mbox series

[ibacm,0/4] Allocate EP addresses dynamically

Message ID cover.1542985774.git.Haakon.Bugge@oracle.com (mailing list archive)
Headers show
Series Allocate EP addresses dynamically | expand

Message

Haakon Bugge Nov. 23, 2018, 3:40 p.m. UTC
This patch series is based on the patch series "Ibacm fix address
compare issues", which currently is a PR in github.

This in order to be able to test the series and avoid too much
conflict resolution.

The series removes the restriction that the number of addresses
associated with an end-point is limited to four (4).

Documentation is updated.

The series (*) has been tested by adding 250 IP addresses to an ipoib
device whilst ibacm is running, reading them out with "ib_acme -e",
thereafter deleting them. It is also tested by having the 250 addresses
already present when ibacm starts.

No regressions have been detected by our test suite.

* This series on top of the github's PRs "Ibacm fix address compare
  issues" (PR #432) and "ibacm: Fix improper refcnt" (PR #431).



Håkon Bugge (4):
  ibacm: Allocate end-point addresses dynamically - v2
  ibacm: Allocate end-point addresses dynamically in provider - v2
  ibacm: Remove trailing blanks
  ibacm: Update documentation

 ibacm/man/ibacm.1          |  28 ++++----
 ibacm/prov/acmp/src/acmp.c |  55 +++++++++++-----
 ibacm/src/acm.c            | 155 +++++++++++++++++++++++++++++++--------------
 ibacm/src/acm_util.c       |   2 +-
 ibacm/src/libacm.c         |  50 ++++++++++-----
 5 files changed, 193 insertions(+), 97 deletions(-)

--
2.14.3

Comments

Haakon Bugge Nov. 30, 2018, 5:53 p.m. UTC | #1
We discussed previously on this list whether an ibacm endpoint should be associated with the Port GUID or the Node GUID. The latter has an issue when for example VFs are enabled, since the VFs associated addresses will still be related to Node GUID. Since ibacm has a fixed number of end-point addresses(4), this is a problem.

In the previous discussion, a concern about backward compatibility of ibacm's address file was raised, if the association was changed from Node GUID to Port GUID.

IIRC, Jason suggested to "simply" allocate the end-point addresses dynamically.

Well, it was not straight forward, but now its done.

Reviews/comments welcome!


Thxs, Håkon




> On 23 Nov 2018, at 16:40, Håkon Bugge <haakon.bugge@oracle.com> wrote:
> 
> This patch series is based on the patch series "Ibacm fix address
> compare issues", which currently is a PR in github.
> 
> This in order to be able to test the series and avoid too much
> conflict resolution.
> 
> The series removes the restriction that the number of addresses
> associated with an end-point is limited to four (4).
> 
> Documentation is updated.
> 
> The series (*) has been tested by adding 250 IP addresses to an ipoib
> device whilst ibacm is running, reading them out with "ib_acme -e",
> thereafter deleting them. It is also tested by having the 250 addresses
> already present when ibacm starts.
> 
> No regressions have been detected by our test suite.
> 
> * This series on top of the github's PRs "Ibacm fix address compare
>  issues" (PR #432) and "ibacm: Fix improper refcnt" (PR #431).
> 
> 
> 
> Håkon Bugge (4):
>  ibacm: Allocate end-point addresses dynamically - v2
>  ibacm: Allocate end-point addresses dynamically in provider - v2
>  ibacm: Remove trailing blanks
>  ibacm: Update documentation
> 
> ibacm/man/ibacm.1          |  28 ++++----
> ibacm/prov/acmp/src/acmp.c |  55 +++++++++++-----
> ibacm/src/acm.c            | 155 +++++++++++++++++++++++++++++++--------------
> ibacm/src/acm_util.c       |   2 +-
> ibacm/src/libacm.c         |  50 ++++++++++-----
> 5 files changed, 193 insertions(+), 97 deletions(-)
> 
> --
> 2.14.3
>