Re: FPU exception, kernel panic
Sun, May 06, 2001 at 17:14:08, rakshe (Rohit Rakshe) wrote about Re: FPU exception,
kernel panic:
(I cannot even guarantree 50% this is the same problem, but...)
There were some reports in current@ about incorrect usage of i586_bzero()
which uses FPU for zero-filling. It generated random kernel panics mainly
on K6-2, also on P5-MMX and similars. You can try to disable such bzero,
but IMHO it is reasonable to upgrade to 4.3-RELEASE (RELENG_4_3) first.
I guessed I asked too much. Can someone please tell me just this:
How do I translate 48 bit address (segment descriptor:offset) to a
linear address which gdb can understand ?
I modified some TCP and UDP code in FreeBSD 4.1 and suddenly started
getting (almost repeatable) kernel panics. This is how it looks like
from remote gdb:
(kgdb) bt
#0 0xc0192e58 in panic (fmt=0xc035c076 npxintr from nowhere) at
../../kern/kern_shutdown.c:550
#1 0xc0303c6c in npx_intr (dummy=0x0) at ../../i386/isa/npx.c:721
#2 0xc02e1052 in Xfpu ()
#3 0xc019824d in softclock () at ../../kern/kern_timeout.c:131
.
.
.
(kgdb) list
126 } else {
127 c-c_flags =
128 (c-c_flags
~CALLOUT_PENDING);
129 }
130 splx(s);
131 c_func(c_arg);
132 s = splhigh();
133 steps = 0;
134 c = nextsoftcheck;
135 }
.
.
.
(kgdb)
#1 0xc0303c6c in npx_intr (dummy=0x0) at ../../i386/isa/npx.c:721
721 panic(npxintr from nowhere);
(kgdb) p npxproc
$5 = 0
(kgdb) p npx_exists
$6 = 1 '\001'
.
.
.
(kgdb) info all-registers
eax0x12 18
ecx0xc03a60a0 -1069916000
edx0x40 4194304
ebx0xcbc56540 -876255936
esp0xc0361354 0xc0361354
ebp0xc0361410 0xc0361410
esi0x40 4194304
edi0x40 4194304
eip0xc019824d 0xc019824d
eflags 0x346838
cs 0x8 8
ss 0x10 16
ds 0x400010 4194320
es 0x10 16
fs 0x0 0
gs 0x0 0
(kgdb) info float
status 0xc8f5: exceptions: INVALID DIVZ UNDERF LOS FPSTACK; flags: 1000;
top 1
control 0x6620: compute to 53 bits; round DOWN; mask: LOS;
warning: reserved bits on: 0x6000
last instruction: opcode 0x2825; pc 0x71a4:0x2824c316; operand
0xf800:0x8176640
regno tag msb lsb value
%st(7)valid 08179a802825 Denormal (0 as a double)
%st(6)valid 71a42824c316 Unnormal (NaN)
%st(5)valid 08176640282571a42824 Unnormal (NaN)
%st(4)valid c8910818f9380818f800 Unnormal (NaN)
%st(3)valid 282571a42824c316bfbf Unnormal (NaN)
%st(2)valid e82808176620089c3000 Unnormal (NaN)
%st(1)valid 003200010001 Denormal (0 as a double)
%st(0) = valid 00240818f800 Denormal (0 as a double)
So, this means that there was a FPU exception in kernel, right ?
1. The code which I added in kernel does not use any floats. So I am
wondering why this problem should happen at all.
2. pc register in FPU should give address of the instruction which caused
this exception, right ?
3. If yes, how do I translate this 48 bit address in a linear address
which gdb can understand ?
/netch
To Unsubscribe: send mail to [EMAIL PROTECTED]
with unsubscribe freebsd-hackers in the body of the message
Re: FPU exception, kernel panic
On Mon, May 07, 2001 at 08:32:06AM +0300, Valentin Nechayev wrote: Sun, May 06, 2001 at 17:14:08, rakshe (Rohit Rakshe) wrote about Re: FPU exception, kernel panic: (I cannot even guarantree 50% this is the same problem, but...) There were some reports in current@ about incorrect usage of i586_bzero() which uses FPU for zero-filling. It generated random kernel panics mainly on K6-2, also on P5-MMX and similars. You can try to disable such bzero, but IMHO it is reasonable to upgrade to 4.3-RELEASE (RELENG_4_3) first. This problem should not happen in 4.X as it is a side effect of the SMP-NG stuff in -current. David. To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-hackers in the body of the message
Re: FPU exception, kernel panic
Valentin Nechayev [EMAIL PROTECTED] writes: (I cannot even guarantree 50% this is the same problem, but...) There were some reports in current@ about incorrect usage of i586_bzero() which uses FPU for zero-filling. It generated random kernel panics mainly on K6-2, also on P5-MMX and similars. You can try to disable such bzero, but IMHO it is reasonable to upgrade to 4.3-RELEASE (RELENG_4_3) first. 1) this bug only occurs on -CURRENT systems that have interrupt preemption enabled (i.e. anything newer than february) 2) the symptoms are nowhere near what's described here anyway - you'd see the kernel stack getting smashed, not an FPU exception. DES -- Dag-Erling Smorgrav - [EMAIL PROTECTED] To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-hackers in the body of the message
Re: FPU exception, kernel panic
I guessed I asked too much. Can someone please tell me just this:
How do I translate 48 bit address (segment descriptor:offset) to a
linear address which gdb can understand ?
Thanks
- Rohit
On Sat, 5 May 2001, Rohit Rakshe wrote:
Hi !
I modified some TCP and UDP code in FreeBSD 4.1 and suddenly started
getting (almost repeatable) kernel panics. This is how it looks like
from remote gdb:
(kgdb) bt
#0 0xc0192e58 in panic (fmt=0xc035c076 npxintr from nowhere) at
../../kern/kern_shutdown.c:550
#1 0xc0303c6c in npx_intr (dummy=0x0) at ../../i386/isa/npx.c:721
#2 0xc02e1052 in Xfpu ()
#3 0xc019824d in softclock () at ../../kern/kern_timeout.c:131
.
.
.
(kgdb) list
126 } else {
127 c-c_flags =
128 (c-c_flags
~CALLOUT_PENDING);
129 }
130 splx(s);
131 c_func(c_arg);
132 s = splhigh();
133 steps = 0;
134 c = nextsoftcheck;
135 }
.
.
.
(kgdb)
#1 0xc0303c6c in npx_intr (dummy=0x0) at ../../i386/isa/npx.c:721
721 panic(npxintr from nowhere);
(kgdb) p npxproc
$5 = 0
(kgdb) p npx_exists
$6 = 1 '\001'
.
.
.
(kgdb) info all-registers
eax0x12 18
ecx0xc03a60a0 -1069916000
edx0x40 4194304
ebx0xcbc56540 -876255936
esp0xc0361354 0xc0361354
ebp0xc0361410 0xc0361410
esi0x40 4194304
edi0x40 4194304
eip0xc019824d 0xc019824d
eflags 0x346838
cs 0x8 8
ss 0x10 16
ds 0x400010 4194320
es 0x10 16
fs 0x0 0
gs 0x0 0
(kgdb) info float
status 0xc8f5: exceptions: INVALID DIVZ UNDERF LOS FPSTACK; flags: 1000;
top 1
control 0x6620: compute to 53 bits; round DOWN; mask: LOS;
warning: reserved bits on: 0x6000
last instruction: opcode 0x2825; pc 0x71a4:0x2824c316; operand
0xf800:0x8176640
regno tag msb lsb value
%st(7)valid 08179a802825 Denormal (0 as a double)
%st(6)valid 71a42824c316 Unnormal (NaN)
%st(5)valid 08176640282571a42824 Unnormal (NaN)
%st(4)valid c8910818f9380818f800 Unnormal (NaN)
%st(3)valid 282571a42824c316bfbf Unnormal (NaN)
%st(2)valid e82808176620089c3000 Unnormal (NaN)
%st(1)valid 003200010001 Denormal (0 as a double)
%st(0) = valid 00240818f800 Denormal (0 as a double)
So, this means that there was a FPU exception in kernel, right ?
1. The code which I added in kernel does not use any floats. So I am
wondering why this problem should happen at all.
2. pc register in FPU should give address of the instruction which caused
this exception, right ?
3. If yes, how do I translate this 48 bit address in a linear address
which gdb can understand ?
Thanks for help !
- Rohit
To Unsubscribe: send mail to [EMAIL PROTECTED]
with unsubscribe freebsd-hackers in the body of the message
To Unsubscribe: send mail to [EMAIL PROTECTED]
with unsubscribe freebsd-hackers in the body of the message
Re: FPU exception, kernel panic
Rohit Rakshe [EMAIL PROTECTED] writes: 1. The code which I added in kernel does not use any floats. So I am wondering why this problem should happen at all. The kernel uses the FPU to optimize certain bcopy- or bzero-like operations. 2. pc register in FPU should give address of the instruction which caused this exception, right ? Possibly. I'm not very familiar with the FPU. 3. If yes, how do I translate this 48 bit address in a linear address which gdb can understand ? You don't. FreeBSD uses a flat address space, you can disregard the segment descriptor. DES -- Dag-Erling Smorgrav - [EMAIL PROTECTED] To Unsubscribe: send mail to [EMAIL PROTECTED] with unsubscribe freebsd-hackers in the body of the message
FPU exception, kernel panic
Hi !
I modified some TCP and UDP code in FreeBSD 4.1 and suddenly started
getting (almost repeatable) kernel panics. This is how it looks like
from remote gdb:
(kgdb) bt
#0 0xc0192e58 in panic (fmt=0xc035c076 npxintr from nowhere) at
../../kern/kern_shutdown.c:550
#1 0xc0303c6c in npx_intr (dummy=0x0) at ../../i386/isa/npx.c:721
#2 0xc02e1052 in Xfpu ()
#3 0xc019824d in softclock () at ../../kern/kern_timeout.c:131
.
.
.
(kgdb) list
126 } else {
127 c-c_flags =
128 (c-c_flags
~CALLOUT_PENDING);
129 }
130 splx(s);
131 c_func(c_arg);
132 s = splhigh();
133 steps = 0;
134 c = nextsoftcheck;
135 }
.
.
.
(kgdb)
#1 0xc0303c6c in npx_intr (dummy=0x0) at ../../i386/isa/npx.c:721
721 panic(npxintr from nowhere);
(kgdb) p npxproc
$5 = 0
(kgdb) p npx_exists
$6 = 1 '\001'
.
.
.
(kgdb) info all-registers
eax0x12 18
ecx0xc03a60a0 -1069916000
edx0x40 4194304
ebx0xcbc56540 -876255936
esp0xc0361354 0xc0361354
ebp0xc0361410 0xc0361410
esi0x40 4194304
edi0x40 4194304
eip0xc019824d 0xc019824d
eflags 0x346838
cs 0x8 8
ss 0x10 16
ds 0x400010 4194320
es 0x10 16
fs 0x0 0
gs 0x0 0
(kgdb) info float
status 0xc8f5: exceptions: INVALID DIVZ UNDERF LOS FPSTACK; flags: 1000;
top 1
control 0x6620: compute to 53 bits; round DOWN; mask: LOS;
warning: reserved bits on: 0x6000
last instruction: opcode 0x2825; pc 0x71a4:0x2824c316; operand
0xf800:0x8176640
regno tag msb lsb value
%st(7)valid 08179a802825 Denormal (0 as a double)
%st(6)valid 71a42824c316 Unnormal (NaN)
%st(5)valid 08176640282571a42824 Unnormal (NaN)
%st(4)valid c8910818f9380818f800 Unnormal (NaN)
%st(3)valid 282571a42824c316bfbf Unnormal (NaN)
%st(2)valid e82808176620089c3000 Unnormal (NaN)
%st(1)valid 003200010001 Denormal (0 as a double)
%st(0) = valid 00240818f800 Denormal (0 as a double)
So, this means that there was a FPU exception in kernel, right ?
1. The code which I added in kernel does not use any floats. So I am
wondering why this problem should happen at all.
2. pc register in FPU should give address of the instruction which caused
this exception, right ?
3. If yes, how do I translate this 48 bit address in a linear address
which gdb can understand ?
Thanks for help !
- Rohit
To Unsubscribe: send mail to [EMAIL PROTECTED]
with unsubscribe freebsd-hackers in the body of the message
