confirmed|0   |1

--- Comment #1 from Mark Wielaard  ---
Petar could you take a peek at the above issue and fix/workaround?

-- 
You are receiving this mail because:
You are watching all bug changes.