Trouble unloading a module..

Trouble unloading a module..

[James Hansen]

I've having trouble unloading a kernel module for a PCI card that I'm 
porting to 64bit linux.  It doesn't seem to be unloading correctly.  
It's tested, stable and unloads perfectly on a 32bit machine, running 
2.6.10 (and all other 32bit kernels I've tried).

However, on a 64bit machine (unofficial amd/emt64 debian with 2.6.8), 
even though it seems to unload corrrectly, I get a kernel oops if an 
application that uses the driver attempts to connect to the device after 
it's been unloaded.

Being relatively inexperienced at kernel programming, does this point to 
anything that might be obvious to any of you, yet not to me? :)   Or are 
there any common stumbling blocks in terms of unloading modules on 64bit 
linux.

Thanks for any suggestions.

James

Re: Trouble unloading a module..

[Robert Hancock]

James Hansen wrote:
> However, on a 64bit machine (unofficial amd/emt64 debian with 2.6.8), 
> even though it seems to unload corrrectly, I get a kernel oops if an 
> application that uses the driver attempts to connect to the device after 
> it's been unloaded.

I'm guessing the driver isn't properly unregistering itself from 
wherever it is it has registered with (block/character device). Anything 
where the driver is registering function pointers, references, etc. to 
itself needs to be cleaned up on module removal.

The oops stacktrace may point to what it is that has been left behind..

-- 
Robert Hancock      Saskatoon, SK, Canada
To email, remove "nospam" from hancockr@nospamshaw.ca
Home Page: http://www.roberthancock.com/

Re: Trouble unloading a module..

[Arjan van de Ven]

> Being relatively inexperienced at kernel programming, does this point to 
> anything that might be obvious to any of you, yet not to me? :)   Or are 
> there any common stumbling blocks in terms of unloading modules on 64bit 
> linux.

you
1) didn't give the oops here
2) didn't post a URL to the driver source

so.. how is anyone supposed to help?

Re: Trouble unloading a module..

[James Hansen]

Arjan van de Ven wrote:

>>Being relatively inexperienced at kernel programming, does this point to 
>>anything that might be obvious to any of you, yet not to me? :)   Or are 
>>there any common stumbling blocks in terms of unloading modules on 64bit 
>>linux.
>>    
>>
>
>you
>1) didn't give the oops here
>  
>
Sorry, it looks to be failing within chrdev_open.  Could this be caused 
by not correctly unloading the driver?  Here's the oops:

Nov  1 11:39:06 localhost -- MARK --
Nov  1 11:59:06 localhost -- MARK --
Nov  1 12:09:01 localhost kernel:  <1>Unable to handle kernel paging 
request at ffffffffa0261b80 RIP:
Nov  1 12:09:01 localhost kernel: PML4 103027 PGD 105027 PMD 3e765067 PTE 0
Nov  1 12:09:01 localhost kernel: CPU 0
Nov  1 12:09:01 localhost kernel: Modules linked in: ipv6 hw_random 
shpchp pciehp pci_hotplug c4 b1 kernelcapi psmouse ext3 jbd raid5 xor 
raid1 raid0 md e1000 ds yenta_socket pcmcia_core sd_mod ide_cd cdrom 
ide_disk ide_generic pdc202xx_new aec62xx alim15x3 amd74xx atiixp cmd64x 
cs5520 cs5530 cy82c693 generic hpt34x ns87415 opti621 pdc202xx_old 
rz1000 sc1200 serverworks siimage sis5513 slc90e66 triflex trm290 
via82cxxx floppy usb_storage piix ide_core fbcon vga16fb vgastate 
usbserial usbkbd ehci_hcd uhci_hcd thermal processor fan ata_piix libata 
scsi_mod unix font vesafb cfbcopyarea cfbimgblt cfbfillrect
Nov  1 12:09:01 localhost kernel: Pid: 5612, comm: hardserver Not 
tainted 2.6.8-11-amd64-generic
Nov  1 12:09:01 localhost kernel: RIP: 0010:[cdev_get+14/73] 
<ffffffff80163453>{cdev_get+14}
Nov  1 12:09:01 localhost kernel: RSP: 0018:000001003628de48  EFLAGS: 
00010246
Nov  1 12:09:01 localhost kernel: RAX: 0000000000000000 RBX: 
ffffffffa0261b80 RCX: 0000000000000000
Nov  1 12:09:01 localhost kernel: RDX: 000001003b5b9080 RSI: 
000001003b5b9080 RDI: 0000010002177680
Nov  1 12:09:01 localhost kernel: RBP: 000001003c45bd38 R08: 
000001003ea831a8 R09: 000001003ea831a8
Nov  1 12:09:01 localhost kernel: R10: 000001003b5b9080 R11: 
00000000000000f0 R12: 0000000000000000
Nov  1 12:09:01 localhost kernel: R13: 0000000000000000 R14: 
000001003b5b9080 R15: 0000000000000000
Nov  1 12:09:01 localhost kernel: FS:  0000002a959c2090(0000) 
GS:ffffffff80391580(0000) knlGS:0000000000000000
Nov  1 12:09:01 localhost kernel: CS:  0010 DS: 0000 ES: 0000 CR0: 
000000008005003b
Nov  1 12:09:01 localhost kernel: CR2: ffffffffa0261b80 CR3: 
0000000000101000 CR4: 00000000000006e0
Nov  1 12:09:01 localhost kernel: Process hardserver (pid: 5612, 
threadinfo 000001003628c000, task 0000010038bdc230)
Nov  1 12:09:01 localhost kernel: Stack: 0000000000000000 
0000010002177680 000001003c45bd38 ffffffff80163573
Nov  1 12:09:01 localhost kernel:        000001003ea831a8 
000001003ea831a8 000001003b5b9080 000001003c45bd38
Nov  1 12:09:01 localhost kernel:        000001003fbbeec0 0000000000000000
Nov  1 12:09:01 localhost kernel: Call 
Trace:<ffffffff80163573>{chrdev_open+180} 
<ffffffff8015bd2c>{dentry_open+205}
Nov  1 12:09:01 localhost kernel:        
<ffffffff8015be53>{filp_open+62} <ffffffff8023252e>{tcp_listen_start+325}
Nov  1 12:09:01 localhost kernel:        <ffffffff80166c0a>{getname+130} 
<ffffffff8015c04e>{sys_open+57}
Nov  1 12:09:01 localhost kernel:        
<ffffffff8010fc92>{system_call+126}
Nov  1 12:09:01 localhost kernel:
Nov  1 12:09:01 localhost kernel: Code: 83 3b 02 74 32 ff 83 00 01 00 00 
e8 41 38 04 00 48 85 c0 48
Nov  1 12:09:06 localhost kernel:  <6>ACPI: PCI interrupt 
0000:02:03.0[A] -> GSI 24 (level, low) -> IRQ 24
Nov  1 12:09:06 localhost kernel: ACPI: PCI interrupt 0000:07:06.0[A] -> 
GSI 18 (level, low) -> IRQ 18
Nov  1 12:19:06 localhost -- MARK --
Nov  1 12:39:06 localhost -- MARK --

>2) didn't post a URL to the driver source
>  
>
Sorry again.  Here is the source.  Right at the bottom, it appears to be 
calling pci_unregister_driver and unregister_chrdev as it should.

http://www.f0rmula.com/stuff/hostif.c

/var/log/messages says:

Nov  2 13:11:54 localhost kernel: nfdrv 2.9.25+: nfp_dev_destroy: entered
Nov  2 13:11:54 localhost kernel: nfdrv 2.9.25+: nfp_dev_destroy: 
freeing irq, 18
Nov  2 13:11:54 localhost kernel: nfdrv 2.9.25+: nfp_dev_destroy: 
freeing IO BAR, 1
Nov  2 13:11:54 localhost kernel: nfdrv 2.9.25+: nfp_dev_destroy: 
freeing MEM BAR, 2
Nov  2 13:11:54 localhost kernel: nfdrv 2.9.25+: nfp_dev_destroy: 
freeing pdev
Nov  2 13:11:54 localhost kernel: nfdrv 2.9.25+: nfp_pci_remove: entered
Nov  2 13:11:54 localhost kernel: nfdrv 2.9.25+: i21285_destroy: entered
Nov  2 13:11:54 localhost kernel: nfdrv 2.9.25+: nfp_outl: addr ec34, data c
Nov  2 13:11:54 localhost kernel: nfdrv 2.9.25+: nfp_dev_destroy: entered
Nov  2 13:11:54 localhost kernel: nfdrv 2.9.25+: nfp_dev_destroy: 
freeing irq, 12
Nov  2 13:11:54 localhost kernel: nfdrv 2.9.25+: nfp_dev_destroy: 
freeing IO BAR, 1
Nov  2 13:11:54 localhost kernel: nfdrv 2.9.25+: nfp_dev_destroy: 
freeing MEM BAR, 2
Nov  2 13:11:54 localhost kernel: nfdrv 2.9.25+: nfp_dev_destroy: 
freeing pdev
Nov  2 13:11:54 localhost kernel: nfdrv 2.9.25+: cleanup_module: Module 
unloaded

I thought that should prevent the driver from being able to kernel 
oops.  Is there anything else that I should be calling?  Or could it be 
the parameters I'm calling these functions with? (bearing in mind this 
works fine on a similar 32bit kernel).

>so.. how is anyone supposed to help?
>  
>
I was thinking that there may have been a common issue that would allow 
a driver oops the kernel if not unloaded properly.  Obviously not.

Thanks for any advice, it's much appreciated.

James

>
>
>
>  
>

Re: Trouble unloading a module..

[Arjan van de Ven]

On Wed, 2005-11-02 at 12:13 +0000, James Hansen wrote:
> I was thinking that there may have been a common issue that would allow 
> a driver oops the kernel if not unloaded properly.  Obviously not.
> 
> Thanks for any advice, it's much appreciated.

static struct pci_device_id nfp_pci_tbl[] __devinitdata = {

you probably don't want that __devinitdata there....

Re: Trouble unloading a module..

[James Hansen]

Update:

Just tried the driver with a 2.6.14 and it doesn't suffer the same 
problem.  As it looks to be doing what is required to unregister the 
chrdev, I'm loath to suggest it, but could this be a bug in the 2.6.8 
kernel?

Just a thought, so please shoot me down if I'm completly wrong :)

Cheers,

James


Arjan van de Ven wrote:

>On Wed, 2005-11-02 at 12:13 +0000, James Hansen wrote:
>  
>
>>I was thinking that there may have been a common issue that would allow 
>>a driver oops the kernel if not unloaded properly.  Obviously not.
>>
>>Thanks for any advice, it's much appreciated.
>>    
>>
>
>static struct pci_device_id nfp_pci_tbl[] __devinitdata = {
>
>you probably don't want that __devinitdata there....
>
>
>
>
>.
>
>  
>