InfiniBand core update
PATCH 1/1 InfiniBand core update
This is version 2 of a patch series to get the Infiniband core up to date. Aside from bug fixes, the following new functionality is also introduced:
MAD (Management Datagram) support for RMPP
(Reliable MultiPacket Protocol)
CM (Communications Manager) support User CM support
User MAD changes for RMPP
Service Record support in SA query
Thanks.
-- Hal
Quoted 3 lines · Hal Rosenstock
This is version 2 of a patch series to get the Infiniband core up to date.
Well that was interesting.
- All the patches had mangled headers:
-- linux-2.6.13-rc2-mm1-16/...
+++ linux-2.6.13-rc2-mm1-17/...
instead of
--- linux-2.6.13-rc2-mm1-16/...
+++ linux-2.6.13-rc2-mm1-17/...
Which I fixed up.
- The second patch didn't apply. I fixed that too.
- The patches add lots of trailing whitespace. I habitually trim that off, figuring that any downstream merging hassle which that causes is just punishment ;)
Please check that it all landed OK in the next -mm.
I've hung on to Tom Duffy's patch pending a test compilation.
I'll tentatively consider this material to be not-for-2.6.13?
...
Quoted 16 lines · Andrew Morton
Quoted 3 lines · Hal Rosenstock
This is version 2 of a patch series to get the Infiniband core up to date.
Well that was interesting.
- All the patches had mangled headers:
-- linux-2.6.13-rc2-mm1-16/...
+++ linux-2.6.13-rc2-mm1-17/...instead of
... +0
-0
Not sure how that happened.
Quoted 1 line
Which I fixed up.
Thanks.
Quoted 1 line
- The second patch didn't apply. I fixed that too.
Not sure how this was broken. Thanks for fixing this.
Quoted 3 lines
- The patches add lots of trailing whitespace. I habitually trim that off, figuring that any downstream merging hassle which that causes is just punishment ;)
I'll work on merging this back downstream :-(
Quoted 1 line
Please check that it all landed OK in the next -mm.
Will do.
Quoted 3 lines
I've hung on to Tom Duffy's patch pending a test compilation.
I'll tentatively consider this material to be not-for-2.6.13?
Presuming that "this material" refers to the patch to add the kernel CM implementation, if kernel CM does not make 2.6.13, then user CM should not either as it is dependent on it.
-- Hal
Quoted 6 lines · Hal Rosenstock
Quoted 2 lines
I'll tentatively consider this material to be not-for-2.6.13?
Presuming that "this material" refers to the patch to add the kernel CM implementation, if kernel CM does not make 2.6.13, then user CM should not either as it is dependent on it.
Well I was asking. Do you guys think that this material is appropriate to and safe enough for 2.6.13?
What are "user CM" and "kernel CM"?
Quoted 2 lines · Andrew Morton
Well I was asking. Do you guys think that this material is appropriate to and safe enough for 2.6.13?
We think so.
Quoted 1 line
What are "user CM" and "kernel CM"?
CM is the InfiniBand Communications Manager. It is (primarily) responsible for setting up and managing connections between endpoints. Connections. IB communication occurs over the queue pairs (QPs) set up by the CM protocol.
kernel CM is CM functionality in the kernel. user CM is to enable userspace applications to utilize this functionality.
-- Hal
Quoted 2 lines · Andrew Morton
Well I was asking. Do you guys think that this material is appropriate to and safe enough for 2.6.13?
I used your versions of the patches (Tom's ucm one is needed and you added that). I also back ported the trailing whitespace elimination changes.
I just completed a regression test including uCM, CM, RMPP, OpenSM, IPoIB and it looks good to me.
I'll check this again when the next -mm comes out.
Thanks.
-- Hal
Quoted 5 lines · Hal Rosenstock
Quoted 2 lines · Andrew Morton
Well I was asking. Do you guys think that this material is appropriate to and safe enough for 2.6.13?
Quoted 7 lines
I used your versions of the patches (Tom's ucm one is needed and you added that). I also back ported the trailing whitespace elimination changes.
I just completed a regression test including uCM, CM, RMPP, OpenSM, IPoIB and it looks good to me.
OK, well the timing of a merge is mainly up to you guys, especially as the subsystem is pretty raw and you're the only people who use it ;)
Two things from a quick scan:
a) In many places the patch does
if (p) kfree(p);
But kfree(0) is permitted. The cleanup police will be after you at some stage - it'd be best to fix those things up immediately.
b) The patch exports a ton of symbols to non-GPL modules:
+EXPORT_SYMBOL(ib_create_ah_from_wc);
+EXPORT_SYMBOL(ib_modify_mad);
+EXPORT_SYMBOL(ib_create_send_mad);
+EXPORT_SYMBOL(ib_free_send_mad);
+EXPORT_SYMBOL(ib_coalesce_recv_mad);
+EXPORT_SYMBOL(ib_sa_service_rec_query);
+EXPORT_SYMBOL(ib_create_cm_id);
+EXPORT_SYMBOL(ib_destroy_cm_id);
+EXPORT_SYMBOL(ib_cm_listen);
+EXPORT_SYMBOL(ib_send_cm_req);
+EXPORT_SYMBOL(ib_send_cm_rep);
+EXPORT_SYMBOL(ib_send_cm_rtu);
+EXPORT_SYMBOL(ib_send_cm_dreq);
+EXPORT_SYMBOL(ib_send_cm_drep);
+EXPORT_SYMBOL(ib_send_cm_rej);
+EXPORT_SYMBOL(ib_send_cm_mra);
+EXPORT_SYMBOL(ib_send_cm_lap);
+EXPORT_SYMBOL(ib_send_cm_apr);
+EXPORT_SYMBOL(ib_send_cm_sidr_req);
+EXPORT_SYMBOL(ib_send_cm_sidr_rep);
+EXPORT_SYMBOL(ib_cm_establish);
+EXPORT_SYMBOL(ib_cm_init_qp_attr);
Why?
Quoted 5 lines · Andrew Morton
OK, well the timing of a merge is mainly up to you guys, especially as the subsystem is pretty raw and you're the only people who use it ;)
Two things from a quick scan:
Quoted 7 lines
a) In many places the patch does
if (p) kfree(p);But kfree(0) is permitted. The cleanup police will be after you at some stage - it'd be best to fix those things up immediately.
I'll/We'll work on eradicating these and pushing a patch for this upstream.
Quoted 16 lines
b) The patch exports a ton of symbols to non-GPL modules:
(quoted diff) +14
-0+EXPORT_SYMBOL(ib_create_ah_from_wc); +EXPORT_SYMBOL(ib_modify_mad); +EXPORT_SYMBOL(ib_create_send_mad); +EXPORT_SYMBOL(ib_free_send_mad); +EXPORT_SYMBOL(ib_coalesce_recv_mad); +EXPORT_SYMBOL(ib_sa_service_rec_query); +EXPORT_SYMBOL(ib_create_cm_id); +EXPORT_SYMBOL(ib_destroy_cm_id); +EXPORT_SYMBOL(ib_cm_listen); +EXPORT_SYMBOL(ib_send_cm_req); +EXPORT_SYMBOL(ib_send_cm_rep); +EXPORT_SYMBOL(ib_send_cm_rtu); +EXPORT_SYMBOL(ib_send_cm_dreq); +EXPORT_SYMBOL(ib_send_cm_drep);
Review
Quoted 10 lines
(quoted diff) +8
-0+EXPORT_SYMBOL(ib_send_cm_rej); +EXPORT_SYMBOL(ib_send_cm_mra); +EXPORT_SYMBOL(ib_send_cm_lap); +EXPORT_SYMBOL(ib_send_cm_apr); +EXPORT_SYMBOL(ib_send_cm_sidr_req); +EXPORT_SYMBOL(ib_send_cm_sidr_rep); +EXPORT_SYMBOL(ib_cm_establish); +EXPORT_SYMBOL(ib_cm_init_qp_attr); Why?
I think that is because OpenIB has a dual GPL/BSD license.
-- Hal
Please acknowledge that you understand how inappropriate and problem causing your huge patch bomb was today to this mailing list.
It is nearly 8 hours later, and vger.kernel.org is still trying mightily to spit out all of your patches to the 5000+ people subscribed to linux-kernel.
There is about ~6 hour posting latency as a result of all of this.
Quoted 10 lines · David S. Miller
Please acknowledge that you understand how inappropriate and problem causing your huge patch bomb was today to this mailing list.
It is nearly 8 hours later, and vger.kernel.org is still trying mightily to spit out all of your patches to the 5000+ people subscribed to linux-kernel.
There is about ~6 hour posting latency as a result of all of this.
I understand the problem it caused. The MAD layer of InfiniBand was quite a bit behind and we wanted to catch up. I tried to make them in smaller manageable pieces and that was why there were so many patches in the series. In the future, I will endeavor to make the changes smaller. Sorry for the problems this caused.
-- Hal
NO!!!! No more huge patch bombs to [email protected] people! Condense your changes into more manageable pieces or only submit smaller groups of changes at a time.
You people doing this are absolutely killing vger.kernel.org.
Every patch you post has to go to 5000+ subscribers, so please keep this in mind when posting multiple patches. If you feel the need to send, say, more than 15 patches at once, reconsider.