Skip to contentSkip to conversation

oops in old isdn4linux and 2.6.2-rc3 (was in -rc2 too)

DISCUSSION Applied Todor Todorov 6 participants 12 messages Feb 2, 2004 – Feb 10, 2004

lore
Share & tools
lore
#0 lore raw

Hello everyone,

didn't find any more applicabale mailing list, so here it goes:

card: AVM A1 PCMCIA
kernel: 2.6.2-rc3
isdn options:
CONFIG_ISDN_BOOL=y
CONFIG_ISDN=m
CONFIG_ISDN_NET_SIMPLE=y
CONFIG_ISDN_NET_CISCO=y
CONFIG_ISDN_PPP=y
CONFIG_ISDN_PPP_VJ=y
CONFIG_ISDN_MPP=y
CONFIG_ISDN_PPP_BSDCOMP=m
CONFIG_ISDN_AUDIO=y
CONFIG_ISDN_TTY_FAX=y
CONFIG_ISDN_DRV_LOOP=m
CONFIG_ISDN_DRV_HISAX=m
CONFIG_HISAX_EURO=y
CONFIG_DE_AOC=y
CONFIG_HISAX_MAX_CARDS=2
CONFIG_HISAX_AVM_A1_PCMCIA=y
CONFIG_HISAX_AVM_A1_CS=m

pcmcia options:
CONFIG_PCMCIA=m
CONFIG_YENTA=m
CONFIG_CARDBUS=y
CONFIG_PCMCIA_PROBE=y

When I insert the ISDN card I get the followin oops and the cardmgr proces does not respond anymore. oops output:

Feb  2 13:44:11 styx kernel: cs: memory probe 0xa0000000-0xa0ffffff: clean.
Feb  2 13:44:11 styx kernel: CSLIP: code copyright 1989 Regents of the University of California
Feb  2 13:44:12 styx kernel: ISDN subsystem initialized
Feb  2 13:44:12 styx kernel: HiSax: Linux Driver for passive ISDN cards
Feb  2 13:44:12 styx kernel: HiSax: Version 3.5 (module)
Feb  2 13:44:12 styx kernel: HiSax: Layer1 Revision 2.41.6.5
Feb  2 13:44:12 styx kernel: HiSax: Layer2 Revision 2.25.6.4
Feb  2 13:44:12 styx kernel: HiSax: TeiMgr Revision 2.17.6.3
Feb  2 13:44:12 styx kernel: HiSax: Layer3 Revision 2.17.6.5
Feb  2 13:44:12 styx kernel: HiSax: LinkLayer Revision 2.51.6.6
Feb  2 13:44:12 styx kernel: avma1_cs: testing i/o 0x140-0x147
Feb  2 13:44:12 styx kernel: avma1_cs: checking at i/o 0x140, irq 3
Feb  2 13:44:12 styx kernel: get_drv 0: 0 -> 1
Feb  2 13:44:12 styx kernel: HiSax: Card 1 Protocol EDSS1 Id=HiSax (0)
Feb  2 13:44:12 styx kernel: HiSax: AVM A1 PCMCIA driver Rev. 2.7.6.2
Feb  2 13:44:12 styx kernel:  printing eip:
Feb  2 13:44:12 styx kernel: d19cd612
Feb  2 13:44:12 styx kernel: Oops: 0000 [#1]
Feb  2 13:44:12 styx kernel: CPU:    0
Feb  2 13:44:12 styx kernel: EIP: 0060:[__crc_blk_attempt_remerge+2649154/3591710]    Not tainted
Feb  2 13:44:12 styx kernel: EFLAGS: 00010282
Feb  2 13:44:12 styx kernel: EIP is at 0xd19cd612
Feb  2 13:44:12 styx kernel: eax: c4e31000   ebx: d3e9f9a0   ecx: 00000001   edx: c0315598
Feb  2 13:44:12 styx kernel: esi: d3e96987   edi: cef7385d   ebp: cef73894   esp: cef7383c
Feb  2 13:44:12 styx kernel: ds: 007b   es: 007b   ss: 0068
Feb  2 13:44:12 styx kernel: Process cardmgr (pid: 975, threadinfo=cef72000 task=cedab900)
Feb  2 13:44:12 styx kernel: Stack: d3e8ebfc c4e31000 d3e9f9a0 76655224 6f697369 32203a6e 362e372e 2400322e
Feb  2 13:44:12 styx kernel:        cef73800 00000246 cffad758 c4e31000 cef738ee c4e310d6 cef73894 d3e78c78
Feb  2 13:44:12 styx kernel:        d3e97240 00000001 d3e94294 c4e31000 cef738ee c4e310d6 cef738bc d3e78e36
Feb  2 13:44:12 styx kernel: Call Trace:
Feb  2 13:44:12 styx kernel: [__crc_journal_set_features+2091499/2437465] setup_avm_a1_pcmcia+0x4a/0x62 [hisax]
Feb  2 13:44:12 styx kernel: [__crc_journal_set_features+2001511/2437465] do_register_isdn+0xe3/0xe9 [hisax]
Feb  2 13:44:12 styx kernel: [__crc_journal_set_features+2001957/2437465] checkcard+0x168/0x1ad [hisax]
Feb  2 13:44:12 styx kernel: [__crc_journal_set_features+2002290/2437465] HiSax_inithardware+0xc3/0x17e [hisax]
Feb  2 13:44:12 styx kernel: [__crc_journal_set_features+2003505/2437465] avm_a1_init_pcmcia+0x105/0x125 [hisax]
Feb  2 13:44:12 styx kernel: [__crc_blk_attempt_remerge+2665386/3591710] avma1cs_config+0x269/0x38c [avma1_cs]
Feb  2 13:44:12 styx kernel: [__crc_blk_attempt_remerge+2483938/3591710] set_cis_map+0x3e/0x10b [pcmcia_core]
Feb  2 13:44:12 styx kernel: [__crc_blk_attempt_remerge+2484415/3591710] read_cis_mem+0x110/0x182 [pcmcia_core]
Feb  2 13:44:12 styx kernel: [__crc_blk_attempt_remerge+2488313/3591710] parse_vers_1+0x5c/0x63 [pcmcia_core]
Feb  2 13:44:12 styx kernel: [__crc_blk_attempt_remerge+2487241/3591710] pcmcia_get_next_tuple+0x249/0x2a3 [pcmcia_core]
Feb  2 13:44:12 styx kernel:  [__call_console_drivers+87/89] __call_console_drivers+0x57/0x59
Feb  2 13:44:12 styx kernel:  [__wake_up_common+49/80] __wake_up_common+0x31/0x50
Feb  2 13:44:12 styx kernel:  [printk+299/380] printk+0x12b/0x17c
Feb  2 13:44:12 styx kernel: [__crc_blk_attempt_remerge+2495383/3591710] do_mem_probe+0xfc/0x1c2 [pcmcia_core]
Feb  2 13:44:12 styx kernel: [__crc_blk_attempt_remerge+2487241/3591710] pcmcia_get_next_tuple+0x249/0x2a3 [pcmcia_core]
Feb  2 13:44:12 styx kernel: [__crc_blk_attempt_remerge+2486012/3591710] pcmcia_get_first_tuple+0x94/0x130 [pcmcia_core]
Feb  2 13:44:12 styx kernel: [__crc_blk_attempt_remerge+2665883/3591710] avma1cs_event+0x59/0xb6 [avma1_cs]
Feb  2 13:44:12 styx kernel: [__crc_blk_attempt_remerge+2510239/3591710] pcmcia_register_client+0x232/0x282 [pcmcia_core]
Feb  2 13:44:12 styx kernel: [__crc_blk_attempt_remerge+2483938/3591710] set_cis_map+0x3e/0x10b [pcmcia_core]
Feb  2 13:44:12 styx kernel: [__crc_blk_attempt_remerge+2664285/3591710] avma1cs_attach+0x12d/0x1a4 [avma1_cs]
Feb  2 13:44:12 styx kernel: [__crc_blk_attempt_remerge+2665794/3591710] avma1cs_event+0x0/0xb6 [avma1_cs]
Feb  2 13:44:12 styx kernel: [__crc_blk_attempt_remerge+2507122/3591710] pcmcia_bind_device+0x68/0xa7 [pcmcia_core]
Feb  2 13:44:12 styx kernel: [__crc_blk_attempt_remerge+2226992/3591710] bind_request+0x108/0x228 [ds]
Feb  2 13:44:12 styx kernel:  [capable+36/67] capable+0x24/0x43
Feb  2 13:44:12 styx kernel: [__crc_blk_attempt_remerge+2230000/3591710] ds_ioctl+0x52a/0x660 [ds]
Feb  2 13:44:12 styx kernel:  [sock_def_readable+128/130] sock_def_readable+0x80/0x82
Feb  2 13:44:12 styx kernel:  [unix_dgram_sendmsg+1134/1364] unix_dgram_sendmsg+0x46e/0x554
Feb  2 13:44:12 styx kernel:  [__free_pages_ok+175/183] __free_pages_ok+0xaf/0xb7
Feb  2 13:44:12 styx kernel:  [__free_pages+58/72] __free_pages+0x3a/0x48
Feb  2 13:44:12 styx kernel:  [slab_destroy+199/264] slab_destroy+0xc7/0x108
Feb  2 13:44:12 styx kernel:  [free_block+185/205] free_block+0xb9/0xcd
Feb  2 13:44:12 styx kernel:  [cache_flusharray+78/236] cache_flusharray+0x4e/0xec
Feb  2 13:44:12 styx kernel:  [zap_pmd_range+75/101] zap_pmd_range+0x4b/0x65
Feb  2 13:44:12 styx kernel:  [unmap_page_range+65/103] unmap_page_range+0x41/0x67
Feb  2 13:44:12 styx kernel:  [unmap_vmas+229/522] unmap_vmas+0xe5/0x20a
Feb  2 13:44:12 styx kernel:  [unmap_vma+65/123] unmap_vma+0x41/0x7b
Feb  2 13:44:12 styx kernel:  [unmap_vma_list+29/42] unmap_vma_list+0x1d/0x2a
Feb  2 13:44:12 styx kernel:  [do_munmap+321/387] do_munmap+0x141/0x183
Feb  2 13:44:12 styx kernel:  [sys_ioctl+286/675] sys_ioctl+0x11e/0x2a3
Feb  2 13:44:12 styx kernel:  [syscall_call+7/11] syscall_call+0x7/0xb
Feb  2 13:44:12 styx kernel:
Feb  2 13:44:12 styx kernel: Code:  Bad EIP value.

Please help with this issue. TIA!

Greetings,
T.Todorov

#1 lore raw
Quoted 4 lines · Todor Todorov

Hello everyone,

didn't find any more applicabale mailing list, so here it goes:

try the actual I4L for 2.6 patch in ftp://ftp.isdn4linux.de/pub/isdn4linux/kernel/v2.6

Signature · 3 lines
Karsten Keil
SuSE Labs
ISDN development
#2 lore raw

Hi Karsten,

Quoted 2 lines

try the actual I4L for 2.6 patch in ftp://ftp.isdn4linux.de/pub/isdn4linux/kernel/v2.6

what is your current plan for including these patches into mainstream once 2.6.2 is out?

Regards

Marcel

#3 lore raw
Quoted 8 lines · Marcel Holtmann

Hi Karsten,

Quoted 2 lines

try the actual I4L for 2.6 patch in ftp://ftp.isdn4linux.de/pub/isdn4linux/kernel/v2.6

what is your current plan for including these patches into mainstream once 2.6.2 is out?

Was sent to Linus a week ago, but no answer so far.

Signature · 3 lines
Karsten Keil
SuSE Labs
ISDN development
#4 lore raw

tir, 03.02.2004 kl. 08.48 skrev Karsten Keil:

Quoted 11 lines · Marcel Holtmann
Quoted 8 lines

Hi Karsten,

Quoted 2 lines

try the actual I4L for 2.6 patch in ftp://ftp.isdn4linux.de/pub/isdn4linux/kernel/v2.6

what is your current plan for including these patches into mainstream once 2.6.2 is out?

Was sent to Linus a week ago, but no answer so far.

Now that 2.6.2 is out, and it still hasn't been applied, is it inappropriate to beg you to resend it to Linus and/or Andrew? It is really crucial for me, without this patch Hisax with TelesPCI support won't even compile. And probably lots of others as well.

Thanks for your great work!

Best regards,
Stian

#5 lore raw

2.6.2-rc3 with isdn patch + devfs doesn't compile

In reply to Karsten Keil
Quoted 8 lines · Karsten Keil
Quoted 4 lines · Todor Todorov

Hello everyone,

didn't find any more applicabale mailing list, so here it goes:

try the actual I4L for 2.6 patch in ftp://ftp.isdn4linux.de/pub/isdn4linux/kernel/v2.6

FYI:
I'm getting the following compile error in 2.6.2-rc3 with this patch applied and CONFIG_DEVFS_FS=n:

<-- snip -->

...
CC drivers/isdn/capi/capidrv.o
In file included from drivers/isdn/capi/capidrv.c:25:
include/linux/isdn.h:637: error: parse error before "devfs_handle_t" include/linux/isdn.h:637: warning: no semicolon at end of struct or union include/linux/isdn.h:638: warning: type defaults to `int' in declaration
of `devfs_handle_isdnctrl'
include/linux/isdn.h:638: warning: data definition has no type or storage class include/linux/isdn.h:639: error: parse error before "devfs_handle_isdnX" include/linux/isdn.h:639: warning: type defaults to int' in declaration of devfs_handle_isdnX' include/linux/isdn.h:639: warning: data definition has no type or storage class include/linux/isdn.h:640: error: parse error before "devfs_handle_isdnctrlX" include/linux/isdn.h:640: warning: type defaults to int' in declaration of devfs_handle_isdnctrlX' include/linux/isdn.h:640: warning: data definition has no type or storage class include/linux/isdn.h:642: error: parse error before "devfs_handle_ipppX" include/linux/isdn.h:642: warning: type defaults to int' in declaration of devfs_handle_ipppX' include/linux/isdn.h:642: warning: data definition has no type or storage class include/linux/isdn.h:645: error: parse error before '}' token include/linux/isdn.h:645: warning: type defaults to int' in declaration of isdn_dev' include/linux/isdn.h:645: warning: data definition has no type or storage class include/linux/isdn.h:647: error: parse error before '*' token include/linux/isdn.h:647: warning: type defaults to int' in declaration of dev' include/linux/isdn.h:647: warning: data definition has no type or storage class drivers/isdn/capi/capidrv.c:2111:10: warning: #warning FIXME: maybe a
race condition the card should be removed here from global list /kkeil
make[3]: *** [drivers/isdn/capi/capidrv.o] Error 1

<-- snip -->

Quoted 1 line

Karsten Keil

cu
Adrian

Signature · 4 lines
       "Is there not promise of rain?" Ling Tan asked suddenly out
        of the darkness. There had been need of rain for many days.
       "Only a promise," Lao Er said.
                                       Pearl S. Buck - Dragon Seed
#6 lore raw

Re: 2.6.2-rc3 with isdn patch + devfs doesn't compile

Quoted 7 lines · Adrian Bunk
Quoted 6 lines · Karsten Keil
Quoted 4 lines · Todor Todorov

Hello everyone,

didn't find any more applicabale mailing list, so here it goes:

Quoted 7 lines
Quoted 2 lines

try the actual I4L for 2.6 patch in ftp://ftp.isdn4linux.de/pub/isdn4linux/kernel/v2.6

FYI:
I'm getting the following compile error in 2.6.2-rc3 with this patch applied and CONFIG_DEVFS_FS=n:

Thank you.
Next version of the patch will remove devfs support from I4L for 2.6 kernel versions.

Signature · 3 lines
Karsten Keil
SuSE Labs
ISDN development
#7 lore raw
In reply to Karsten Keil
Quoted 2 lines · Karsten Keil

try the actual I4L for 2.6 patch in ftp://ftp.isdn4linux.de/pub/isdn4linux/kernel/v2.6

I applied isdn-2.6.2-rc2.diff.gz to 2.6.3-rc1 (actually to BK as of about Friday). Works fine for dialup, including isdn_lzscomp, on SMP.

Dies horribly upon an incoming v.110 call though...

(akpm/linus please note v.110 is fairly esoteric and it won't even _compile_ in 2.6 at the moment so this is not a reason not to merge Karsten's patch, which does at least make ISDN usable again in 2.6)

NMI Watchdog detected LOCKUP on CPU1, eip f8a5cd65, registers:

CPU:    1
EIP:    0060:[<f8a5cd65>]    Not tainted
EFLAGS: 00000086

EIP is at .text.lock.hfc_pci+0xfb/0x186 [hisax]
eax: f6531000 ebx: 00000120 ecx: 00000000 edx: f6288a40
esi: f6531614 edi: f652ac00 ebp: f6288a40 esp: f7f81b40
ds: 007b es: 007b ss: 0068
Process events/1 (pid: 7, threadinfo=f7f80000 task=f7f94ce0)

Stack: f6288800 00000028 c0224fea 00000097 f6531210 f652ac00 00000028 f6288800
       f8a533e6 f652ac00 00000120 f6288a40 0000003c 00000020 f75b7810 f6d52000
       f6288800 00000000 00000001 f89fb58f 00000000 00000001 00000001 f6288800
Call Trace:
 [<c0224fea>] skb_clone+0xaa/0x180
 [<f8a533e6>] HiSax_writebuf_skb+0x156/0x230 [hisax]
 [<f89fb58f>] isdn_v110_stat_callback+0x2bf/0x330 [isdn]
 [<f89fc57e>] isdn_status_callback+0x76e/0x8e0 [isdn]
 [<f89fc09c>] isdn_status_callback+0x28c/0x8e0 [isdn]
 [<c01b5937>] vsprintf+0x27/0x30
 [<f8a52231>] ll_writewakeup+0x61/0xa0 [hisax]
 [<c01b567d>] vsnprintf+0x23d/0x4a0
 [<c011afea>] __wake_up_common+0x3a/0x60
 [<f8a5af40>] hfcpci_fill_fifo+0x4e0/0x6a0 [hisax]
 [<f8a5c2d2>] hfcpci_send_data+0x22/0x50 [hisax]
 [<f8a5c5d5>] hfcpci_l2l1+0xa5/0x200 [hisax]
 [<c0224fea>] skb_clone+0xaa/0x180
 [<f8a533e6>] HiSax_writebuf_skb+0x156/0x230 [hisax]
 [<f89fb3dc>] isdn_v110_stat_callback+0x10c/0x330 [isdn]
 [<f89fc109>] isdn_status_callback+0x2f9/0x8e0 [isdn]
 [<c01b567d>] vsnprintf+0x23d/0x4a0
 [<f8a50370>] lli_go_active+0x0/0xd0 [hisax]
 [<c01b5937>] vsprintf+0x27/0x30
 [<f8a50370>] lli_go_active+0x0/0xd0 [hisax]
 [<f8a4ff24>] link_debug+0x64/0x80 [hisax]
 [<f8a50370>] lli_go_active+0x0/0xd0 [hisax]
 [<f8a503e7>] lli_go_active+0x77/0xd0 [hisax]
 [<f8a50370>] lli_go_active+0x0/0xd0 [hisax]
 [<c01b5937>] vsprintf+0x27/0x30
 [<f8a50370>] lli_go_active+0x0/0xd0 [hisax]
 [<f8a51c60>] callc_debug+0x50/0x60 [hisax]
 [<f8a44374>] l1m_debug+0x54/0x70 [hisax]
 [<f8a53640>] FsmEvent+0x90/0xf0 [hisax]
 [<f8a45130>] l1b_timer_act+0x0/0x40 [hisax]
 [<f8a45164>] l1b_timer_act+0x34/0x40 [hisax]
 [<f8a53640>] FsmEvent+0x90/0xf0 [hisax]
 [<f8a448b4>] BChannel_proc_rcv+0xa4/0xb0 [hisax]
 [<f8a4490d>] BChannel_bh+0x4d/0x50 [hisax]
 [<c012edea>] worker_thread+0x1aa/0x220
 [<f8a448c0>] BChannel_bh+0x0/0x50 [hisax]
 [<c011af90>] default_wake_function+0x0/0x20
 [<c0108e82>] ret_from_fork+0x6/0x14
 [<c011af90>] default_wake_function+0x0/0x20
 [<c012ec40>] worker_thread+0x0/0x220
 [<c0106e69>] kernel_thread_helper+0x5/0xc
 
Code: 80 38 00 7e f9 e9 fe f7 ff ff f3 90 80 7e 34 00 7e f8 e9 07

console shuts up ...

Signature · 1 line
dwmw2
#8 lore raw
Quoted 7 lines · David Woodhouse

Dies horribly upon an incoming v.110 call though...

NMI Watchdog detected LOCKUP on CPU1, eip f8a5cd65, registers:

CPU:    1
EIP:    0060:[<f8a5cd65>]    Not tainted
EFLAGS: 00000086

EIP is at .text.lock.hfc_pci+0xfb/0x186 [hisax]

This is line 1391 of hfc_pci.c, in hfcpci_l2l1().
It's trying to relock &bcs->cs->lock which is evidently already locked.

This is because isdn_v110_stat_callback() is being called with the spinlock held when we've transmitted a skb... and it's trying to transmit another immediately.

Quoted 8 lines

eax: f6531000 ebx: 00000120 ecx: 00000000 edx: f6288a40
esi: f6531614 edi: f652ac00 ebp: f6288a40 esp: f7f81b40
ds: 007b es: 007b ss: 0068
Process events/1 (pid: 7, threadinfo=f7f80000 task=f7f94ce0)

Stack: f6288800 00000028 c0224fea 00000097 f6531210 f652ac00 00000028 f6288800
       f8a533e6 f652ac00 00000120 f6288a40 0000003c 00000020 f75b7810 f6d52000
       f6288800 00000000 00000001 f89fb58f 00000000 00000001 00000001 f6288800
Call Trace:
	<hfcpci_l2l1>
Quoted 11 lines
 [<f8a533e6>] HiSax_writebuf_skb+0x156/0x230 [hisax]
 [<f89fb58f>] isdn_v110_stat_callback+0x2bf/0x330 [isdn]
 [<f89fc57e>] isdn_status_callback+0x76e/0x8e0 [isdn]
 [<f89fc09c>] isdn_status_callback+0x28c/0x8e0 [isdn]
 [<f8a52231>] ll_writewakeup+0x61/0xa0 [hisax]
 [<f8a5af40>] hfcpci_fill_fifo+0x4e0/0x6a0 [hisax]
 [<f8a5c2d2>] hfcpci_send_data+0x22/0x50 [hisax]
 [<f8a5c5d5>] hfcpci_l2l1+0xa5/0x200 [hisax]
 [<f8a533e6>] HiSax_writebuf_skb+0x156/0x230 [hisax]
 [<f89fb3dc>] isdn_v110_stat_callback+0x10c/0x330 [isdn]
 [<f89fc109>] isdn_status_callback+0x2f9/0x8e0 [isdn]

I'm going to try an _evil_ hack like this...

isdn/hisax/hfc_pci.c renamed from isdn/hisax/hfc_pci.c.orig +8 -2
@@ -597,7 +597,7 @@
 	u_char new_f1, *src, *dst;
 	unsigned short *z1t, *z2t;
 
-	if (!bcs->tx_skb)
+	if (!bcs->tx_skb && !(bcs->tx_skb = skb_dequeue(&bcs->squeue)))
 		return;
 	if (bcs->tx_skb->len <= 0)
 		return;
@@ -1388,7 +1388,13 @@
 
 	switch (pr) {
 		case (PH_DATA | REQUEST):
-			spin_lock_irqsave(&bcs->cs->lock, flags);
+			local_irq_save(flags);
+			if (!spin_trylock(&bcs->cs->lock)) {
+				/* Possible recursion. Queue it and return */
+				local_irq_restore(flags);
+				skb_queue_tail(&bcs->squeue, skb);
+				break;
+			}
 			if (bcs->tx_skb) {
 				skb_queue_tail(&bcs->squeue, skb);
 			} else {
Signature · 1 line
dwmw2
#9 lore raw
Quoted 9 lines · David Woodhouse
Quoted 7 lines · David Woodhouse

Dies horribly upon an incoming v.110 call though...

NMI Watchdog detected LOCKUP on CPU1, eip f8a5cd65, registers:

CPU:    1
EIP:    0060:[<f8a5cd65>]    Not tainted
EFLAGS: 00000086

EIP is at .text.lock.hfc_pci+0xfb/0x186 [hisax]

Quoted 6 lines

This is line 1391 of hfc_pci.c, in hfcpci_l2l1().
It's trying to relock &bcs->cs->lock which is evidently already locked.

This is because isdn_v110_stat_callback() is being called with the spinlock held when we've transmitted a skb... and it's trying to transmit another immediately.

Your analyse is correct, but the error is, that
ll_writewakeup is called with the lock.
Upper layers should not be called in this path.

Quoted 22 lines
Quoted 9 lines

eax: f6531000 ebx: 00000120 ecx: 00000000 edx: f6288a40
esi: f6531614 edi: f652ac00 ebp: f6288a40 esp: f7f81b40
ds: 007b es: 007b ss: 0068
Process events/1 (pid: 7, threadinfo=f7f80000 task=f7f94ce0)

Stack: f6288800 00000028 c0224fea 00000097 f6531210 f652ac00 00000028 f6288800
       f8a533e6 f652ac00 00000120 f6288a40 0000003c 00000020 f75b7810 f6d52000
       f6288800 00000000 00000001 f89fb58f 00000000 00000001 00000001 f6288800
Call Trace:
	<hfcpci_l2l1>
Quoted 11 lines
 [<f8a533e6>] HiSax_writebuf_skb+0x156/0x230 [hisax]
 [<f89fb58f>] isdn_v110_stat_callback+0x2bf/0x330 [isdn]
 [<f89fc57e>] isdn_status_callback+0x76e/0x8e0 [isdn]
 [<f89fc09c>] isdn_status_callback+0x28c/0x8e0 [isdn]
 [<f8a52231>] ll_writewakeup+0x61/0xa0 [hisax]
 [<f8a5af40>] hfcpci_fill_fifo+0x4e0/0x6a0 [hisax]
 [<f8a5c2d2>] hfcpci_send_data+0x22/0x50 [hisax]
 [<f8a5c5d5>] hfcpci_l2l1+0xa5/0x200 [hisax]
 [<f8a533e6>] HiSax_writebuf_skb+0x156/0x230 [hisax]
 [<f89fb3dc>] isdn_v110_stat_callback+0x10c/0x330 [isdn]
 [<f89fc109>] isdn_status_callback+0x2f9/0x8e0 [isdn]
Quoted 2 lines

I'm going to try an _evil_ hack like this...

Yes this should help you for the moment.

Quoted 31 lines
isdn/hisax/hfc_pci.c renamed from isdn/hisax/hfc_pci.c.orig +8 -2
@@ -597,7 +597,7 @@
 	u_char new_f1, *src, *dst;
 	unsigned short *z1t, *z2t;
 
-	if (!bcs->tx_skb)
+	if (!bcs->tx_skb && !(bcs->tx_skb = skb_dequeue(&bcs->squeue)))
 		return;
 	if (bcs->tx_skb->len <= 0)
 		return;
@@ -1388,7 +1388,13 @@
 
 	switch (pr) {
 		case (PH_DATA | REQUEST):
-			spin_lock_irqsave(&bcs->cs->lock, flags);
+			local_irq_save(flags);
+			if (!spin_trylock(&bcs->cs->lock)) {
+				/* Possible recursion. Queue it and return */
+				local_irq_restore(flags);
+				skb_queue_tail(&bcs->squeue, skb);
+				break;
+			}
 			if (bcs->tx_skb) {
 				skb_queue_tail(&bcs->squeue, skb);
 			} else {
Signature · 2 lines
-- 
dwmw2
Signature · 3 lines
Karsten Keil
SuSE Labs
ISDN development
#10 lore raw
Quoted 3 lines · Karsten Keil

Your analyse is correct, but the error is, that
ll_writewakeup is called with the lock.
Upper layers should not be called in this path.

That's an evidently sensible answer, yes -- but it looks like it's always been like this, and fixing it really wasn't not a task I particularly wanted to undertake at 2am this morning... I think I might leave that to you :)

Signature · 1 line
dwmw2
#11 lore raw
Quoted 9 lines · David Woodhouse
Quoted 3 lines · Karsten Keil

Your analyse is correct, but the error is, that
ll_writewakeup is called with the lock.
Upper layers should not be called in this path.

That's an evidently sensible answer, yes -- but it looks like it's always been like this, and fixing it really wasn't not a task I particularly wanted to undertake at 2am this morning... I think I might leave that to you :)

I don't want give you the impression of a "evidently sensible answer", (maybe ist was also a little bit late or early ... :-)
I only want to point out that here is a deeper design flaw in this code and it is here for a long time, but the old none spinlock code with a big CLI (bad,bad,bad) make it working well enough. During porting I saw this problem, but forget to mark it for fixing later.

These things (missing separation of layers) was one reason for me to begin a new driver from scratch some time ago (mISDN).

Signature · 3 lines
Karsten Keil
SuSE Labs
ISDN development