Re: [gentoo-alpha] 2.6 kernel on Alpha ?

["Jean-Sébastien Guay" <jean_seb@videotron.ca>]

Hello Matthew,

>I am able to compile and boot a UP kernel, but I continuously
>get these messages on screen during boot and later in the
>system logs
>
>[kernel] sd 0:0:0:0: M_REJECT to send for : 1-2-3-1.
>
>seems to come from the SCSI SYM53C8XX_2 driver which handles the
>Symbios 53c810 controller chip. I am using these settings in
>the kernel config
>CONFIG_SCSI_SYM53C8XX_2=y
>CONFIG_SCSI_SYM53C8XX_DMA_ADDRESSING_MODE=0
>CONFIG_SCSI_SYM53C8XX_DEFAULT_TAGS=16
>CONFIG_SCSI_SYM53C8XX_MAX_TAGS=16
># CONFIG_SCSI_SYM53C8XX_IOMAPPED is not set
>(I tried using IOMAPPED and different DMA modes and number of TAGs
>also, but the message still shows up)
>
>in 2.4.30, the driver version is sym-2.1.17a and it works fine
>(without any of these REJECT messages), while 2.6.12.5, the
>driver version is 2.2.x I believe.
>
>does anyone on the list know of an issue which could cause this ?
>is the REJECT message a cause for concern ? (system seem to work
>fine as before)
>  
>
I have this problem with my SYM53C810 controller too, and the only 
solution I've found (that worked) was to manually modify the kernel's 
SCSI driver. See this page for the details :
http://lists.debian.org/debian-alpha/2005/05/msg00041.html

 From what I understand (which may well be a simplistic understanding of 
the real situation), the problem is that the drive connected to the 
controller is FAST-WIDE SCSI 2, but the controller can't do WIDE. The 
driver tries to set it to WIDE, then send some data to see if it works, 
and should fail (the second if in the code in the above page) but 
doesn't, so the controller is left in WIDE mode even though it doesn't 
support it and that's why it gives those messages. The mystery is why 
the WIDE mode transfer doesn't fail as it should.

What I've done is comment out the whole section of code (starting with 
the if right after the /* test width */ comment and ending with the 
closing curly brace for that if) and then compile the kernel. That way 
the driver never sets the controller to WIDE mode at all. Works great. 
I'd really like someone to find a more sophisticated solution that 
commenting out the whole lot, maybe fixing the test that should fail but 
doesn't? Unfortunately my knowledge of Linux kernel drivers is too 
limited to do that.

This modification affects the SCSI driver, though, which in my case 
isn't a problem since I'm only using this controller, but if your 
machine has another controller that should support WIDE mode, then this 
change will probably affect it too... Watch out for that.

I wonder also why the patch YosWinK pointed to affects a totally 
different source file... I'm not sure it's the same problem, because I'm 
on the RedHat-Alpha mailing list too and haven't heard of this specific 
problem. Perhaps it addresses another problem in the 53C810 code on Alpha.

Other than that minor problem, I've been running a 2.6 kernel for a 
couple months now on my AlphaXL 366 with a 50GB SCSI-2 drive. Runs 
great. Of course, it's a single processor machine so I can't help you 
with the other problem you're having.

Hope this helps,

J-S

-- 
___________________________________________
Jean-Sébastien Guay   jean_seb@videotron.ca
             http://whitestar02.webhop.org/

-- 
gentoo-alpha@gentoo.org mailing list