[openib-general] libehca causes segfault when not physically present..

[Christoph Raisch]

The secret of 0036 was that we managed to build a driver which didn't work 
in all cases.

I would guess in these changes
>  33 files changed, 945 insertions(+), 1163 deletions(-)

There are about ~80 lines of new code, rest of all that were modifications 
which don't change any algorithm but are desperately needed to be kernel 
coding style compliant:
That's the mostly complete list what we've changed
removing+renaming already existing assembly macros in ehca_asm.h
changed the ehca_module pointer to a ehca_module struct
removed EHCA_MEMPAGESIZE
replaced quite a lot of typedef struct by struct
capitalize DEFINES, changed most struct members to small letters
replaced all ehca_retcode_t by u64
replaced the  ehca_sleep() by appropriate kernel function
replaced the assert() by BUG_ON()
replaced ntohd()
some naming and comment cleanup on struct hcp_modify_qp_control_block

Roland, in case you're missing some some changes in there, we'll add these 
to one of the next releases to seperate the
coding style cleanups from the functional changes.


Gruss / Regards . . . Christoph R.


Roland Dreier <> wrote on 03.11.2005 20:13:58:

>     Heiko> this bug should be fixed in OpenIB trunk 3960.
> 
> It's good to see this fixed and all the other cleanups in this
> checkin.  I'll have to go back to my ehca code reviewing....
> 
> However, when this code moves upstream, you'll have to make your
> changes in smaller digestible chunks.  The diff between r3959 and
> r3960 is rather gigantic:
> 
>  33 files changed, 945 insertions(+), 1163 deletions(-)
> 
> And this piece:
> 
>     > -MODULE_VERSION("EHCA2_0035");
>     > +MODULE_VERSION("EHCA2_0037");
> 
> indicates that there was a 0036 that you never let anyone see.
> 
> I would suggest you try to use the openib.org svn tree as your real
> development repository.  This will be the way you will have to work
> once your driver is in the upstream kernel, and even now you will get
> benefit from getting better patch review and having users better able
> to pin down when a regression might have been introduced.
> 
> For your latest checkin, it would have been better to see a series of
> changesets with commit logs like:
> 
>     - remove asm_sync_mem() and mftb(), which duplicate existing
>       definitions in include/asm-ppc64
>     - make sure device is an eHCA in libehca's openib_driver_init()
>     - update Kconfig help text
> 
> and so on...
> 
> Thanks,
>   Roland
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openfabrics.org/pipermail/general/attachments/20051104/8e843805/attachment.html>