refresh (create temporary patch)

refresh (create temporary patch)

Refresh of errmsg

new

new

librdmacm: Report error level in error messages

Report error messages as either 'Warning' or 'Fatal'.

Signed-off-by: Yann Droneaud <ydroneaud@opteya.com>
Signed-off-by: Sean Hefty <sean.hefty@intel.com>

refresh

refresh

refresh (create temporary patch)

refresh (create temporary patch)

Refresh of pfx

librdmacm: Use common prefix for error messages

Signed-off-by: Yann Droneaud <ydroneaud@opteya.com>
Signed-off-by: Sean Hefty <sean.hefty@intel.com>

new

new

librdmacm: Use common prefix for error messages

Signed-off-by: Yann Droneaud <ydroneaud@opteya.com>
Signed-off-by: Sean Hefty <sean.hefty@intel.com>

refresh

refresh

refresh (create temporary patch)

refresh (create temporary patch)

Refresh of stder

refresh

refresh

refresh (create temporary patch)

refresh (create temporary patch)

Refresh of stder

librdmacm: Report error messages on stderr

Signed-off-by: Yann Droneaud <ydroneaud@opteya.com>
Signed-off-by: Sean Hefty <sean.hefty@intel.com>

librdmacm: Report error messages on stderr

Signed-off-by: Yann Droneaud <ydroneaud@opteya.com>
Signed-off-by: Sean Hefty <sean.hefty@intel.com>

new

new

librdmacm: Report error messages on stderr

Signed-off-by: Yann Droneaud <ydroneaud@opteya.com>
Signed-off-by: Sean Hefty <sean.hefty@intel.com>

pop

pop

pop

pop

refresh

refresh

refresh (create temporary patch)

refresh (create temporary patch)

Refresh of af_ib

refresh

refresh

refresh (create temporary patch)

refresh (create temporary patch)

Refresh of af_ib

refresh

refresh

refresh (create temporary patch)

refresh (create temporary patch)

Refresh of af_ib

librdmacm: Enable AF_IB support

Signed-off-by: Sean Hefty <sean.hefty@intel.com>

librdmacm: Enable AF_IB support

Signed-off-by: Sean Hefty <sean.hefty@intel.com>

librdmacm: Enable AF_IB support

Signed-off-by: Sean Hefty <sean.hefty@intel.com>

pop

pop

librdmacm: Enable AF_IB support

Signed-off-by: Sean Hefty <sean.hefty@intel.com>

librdmacm: Set address family for source address returned by ACM

Set the sa_family type when saving the source address returnd
by ACM.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>

pop

pop

pop

pop

pop

pop

refresh

refresh

refresh (create temporary patch)

refresh (create temporary patch)

Refresh of sa_family

refresh

refresh

refresh (create temporary patch)

refresh (create temporary patch)

Refresh of sa_family

librdmacm: Set address family for source address returned by ACM

Set the sa_family type when saving the source address returnd
by ACM.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>

new

new

librdmacm: Set address family for source address returned by ACM

Set the sa_family type when saving the source address returnd
by ACM.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>

pop

pop

commit

commit

pop

pop

librdmacm: Enable AF_IB support

Signed-off-by: Sean Hefty <sean.hefty@intel.com>

pop

pop

refresh

refresh

refresh (create temporary patch)

refresh (create temporary patch)

Refresh of fork-fix

refresh

refresh

refresh (create temporary patch)

refresh (create temporary patch)

Refresh of fork-fix

rspreload: Avoid rsocket calls until after fork

When an rsocket call is made before an application calls fork(),
the forked applications can hang.  This can be seen by running
netserver and two netperf clients simultaneously.  The second
netperf client will eventually stop performing data transfers.

LD_PRELOAD=librspreload.so netserver -D

LD_PRELOAD=librspreload.so netperf -v2 -c -C -H 192.168.0.101 -l30
LD_PRELOAD=librspreload.so netperf -v2 -c -C -H 192.168.0.101 -l30

It's not clear what the specific problem is.  The best guess is
that libibverbs or the provider library (e.g. libmlx4) perform
some initialization, such as mmap'ing device memory, which does not
work when fork is called.

As a work-around, avoid calling rsocket routines until immediately
before they are needed.  This allows the process to fork before
the libraries are initialized.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>

new

new

rspreload: Avoid rsocket calls until after fork

When an rsocket call is made before an application calls fork(),
the forked applications can hang.  This can be seen by running
netserver and two netperf clients simultaneously.  The second
netperf client will eventually stop performing data transfers.

LD_PRELOAD=librspreload.so netserver -D

LD_PRELOAD=librspreload.so netperf -v2 -c -C -H 192.168.0.101 -l30
LD_PRELOAD=librspreload.so netperf -v2 -c -C -H 192.168.0.101 -l30

It's not clear what the specific problem is.  The best guess is
that libibverbs or the provider library (e.g. libmlx4) perform
some initialization, such as mmap'ing device memory, which does not
work when fork is called.

As a work-around, avoid calling rsocket routines until immediately
before they are needed.  This allows the process to fork before
the libraries are initialized.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>

pop