[PATCH] freeimpi in embedded world

[Fillod Stephane-2]

Hi,

I don't know whether it's been done already, me and a colleague,
we're trying to use freeipmi in an embedded machine. The machine
is running Linux 2.6 nodes on powerpc SoC. The cross-compilation
tool chain is based on uClibc, and we have to keep the library
dependency low.

Please find attached a patch against freeipmi 0.7.10 with following
changes:

* Allow FreeIPMI to build without libgcrypt and libgpg-error.
  This proves useful for embedded machines using only SSIF/openipmi.
  Feel free to comment on the way how to make libgcrypt optional.
  BTW, there was a typo in configure.ac: have_gcrypte_header.
* Check for cbrt(), which may be missing in uClibc.
* common/src/toolcommon/tool-cmdline-common.c: no need to check X_OK
  file access right upon driver device, this is too restrictive.

Thanks for making available freeimpi.
--
Stephane


_______________________________________________
Freeipmi-devel mailing list
Freeipmi-devel@...
http://lists.gnu.org/mailman/listinfo/freeipmi-devel

[Al Chu11]

Hi Fillod,

Thanks for the patch and a number of the fixes within it.  Thinking
about it a little bit, instead of just checking for gcrypt, I'm actually
thinking a compile option for "in-band only" might be more appropriate.
I assume you won't do IPMI over LAN at all??  If so, I can begin working
on a patch for 0.7.11 for this.  Please LMK.

I'll get the remaining parts of your patch in the HEAD and 0.7.X branch.

Thanks,

Al

On Thu, 2009-07-09 at 17:41 +0200, Fillod Stephane wrote:

> Hi,
>
> I don't know whether it's been done already, me and a colleague,
> we're trying to use freeipmi in an embedded machine. The machine
> is running Linux 2.6 nodes on powerpc SoC. The cross-compilation
> tool chain is based on uClibc, and we have to keep the library
> dependency low.
>
> Please find attached a patch against freeipmi 0.7.10 with following
> changes:
>
> * Allow FreeIPMI to build without libgcrypt and libgpg-error.
>   This proves useful for embedded machines using only SSIF/openipmi.
>   Feel free to comment on the way how to make libgcrypt optional.
>   BTW, there was a typo in configure.ac: have_gcrypte_header.
> * Check for cbrt(), which may be missing in uClibc.
> * common/src/toolcommon/tool-cmdline-common.c: no need to check X_OK
>   file access right upon driver device, this is too restrictive.
>
> Thanks for making available freeimpi.
> _______________________________________________
> Freeipmi-devel mailing list
> Freeipmi-devel@...
> http://*lists.gnu.org/mailman/listinfo/freeipmi-devel

--
Albert Chu
chu11@...
Computer Scientist
High Performance Systems Division
Lawrence Livermore National Laboratory



_______________________________________________
Freeipmi-devel mailing list
Freeipmi-devel@...
http://lists.gnu.org/mailman/listinfo/freeipmi-devel

[Al Chu11]

Hey Fillod,

Ugh.  I began to work on a patch for "--without-ipmioverlan" (as I
called it), but ultimately, I'm finding this patch too cumbersome to
write.  LAN/networking code is *everywhere* and the #ifdef-ing is
ridiculously invasive.

So, for now, I think I'm going to clean up your original patch to make a
configure option "--without-encryption", which will not link gcrypt and
effectively make IPMI 2.0 useless.  Users who run the tools/use the libs
under this conditions will get some type of error.  I'll try to get a
tar.gz out to you semi-soon to see if it's fine.

Al

On Thu, 2009-07-09 at 09:47 -0700, Al Chu wrote:

> Hi Fillod,
>
> Thanks for the patch and a number of the fixes within it.  Thinking
> about it a little bit, instead of just checking for gcrypt, I'm actually
> thinking a compile option for "in-band only" might be more appropriate.
> I assume you won't do IPMI over LAN at all??  If so, I can begin working
> on a patch for 0.7.11 for this.  Please LMK.
>
> I'll get the remaining parts of your patch in the HEAD and 0.7.X branch.
>
> Thanks,
>
> Al
>
> On Thu, 2009-07-09 at 17:41 +0200, Fillod Stephane wrote:
> > Hi,
> >
> > I don't know whether it's been done already, me and a colleague,
> > we're trying to use freeipmi in an embedded machine. The machine
> > is running Linux 2.6 nodes on powerpc SoC. The cross-compilation
> > tool chain is based on uClibc, and we have to keep the library
> > dependency low.
> >
> > Please find attached a patch against freeipmi 0.7.10 with following
> > changes:
> >
> > * Allow FreeIPMI to build without libgcrypt and libgpg-error.
> >   This proves useful for embedded machines using only SSIF/openipmi.
> >   Feel free to comment on the way how to make libgcrypt optional.
> >   BTW, there was a typo in configure.ac: have_gcrypte_header.
> > * Check for cbrt(), which may be missing in uClibc.
> > * common/src/toolcommon/tool-cmdline-common.c: no need to check X_OK
> >   file access right upon driver device, this is too restrictive.
> >
> > Thanks for making available freeimpi.
> > _______________________________________________
> > Freeipmi-devel mailing list
> > Freeipmi-devel@...
> > http://**lists.gnu.org/mailman/listinfo/freeipmi-devel

--
Albert Chu
chu11@...
Computer Scientist
High Performance Systems Division
Lawrence Livermore National Laboratory



_______________________________________________
Freeipmi-devel mailing list
Freeipmi-devel@...
http://lists.gnu.org/mailman/listinfo/freeipmi-devel

[Al Chu11]

Hey Fillod,

Beta is released here:

http://ftp.zresearch.com/pub/freeipmi/qa-release/freeipmi-0.7.11.beta1.tar.gz

LMK if it's ok for you guys.  The configure option is
--without-encryption.

Al

On Thu, 2009-07-09 at 15:17 -0700, Al Chu wrote:

> Hey Fillod,
>
> Ugh.  I began to work on a patch for "--without-ipmioverlan" (as I
> called it), but ultimately, I'm finding this patch too cumbersome to
> write.  LAN/networking code is *everywhere* and the #ifdef-ing is
> ridiculously invasive.
>
> So, for now, I think I'm going to clean up your original patch to make a
> configure option "--without-encryption", which will not link gcrypt and
> effectively make IPMI 2.0 useless.  Users who run the tools/use the libs
> under this conditions will get some type of error.  I'll try to get a
> tar.gz out to you semi-soon to see if it's fine.
>
> Al
>
> On Thu, 2009-07-09 at 09:47 -0700, Al Chu wrote:
> > Hi Fillod,
> >
> > Thanks for the patch and a number of the fixes within it.  Thinking
> > about it a little bit, instead of just checking for gcrypt, I'm actually
> > thinking a compile option for "in-band only" might be more appropriate.
> > I assume you won't do IPMI over LAN at all??  If so, I can begin working
> > on a patch for 0.7.11 for this.  Please LMK.
> >
> > I'll get the remaining parts of your patch in the HEAD and 0.7.X branch.
> >
> > Thanks,
> >
> > Al
> >
> > On Thu, 2009-07-09 at 17:41 +0200, Fillod Stephane wrote:
> > > Hi,
> > >
> > > I don't know whether it's been done already, me and a colleague,
> > > we're trying to use freeipmi in an embedded machine. The machine
> > > is running Linux 2.6 nodes on powerpc SoC. The cross-compilation
> > > tool chain is based on uClibc, and we have to keep the library
> > > dependency low.
> > >
> > > Please find attached a patch against freeipmi 0.7.10 with following
> > > changes:
> > >
> > > * Allow FreeIPMI to build without libgcrypt and libgpg-error.
> > >   This proves useful for embedded machines using only SSIF/openipmi.
> > >   Feel free to comment on the way how to make libgcrypt optional.
> > >   BTW, there was a typo in configure.ac: have_gcrypte_header.
> > > * Check for cbrt(), which may be missing in uClibc.
> > > * common/src/toolcommon/tool-cmdline-common.c: no need to check X_OK
> > >   file access right upon driver device, this is too restrictive.
> > >
> > > Thanks for making available freeimpi.
> > > _______________________________________________
> > > Freeipmi-devel mailing list
> > > Freeipmi-devel@...
> > > http://***lists.gnu.org/mailman/listinfo/freeipmi-devel

--
Albert Chu
chu11@...
Computer Scientist
High Performance Systems Division
Lawrence Livermore National Laboratory



_______________________________________________
Freeipmi-devel mailing list
Freeipmi-devel@...
http://lists.gnu.org/mailman/listinfo/freeipmi-devel

[Fillod Stephane-2]

Hi Al,

Thank you for your impressive responsiveness.

Al Chu wrote:
[...]
> Ugh.  I began to work on a patch for "--without-ipmioverlan" (as I
> called it), but ultimately, I'm finding this patch too cumbersome to
> write.  LAN/networking code is *everywhere* and the #ifdef-ing is
> ridiculously invasive.

Indeed.

> So, for now, I think I'm going to clean up your original patch to make
a
> configure option "--without-encryption", which will not link gcrypt
and
> effectively make IPMI 2.0 useless.  Users who run the tools/use the
libs
> under this conditions will get some type of error.  I'll try to get a
> tar.gz out to you semi-soon to see if it's fine.

This new configure option is very appropriate since we won't do IPMI
over LAN at all. I have downloaded 0.7.11.beta1, and it cross-builds
fine, thanks!

Some patch for doing in-band only (ipmi_ssif_cmd_api_ipmb) will follow,
but we will have first to discuss the best way to turn the Linux kernel
into an I2C multi-master with slave receive handling. If anyone is
interested, please LMK. I'll start a new thread next week.

Cheers
--
Stephane

PS to all readers: sorry, my MUA stinks and doesn't honour References:
mail
header.


_______________________________________________
Freeipmi-devel mailing list
Freeipmi-devel@...
http://lists.gnu.org/mailman/listinfo/freeipmi-devel