Skip to contentSkip to conversation

BUG on 2.6.20-rc1 when using gdb

PATCH In review Andrew J. Barr 6 participants 12 messages Dec 18, 2006 – Dec 20, 2006 ptrace support

lore
Share & tools
lore
#0 lore raw

When I was using gdb to debug xchat-gnome, I got a kernel BUG and stack trace as the program was running (e.g. I had typed 'run' in gdb):

WARNING at kernel/softirq.c:137 local_bh_enable()

 [<c0103cd6>] dump_trace+0x68/0x1d9
 [<c0103e5f>] show_trace_log_lvl+0x18/0x2c
 [<c01044d3>] show_trace+0xf/0x11
 [<c010455e>] dump_stack+0x12/0x14
 [<c011cc7d>] local_bh_enable+0x44/0x94
 [<c02871b9>] unix_release_sock+0x6e/0x1fe
 [<c02887eb>] unix_stream_connect+0x3b4/0x3cf
 [<c0232dee>] sys_connect+0x82/0xad
 [<c0233641>] sys_socketcall+0xac/0x261
 [<c0102d38>] syscall_call+0x7/0xb
 [<b7f70822>] 0xb7f70822
 =======================
------------[ cut here ]------------
kernel BUG at fs/buffer.c:1235!
invalid opcode: 0000 [#1]
PREEMPT
Modules linked in: binfmt_misc rfcomm l2cap i915 drm bluetooth nfs nfsd

exportfs lockd nfs_acl sunrpc nvram uinput ipv6 ppdev lp button ac battery dm_crypt dm_snapshot dm_mirror dm_mod fuse cpufreq_conservative cpufreq_ondemand cpufreq_performance cpufreq_powersave speedstep_centrino freq_table ibm_acpi loop snd_intel8x0m snd_pcm_oss snd_mixer_oss snd_intel8x0 snd_ac97_codec pcmcia ac97_bus irtty_sir sir_dev ipw2200 snd_pcm snd_timer irda ieee80211 ieee80211_crypt crc_ccitt rtc parport_pc parport 8250_pnp snd soundcore 8250_pci 8250 serial_core firmware_class i2c_i801 yenta_socket rsrc_nonstatic pcmcia_core snd_page_alloc i2c_core intel_agp agpgart evdev tsdev joydev ext3 jbd mbcache ide_cd cdrom ide_disk ide_generic e100 mii generic piix ide_core ehci_hcd uhci_hcd usbcore

CPU:    0
EIP:    0060:[<c0179266>]    Not tainted VLI
EFLAGS: 00010046   (2.6.20-rc1 #1)

EIP is at __find_get_block+0x1c/0x16f
eax: 00000086 ebx: 00000000 ecx: 00000000 edx: 0088a800
esi: 0088a800 edi: 00000000 ebp: dfffd040 esp: cad2dd30
ds: 007b es: 007b ss: 0068
Process xchat-gnome (pid: 4322, ti=cad2c000 task=d0cd3ab0 task.ti=cad2c000)

Stack: cad2dd58 c02caa0b 00000002 0000000e 0000000b 00000001 e8836580
0088a800
       00000000 00000000 e8836610 00000000 c01793dc 00001000 c03ab3e0
f3cadd80
       00000086 c90d41b0 0088a800 00000000 dfffd040 00008000 00000000
00000002
Call Trace:
 [<c01793dc>] __getblk+0x23/0x268
 [<f040d4c6>] ext3_getblk+0x10b/0x244 [ext3]
 [<f040e364>] ext3_bread+0x19/0x70 [ext3]
 [<f04106f3>] dx_probe+0x43/0x2c9 [ext3]
 [<f04119b3>] ext3_htree_fill_tree+0x99/0x1ba [ext3]
 [<f040ab77>] ext3_readdir+0x1d4/0x5ed [ext3]
 [<c0167b29>] vfs_readdir+0x63/0x8d
 [<c0167bb6>] sys_getdents64+0x63/0xa5
 [<c0102d38>] syscall_call+0x7/0xb
 [<b7f70822>] 0xb7f70822
 =======================
Code: 8b 40 08 a8 08 74 05 e8 02 2f 11 00 5b 5e c3 55 89 c5 57 89 cf 56

89 d6 53 83 ec 20 9c 58 90 8d b4 26 00 00 00 00 f6 c4 02 75 04 <0f> 0b eb fe 89 e0 25 00 e0 ff ff ff 40 14 31 c9 8b 1c 8d a0 74
EIP: [<c0179266>] __find_get_block+0x1c/0x16f SS:ESP 0068:cad2dd30

This happens on 2.6.20-rc1 but not 2.6.19.

Andrew Barr
[email protected]

#1 lore raw
Quoted 60 lines · On Sun, 17 Dec 2006 20:55:18 -0500 "Andrew J. Barr"…

When I was using gdb to debug xchat-gnome, I got a kernel BUG and stack trace as the program was running (e.g. I had typed 'run' in gdb):

WARNING at kernel/softirq.c:137 local_bh_enable()

 [<c0103cd6>] dump_trace+0x68/0x1d9
 [<c0103e5f>] show_trace_log_lvl+0x18/0x2c
 [<c01044d3>] show_trace+0xf/0x11
 [<c010455e>] dump_stack+0x12/0x14
 [<c011cc7d>] local_bh_enable+0x44/0x94
 [<c02871b9>] unix_release_sock+0x6e/0x1fe
 [<c02887eb>] unix_stream_connect+0x3b4/0x3cf
 [<c0232dee>] sys_connect+0x82/0xad
 [<c0233641>] sys_socketcall+0xac/0x261
 [<c0102d38>] syscall_call+0x7/0xb
 [<b7f70822>] 0xb7f70822
 =======================
------------[ cut here ]------------
kernel BUG at fs/buffer.c:1235!
invalid opcode: 0000 [#1]
PREEMPT
Modules linked in: binfmt_misc rfcomm l2cap i915 drm bluetooth nfs nfsd

exportfs lockd nfs_acl sunrpc nvram uinput ipv6 ppdev lp button ac battery dm_crypt dm_snapshot dm_mirror dm_mod fuse cpufreq_conservative cpufreq_ondemand cpufreq_performance cpufreq_powersave speedstep_centrino freq_table ibm_acpi loop snd_intel8x0m snd_pcm_oss snd_mixer_oss snd_intel8x0 snd_ac97_codec pcmcia ac97_bus irtty_sir sir_dev ipw2200 snd_pcm snd_timer irda ieee80211 ieee80211_crypt crc_ccitt rtc parport_pc parport 8250_pnp snd soundcore 8250_pci 8250 serial_core firmware_class i2c_i801 yenta_socket rsrc_nonstatic pcmcia_core snd_page_alloc i2c_core intel_agp agpgart evdev tsdev joydev ext3 jbd mbcache ide_cd cdrom ide_disk ide_generic e100 mii generic piix ide_core ehci_hcd uhci_hcd usbcore

CPU:    0
EIP:    0060:[<c0179266>]    Not tainted VLI
EFLAGS: 00010046   (2.6.20-rc1 #1)

EIP is at __find_get_block+0x1c/0x16f
eax: 00000086 ebx: 00000000 ecx: 00000000 edx: 0088a800
esi: 0088a800 edi: 00000000 ebp: dfffd040 esp: cad2dd30
ds: 007b es: 007b ss: 0068
Process xchat-gnome (pid: 4322, ti=cad2c000 task=d0cd3ab0 task.ti=cad2c000)

Stack: cad2dd58 c02caa0b 00000002 0000000e 0000000b 00000001 e8836580
0088a800
       00000000 00000000 e8836610 00000000 c01793dc 00001000 c03ab3e0
f3cadd80
       00000086 c90d41b0 0088a800 00000000 dfffd040 00008000 00000000
00000002
Call Trace:
 [<c01793dc>] __getblk+0x23/0x268
 [<f040d4c6>] ext3_getblk+0x10b/0x244 [ext3]
 [<f040e364>] ext3_bread+0x19/0x70 [ext3]
 [<f04106f3>] dx_probe+0x43/0x2c9 [ext3]
 [<f04119b3>] ext3_htree_fill_tree+0x99/0x1ba [ext3]
 [<f040ab77>] ext3_readdir+0x1d4/0x5ed [ext3]
 [<c0167b29>] vfs_readdir+0x63/0x8d
 [<c0167bb6>] sys_getdents64+0x63/0xa5
 [<c0102d38>] syscall_call+0x7/0xb
 [<b7f70822>] 0xb7f70822
 =======================
Code: 8b 40 08 a8 08 74 05 e8 02 2f 11 00 5b 5e c3 55 89 c5 57 89 cf 56
Quoted 6 lines

89 d6 53 83 ec 20 9c 58 90 8d b4 26 00 00 00 00 f6 c4 02 75 04 <0f> 0b eb fe 89 e0 25 00 e0 ff ff ff 40 14 31 c9 8b 1c 8d a0 74
EIP: [<c0179266>] __find_get_block+0x1c/0x16f SS:ESP 0068:cad2dd30

This happens on 2.6.20-rc1 but not 2.6.19.

And it's repeatable, yes?

And you're sure that use of gdb triggers it?

Something is forgetting to reenable local interrupts.

#2 lore raw
Quoted 64 lines · Andrew Morton
Quoted 64 lines

When I was using gdb to debug xchat-gnome, I got a kernel BUG and stack trace as the program was running (e.g. I had typed 'run' in gdb):

WARNING at kernel/softirq.c:137 local_bh_enable()

 [<c0103cd6>] dump_trace+0x68/0x1d9
 [<c0103e5f>] show_trace_log_lvl+0x18/0x2c
 [<c01044d3>] show_trace+0xf/0x11
 [<c010455e>] dump_stack+0x12/0x14
 [<c011cc7d>] local_bh_enable+0x44/0x94
 [<c02871b9>] unix_release_sock+0x6e/0x1fe
 [<c02887eb>] unix_stream_connect+0x3b4/0x3cf
 [<c0232dee>] sys_connect+0x82/0xad
 [<c0233641>] sys_socketcall+0xac/0x261
 [<c0102d38>] syscall_call+0x7/0xb
 [<b7f70822>] 0xb7f70822
 =======================
------------[ cut here ]------------
kernel BUG at fs/buffer.c:1235!
invalid opcode: 0000 [#1]
PREEMPT
Modules linked in: binfmt_misc rfcomm l2cap i915 drm bluetooth nfs nfsd

exportfs lockd nfs_acl sunrpc nvram uinput ipv6 ppdev lp button ac battery dm_crypt dm_snapshot dm_mirror dm_mod fuse cpufreq_conservative cpufreq_ondemand cpufreq_performance cpufreq_powersave speedstep_centrino freq_table ibm_acpi loop snd_intel8x0m snd_pcm_oss snd_mixer_oss snd_intel8x0 snd_ac97_codec pcmcia ac97_bus irtty_sir sir_dev ipw2200 snd_pcm snd_timer irda ieee80211 ieee80211_crypt crc_ccitt rtc parport_pc parport 8250_pnp snd soundcore 8250_pci 8250 serial_core firmware_class i2c_i801 yenta_socket rsrc_nonstatic pcmcia_core snd_page_alloc i2c_core intel_agp agpgart evdev tsdev joydev ext3 jbd mbcache ide_cd cdrom ide_disk ide_generic e100 mii generic piix ide_core ehci_hcd uhci_hcd usbcore

CPU:    0
EIP:    0060:[<c0179266>]    Not tainted VLI
EFLAGS: 00010046   (2.6.20-rc1 #1)

EIP is at __find_get_block+0x1c/0x16f
eax: 00000086 ebx: 00000000 ecx: 00000000 edx: 0088a800
esi: 0088a800 edi: 00000000 ebp: dfffd040 esp: cad2dd30
ds: 007b es: 007b ss: 0068
Process xchat-gnome (pid: 4322, ti=cad2c000 task=d0cd3ab0 task.ti=cad2c000)

Stack: cad2dd58 c02caa0b 00000002 0000000e 0000000b 00000001 e8836580
0088a800
       00000000 00000000 e8836610 00000000 c01793dc 00001000 c03ab3e0
f3cadd80
       00000086 c90d41b0 0088a800 00000000 dfffd040 00008000 00000000
00000002
Call Trace:
 [<c01793dc>] __getblk+0x23/0x268
 [<f040d4c6>] ext3_getblk+0x10b/0x244 [ext3]
 [<f040e364>] ext3_bread+0x19/0x70 [ext3]
 [<f04106f3>] dx_probe+0x43/0x2c9 [ext3]
 [<f04119b3>] ext3_htree_fill_tree+0x99/0x1ba [ext3]
 [<f040ab77>] ext3_readdir+0x1d4/0x5ed [ext3]
 [<c0167b29>] vfs_readdir+0x63/0x8d
 [<c0167bb6>] sys_getdents64+0x63/0xa5
 [<c0102d38>] syscall_call+0x7/0xb
 [<b7f70822>] 0xb7f70822
 =======================
Code: 8b 40 08 a8 08 74 05 e8 02 2f 11 00 5b 5e c3 55 89 c5 57 89 cf 56

89 d6 53 83 ec 20 9c 58 90 8d b4 26 00 00 00 00 f6 c4 02 75 04 <0f> 0b eb fe 89 e0 25 00 e0 ff ff ff 40 14 31 c9 8b 1c 8d a0 74
EIP: [<c0179266>] __find_get_block+0x1c/0x16f SS:ESP 0068:cad2dd30

Quoted 8 lines
Quoted 2 lines

This happens on 2.6.20-rc1 but not 2.6.19.

And it's repeatable, yes?

And you're sure that use of gdb triggers it?

Something is forgetting to reenable local interrupts.

I've managed to get nearly the same thing on a test system I built yesterday, my app when running under gdb would also blow up in __find_get_block.

I was using close to Linus's git head...

Dave.

#3 lore raw
Quoted 72 lines · Dave Airlie
Quoted 71 lines · Andrew Morton
Quoted 60 lines

When I was using gdb to debug xchat-gnome, I got a kernel BUG and stack trace as the program was running (e.g. I had typed 'run' in gdb):

WARNING at kernel/softirq.c:137 local_bh_enable()

 [<c0103cd6>] dump_trace+0x68/0x1d9
 [<c0103e5f>] show_trace_log_lvl+0x18/0x2c
 [<c01044d3>] show_trace+0xf/0x11
 [<c010455e>] dump_stack+0x12/0x14
 [<c011cc7d>] local_bh_enable+0x44/0x94
 [<c02871b9>] unix_release_sock+0x6e/0x1fe
 [<c02887eb>] unix_stream_connect+0x3b4/0x3cf
 [<c0232dee>] sys_connect+0x82/0xad
 [<c0233641>] sys_socketcall+0xac/0x261
 [<c0102d38>] syscall_call+0x7/0xb
 [<b7f70822>] 0xb7f70822
 =======================
------------[ cut here ]------------
kernel BUG at fs/buffer.c:1235!
invalid opcode: 0000 [#1]
PREEMPT
Modules linked in: binfmt_misc rfcomm l2cap i915 drm bluetooth nfs nfsd

exportfs lockd nfs_acl sunrpc nvram uinput ipv6 ppdev lp button ac battery dm_crypt dm_snapshot dm_mirror dm_mod fuse cpufreq_conservative cpufreq_ondemand cpufreq_performance cpufreq_powersave speedstep_centrino freq_table ibm_acpi loop snd_intel8x0m snd_pcm_oss snd_mixer_oss snd_intel8x0 snd_ac97_codec pcmcia ac97_bus irtty_sir sir_dev ipw2200 snd_pcm snd_timer irda ieee80211 ieee80211_crypt crc_ccitt rtc parport_pc parport 8250_pnp snd soundcore 8250_pci 8250 serial_core firmware_class i2c_i801 yenta_socket rsrc_nonstatic pcmcia_core snd_page_alloc i2c_core intel_agp agpgart evdev tsdev joydev ext3 jbd mbcache ide_cd cdrom ide_disk ide_generic e100 mii generic piix ide_core ehci_hcd uhci_hcd usbcore

CPU:    0
EIP:    0060:[<c0179266>]    Not tainted VLI
EFLAGS: 00010046   (2.6.20-rc1 #1)

EIP is at __find_get_block+0x1c/0x16f
eax: 00000086 ebx: 00000000 ecx: 00000000 edx: 0088a800
esi: 0088a800 edi: 00000000 ebp: dfffd040 esp: cad2dd30
ds: 007b es: 007b ss: 0068
Process xchat-gnome (pid: 4322, ti=cad2c000 task=d0cd3ab0 task.ti=cad2c000)

Stack: cad2dd58 c02caa0b 00000002 0000000e 0000000b 00000001 e8836580
0088a800
       00000000 00000000 e8836610 00000000 c01793dc 00001000 c03ab3e0
f3cadd80
       00000086 c90d41b0 0088a800 00000000 dfffd040 00008000 00000000
00000002
Call Trace:
 [<c01793dc>] __getblk+0x23/0x268
 [<f040d4c6>] ext3_getblk+0x10b/0x244 [ext3]
 [<f040e364>] ext3_bread+0x19/0x70 [ext3]
 [<f04106f3>] dx_probe+0x43/0x2c9 [ext3]
 [<f04119b3>] ext3_htree_fill_tree+0x99/0x1ba [ext3]
 [<f040ab77>] ext3_readdir+0x1d4/0x5ed [ext3]
 [<c0167b29>] vfs_readdir+0x63/0x8d
 [<c0167bb6>] sys_getdents64+0x63/0xa5
 [<c0102d38>] syscall_call+0x7/0xb
 [<b7f70822>] 0xb7f70822
 =======================
Code: 8b 40 08 a8 08 74 05 e8 02 2f 11 00 5b 5e c3 55 89 c5 57 89 cf 56
Quoted 6 lines

89 d6 53 83 ec 20 9c 58 90 8d b4 26 00 00 00 00 f6 c4 02 75 04 <0f> 0b eb fe 89 e0 25 00 e0 ff ff ff 40 14 31 c9 8b 1c 8d a0 74
EIP: [<c0179266>] __find_get_block+0x1c/0x16f SS:ESP 0068:cad2dd30

This happens on 2.6.20-rc1 but not 2.6.19.

And it's repeatable, yes?

And you're sure that use of gdb triggers it?

Quoted 7 lines
Quoted 1 line

Something is forgetting to reenable local interrupts.

I've managed to get nearly the same thing on a test system I built yesterday, my app when running under gdb would also blow up in __find_get_block.

I was using close to Linus's git head...

And of course it was on a fresh 32-bit x86 with FC6 on it.

Dave.

#4 lore raw
In reply to Andrew Morton
Quoted 72 lines · Andrew Morton
Quoted 60 lines · On Sun, 17 Dec 2006 20:55:18 -0500 "Andrew J. Barr"…

When I was using gdb to debug xchat-gnome, I got a kernel BUG and stack trace as the program was running (e.g. I had typed 'run' in gdb):

WARNING at kernel/softirq.c:137 local_bh_enable()

 [<c0103cd6>] dump_trace+0x68/0x1d9
 [<c0103e5f>] show_trace_log_lvl+0x18/0x2c
 [<c01044d3>] show_trace+0xf/0x11
 [<c010455e>] dump_stack+0x12/0x14
 [<c011cc7d>] local_bh_enable+0x44/0x94
 [<c02871b9>] unix_release_sock+0x6e/0x1fe
 [<c02887eb>] unix_stream_connect+0x3b4/0x3cf
 [<c0232dee>] sys_connect+0x82/0xad
 [<c0233641>] sys_socketcall+0xac/0x261
 [<c0102d38>] syscall_call+0x7/0xb
 [<b7f70822>] 0xb7f70822
 =======================
------------[ cut here ]------------
kernel BUG at fs/buffer.c:1235!
invalid opcode: 0000 [#1]
PREEMPT
Modules linked in: binfmt_misc rfcomm l2cap i915 drm bluetooth nfs nfsd

exportfs lockd nfs_acl sunrpc nvram uinput ipv6 ppdev lp button ac battery dm_crypt dm_snapshot dm_mirror dm_mod fuse cpufreq_conservative cpufreq_ondemand cpufreq_performance cpufreq_powersave speedstep_centrino freq_table ibm_acpi loop snd_intel8x0m snd_pcm_oss snd_mixer_oss snd_intel8x0 snd_ac97_codec pcmcia ac97_bus irtty_sir sir_dev ipw2200 snd_pcm snd_timer irda ieee80211 ieee80211_crypt crc_ccitt rtc parport_pc parport 8250_pnp snd soundcore 8250_pci 8250 serial_core firmware_class i2c_i801 yenta_socket rsrc_nonstatic pcmcia_core snd_page_alloc i2c_core intel_agp agpgart evdev tsdev joydev ext3 jbd mbcache ide_cd cdrom ide_disk ide_generic e100 mii generic piix ide_core ehci_hcd uhci_hcd usbcore

CPU:    0
EIP:    0060:[<c0179266>]    Not tainted VLI
EFLAGS: 00010046   (2.6.20-rc1 #1)

EIP is at __find_get_block+0x1c/0x16f
eax: 00000086 ebx: 00000000 ecx: 00000000 edx: 0088a800
esi: 0088a800 edi: 00000000 ebp: dfffd040 esp: cad2dd30
ds: 007b es: 007b ss: 0068
Process xchat-gnome (pid: 4322, ti=cad2c000 task=d0cd3ab0 task.ti=cad2c000)

Stack: cad2dd58 c02caa0b 00000002 0000000e 0000000b 00000001 e8836580
0088a800
       00000000 00000000 e8836610 00000000 c01793dc 00001000 c03ab3e0
f3cadd80
       00000086 c90d41b0 0088a800 00000000 dfffd040 00008000 00000000
00000002
Call Trace:
 [<c01793dc>] __getblk+0x23/0x268
 [<f040d4c6>] ext3_getblk+0x10b/0x244 [ext3]
 [<f040e364>] ext3_bread+0x19/0x70 [ext3]
 [<f04106f3>] dx_probe+0x43/0x2c9 [ext3]
 [<f04119b3>] ext3_htree_fill_tree+0x99/0x1ba [ext3]
 [<f040ab77>] ext3_readdir+0x1d4/0x5ed [ext3]
 [<c0167b29>] vfs_readdir+0x63/0x8d
 [<c0167bb6>] sys_getdents64+0x63/0xa5
 [<c0102d38>] syscall_call+0x7/0xb
 [<b7f70822>] 0xb7f70822
 =======================
Code: 8b 40 08 a8 08 74 05 e8 02 2f 11 00 5b 5e c3 55 89 c5 57 89 cf 56
Quoted 7 lines

89 d6 53 83 ec 20 9c 58 90 8d b4 26 00 00 00 00 f6 c4 02 75 04 <0f> 0b eb fe 89 e0 25 00 e0 ff ff ff 40 14 31 c9 8b 1c 8d a0 74
EIP: [<c0179266>] __find_get_block+0x1c/0x16f SS:ESP 0068:cad2dd30

This happens on 2.6.20-rc1 but not 2.6.19.

Quoted 5 lines

And it's repeatable, yes?

And you're sure that use of gdb triggers it?

Something is forgetting to reenable local interrupts.

"walt" <[email protected]> reported a similar problem which he bisected down to the PDA changeset which touches ptrace (66e10a44d724f1464b5e8b5a3eae1e2cbbc2cca6). I haven't managed to repo the problem, but I guess there's something nasty going on in ptrace - maybe its screwing up eflags on the stack or something. Need to double-check all the conversions from kernel<->usermode registers. Hm, wonder if its fixed with the %gs->%fs conversion patch applied?

J

#5 lore raw

PATCH 1/1 ptrace: Fix EFL_OFFSET value according to i386 pda changes (was Re: BUG on 2.6.20-rc1 when using gdb)

Quoted 8 lines · Jeremy Fitzhardinge

"walt" <[email protected]> reported a similar problem which he bisected down to the PDA changeset which touches ptrace (66e10a44d724f1464b5e8b5a3eae1e2cbbc2cca6). I haven't managed to repo the problem, but I guess there's something nasty going on in ptrace - maybe its screwing up eflags on the stack or something. Need to double-check all the conversions from kernel<->usermode registers. Hm, wonder if its fixed with the %gs->%fs conversion patch applied?

Hi Jeremy,

Same problems here with 2.6.20-rc1-mm1 (ie with the %gs->%fs patch). It seems to me that the problem comes from the EFL_OFFSET no longer beeing accurate.
The following patch fixes the problem for me.

Regards,
Frederik

arch/i386/kernel/ptrace.c +1 -1
index 7f7d830..00d8a5a 100644
@@ -45,7 +45,7 @@
 /*
  * Offset of eflags on child stack..
  */
-#define EFL_OFFSET ((EFL-2)*4-sizeof(struct pt_regs))
+#define EFL_OFFSET ((EFL-1)*4-sizeof(struct pt_regs))
 
 static inline struct pt_regs *get_child_regs(struct task_struct *task)
 {
#6 lore raw

Re: [-mm patch] ptrace: Fix EFL_OFFSET value according to i386 pda changes (was Re: BUG on 2.6.20-rc1 when using gdb)

Quoted 9 lines · Frederik Deweerdt
Quoted 8 lines · Jeremy Fitzhardinge

"walt" <[email protected]> reported a similar problem which he bisected down to the PDA changeset which touches ptrace (66e10a44d724f1464b5e8b5a3eae1e2cbbc2cca6). I haven't managed to repo the problem, but I guess there's something nasty going on in ptrace - maybe its screwing up eflags on the stack or something. Need to double-check all the conversions from kernel<->usermode registers. Hm, wonder if its fixed with the %gs->%fs conversion patch applied?

Quoted 6 lines

Hi Jeremy,

Same problems here with 2.6.20-rc1-mm1 (ie with the %gs->%fs patch). It seems to me that the problem comes from the EFL_OFFSET no longer beeing accurate.
The following patch fixes the problem for me.

Me too. Thanks.

Andrew

Quoted 18 lines

Regards,
Frederik

arch/i386/kernel/ptrace.c +1 -1
index 7f7d830..00d8a5a 100644
@@ -45,7 +45,7 @@
 /*
  * Offset of eflags on child stack..
  */
-#define EFL_OFFSET ((EFL-2)*4-sizeof(struct pt_regs))
+#define EFL_OFFSET ((EFL-1)*4-sizeof(struct pt_regs))
 
 static inline struct pt_regs *get_child_regs(struct task_struct *task)
 {
#7 lore raw

Re: [-mm patch] ptrace: Fix EFL_OFFSET value according to i386 pda changes (was Re: BUG on 2.6.20-rc1 when using gdb)

In reply to Frederik Deweerdt
Quoted 5 lines · Frederik Deweerdt

Same problems here with 2.6.20-rc1-mm1 (ie with the %gs->%fs patch). It seems to me that the problem comes from the EFL_OFFSET no longer beeing accurate.
The following patch fixes the problem for me.

Thanks Frederik; that's exactly the kind of thing I thought it might be. I wonder if there's some way we can make this more robust though... Does this work for you? I did a slightly larger cleanup which should make it less fragile and more comprehensible.

J

arch/i386/kernel/ptrace.c +10 -11
diff -r e775f6e42258 arch/i386/kernel/ptrace.c
@@ -45,7 +45,7 @@
 /*
  * Offset of eflags on child stack..
  */
-#define EFL_OFFSET ((EFL-2)*4-sizeof(struct pt_regs))
+#define EFL_OFFSET offsetof(struct pt_regs, eflags)
 
 static inline struct pt_regs *get_child_regs(struct task_struct *task)
 {
@@ -54,24 +54,24 @@ static inline struct pt_regs *get_child_
 }
 
 /*
- * this routine will get a word off of the processes privileged stack. 
- * the offset is how far from the base addr as stored in the TSS.  
- * this routine assumes that all the privileged stacks are in our
+ * This routine will get a word off of the processes privileged stack. 
+ * the offset is bytes into the pt_regs structure on the stack.
+ * This routine assumes that all the privileged stacks are in our
  * data space.
  */   
 static inline int get_stack_long(struct task_struct *task, int offset)
 {
 	unsigned char *stack;
 
-	stack = (unsigned char *)task->thread.esp0;
+	stack = (unsigned char *)task->thread.esp0 - sizeof(struct pt_regs);
 	stack += offset;
 	return (*((int *)stack));
 }
 
 /*
- * this routine will put a word on the processes privileged stack. 
- * the offset is how far from the base addr as stored in the TSS.  
- * this routine assumes that all the privileged stacks are in our
+ * This routine will put a word on the processes privileged stack. 
+ * the offset is bytes into the pt_regs structure on the stack.
+ * This routine assumes that all the privileged stacks are in our
  * data space.
  */
 static inline int put_stack_long(struct task_struct *task, int offset,
@@ -79,7 +79,7 @@ static inline int put_stack_long(struct 
 {
 	unsigned char * stack;
 
-	stack = (unsigned char *) task->thread.esp0;
+	stack = (unsigned char *)task->thread.esp0 - sizeof(struct pt_regs);
 	stack += offset;
 	*(unsigned long *) stack = data;
 	return 0;
@@ -114,7 +114,7 @@ static int putreg(struct task_struct *ch
 	}
 	if (regno > ES*4)
 		regno -= 1*4;
-	put_stack_long(child, regno - sizeof(struct pt_regs), value);
+	put_stack_long(child, regno, value);
 	return 0;
 }
 
@@ -137,7 +137,6 @@ static unsigned long getreg(struct task_
 		default:
 			if (regno > ES*4)
 				regno -= 1*4;
-			regno = regno - sizeof(struct pt_regs);
 			retval &= get_stack_long(child, regno);
 	}
 	return retval;
#8 lore raw

Re: [-mm patch] ptrace: Fix EFL_OFFSET value according to i386 pda changes (was Re: BUG on 2.6.20-rc1 when using gdb)

Quoted 11 lines · Jeremy Fitzhardinge
Quoted 5 lines · Frederik Deweerdt

Same problems here with 2.6.20-rc1-mm1 (ie with the %gs->%fs patch). It seems to me that the problem comes from the EFL_OFFSET no longer beeing accurate.
The following patch fixes the problem for me.

Thanks Frederik; that's exactly the kind of thing I thought it might be. I wonder if there's some way we can make this more robust though... Does this work for you? I did a slightly larger cleanup which should make it less fragile and more comprehensible.

<patch snipped>

Hi Jeremy,

Your patch works fine for me. (I didn't try the first patch, but I will if anyone wants.) Thanks!

#9 lore raw

Re: [-mm patch] ptrace: Fix EFL_OFFSET value according to i386 pda changes (was Re: BUG on 2.6.20-rc1 when using gdb)

In reply to Jeremy Fitzhardinge
Quoted 7 lines · Jeremy Fitzhardinge
Quoted 5 lines · Frederik Deweerdt

Same problems here with 2.6.20-rc1-mm1 (ie with the %gs->%fs patch). It seems to me that the problem comes from the EFL_OFFSET no longer beeing accurate.
The following patch fixes the problem for me.

Quoted 5 lines

Thanks Frederik; that's exactly the kind of thing I thought it might be. I wonder if there's some way we can make this more robust though... Does this work for you? I did a slightly larger cleanup which should make it less fragile and more comprehensible.

It works too, thanks. BTW, I wondered if the "case GS:" in getreg() made sense now?

Frederik

Quoted 79 lines

J

diff -r e775f6e42258 arch/i386/kernel/ptrace.c

arch/i386/kernel/ptrace.c +10 -12
@@ -45,7 +45,7 @@
 /*
  * Offset of eflags on child stack..
  */
-#define EFL_OFFSET ((EFL-2)*4-sizeof(struct pt_regs))
+#define EFL_OFFSET offsetof(struct pt_regs, eflags)
 
 static inline struct pt_regs *get_child_regs(struct task_struct *task)
 {
@@ -54,24 +54,24 @@ static inline struct pt_regs *get_child_
 }
 
 /*
- * this routine will get a word off of the processes privileged stack. 
- * the offset is how far from the base addr as stored in the TSS.  
- * this routine assumes that all the privileged stacks are in our
+ * This routine will get a word off of the processes privileged stack. 
+ * the offset is bytes into the pt_regs structure on the stack.
+ * This routine assumes that all the privileged stacks are in our
  * data space.
  */   
 static inline int get_stack_long(struct task_struct *task, int offset)
 {
 	unsigned char *stack;
 
-	stack = (unsigned char *)task->thread.esp0;
+	stack = (unsigned char *)task->thread.esp0 - sizeof(struct pt_regs);
 	stack += offset;
 	return (*((int *)stack));
 }
 
 /*
- * this routine will put a word on the processes privileged stack. 
- * the offset is how far from the base addr as stored in the TSS.  
- * this routine assumes that all the privileged stacks are in our
+ * This routine will put a word on the processes privileged stack. 
+ * the offset is bytes into the pt_regs structure on the stack.
+ * This routine assumes that all the privileged stacks are in our
  * data space.
  */
 static inline int put_stack_long(struct task_struct *task, int offset,
@@ -79,7 +79,7 @@ static inline int put_stack_long(struct 
 {
 	unsigned char * stack;
 
-	stack = (unsigned char *) task->thread.esp0;
+	stack = (unsigned char *)task->thread.esp0 - sizeof(struct pt_regs);
 	stack += offset;
 	*(unsigned long *) stack = data;
 	return 0;
@@ -114,7 +114,7 @@ static int putreg(struct task_struct *ch
 	}
 	if (regno > ES*4)
 		regno -= 1*4;
-	put_stack_long(child, regno - sizeof(struct pt_regs), value);
+	put_stack_long(child, regno, value);
 	return 0;
 }
 
@@ -137,7 +137,6 @@ static unsigned long getreg(struct task_
 		default:
 			if (regno > ES*4)
 				regno -= 1*4;
-			regno = regno - sizeof(struct pt_regs);
 			retval &= get_stack_long(child, regno);
 	}
 	return retval;
 
-

To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

#10 lore raw

Re: [-mm patch] ptrace: Fix EFL_OFFSET value according to i386 pda changes (was Re: BUG on 2.6.20-rc1 when using gdb)

Quoted 2 lines · Frederik Deweerdt

It works too, thanks. BTW, I wondered if the "case GS:" in getreg() made sense now?

Sorry, what do you mean? It looks OK to me, but I'm not sure what you're referring to.

J

#11 lore raw

Re: [-mm patch] ptrace: Fix EFL_OFFSET value according to i386 pda changes (was Re: BUG on 2.6.20-rc1 when using gdb)

Quoted 6 lines · Jeremy Fitzhardinge
Quoted 2 lines · Frederik Deweerdt

It works too, thanks. BTW, I wondered if the "case GS:" in getreg() made sense now?

Sorry, what do you mean? It looks OK to me, but I'm not sure what you're referring to.

My bad, that's the code I'm referring to:

121 static unsigned long getreg(struct task_struct *child,
122 unsigned long regno)
[...]
126 switch (regno >> 2) {
127 case GS:
128 retval = child->thread.gs;
129 break;

What seem weird to me is that putreg(GS) will end up putting 'value' in:
child->thread.esp0 - sizeof(struct pt_regs) + (GS - 1)*4 whereas getreg(GS) will return the value of child->thread.gs I must miss something, but the symetry seemed odd to me.

Regards,
Frederik

Quoted 8 lines

J
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/