Skip to content

RFC IBM eHCA InfiniBand adapter driver

Roland Dreier lkml 1 version In review

Get series with b4
b4 am [email protected]

Versions newest first

  1. v1 RFC Latest21 of 22 patches 33 repliesIn review

21 of 22 patches are in this archive. The cover letter announces 22.

Patches in v1 No applies recorded yet

#Subject and filesSizeRepliesApplied
1/22Add powerpc-specific clear_cacheline(), which just compiles to "dcbz".drivers/infiniband/hw/ehca/ehca_asm.h+58 -02 replies
2/22Firmware interface code for IB device.drivers/infiniband/hw/ehca/hcp_if.h+2022 -017 replies
3/22pHype specific stuffdrivers/infiniband/hw/ehca/ehca_galpa.h drivers/infiniband/hw/ehca/hcp_phyp.c drivers/infiniband/hw/ehca/hcp_phyp.h+493 -02 replies
4/22OF adapter probingdrivers/infiniband/hw/ehca/hcp_sense.c drivers/infiniband/hw/ehca/hcp_sense.h+280 -03 replies
5/22HW register abstractionsdrivers/infiniband/hw/ehca/hipz_fns.h drivers/infiniband/hw/ehca/hipz_fns_core.h drivers/infiniband/hw/ehca/hipz_hw.h +1 more+642 -00 replies
6/22Queue handlingdrivers/infiniband/hw/ehca/ipz_pt_fn.c drivers/infiniband/hw/ehca/ipz_pt_fn.h drivers/infiniband/hw/ehca/ipz_pt_fn_core.h+454 -00 replies
7/22Hypercall definitionsdrivers/infiniband/hw/ehca/ehca_common.h+115 -01 reply
8/22Generic ehca headersdrivers/infiniband/hw/ehca/ehca_flightrecorder.h drivers/infiniband/hw/ehca/ehca_kernel.h drivers/infiniband/hw/ehca/ehca_tools.h+640 -01 reply
9/22ehca classesdrivers/infiniband/hw/ehca/ehca_classes.c drivers/infiniband/hw/ehca/ehca_classes.h drivers/infiniband/hw/ehca/ehca_classes_core.h +1 more+889 -00 replies
11/22ehca event queuesdrivers/infiniband/hw/ehca/ehca_eq.c drivers/infiniband/hw/ehca/ehca_eq.h+320 -00 replies
12/22ehca low-level verbsdrivers/infiniband/hw/ehca/ehca_iverbs.h drivers/infiniband/hw/ehca/ehca_qes.h+437 -00 replies
13/22HCA query functionsdrivers/infiniband/hw/ehca/ehca_hca.c+321 -00 replies
14/22ehca completion queue handlingdrivers/infiniband/hw/ehca/ehca_cq.c+416 -00 replies
15/22ehca queue pair handlingdrivers/infiniband/hw/ehca/ehca_qp.c+1528 -00 replies
16/22ehca post send/receive and poll CQdrivers/infiniband/hw/ehca/ehca_reqs.c drivers/infiniband/hw/ehca/ehca_reqs_core.c+821 -00 replies
17/22Special QP functionsdrivers/infiniband/hw/ehca/ehca_sqp.c+135 -00 replies
18/22ehca address vectors, multicast groups, protection domainsdrivers/infiniband/hw/ehca/ehca_av.c drivers/infiniband/hw/ehca/ehca_mcast.c drivers/infiniband/hw/ehca/ehca_pd.c+552 -00 replies
19/22ehca memory regionsdrivers/infiniband/hw/ehca/ehca_mrmw.c drivers/infiniband/hw/ehca/ehca_mrmw.h+2450 -00 replies
20/22ehca userspace verbsdrivers/infiniband/hw/ehca/ehca_uverbs.c+376 -00 replies
21/22ehca main filedrivers/infiniband/hw/ehca/ehca_main.c+1032 -04 replies
22/22ehca Makefile/Kconfig changesdrivers/infiniband/Kconfig drivers/infiniband/Makefile drivers/infiniband/hw/ehca/Kbuild +1 more+17 -00 replies

Reviews trailers per patch

No Reviewed-by, Acked-by or Tested-by trailers for v1 yet.

Read the v1 thread

Cover letter [PATCH 00/22] [RFC] IBM eHCA InfiniBand adapter driver

Here's a series of patches that add an InfiniBand adapter driver for IBM eHCA hardware. Please look it over with an eye towards issues that need to be addressed before merging this upstream.

This patch series is somewhat unusual in that I am not the original author of this driver -- I am just sending it for review for the authors, who are apparently not able to post patches themselves due to internal issues at IBM. However they are cc'ed and will respond to comments in this thread.

In fact I have some issues with the code myself that need to be addressed before this driver is mergeable. I've included most of them in the individual patches, although I have some general comments too. However I would like to get some early feedback for the ehca authors from the wider community. In particular I think its important to run this past the ppc64 experts, since I'm not sure what the standards for this sort of pSeries driver are.

Anyway, my general comments:

- The #ifs that test EHCA_USERDRIVER and __KERNEL__ should be killed. We know that this is kernel code, so there's no reason to include userspace compatibility junk.

- Many of the comments look like they are for some automatic documentation system that is not quite kerneldoc. They should be fixed to be real kerneldoc comments.

- In general there is a huge amount of code in large inline functions in .h files. Things should be reorganized to cut this down to a sane amount.

Thanks,
Roland