[openib-general] Why doesn't the 24-kernel actually compile on 2.4 kernels?

Thu Apr 1 17:36:31 PST 2004

    Mike> 1) There is the redefinition of PDE in
    Mike> infiniband/include/ts_kernel_services.h (easy to work
    Mike> around).

Hmm... didn't realize kernel 2.4.25 added this.  I'll have to put an
#ifdef around that code (since older kernels did not define PDE).

Looks like the Mellanox driver in the drop is out of date, which
explains problems 2 and 3.

    Mike> 4) ipoib fails to build because it cannot find
    Mike> ./TS_OBJ_DIR/ipoib_export.ver,
    Mike> ../core/TS_OBJ_DIR/core_export.ver,
    Mike> ../sa_client/TS_OBJ_DIR/sa_client_export.ver,
    Mike> ../client_query/TS_OBJ_DIR/client_query_export.ver, or
    Mike> ../TS_OBJ_DIR/services_export.ver

I don't see how this is possible... drivers/infiniband/ulp/ipoib/Makefile
define "-DIN_TREE_BUILD", which should turn these includes into an
include of <linux/modversions.h>.  It's an ugly hack to graft to deal
with warts dating back to our internal build system, but I'm not sure
how it's breaking for you.  Can you post the gcc command line that is
actually failing?

    Mike> This is not an issue with the drivers on Mellanox's ftp
    Mike> site...  version ~3.2.  Is the MOSAL code in this stack that
    Mike> out of date?  We probably need a resync on this...

Yes, my fault for including out-of-date Mellanox code.  Once we get
the Subversion repository up, we can fix this (I'm reluctant to post
more tarballs since it's not a very good way to work).

By the way, I also have module unloading fixed in my tree.  It was
just a simple matter of adding try_module_get() calls to bump the
reference count of the HW driver.  (There were no deeper issues, just
a pure module unloading problem)

 - Roland

-- 
To unsubscribe send an email with subject unsubscribe to openib-general at openib.org.
Please contact moderator at openib.org for questions.