On 5/18/08, <b class="gmail_sendername">Vladimir Sokolovsky</b> <<a href="mailto:vlad@dev.mellanox.co.il">vlad@dev.mellanox.co.il</a>> wrote:<div><span class="gmail_quote"></span><blockquote class="gmail_quote" style="border-left: 1px solid rgb(204, 204, 204); margin: 0pt 0pt 0pt 0.8ex; padding-left: 1ex;">
There is a bug in the OFED 1.3 mlx4 driver in mlx4_alloc_fmr which hardcoded<br>
the minimum acceptable page_shift to be 12. However, new mlx4 firmware has a<br>
minimum page_shift of 9 (log_pg_sz of 9 returned by QUERY_DEV_LIM) -- so that<br>
ib_fmr_alloc fails for ULPs using the device minimum when creating FMRs.<br>
<br>
To preserve firmware compatibility with released OFED drivers, the firmware<br>
will continue to return 12 as before for log_page_sz in QUERY_DEV_CAP for these<br>
drivers.</blockquote><div><br>
Hi Vlad, Roland,<br>
<br>
To start with, the bug is in the Linux kernel mlx4 driver, there's
nothing like "OFED 1.3 mlx4 driver" (who's the maintainer? why there's
need to be another instace of a driver merged into the mainline kerel,
etc). This bug was fixed last week or so by a patch sent by someone
from Mellanox.<br>
<br>
To continue with, maybe we just state that kernels < X = 2.6.26 are
not compatible with FW version > Y = 2.3? or have the patch that
fixes the problem be sent to -stable versions of older kernels?<br>
<br>
If those solutions are not enough, I think that the default behaviour
of FW AND the mainline driver would be to get the actual minimal driver
supported, namely nine. <br>
</div></div><br>
Or.<br>