On 5/1/05, Sergey Vlasov <vsu@altlinux.ru> wrote:
>> Hello!
>>
>> http://openipmi.sourceforge.net/ provides patches for 2.4.x kernels
>> with updated IPMI code. However, these patches change the IPMI driver
>> API by adding an additional argument to ipmi_request():
>>
>> int ipmi_request(ipmi_user_t user,
>> struct ipmi_addr *addr,
>> long msgid,
>> struct ipmi_msg *msg,
>> + void *user_msg_data,
>> int priority);
>>
>> (user_msg_data is later returned in the struct ipmi_recv_msg field).
>> This API change breaks compilation of some lm_sensors drivers
>> (i2c-ipmb, i2c-ipmi).
>>
>> The patch below adds compatibility with the new IPMI API to i2c-ipmb
>> and i2c-ipmi drivers. <linux/ipmi.h> does not provide a real
>> API-version #define, therefore IPMI_RESPONSE_RESPONSE_TYPE (also added
>> by the openipmi.sf.net patch) is (ab)used as a new API indicator.
>>
>> Compile tested only due to lack of hardware.
(Yani I.):
Hi Sergey,
Actually there was a patch to the 2.6 i2c-ipmi/bmcsensors for this
problem (after Bene Martin notified me of the problem and how to fix
it), but from what I understand Corey re-included the ipmi_request
method in the next version once he learned of a driver using it (it
was removed because there was thought to be no usage). He was eager to
learn when bmcsensors would make it into the kernel so that he could
stop being nagged by the 'unused function police' as he put it .
It might be wise to apply a patch like this in 2.4 to support the
openipmi that went 'wrong', I changed the 2.6 driver to use the
ipmi_request_settime function to no ill effect, but you should
consider that Corey notes that it is preferred to use the old one:
(extract from some e-mail between Corey, Martin and I about the problem):
>>>>>>Interestingly enough the ipmi_request_settime still has a comment in
>>>>>>the header "Don't use this unless you *really* have to.", making me
>>>>>>wonder if there is a reason not to be using it instead of ipmi_request
>>>>>>if its re-incorporated into the kernel.
>>>>>>
>>>>>>
>>
>>>>
>>>>I guess the only one qualified to answer that one would be corey.
>>>>
>>>>Bye, Martin
>>>>
>>>>
>
>>You wouldn't generally use it because it sets the timeouts and for most
>>purposes the timeouts are defined by the spec (sort of).
>>
>>Do you have any feeling when the BMC sensors package will go into the
>>mainstream kernel? That would be the time to coordinate adding the
>>normal function back in so I'm not constantly fighting the unused
>>function police.
>>
>>-Corey
git-svn-id: http://lm-sensors.org/svn/lm-sensors/trunk@2987 7894878c-1315-0410-8ee3-d5d059ff63e0
Module refcounting using .owner, removing of inc/dec_use.
These require (yet unreleased) i2c 2.8.0+.
Named initializers everywhere. Unified namespace.
Cleanup for 2.5, remove EXPORT_NO_SYMBOLS, declare everything
static. Apply initcalls.
PCI device registration changes for busses started, replicating
from i2c-amd*. While everything does compile, nothing really
works. Well, maybe i2c-amd* ported back from 2.5 tree.
After updating your work directory, run
> cvs diff -u -r LAST-PRE-2-8-I2C > lm-2.8-patches-1
BEWARE: I have not even insmoded any of these yet.
To compile, you need i2c from cvs too.
For 2.4 kernel :
> cvs co -r lk2-4 i2c
For 2.5 kernel :
> cvs co i2c
git-svn-id: http://lm-sensors.org/svn/lm-sensors/trunk@1698 7894878c-1315-0410-8ee3-d5d059ff63e0
Mark D. Studebaker