[openfabrics-ewg] Gen2 basic and resource tracking test fails for Pathscale HCA

[Sharma, Karun]

Hi Robert:
 
Thanks for acknowledging the issue.
I didn't reported other failures with gen2_basic tests, because Mellanox HCA card also fails those tests. Also, these are known failures mentioned in readme file.
 
Thanks
Karun

________________________________

From: Robert Walsh [mailto:rjwalsh at pathscale.com]
Sent: Tue 9/19/2006 7:34 PM
To: Sharma, Karun
Cc: dotanb at mellanox.co.il; openfabrics-ewg at openib.org
Subject: Re: [openfabrics-ewg] Gen2 basic and resource tracking test fails for Pathscale HCA



-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Sharma, Karun wrote:
> Hi Dotan:
> 
> I am testing Pathscale HCA with OFED1.1 RC5. I performed gen2_basic,
> qp_test, multicast_test, mem_lock_check and resource_tracking tests.
> Following are the failures encountered.
> 
> 1. gen2_basic test fails for POLL_POST component. With Mellanox HCA,
> this is passing.
> 2. When I specify -n option with resource_tracking_test, the test fails
> for Pathscale HCA. Again it passes for Mellanox HCA.
> 
> I have attached the logs below.

Hi Karun,

We've run the gen2_basic tests here and have fixed a number of problems
in the test suite itself.  I will be sending patches out shortly, after
I've had time to review them.  We now have most tests passing.  The
tests that still fails are:

  AV (try to create max ah + 1)
  CQ (resize CQ to more than initial size)
  CQ (modify with outstanding WR to more than outstanding)
  POLL_POST (post sr to RC QP)
  QP (test 3, test 5 and test 6)

The first test fails because the ib_mad module has an AH already
created, so we cannot get as far as creating max AH.  I'm not sure how
to fix this one, but that seems like a test shortcoming.

The second and third tests fail because, for reasons I haven't figured
out yet, resize_cq support got dropped from our libipathverbs.  I'm
still tracking this down and hope to have a fix for RC6.

I haven't figured out yet why the fourth test fails.  It's looking like
a memory corruption problem (stack smashing), and I think it's in the
test itself, but I haven't figured it out for certain yet.

The QP tests are failing because of core dumps in the tests.  Again, I
think these are failures in the tests themselves, but I haven't figured
them out yet.

We have not yet run a full resource test, as we had some early problems
with that running out of memory.  We are still working on this.

Regards,
 Robert.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.5 (GNU/Linux)
Comment: Using GnuPG with Fedora - http://enigmail.mozdev.org <http://enigmail.mozdev.org/> 

iQEVAwUBRRB+b/zvnpzTd9fxAQKcjwgApa3l92R6FLv9D37zlHf6hPwqJNoOtGgG
uAGYfMWpfOai0cHx0fgTNTEFBxxeFPdTcG6CEEmyO53piNDUxEpB2OqBJSd5dBys
sKjHNscmmGxktHaUPnuK1MF9defvCoGv0Wd+yGBhjuzhazNaOlLBRVkXPXpe9fX7
UMeCdiRgGOoTbcAUt+Rwq8eDYNIkCO7U+EapsRxeIsgpFDkiVQ/RsQsh+QSfLcQb
SX6sY0tm+WZx7WHesOcHGP9i8JtIb7rUNf1n05xhmdaaaEgp/kR/cYIROvwoI+Oq
63+z95vDGFKG0BM3IAvelSCQL/wqLm9QlzsPTLGOgPyE5ymnyF4uhw==
=2Lk6
-----END PGP SIGNATURE-----


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openfabrics.org/pipermail/ewg/attachments/20060920/072ecdac/attachment.html>