On Mon, Jun 27, 2005 at 09:24:45AM -0700, Roland Dreier wrote:
Gleb This is what happens: ibv_close_device() close cmd_fd and
Gleb then calls free_context(). free_context() calls munmap to
Gleb unmap doorbell registers. In kernel sys_munmap gets
Gleb mm-mmap_sem semaphore and
On Mon, Jun 27, 2005 at 04:59:54PM -0700, Roland Dreier wrote:
Something like this should work...
Yes, but don't forget the first part of my patch (for libibverbs/src/device.c)
I think it should be applied to.
[snip]
-void ib_umem_release(struct ib_device *dev, struct ib_umem *umem)
Am Montag, den 27.06.2005, 11:35 -0700 schrieb Libor Michalek:
On Wed, Jun 22, 2005 at 10:39:43AM +0200, Arne Redlich wrote:
Hi,
I'm trying to use SDP from within the kernel. My problem is that the
code relies on sk_data_ready() (this callback is modified to wake up a
Rx thread before
Quoting r. Roland Dreier [EMAIL PROTECTED]:
Subject: Re: [PATCH] process locked in D state.
Something like this should work...
Note that there's a window when close() has returned but
vm_locked isnt updated yet.
Actually, a process that doesnt close memory regions before
closing the device
On Tue, Jun 28, 2005 at 10:22:22AM +0300, Michael S. Tsirkin wrote:
Quoting r. Roland Dreier [EMAIL PROTECTED]:
Subject: Re: [PATCH] process locked in D state.
Something like this should work...
Note that there's a window when close() has returned but
vm_locked isnt updated yet.
This
By your patch you completely removed locked_vm accounting on deregister
path. I don't think this was your intention.
Of course. Closing the file should be handled separately from regular
deregistration. A new patch (below) fixes that.
---
Dont touch locked_vm when file is being closed, since
On Tue, 2005-06-28 at 01:29, Eitan Zahavi wrote:
I was not aware of an issue with multiple HCAs in the OpenIB (gen2)
stack.
The Gen1 stack had this issue and it was resolved. I hope to be able
to focus on OpenIB stack in the coming months such that I can help Hal
in fixing these kind of
On Mon, 2005-06-27 at 15:27, Eitan Zahavi wrote:
The current OpenIB topology file has a place where these annotations
can
be made (and displayed).
[EZ] How would you define the internal structure of a 288port switch
in the existing topology file?
Would it support writing code that is
Title: RE: IB Diagnositic Tools
[EZ] How would you define the internal structure of a 288port switch
in the existing topology file?
Would it support writing code that is able to report something like
board spine2 of system mySwitch is missing?
I think it could.
[EZ] Anyway this
Quoting r. Hal Rosenstock [EMAIL PROTECTED]:
Will the Linux distros take it this way (with #ifdef OS)
I never looked at opensm vendor layer, but generally
in userspace code, it is not always worth it the effort to get rid of all
os-dependent code.
If you want an application to be portable, you
Hi!
I have uploaded the patches to make svn trunk build on 2.6.11.
They can be found in
https://openib.org/svn/gen2/branches/backport-to-2.6.11
Enjoy!
--
MST
___
openib-general mailing list
openib-general@openib.org
On Tue, 2005-06-28 at 07:30, Eitan Zahavi wrote:
[EZ] How would you define the internal structure of a 288port
switch
in the existing topology file?
Would it support writing code that is able to report something
like
board spine2 of system mySwitch is missing?
I think it could.
Quoting r. Hal Rosenstock [EMAIL PROTECTED]:
Subject: Re: [openib-general] backport patch for 2.6.11
On Tue, 2005-06-28 at 08:39, Michael S. Tsirkin wrote:
Hi!
I have uploaded the patches to make svn trunk build on 2.6.11.
They can be found in
On Mon, 2005-06-27 at 17:30, Hal Rosenstock wrote:
On Mon, 2005-06-27 at 17:13, Bernhard Fischer wrote:
Hal,
This is kinda offtopic,
Yes, this is different topic(s).
but (iirc) i once stumbled over the issue of
port vs. mgmt port [back then i had access to two 2-port cards]
At 05:34 PM 6/27/2005, Roland Dreier wrote:
I'm not sure I understand this. At best, ATS can give you back a list
of IPs. How do you decide which one to check against the exports?
Any or all of them. Exports is a fairly simple access list, and membership
by the client is all that's required. It
Arlin,
Would this be a more standard way to do the cast:
On Mon, 27 Jun 2005, Arlin Davis wrote:
Signed-off-by: Arlin Davis [EMAIL PROTECTED]
Index: dapl/openib/dapl_ib_dto.h
===
--- dapl/openib/dapl_ib_dto.h (revision 2720)
On Tue, 2005-06-28 at 09:34, Eitan Zahavi wrote:
On Tue, 2005-06-28 at 07:30, Eitan Zahavi wrote:
[EZ] How would you define the internal structure of a 288port
switch
in the existing topology file?
Would it support writing code that is able to report something
like
Hi Eitan,
First, a nit. Should IBMgtSim/src/.autosave be removed from the tree ?
Also, have you gone through the instructions in README on a fresh
machine ? It fails for me on the first step as follows:
./config/bootstrap
/home/hal/openib/utils/src/linux-user/IBMgtSim/config/libtoolize: line 1:
At 06:25 PM 6/27/2005, Tom Duffy wrote:
I have done some initial work to port nfsrdma to 2.6 and to James's
kDAPL. This builds now inside the kernel.
Tom, thanks for starting this and I'll take a look at your approach.
In fact we already have a version working on 2.6.11, the main change
merging
Title: RE: Simulator building
This is all good input. I will go through and revise the material.
Also I will upload the newest and greatest code.
This should be done during the weekend and be there for Monday.
Eitan Zahavi
Design Technology Director
Mellanox Technologies LTD
Gleb Yes, but don't forget the first part of my patch (for
Gleb libibverbs/src/device.c) I think it should be applied to.
Yes, I guess it makes sense.
- mm = get_task_mm(current);
Gleb Are you sure that ib_umem_account will run on behalf the
Gleb process that schedules it?
Gleb This is the case only if close() can return before
Gleb workqueues are ran. Is this possible? If yes perhaps it is
Gleb better to use tascklets.
Yes, it's possible. Unfortunately a tasklet can't be used because
tasklet context can't do down_write().
- R.
Michael Of course. Closing the file should be handled separately
Michael from regular deregistration. A new patch (below) fixes
Michael that.
This looks better than my approach. Let me make sure it covers
everything, but it definitely seems simpler.
- R.
On Tue, 2005-06-28 at 09:58 -0400, Talpey, Thomas wrote:
At 06:25 PM 6/27/2005, Tom Duffy wrote:
I have done some initial work to port nfsrdma to 2.6 and to James's
kDAPL. This builds now inside the kernel.
Tom, thanks for starting this and I'll take a look at your approach.
In fact we
At 11:13 AM 6/28/2005, Tom Duffy wrote:
I am sure I got the RPC stuff wrong. I just wanted to make it compile
against James's kDAPL and inside the drivers/infiniband directory.
Where can I find the 2.6.11 version of the patch?
Here:
On Tue, Jun 28, 2005 at 07:10:13AM -0700, Roland Dreier wrote:
Gleb Yes, but don't forget the first part of my patch (for
Gleb libibverbs/src/device.c) I think it should be applied to.
Yes, I guess it makes sense.
- mm = get_task_mm(current);
Gleb Are you sure that
I was able to build udapltest for both x86 and x86-64 with the following
change:
uDAPL: Fix compile of dapl_bpool.c on x86_64
Signed-off-by: Hal Rosenstock [EMAIL PROTECTED]
--- userspace/test/dapltest/test/dapl_bpool.c.orig 2005-01-18
13:20:31.0 -0500
+++
On Tue, Jun 28, 2005 at 07:11:23AM -0700, Roland Dreier wrote:
Gleb This is the case only if close() can return before
Gleb workqueues are ran. Is this possible? If yes perhaps it is
Gleb better to use tascklets.
Yes, it's possible. Unfortunately a tasklet can't be used because
On Tue, Jun 28, 2005 at 07:11:54AM -0700, Roland Dreier wrote:
Michael Of course. Closing the file should be handled separately
Michael from regular deregistration. A new patch (below) fixes
Michael that.
This looks better than my approach. Let me make sure it covers
Jordan, Bill wrote:
Sean,
Here is a to fix a bug in service_id/service_mask
handling in CM.
Previously, if you tried to listen on service_id:
service_mask of 0x00:0x0F, followed by a listen
on 0xF0:0xF0, the second listen would fail because
the two ranges overlap (0xF0 appears in both ranges).
The Sun implementation uses a Service ID from the Local OS
Administrative range when installing an entry in the SA. So, the first
byte should be 0x2 in the ServiceID component in the SA record. The
Service Name in the SA record contained the IP address with the NFS port
number appended at the
Hi Sean,
I'm in the process of enabling the receive side RMPP from user space and
this is what I'm seeing in terms of RMPP right now. I have a question
about the OpenSM side.
SA client OpenSM
SA GetTable (PortInfoRecord) --
-- SA GetTableResp (PortInfoRecord)
RMPP
Hal Rosenstock wrote:
Hi Sean,
I'm in the process of enabling the receive side RMPP from user space and
this is what I'm seeing in terms of RMPP right now. I have a question
about the OpenSM side.
SA client OpenSM
SA GetTable (PortInfoRecord) --
-- SA GetTableResp
On Tue, 2005-06-28 at 13:44, Sean Hefty wrote:
Hal Rosenstock wrote:
Hi Sean,
I'm in the process of enabling the receive side RMPP from user space and
this is what I'm seeing in terms of RMPP right now. I have a question
about the OpenSM side.
SA client OpenSM
SA GetTable
On Sun, Jun 19, 2005 at 10:33:36AM +0300, Michael S. Tsirkin wrote:
Here's an update to the latest bits: Libor, does this make sense?
Make sdp_cm_actv_error and sdp_conn_inet_error accept standard, negative
error values (esp. for sdp_cm_actv_error it was sufficiently confusing to
be passing
Michael Of course. Closing the file should be handled separately
Michael from regular deregistration. A new patch (below) fixes
Michael that.
Thinking about this some more, I really don't like having such a large
hole where vm_locked accounting can be wrong. I would prefer to stick
The attached PDF, prepared by Itamar Rabenstein of Mellanox, contains
a proposed mapping from DAT_RETURN values to ERRNO values.
Please send me suggestions for alternative mappings by the end of
the week.
james
errno.pdf
Description: errno.pdf
On Tue, 2005-06-28 at 13:48, Hal Rosenstock wrote:
On Tue, 2005-06-28 at 13:44, Sean Hefty wrote:
Hal Rosenstock wrote:
Hi Sean,
I'm in the process of enabling the receive side RMPP from user space and
this is what I'm seeing in terms of RMPP right now. I have a question
about
On Mon, Jun 27, 2005 at 04:06:11PM -0700, Libor Michalek wrote:
On Mon, Jun 27, 2005 at 02:27:54PM -0700, Tom Duffy wrote:
On Mon, 2005-06-27 at 11:17 -0700, Libor Michalek wrote:
The problem is that each call to sk_alloc() is grabbing a reference to
the module, but it checks to make
While all qps and cqs are closed first, so HCA doesnt have a reason to
write to the mr, I think its still cleaner to first deregister
the mr, and then release the memory.
It might become more important if/when shared mrs are implemented.
---
Deregister memory before ib_umem_release, to make sure
On Tue, 2005-06-28 at 14:07, Hal Rosenstock wrote:
On Tue, 2005-06-28 at 13:48, Hal Rosenstock wrote:
On Tue, 2005-06-28 at 13:44, Sean Hefty wrote:
Hal Rosenstock wrote:
Hi Sean,
I'm in the process of enabling the receive side RMPP from user space and
this is what I'm
Yes, good catch. Applied, thanks.
- R.
___
openib-general mailing list
openib-general@openib.org
http://openib.org/mailman/listinfo/openib-general
To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
Michael Apropos that, do you have the kernel patch implementing
Michael something like mprotect(, PROT_DONTCOPY) (needed for
Michael fork support)?
Not yet -- I am deferring that a little.
- R.
___
openib-general mailing list
Quoting r. Roland Dreier [EMAIL PROTECTED]:
Subject: Re: [PATCH] process locked in D state.
I would prefer to stick
with the current code, which has at worst a short window where
vm_locked is too high.
Right, but it seems this introduces the race for apps doing deregister_mr
properly, too,
user_mad: In send_handler, only queue_packet on timeout if not response
Signed-off-by: Hal Rosenstock [EMAIL PROTECTED]
Index: user_mad.c
===
--- user_mad.c (revision 2731)
+++ user_mad.c (working copy)
@@ -146,12 +146,14 @@
I'd like to summarize the discussion we've been having around
addressing and start a new email thread with a more appropriate title.
First off, here is there requirement we are trying to satisfy:
kDAPL consumers use an Internet Protocol (IP) addresses to
identify remote nodes in an
Hal user_mad: In send_handler, only queue_packet on timeout if
Hal not response
Why do we want to prevent userspace from seeing these events? What's
special about trap repress and resp methods?
- R.
___
openib-general mailing list
On Tue, Jun 28, 2005 at 03:24:35PM -0400, James Lentini wrote:
I'd like to summarize the discussion we've been having around
addressing and start a new email thread with a more appropriate title.
First off, here is there requirement we are trying to satisfy:
kDAPL consumers use an
Michael Right, but it seems this introduces the race for apps
Michael doing deregister_mr properly, too, which used to work
Michael fine.
True.
Michael Right, but at least lets not introduce the hole for the
Michael good case. When I call deregister_mr I have the right to
Hi Bernhard,
The changes look fine. Why the additional copyright? I need to be able
to explain it to my legal department.
james
On Mon, 27 Jun 2005, Bernhard Fischer wrote:
Hi James,
untested.
- cleanup dapl_cookie.c: remove unneeded local variables and simplify
branches to be
On Tue, 2005-06-28 at 15:26, Roland Dreier wrote:
Hal user_mad: In send_handler, only queue_packet on timeout if
Hal not response
Why do we want to prevent userspace from seeing these events? What's
special about trap repress and resp methods?
Why does user space need to see all
Hal Why does user space need to see all the timeouts ? I thought
Hal it was only request timeouts which were needed to be
Hal seen. The semantics for RMPP timeouts appear to be different.
Well, can the MAD core even generate timeouts for trap repress or
response messages? If so, I
On Tue, 2005-06-28 at 15:43, Roland Dreier wrote:
Hal Why does user space need to see all the timeouts ? I thought
Hal it was only request timeouts which were needed to be
Hal seen. The semantics for RMPP timeouts appear to be different.
Well, can the MAD core even generate
Hal I'm not sure what a consumer would want a timeout to a non
Hal request method.
I guess my view would be that if a consumer would never want a
timeout, then the kernel shouldn't generate them. We be consistent
between kernel space and user space though -- the umad module
shouldn't be
On Tue, 2005-06-28 at 15:24, James Lentini wrote:
+ IPoIB
IPoIB encapsulates IP packets in InfiniBand messages. There have been
proposals to use the address resolution mechanisms in IPoIB to
implement these features. IPv4 subnets use ARP and IPv6 subnets use
Neighbor
Libor, here's a stub at solving an old problem.
Most of the patch is passing sdp_opt to iocb complete and cancel functions.
I didnt yet test it, and I wont have the time today,
but maybe you could tell me whether I'm going in the right
direction with this.
---
Use users in the lock to prevent
Hal Rosenstock wrote:
On Tue, 2005-06-28 at 15:43, Roland Dreier wrote:
Hal Why does user space need to see all the timeouts ? I thought
Hal it was only request timeouts which were needed to be
Hal seen. The semantics for RMPP timeouts appear to be different.
Well, can the MAD core
Hal Rosenstock wrote:
I'm not sure what a consumer would want a timeout to a non request
method.
Since RMPP provides reliability, I think that it makes sense to indicate to
the user when an RMPP MAD is not acknowledged. If no error is generated,
the user could mistakenly assume that the
On Tue, 2005-06-28 at 16:12, Roland Dreier wrote:
Hal I'm not sure what a consumer would want a timeout to a non
Hal request method.
I guess my view would be that if a consumer would never want a
timeout, then the kernel shouldn't generate them. We be consistent
between kernel
Hal I still think there's a semantic difference with timeouts
Hal which needs more consistency.
Could you elaborate? What are the different situations where timeouts
behave differently?
- R.
___
openib-general mailing list
On 6/28/05, Roland Dreier [EMAIL PROTECTED] wrote:
James First off, here [are the] requirement we are trying to satisfy:
JamesOn the passive side of a connection, a InfiniBand kDAPL
James provider must determine a source IB address for an
James InfiniBand connection
Hi, James!
I dont know much about dapl, so forgive me if the question is naive:
Quoting r. James Lentini [EMAIL PROTECTED]:
+ CM Private Data
The active side of an IB connection could place its source IP
address in the CM's private data. The passive side would retrieve
the
From: Hal Rosenstock [mailto:[EMAIL PROTECTED]
Sent: Tuesday, June 28, 2005 11:47 AM
I think I see what is going on here...
In user_mad.c::send_handler
if (send_wc-status == IB_WC_RESP_TIMEOUT_ERR) {
packet-mad.hdr.status = ETIMEDOUT;
if
Hal user_mad: In send_handler, only return header on timeout
This seems reasonable. Is there anything more self-documenting we can
use instead of hard-coding the magic number of 24?
- R.
___
openib-general mailing list
openib-general@openib.org
Hi Libor,
I have a couple of uCM questions regarding create_id and events...
Is it possible for a consumer of uCM to provide a context with the
create_id that could be returned with the event? I will have some scale
up issues if I have to walk a list looking for a uCM provided connection
ID
On Tue, 2005-06-28 at 16:58, Roland Dreier wrote:
Hal user_mad: In send_handler, only return header on timeout
This seems reasonable. Is there anything more self-documenting we can
use instead of hard-coding the magic number of 24?
I think so. I had a feeling that would be a comment.
On Tue, 2005-06-28 at 17:17, [EMAIL PROTECTED] wrote:
Medium projects (well understood but require a fair amount of code):
MW support: ib_mthca does not support memory windows.
-FMR support: ib_mthca does not support Mellanox-style fast
-memory regions as used by SDP and
Committed in revision 2739.
On Tue, 28 Jun 2005, Arlin Davis wrote:
ardavis James Lentini wrote:
ardavis
ardavis
ardavis Arlin,
ardavis
ardavis Would this be a more standard way to do the cast:
ardavis
ardavis yes, I agree.
ardavis
ardavis
ardavis On Mon, 27 Jun 2005, Arlin Davis
At 10:30 AM 6/24/2005, Roland Dreier wrote:
Thomas As
I said - I am not attached to ATS. I would welcome an
Thomas alternative.
Sure, understood. I'm suggesting a slight tweak to the IB wire
protocol. I don't think there's a difference in the security
provided, and carrying the peer address in
On Tue, 2005-06-28 at 17:17, [EMAIL PROTECTED] wrote:
Medium projects (well understood but require a fair amount of code):
MW support: ib_mthca does not support memory windows.
-FMR support: ib_mthca does not support Mellanox-style fast
-memory regions as used by SDP and
On 6/28/05, Michael S. Tsirkin [EMAIL PROTECTED] wrote:
Hi, James!
I dont know much about dapl, so forgive me if the question is naive:
Quoting r. James Lentini [EMAIL PROTECTED]:
+ CM Private Data
The active side of an IB connection could place its source IP
address in the
sa_query: Add service record support
Signed-off-by: Hal Rosenstock [EMAIL PROTECTED]
Index: include/ib_sa.h
===
-- include/ib_sa.h (revision 2732)
+++ include/ib_sa.h (working copy)
@@ -1,5 +1,6 @@
/*
* Copyright (c)
Hal Should SMR support be added to this list ?
Sure, I just did that.
- R.
___
openib-general mailing list
openib-general@openib.org
http://openib.org/mailman/listinfo/openib-general
To unsubscribe, please visit
Quoting r. Hal Rosenstock [EMAIL PROTECTED]:
Subject: [PATCH] sa_query: Add service record support
sa_query: Add service record support
Signed-off-by: Hal Rosenstock [EMAIL PROTECTED]
...
+struct ib_sa_srvrec_query {
+ void (*callback)(int, struct ib_sa_service_rec *, void *);
+
That would require systems to be configured using IB addresses. Can
that be made feasible in non-trivial configurations?
For example, how could name resolution be made scalable? If you want
to resolve a hostname like foo.bar.com to an IB GID, then information
about IB GIDs needs to added to
On Tue, 28 Jun 2005, Roland Dreier wrote:
James First off, here [are the] requirement we are trying to satisfy:
JamesOn the passive side of a connection, a InfiniBand kDAPL
James provider must determine a source IB address for an
James InfiniBand connection request. This
On Tue, 2005-06-28 at 17:34, Michael S. Tsirkin wrote:
Quoting r. Hal Rosenstock [EMAIL PROTECTED]:
Subject: [PATCH] sa_query: Add service record support
sa_query: Add service record support
Signed-off-by: Hal Rosenstock [EMAIL PROTECTED]
...
+struct ib_sa_srvrec_query {
+
Thanks, committed.
___
openib-general mailing list
openib-general@openib.org
http://openib.org/mailman/listinfo/openib-general
To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
Quoting r. Hal Rosenstock [EMAIL PROTECTED]:
Can this be renamed to struct ib_sa_service_rec_query, please?
Or does it mean something else?
Sure but ib_sa_service_query would be more in keeping.
The functions are called ib_sa_service_rec_query, I guess you'll rename
that too?
--
MST
Michael Can this be renamed to struct ib_sa_service_rec_query,
Michael please? Or does it mean something else?
Makes sense. The style I tried to follow in the SA client module was
that everything was named the same was as the IBA spec, except that
StudlyCaps are replaced with
James Which two points? The ability to determine a source IB
James address for an InfiniBand connection request and
James interoperability?
James I don't think those two are contradictory by definition.
Yes, those two points I quoted. As I said, the IB communication
management
On Tue, 28 Jun 2005, Hal Rosenstock wrote:
On Tue, 2005-06-28 at 15:24, James Lentini wrote:
+ IPoIB
IPoIB encapsulates IP packets in InfiniBand messages. There have been
proposals to use the address resolution mechanisms in IPoIB to
implement these features. IPv4 subnets use ARP
Quoting r. Roland Dreier [EMAIL PROTECTED]:
Subject: Re: [PATCH] process locked in D state.
Thanks, committed.
Hmm, but why are we still testing
if (current-mm) ?
I think current-mm is never NULL when we are inside a system call,
except sys_exit.
Am I wrong?
--
MST
Site-local GIDs are just as manageable as site-local IPv6 addresses.
In fact they look just like them to /etc/hosts, NIS, DNS, etc.
The only real work required of the local network administrator(s)
is that they assign *different* site-local network IDs for IPv6 and IB.
There are 64K total
Are there any tools available for a network administrator to assign a
GID? Does OpenSM provider this capability?
On Tue, 28 Jun 2005, Caitlin Bestler wrote:
Site-local GIDs are just as manageable as site-local IPv6 addresses.
In fact they look just like them to /etc/hosts, NIS, DNS, etc.
Hal,
This brings up a good point. I need to get the uDAPL dapltest into
the OpenIB tree. I'll work on doing that tomorrow.
james
On Tue, 28 Jun 2005, Hal Rosenstock wrote:
halr I was able to build udapltest for both x86 and x86-64 with the following
halr change:
halr
halr uDAPL: Fix compile
Michael Hmm, but why are we still testing if (current-mm) ?
Michael I think current-mm is never NULL when we are inside a
Michael system call, except sys_exit. Am I wrong?
Good question. I don't think we need this.
- R.
___
On Tue, Jun 28, 2005 at 09:15:21AM +0200, Arne Redlich wrote:
Am Montag, den 27.06.2005, 11:35 -0700 schrieb Libor Michalek:
On Wed, Jun 22, 2005 at 10:39:43AM +0200, Arne Redlich wrote:
Hi,
I'm trying to use SDP from within the kernel. My problem is that the
code relies on
James I found this limitation in section 3.3 of the IP over
James InfiniBand(IPoIB) Architecture draft (April, 2004 version,
James
http://www.ietf.org/internet-drafts/draft-ietf-ipoib-architecture-04.txt)
James Until the above conditions are met it is not
James
Here is a series of patches that adds support for direct userspace access to
InfiniBand hardware -- so-called userspace verbs. I believe these patches
are ready to merge, but a final review would be useful.
These patches should incorporate all of the feedback from the discussion when
I posted an
Add support for userspace protection domains (PDs) to mthca.
Signed-off-by: Roland Dreier [EMAIL PROTECTED]
---
drivers/infiniband/hw/mthca/mthca_dev.h |3 ++-
drivers/infiniband/hw/mthca/mthca_main.c |2 +-
drivers/infiniband/hw/mthca/mthca_pd.c | 24
Add support for userspace memory regions (MRs) to mthca.
Signed-off-by: Roland Dreier [EMAIL PROTECTED]
---
drivers/infiniband/hw/mthca/mthca_provider.c | 82 +++
1 files changed, 82 insertions(+)
--- linux.orig/drivers/infiniband/hw/mthca/mthca_provider.c
Update kernel InfiniBand midlayer to compile against the updated API
for low-level drivers. This just amounts to passing NULL for all
userspace-related parameters, and setting userspace-related structure
members to NULL.
Signed-off-by: Roland Dreier [EMAIL PROTECTED]
---
Add support for userspace queue pairs (QPs) to mthca.
Signed-off-by: Roland Dreier [EMAIL PROTECTED]
---
drivers/infiniband/hw/mthca/mthca_dev.h |2
drivers/infiniband/hw/mthca/mthca_provider.c | 80 --
drivers/infiniband/hw/mthca/mthca_qp.c | 215
Add support for mmap() method to mthca, so that userspace can get
access to doorbell registers. This allows userspace to get direct
access to the HCA for data path operations.
Each userspace context gets its own copy of the doorbell registers and
is only allowed to use resources that the kernel
Update mthca to compile against the updated API for low-level drivers.
Signed-off-by: Roland Dreier [EMAIL PROTECTED]
---
drivers/infiniband/hw/mthca/mthca_provider.c | 13 ++---
1 files changed, 10 insertions(+), 3 deletions(-)
---
Add the ib_user_verbs.h header file, which defines the ABI used by
InfiniBand userspace verbs for kernel/user communication.
Signed-off-by: Roland Dreier [EMAIL PROTECTED]
---
drivers/infiniband/include/ib_user_verbs.h | 389 +
1 files changed, 389 insertions(+)
Add documentation for InfiniBand userspace verbs.
Signed-off-by: Roland Dreier [EMAIL PROTECTED]
---
Documentation/infiniband/user_verbs.txt | 69
1 files changed, 69 insertions(+)
--- /dev/null 2005-06-23 14:14:38.423479552 -0700
+++
Hook up InfiniBand userspace verbs to Kconfig and the make system.
Signed-off-by: Roland Dreier [EMAIL PROTECTED]
---
drivers/infiniband/Kconfig | 10 ++
drivers/infiniband/core/Makefile |5 -
2 files changed, 14 insertions(+), 1 deletion(-)
---
Roland Dreier [EMAIL PROTECTED] wrote:
Add support for pinning userspace memory regions and returning a list
of pages in the region. This includes tracking pinned memory against
vm_locked and preventing unprivileged users from exceeding RLIMIT_MEMLOCK.
Can you tell us a bit more about the
1 - 100 of 103 matches
Mail list logo