[ipxe-devel] arbel sanboot ib_srp

Michael Brown mbrown at fensystems.co.uk
Thu May 19 08:34:11 UTC 2011


On Wednesday 18 May 2011 20:28:43 Lee Staples wrote:
> Have just reinstalled SCST and built latest ipxe commit
> (c49659c4f26e23f3fc234c2068786872554daa69) with the arbel queue pair
> patch from last year.
> 
> Would be grateful if you could take a look at the error I'm getting and
> advise as it appears to login successfully to the server
> 
>    Could not open SAN device: Input/output error (http://ipxe.org/1d714039)
>    srp boot failedCMRC 0x23354 shutting down
>    Arbel 0x215b4 issuing command 0021
>    CMRC 0x23354 send error: Operation canceled (http://ipxe.org/0b1360a0)

Login is handled via a pair of management datagrams.  The RC queue pair is not 
used until the first packet after login completes.  Your client.log shows that 
the RC queue pair is transitioning into an error state as soon as the first 
send WQE is posted:

  QPN 0xd75405 context before doorbell:
  Arbel 0x215b4 issuing command 0022
  Arbel 0x215b4 QPN 0xd75405 context:
  00000000 : 30 00 19 00 ff 3e 3f 16-9f 0a 13 00 00 00 00 01
  ...
  QPN 0xd75405 context after doorbell:
  Arbel 0x215b4 issuing command 0022
  Arbel 0x215b4 QPN 0xd75405 context:
  00000000 : 60 00 19 00 f8 3c 28 14-9f 0a 13 00 00 00 00 01
  ...

The first nibble of this hex dump is the queue state - 3="ready to send", 
6="error".

Something must be wrong with either the queue pair context or the send WQE 
created by iPXE for Arbel.  I have examined both in excruciating detail the 
last time I worked on this, and I couldn't find any problem.


Itay: is there a debug version of the firmware available that would provide 
some indication of why the QP is transitioning to ERR?

Michael



More information about the ipxe-devel mailing list