]> git.openfabrics.org - ~shefty/rdma-win.git/log
~shefty/rdma-win.git
14 years agopop
Sean Hefty [Tue, 9 Mar 2010 19:49:17 +0000 (11:49 -0800)]
pop

14 years agopop
Sean Hefty [Tue, 9 Mar 2010 19:49:17 +0000 (11:49 -0800)]
pop

14 years agowinverbs/nd: remove inline data from read
Sean Hefty [Tue, 9 Mar 2010 18:24:22 +0000 (10:24 -0800)]
winverbs/nd: remove inline data from read

RDMA reads do not support inline data.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>
14 years agodelete
Sean Hefty [Tue, 9 Mar 2010 19:49:03 +0000 (11:49 -0800)]
delete

14 years agodelete
Sean Hefty [Tue, 9 Mar 2010 19:49:03 +0000 (11:49 -0800)]
delete

14 years agopop
Sean Hefty [Tue, 9 Mar 2010 19:48:57 +0000 (11:48 -0800)]
pop

14 years agopop
Sean Hefty [Tue, 9 Mar 2010 19:48:57 +0000 (11:48 -0800)]
pop

14 years agonew
Sean Hefty [Tue, 9 Mar 2010 19:36:35 +0000 (11:36 -0800)]
new

14 years agonew
Sean Hefty [Tue, 9 Mar 2010 19:36:35 +0000 (11:36 -0800)]
new

14 years agowinverbs: issue drep if in active disconnect state
Sean Hefty [Tue, 9 Mar 2010 19:36:33 +0000 (11:36 -0800)]
winverbs: issue drep if in active disconnect state

If we've received a DREQ after sending a DREQ, then
we're already actively disconnecting and should issue
a DREQ in response.  This prevents the remote side
from timing out on the DREQ.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>
14 years agopop
Sean Hefty [Tue, 9 Mar 2010 19:34:49 +0000 (11:34 -0800)]
pop

14 years agopop
Sean Hefty [Tue, 9 Mar 2010 19:34:49 +0000 (11:34 -0800)]
pop

14 years agodelete
Sean Hefty [Tue, 9 Mar 2010 19:13:00 +0000 (11:13 -0800)]
delete

14 years agodelete
Sean Hefty [Tue, 9 Mar 2010 19:13:00 +0000 (11:13 -0800)]
delete

14 years agorefresh
Sean Hefty [Tue, 9 Mar 2010 18:26:23 +0000 (10:26 -0800)]
refresh

14 years agorefresh
Sean Hefty [Tue, 9 Mar 2010 18:26:22 +0000 (10:26 -0800)]
refresh

14 years agorefresh (create temporary patch)
Sean Hefty [Tue, 9 Mar 2010 18:26:20 +0000 (10:26 -0800)]
refresh (create temporary patch)

14 years agorefresh (create temporary patch)
Sean Hefty [Tue, 9 Mar 2010 18:26:20 +0000 (10:26 -0800)]
refresh (create temporary patch)

14 years agoRefresh of nd-read
Sean Hefty [Tue, 9 Mar 2010 18:26:17 +0000 (10:26 -0800)]
Refresh of nd-read

14 years agonew
Sean Hefty [Tue, 9 Mar 2010 18:24:25 +0000 (10:24 -0800)]
new

14 years agonew
Sean Hefty [Tue, 9 Mar 2010 18:24:25 +0000 (10:24 -0800)]
new

14 years agowinverbs/nd: remove inline data from read
Sean Hefty [Tue, 9 Mar 2010 18:24:22 +0000 (10:24 -0800)]
winverbs/nd: remove inline data from read

RDMA reads do not support inline data.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>
14 years agowinverbs/nd: remove inline data from read
Sean Hefty [Tue, 9 Mar 2010 18:24:22 +0000 (10:24 -0800)]
winverbs/nd: remove inline data from read

RDMA reads do not support inline data.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>
14 years agorefresh
Sean Hefty [Tue, 9 Mar 2010 01:55:25 +0000 (17:55 -0800)]
refresh

14 years agorefresh
Sean Hefty [Tue, 9 Mar 2010 01:55:24 +0000 (17:55 -0800)]
refresh

14 years agorefresh (create temporary patch)
Sean Hefty [Tue, 9 Mar 2010 01:55:22 +0000 (17:55 -0800)]
refresh (create temporary patch)

14 years agorefresh (create temporary patch)
Sean Hefty [Tue, 9 Mar 2010 01:55:22 +0000 (17:55 -0800)]
refresh (create temporary patch)

14 years agoRefresh of ep-disc
Sean Hefty [Tue, 9 Mar 2010 01:55:19 +0000 (17:55 -0800)]
Refresh of ep-disc

14 years agonew
Sean Hefty [Tue, 9 Mar 2010 01:53:24 +0000 (17:53 -0800)]
new

14 years agonew
Sean Hefty [Tue, 9 Mar 2010 01:53:24 +0000 (17:53 -0800)]
new

14 years agowinverbs: allow WV:Disconnect() to be retried
Sean Hefty [Tue, 9 Mar 2010 01:53:21 +0000 (17:53 -0800)]
winverbs: allow WV:Disconnect() to be retried

ND for some odd reason wants a successful return code for
a failed disconnect call.  If a Disconnect() call fails,
allow it to be retried.  Return STATUS_TIMEOUT - a 'successful'
failure, rather than STATUS_IO_TIMEOUT, which is a 'failed' failure.
(I love Windows, really, I do.)  A subsequent call to EP:
Disconnect() after a timeout will force the QP into the error
state and force the EP into the disconnected state.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>
14 years agowinverbs: allow WV:Disconnect() to be retried
Sean Hefty [Tue, 9 Mar 2010 01:53:21 +0000 (17:53 -0800)]
winverbs: allow WV:Disconnect() to be retried

ND for some odd reason wants a successful return code for
a failed disconnect call.  If a Disconnect() call fails,
allow it to be retried.  Return STATUS_TIMEOUT - a 'successful'
failure, rather than STATUS_IO_TIMEOUT, which is a 'failed' failure.
(I love Windows, really, I do.)  A subsequent call to EP:
Disconnect() after a timeout will force the QP into the error
state and force the EP into the disconnected state.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>
14 years agopop
Sean Hefty [Tue, 9 Mar 2010 01:24:42 +0000 (17:24 -0800)]
pop

14 years agopop
Sean Hefty [Tue, 9 Mar 2010 01:24:42 +0000 (17:24 -0800)]
pop

14 years agowinverbs/nd: do not convert wv_io_timeout to nd_timeout
Sean Hefty [Fri, 5 Mar 2010 18:53:35 +0000 (10:53 -0800)]
winverbs/nd: do not convert wv_io_timeout to nd_timeout

wv_io_timeout is an error value, but nd_timeout is a
success value.  This means that an overlapped request
can be completed in error, but GetOverlappedResult can
return a success value.  Instead, if a timeout is fatal,
report the status as wv_io_timeout, but if the timeout
can be retried, then report the status as wv_timeout.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>
14 years agorefresh
Sean Hefty [Tue, 9 Mar 2010 01:24:31 +0000 (17:24 -0800)]
refresh

14 years agorefresh
Sean Hefty [Tue, 9 Mar 2010 01:24:31 +0000 (17:24 -0800)]
refresh

14 years agorefresh (create temporary patch)
Sean Hefty [Tue, 9 Mar 2010 01:24:30 +0000 (17:24 -0800)]
refresh (create temporary patch)

14 years agorefresh (create temporary patch)
Sean Hefty [Tue, 9 Mar 2010 01:24:29 +0000 (17:24 -0800)]
refresh (create temporary patch)

14 years agoRefresh of nd-conn
Sean Hefty [Tue, 9 Mar 2010 01:24:28 +0000 (17:24 -0800)]
Refresh of nd-conn

14 years agopop
Sean Hefty [Tue, 9 Mar 2010 01:23:57 +0000 (17:23 -0800)]
pop

14 years agopop
Sean Hefty [Tue, 9 Mar 2010 01:23:57 +0000 (17:23 -0800)]
pop

14 years agorename
Sean Hefty [Tue, 9 Mar 2010 01:14:03 +0000 (17:14 -0800)]
rename

14 years agorename
Sean Hefty [Tue, 9 Mar 2010 01:14:03 +0000 (17:14 -0800)]
rename

14 years agopop
Sean Hefty [Tue, 9 Mar 2010 01:13:49 +0000 (17:13 -0800)]
pop

14 years agopop
Sean Hefty [Tue, 9 Mar 2010 01:13:49 +0000 (17:13 -0800)]
pop

14 years agowinverbs/nd: do not convert wv_io_timeout to nd_timeout
Sean Hefty [Fri, 5 Mar 2010 18:53:35 +0000 (10:53 -0800)]
winverbs/nd: do not convert wv_io_timeout to nd_timeout

wv_io_timeout is an error value, but nd_timeout is a
success value.  This means that an overlapped request
can be completed in error, but GetOverlappedResult can
return a success value.  Instead, if a timeout is fatal,
report the status as wv_io_timeout, but if the timeout
can be retried, then report the status as wv_timeout.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>
14 years agowinverbs/nd: allow retrying ND:Connect()
Sean Hefty [Tue, 9 Mar 2010 00:59:27 +0000 (16:59 -0800)]
winverbs/nd: allow retrying ND:Connect()

The ND documentation specifies that ND:Connect() should be retryable.
Add this support to the winverbs ND provider.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>
14 years agowinverbs/nd: allow retrying ND:Connect()
Sean Hefty [Tue, 9 Mar 2010 00:59:27 +0000 (16:59 -0800)]
winverbs/nd: allow retrying ND:Connect()

The ND documentation specifies that ND:Connect() should be retryable.
Add this support to the winverbs ND provider.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>
14 years agowinverbs/ep: allow failed connection requests to be retried
Sean Hefty [Fri, 5 Mar 2010 17:26:50 +0000 (09:26 -0800)]
winverbs/ep: allow failed connection requests to be retried

Allow a user to retry a connection request if it fails.  Report a
'success' timeout value, rather than an error timeout, reset the
EP state, and allow the request to be retried from user space.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>
14 years agodapl: fix ring buffer synchronization
Sean Hefty [Thu, 4 Mar 2010 22:54:46 +0000 (14:54 -0800)]
dapl: fix ring buffer synchronization

The dapl ring buffer implementation is not thread safe and requires
the user to provide all synchronization.  Since the user must provide
synchronization, replace atomic definitions with simple ints to
improve performance and make the code more self-documenting that
no synchronization is provided.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>
14 years agopop
Sean Hefty [Tue, 9 Mar 2010 01:13:32 +0000 (17:13 -0800)]
pop

14 years agopop
Sean Hefty [Tue, 9 Mar 2010 01:13:32 +0000 (17:13 -0800)]
pop

14 years agoibal: move transport type attribute
Sean Hefty [Fri, 5 Mar 2010 19:06:32 +0000 (11:06 -0800)]
ibal: move transport type attribute

Relocate the transport type enum into space currently
reserved by the compiler to avoid binary compatibility issues.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>
14 years agopop
Sean Hefty [Tue, 9 Mar 2010 01:13:23 +0000 (17:13 -0800)]
pop

14 years agopop
Sean Hefty [Tue, 9 Mar 2010 01:13:23 +0000 (17:13 -0800)]
pop

14 years agopop
Sean Hefty [Tue, 9 Mar 2010 01:13:16 +0000 (17:13 -0800)]
pop

14 years agopop
Sean Hefty [Tue, 9 Mar 2010 01:13:16 +0000 (17:13 -0800)]
pop

14 years agopop
Sean Hefty [Tue, 9 Mar 2010 01:12:52 +0000 (17:12 -0800)]
pop

14 years agopop
Sean Hefty [Tue, 9 Mar 2010 01:12:52 +0000 (17:12 -0800)]
pop

14 years agoibal: move transport type attribute
Sean Hefty [Fri, 5 Mar 2010 19:06:32 +0000 (11:06 -0800)]
ibal: move transport type attribute

Relocate the transport type enum into space currently
reserved by the compiler to avoid binary compatibility issues.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>
14 years agopop
Sean Hefty [Tue, 9 Mar 2010 01:12:03 +0000 (17:12 -0800)]
pop

14 years agopop
Sean Hefty [Tue, 9 Mar 2010 01:12:03 +0000 (17:12 -0800)]
pop

14 years agowinverbs/nd: do not convert wv_io_timeout to nd_timeout
Sean Hefty [Fri, 5 Mar 2010 18:53:35 +0000 (10:53 -0800)]
winverbs/nd: do not convert wv_io_timeout to nd_timeout

wv_io_timeout is an error value, but nd_timeout is a
success value.  This means that an overlapped request
can be completed in error, but GetOverlappedResult can
return a success value.  Instead, if a timeout is fatal,
report the status as wv_io_timeout, but if the timeout
can be retried, then report the status as wv_timeout.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>
14 years agodelete
Sean Hefty [Tue, 9 Mar 2010 01:11:51 +0000 (17:11 -0800)]
delete

14 years agodelete
Sean Hefty [Tue, 9 Mar 2010 01:11:51 +0000 (17:11 -0800)]
delete

14 years agorefresh
Sean Hefty [Tue, 9 Mar 2010 01:11:45 +0000 (17:11 -0800)]
refresh

14 years agorefresh
Sean Hefty [Tue, 9 Mar 2010 01:11:45 +0000 (17:11 -0800)]
refresh

14 years agorefresh (create temporary patch)
Sean Hefty [Tue, 9 Mar 2010 01:11:27 +0000 (17:11 -0800)]
refresh (create temporary patch)

14 years agorefresh (create temporary patch)
Sean Hefty [Tue, 9 Mar 2010 01:11:27 +0000 (17:11 -0800)]
refresh (create temporary patch)

14 years agoRefresh of nd-con
Sean Hefty [Tue, 9 Mar 2010 01:11:25 +0000 (17:11 -0800)]
Refresh of nd-con

14 years agorefresh
Sean Hefty [Tue, 9 Mar 2010 01:11:17 +0000 (17:11 -0800)]
refresh

14 years agorefresh
Sean Hefty [Tue, 9 Mar 2010 01:11:17 +0000 (17:11 -0800)]
refresh

14 years agorefresh (create temporary patch)
Sean Hefty [Tue, 9 Mar 2010 01:11:15 +0000 (17:11 -0800)]
refresh (create temporary patch)

14 years agorefresh (create temporary patch)
Sean Hefty [Tue, 9 Mar 2010 01:11:15 +0000 (17:11 -0800)]
refresh (create temporary patch)

14 years agoRefresh of nd-con
Sean Hefty [Tue, 9 Mar 2010 01:11:13 +0000 (17:11 -0800)]
Refresh of nd-con

14 years agonew
Sean Hefty [Tue, 9 Mar 2010 00:59:28 +0000 (16:59 -0800)]
new

14 years agonew
Sean Hefty [Tue, 9 Mar 2010 00:59:28 +0000 (16:59 -0800)]
new

14 years agowinverbs/nd: allow retrying ND:Connect()
Sean Hefty [Tue, 9 Mar 2010 00:59:27 +0000 (16:59 -0800)]
winverbs/nd: allow retrying ND:Connect()

The ND documentation specifies that ND:Connect() should be retryable.
Add this support to the winverbs ND provider.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>
14 years agowinverbs/nd: allow retrying ND:Connect()
Sean Hefty [Tue, 9 Mar 2010 00:59:27 +0000 (16:59 -0800)]
winverbs/nd: allow retrying ND:Connect()

The ND documentation specifies that ND:Connect() should be retryable.
Allow Connect() to be retried.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>
14 years agowinverbs/nd: allow retrying ND:Connect()
Sean Hefty [Tue, 9 Mar 2010 00:59:27 +0000 (16:59 -0800)]
winverbs/nd: allow retrying ND:Connect()

The ND documentation specifies that ND:Connect() should be retryable.
Allow Connect() to be retried.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>
14 years agopop
Sean Hefty [Tue, 9 Mar 2010 00:58:16 +0000 (16:58 -0800)]
pop

14 years agopop
Sean Hefty [Tue, 9 Mar 2010 00:58:16 +0000 (16:58 -0800)]
pop

14 years agopop
Sean Hefty [Tue, 9 Mar 2010 00:53:23 +0000 (16:53 -0800)]
pop

14 years agopop
Sean Hefty [Tue, 9 Mar 2010 00:53:23 +0000 (16:53 -0800)]
pop

14 years agopop
Sean Hefty [Tue, 9 Mar 2010 00:53:12 +0000 (16:53 -0800)]
pop

14 years agopop
Sean Hefty [Tue, 9 Mar 2010 00:53:11 +0000 (16:53 -0800)]
pop

14 years agopop
Sean Hefty [Mon, 8 Mar 2010 23:25:20 +0000 (15:25 -0800)]
pop

14 years agopop
Sean Hefty [Mon, 8 Mar 2010 23:25:20 +0000 (15:25 -0800)]
pop

14 years agopop
Sean Hefty [Mon, 8 Mar 2010 20:41:11 +0000 (12:41 -0800)]
pop

14 years agopop
Sean Hefty [Mon, 8 Mar 2010 20:41:10 +0000 (12:41 -0800)]
pop

14 years agopop
Sean Hefty [Mon, 8 Mar 2010 17:11:11 +0000 (09:11 -0800)]
pop

14 years agopop
Sean Hefty [Mon, 8 Mar 2010 17:11:11 +0000 (09:11 -0800)]
pop

14 years agoibal: move transport type attribute
Sean Hefty [Fri, 5 Mar 2010 19:06:32 +0000 (11:06 -0800)]
ibal: move transport type attribute

Relocate the transport type enum into space currently
reserved by the compiler to avoid binary compatibility issues.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>
14 years agowinverbs/nd: do not convert wv_io_timeout to nd_timeout
Sean Hefty [Fri, 5 Mar 2010 18:53:35 +0000 (10:53 -0800)]
winverbs/nd: do not convert wv_io_timeout to nd_timeout

wv_io_timeout is an error value, but nd_timeout is a
success value.  This means that an overlapped request
can be completed in error, but GetOverlappedResult can
return a success value.  Instead, if a timeout is fatal,
report the status as wv_io_timeout, but if the timeout
can be retried, then report the status as wv_timeout.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>
14 years agowinverbs/nd: allow retrying connect() call
Sean Hefty [Fri, 5 Mar 2010 18:24:11 +0000 (10:24 -0800)]
winverbs/nd: allow retrying connect() call

The ND specification allows for retrying NDConnector:Connect()
calls.  Add this support to the winverbs ND provider.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>
14 years agowinverbs/ep: allow failed connection requests to be retried
Sean Hefty [Fri, 5 Mar 2010 17:26:50 +0000 (09:26 -0800)]
winverbs/ep: allow failed connection requests to be retried

Allow a user to retry a connection request if it fails.  Report a
'success' timeout value, rather than an error timeout, reset the
EP state, and allow the request to be retried from user space.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>
14 years agorefresh
Sean Hefty [Mon, 8 Mar 2010 17:10:46 +0000 (09:10 -0800)]
refresh

14 years agorefresh
Sean Hefty [Mon, 8 Mar 2010 17:10:46 +0000 (09:10 -0800)]
refresh

14 years agorefresh (create temporary patch)
Sean Hefty [Mon, 8 Mar 2010 17:08:30 +0000 (09:08 -0800)]
refresh (create temporary patch)