[ofa-general] Several RDS patches

[Olaf Kirch]

Hi Vlad,

during testing of RDS, a number of issues came up that I had to
address. Some of them were bugs, others were usability issues.

I submitted the following patches to
git://git.openfabrics.org/~okir/ofed_1_3/linux-2.6.git code-drop-20080617

d9fd0d04ba31951bcd7fe4393bccfe98ab8a45d5 RDS: introduce rds_send_get_message
9906c6480cedb05905d23c696671bcde9894b6ed RDS: Track whether RDMA S/G list is DMA mapped
db7a8ed76ac77e6a5352c8d41d28d9fac59e953d RDS: Another fix for RDMA signalling
	These fix another issue with signaling the completion of RDMA operations.

2665a94825a234fdd989ed98ae24c88bd3415368 RDS: Use vmalloc in rds_ib_setup_qp
	On a loaded system, re-establishing a broken RDS connection would
	fail because we tried to kmalloc the send and receive ring, which
	would result in an order-5 allocation.

1c3446c39af17e77fe7d39f32388e4c6462ee550 RDS: increase size of stats counters
	The stats counters interface to user space uses uint64_t, but inside the
	kernel, only "unsigned long" was used. On 32bit boxes this caused counters
	to be truncated.

1955124f7def27d55e98bfaff0c8da1438f2cca6 RDS: Expose info on IB connections
c2d8ba51f925cddca73562611ead156e62c3f93a RDS: Extend IB connection info to include settings
	People asked for a way to identify the IB port a RDS connection would use
	when the application does the load balancing.

804de7023165fd3fe89af6aac94843f195a23e72 RDS: Wake up send worker after TX queue full condition is cleared
	When encountering a full IB send queue, we back off for 2 jiffies. When
	the send CQ handler frees up several entries, we want to kick the
	send worker quickly. The existing code had a bug in it.

23d9822407b96538950e9e3dff5a7dfdbb464efb RDS: Send notification for cancelled RDMA ops
	We need notifications for RDMA operations that got canceled using
	RDS_CANCEL_SENT_TO

5e210431f65089062302c6e5ae9e998d67b9a980 RDS: Add mr_pool_depleted stats counter
	Track how often we were unable to initiate a RDMA because we couldn't
	allocate a FMR from our pool.

b99faff27c2c4e4b1058580645d997ce82d87ec0 RDS: rds-info -k should display inode number
	In order to identify the process holding a specific RDS socket, we want
	rds-info -k to show the socket's inode number, so that we can search
	/proc/*/fd for it

eabede450f3d540605882727a3b71a116919d5c1 RDS: Clarify comment on rds_poll
	Cosmetic - Rick complained about a misleading comment in rds_poll :)

------------------------
The rds-info changes go with several patches to rds-tools, which I also pushed to
my tree:

git://git.openfabrics.org/~okir/ofed_1_3/rds-tools.git code-drop-20080617

e40149599d03c057b99df78608794dbb8803d732 rds-info: introduce address printing functions
4f84065a0b3aae50c2ee346d4b8935b654f9f64c rds-info: add option to display IB connections
	This implements display of RDS/IB connections

15eb9f7dfe6724857c80a698cf339146088c1bc8 rds-info: Add -v option
4b90f7329dc146f01ca5b91eda924bcf0f8aef3f rds-info: Extend IB connection info to include settings
	This displays per-connection IB settings such as max_send_wr if -v is given

794d3504649e5686342232cdad59b1cc6b445b35 rds-info -k should display inode number
	Display the inode number of the RDS socket.

Can you please pull these into OFED-1.3?

Olaf
-- 
Olaf Kirch  |  --- o --- Nous sommes du soleil we love when we play
okir at lst.de |    / | \   sol.dhoop.naytheet.ah kin.ir.samse.qurax