ibportstate speed/width change support for IBSim #2
Conversation
hnrose
left a comment
hnrose
left a comment
There was a problem hiding this comment.
Why are these simulator commands needed ? There is ibsendtrap utility in infiniband-diags which can be used for testing sending trap 144.
hnrose
left a comment
hnrose
left a comment
There was a problem hiding this comment.
This is a number of different changes (thoughts) in a single patch and is best separated out into separate patches for comment. In addition to the primary purpose of this patch, there are the following:
The maximum width supported has been changed to LINKWIDTH_1x_4x_8x_12x looking at the Infiniband specification.
The warning "Port %s:%d overwrite lid table entry for lid %u (was %s:%d)" will not be triggered when the computed port is corresponding to a switch-port without owning a LID.
On the first point, this is changing the default (when ibnetdiscover file is not annotated with link widths as in net-examples). The default was chosen as a common expected value. Perhaps new default should be 1x, 2x, and 4x.
Also, the extended link speeds (espeed) should also be supported in addition to speed. That could be included in primary patch or as additional patch.
|
"Why are these simulator commands needed ? There is ibsendtrap utility in infiniband-diags which can be used for testing sending trap 144." -> This is because we want to change the speed and width through ibportstate. |
|
I can split the pull request into different ones then you can accept what you think is useful for the simulator. I will split it next week. |
|
About the extended speed: Sure, this is currently missing in our changes and this is something we have to add later as well. |
|
Simulator commands aren't needed for that. What am I missing ?
…On Thu, Aug 2, 2018 at 9:39 AM, Jesus Camacho Villanueva < ***@***.***> wrote:
"Why are these simulator commands needed ? There is ibsendtrap utility in
infiniband-diags which can be used for testing sending trap 144." -> This
is because we want to change the speed and width through ibportstate.
—
You are receiving this because you commented.
Reply to this email directly, view it on GitHub
<#2 (comment)>, or mute
the thread
<https://github.com/notifications/unsubscribe-auth/AC2_GOFWV-PU3OF2MC9m83XbuMh4RJgRks5uMwEEgaJpZM4VmfZZ>
.
|
|
Do you mean the next? We started with this and later we implemented the ibportstate part. |
|
On Thu, Aug 2, 2018 at 10:04 AM, Jesus Camacho Villanueva < ***@***.***> wrote:
Do you mean the next?
"Speed "nodeid"[port] : change LinkSpeedEnabled"
"Width "nodeid"[port] : change LinkWidthEnabled"
We started with this and later we implemented the ibportstate part.
We can remove those commands if you refer to that.
Yes, that's (the simulator commands and not the ibportstate part) what I
was referring to.
… —
You are receiving this because you commented.
Reply to this email directly, view it on GitHub
<#2 (comment)>, or mute
the thread
<https://github.com/notifications/unsubscribe-auth/AC2_GN1oNun2PqzMRq_8FbaHtylakbcWks5uMwbWgaJpZM4VmfZZ>
.
|
|
OK, then we will remove that part :). |
Support for using 'ibportstate speed/espeed/width' command together with IBSim. After a port change with the mentioned command and reception of trap 144, a heavy sweep will set the enabled speed and width to every port. To avoid sending a trap 144 every time a port is set, the switch sends a trap only when the speed or width has been effectively changed. Active speed, espeed and width port is automatically adjusted when rewriting the enabled speed and enabled width port, respectively. Here it's an example of the new functiolality (suppose that $IBSIM is the path of your ibsim folder): Run in terminal 1: $ ibsim -s $IBSIM/net-examples/net.2sw2path4hca Run in terminal 2: $ export LD_PRELOAD=$IBSIM/umad2sim/libumad2sim.so $ opensm Run in terminal 3: $ export LD_PRELOAD=$IBSIM/umad2sim/libumad2sim.so $ ibportstate -G 0x200000 3 speed 3 When changing the LinkSpeedEnabled in terminal 3, you will see the next change: Initial Switch PortInfo: LinkSpeedEnabled:................2.5 Gbps After PortInfo set: LinkSpeedEnabled:................2.5 Gbps or 5.0 Gbps $ ibportstate -G 0x200000 3 width 3 Initial Switch PortInfo: LinkWidthEnabled:................4X After PortInfo set: LinkWidthEnabled:................1X or 4X $ ibportstate -G 0x200000 3 espeed 3 Initial Switch PortInfo: LinkSpeedExtEnabled:.............0 After PortInfo set: LinkSpeedExtEnabled:.............14.0625 Gbps or 25.78125 Gbps NOTE: force_link_speed, force_link_speed_ext and force_link_width must be 0. Signed-off-by: jecavil <jesus.camacho@fabriscale.com>
|
|
||
| if (p->state == 4) { | ||
| if (p->lid > 0 && p->lid < maxlinearcap | ||
| if (node->type != SWITCH_NODE && p->lid > 0 && p->lid < maxlinearcap |
There was a problem hiding this comment.
Why is node type != switch needed here ?
There was a problem hiding this comment.
If this is needed, it should be in a separate patch with more info as to why it's needed/how to reproduce.
I'm going to commit this PR without this change.
|
Merged commit cfcefe0 (based on this PR) into master |
|
This is fine!
Jesus
El jue., 20 sept. 2018 18:31, hnrose <notifications@github.com> escribió:
… ***@***.**** commented on this pull request.
------------------------------
In ibsim/sim_mad.c
<#2 (comment)>:
> @@ -540,7 +541,7 @@ do_portinfo(Port * port, unsigned op, uint32_t portnum, uint8_t * data)
return ERR_BAD_PARAM; /* trying to change the state of DOWN port */
if (p->state == 4) {
- if (p->lid > 0 && p->lid < maxlinearcap
+ if (node->type != SWITCH_NODE && p->lid > 0 && p->lid < maxlinearcap
If this is needed, it should be in a separate patch with more info as to
why it's needed/how to reproduce.
I'm going to commit this PR without this change.
—
You are receiving this because you authored the thread.
Reply to this email directly, view it on GitHub
<#2 (comment)>, or mute
the thread
<https://github.com/notifications/unsubscribe-auth/AmzVcK1kKQ1sxhTO4X455RwBU8VDtMIKks5uc8L5gaJpZM4VmfZZ>
.
|
2 participants
No description provided.