Skip to contentSkip to conversation

Solaris developer wants a Linux Mentor for drivers.

DISCUSSION Alan Kilian 3 participants 3 messages Oct 6, 2004

lore
Share & tools
lore
#0 lore raw

Kernel folks,

I am in the process of porting a Sun Solaris PCI bus driver that I wrote over to a 2.4 kernel, and I could use a mentor just to get me over the initial bumps.

I have a module that can be loaded, and detects my card, and responds to ioctl() calls properly, so I'm moving right along, but I do have some problems now.

1) If I "oops" in my module, I cannot unload it with:
# /sbin/rmmod sse
sse: Device or resource busy

I have only figured out that a reboot cleans things up again.

2) An example of using pci_ops read_dword() would be superb.

By the way, this development environment is really slick compared to Solaris. When I "oops" in Solaris, the kernel panics and I'm in for a messy fsck on the way back up. This is a great improvement.

Thanks in advance for any help any of you may provide.

After this, there will be one more Linux PCI bus driver developer in the world, and that can't be a bad thing.

-Alan

Signature · 2 lines
- Alan Kilian <kilian(at)timelogic.com> 
Director of Bioinformatics, TimeLogic Corporation 763-449-7622
#1 lore raw
Quoted 7 lines · Alan Kilian

Kernel folks,

I am in the process of porting a Sun Solaris PCI bus driver that I wrote over to a 2.4 kernel, and I could use a mentor just to get me over the initial bumps.

Quoted 7 lines

I have a module that can be loaded, and detects my card, and responds to ioctl() calls properly, so I'm moving right along, but I do have some problems now.

1) If I "oops" in my module, I cannot unload it with:
# /sbin/rmmod sse
sse: Device or resource busy

somebody is holding a reference to your device, either through your modules use of the MOD_INC_USE_COUNT macro (which I believe is depricated), or through the use of one of the macros that the kernel uses to prevent module remove race conditions (dev_hold is the only one that jumps to mind at the moment). You can only get the module to remove by balancing the MOD_INC/DEC macros, or massaging your kernel to release any references to your module. However, after an oops, its usually best practice to reboot, lest you run a kernel with corrupted memory somewhere.

Quoted 5 lines

I have only figured out that a reboot cleans things up again.

2) An example of using pci_ops read_dword() would be superb.

I don't believe you call the operation in a pci_ops structure directly.
check the kernel for references to
pci_[read|write]_config_[byte|word|dword] instead.

HTH
Neil

Quoted 12 lines

By the way, this development environment is really slick compared to Solaris. When I "oops" in Solaris, the kernel panics and I'm in for a messy fsck on the way back up. This is a great improvement.

Thanks in advance for any help any of you may provide.

After this, there will be one more Linux PCI bus driver developer in the world, and that can't be a bad thing.

-Alan

Signature · 8 lines
/***************************************************
 *Neil Horman
 *Software Engineer
 *Red Hat, Inc.
 *[email protected]
 *gpg keyid: 1024D / 0x92A74FA1
 *http://pgp.mit.edu
 ***************************************************/
#2 lore raw
In reply to Alan Kilian
Quoted 7 lines · Alan Kilian

Kernel folks,

I am in the process of porting a Sun Solaris PCI bus driver that I wrote over to a 2.4 kernel, and I could use a mentor just to get me over the initial bumps.

Why not 2.6? No new Linux distros are shipping 2.4 kernels anymore...

And a PCI bus driver? What kind of hardware is this? Is this a driver for a pci card, or a pci bus controller?

thanks,

greg k-h