| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-cyborg - 2026-07-28 | |||
| 14:06:21 | jgilaber | e.g 'insmod fake_pci_sriov.ko num_pfs=2 pf_mode=sriov,mdev' would create two pfs, the first one sriov, the second with mdevs | |
| 14:06:50 | sean-k-mooney | that shoudl not be via kernel args | |
| 14:07:00 | sean-k-mooney | it shoudl be via per PF files in /sys | |
| 14:07:41 | sean-k-mooney | that the same feedback i gave to chandankumar for the nvm devices as well | |
| 14:08:05 | sean-k-mooney | each PF shoudl have a personality file that the pf reads to determin the type of VF to create | |
| 14:08:17 | sean-k-mooney | which you set before you allocate teh VFs | |
| 14:08:55 | sean-k-mooney | that is similar ot how vGPUs work and other hader deivce that supprot configurable types for there VFs | |
| 14:10:09 | jgilaber | so the user/devstack plugin creates those files? | |
| 14:10:10 | sean-k-mooney | basiclly /sys/bus/pci/devices/<PF>/vf_personality | |
| 14:10:31 | sean-k-mooney | when you mode prope the module it will create teh file | |
| 14:10:35 | opendevreview | Gihong Lee proposed openstack/cyborg master: Add shared lspci helpers to common utils https://review.opendev.org/c/openstack/cyborg/+/998726 | |
| 14:10:35 | opendevreview | Gihong Lee proposed openstack/cyborg master: Migrate accelerator drivers to shared lspci helpers https://review.opendev.org/c/openstack/cyborg/+/998817 | |
| 14:10:40 | sean-k-mooney | and you need to write the desired value to it | |
| 14:11:05 | sean-k-mooney | we shoudl also have a /sys/bus/pci/devices/<PF>/vf_personalitys that list the avlaible type space seperated | |
| 14:11:39 | sean-k-mooney | that or follow the cpu governors nameing convetion of adding _avaialble | |
| 14:12:13 | jgilaber | ok I see | |
| 14:12:21 | sean-k-mooney | [15:11:16]> cat /sys/bus/cpu/devices/cpu0/cpufreq/scaling_governor | |
| 14:12:23 | sean-k-mooney | ─────┬───────────────────────────────────────────────────────────────────────────────────────────── | |
| 14:12:25 | sean-k-mooney | │ File: /sys/bus/cpu/devices/cpu0/cpufreq/scaling_governor | |
| 14:12:27 | sean-k-mooney | ─────┼───────────────────────────────────────────────────────────────────────────────────────────── | |
| 14:12:29 | sean-k-mooney | 1 │ powersave | |
| 14:12:31 | sean-k-mooney | ─────┴───────────────────────────────────────────────────────────────────────────────────────────── | |
| 14:12:33 | sean-k-mooney | ~ via 🐹 v1.25.12 via python: v3.14.6 on ☁️ smooney@redhat.com | |
| 14:12:35 | sean-k-mooney | [15:12:08]> cat /sys/bus/cpu/devices/cpu0/cpufreq/scaling_available_governors | |
| 14:12:37 | sean-k-mooney | ─────┬───────────────────────────────────────────────────────────────────────────────────────────── | |
| 14:12:39 | sean-k-mooney | │ File: /sys/bus/cpu/devices/cpu0/cpufreq/scaling_available_governors | |
| 14:12:40 | jgilaber | from UX side it seems a bit more cumbersome | |
| 14:12:41 | sean-k-mooney | ─────┼───────────────────────────────────────────────────────────────────────────────────────────── | |
| 14:12:43 | sean-k-mooney | 1 │ performance powersave | |
| 14:12:45 | sean-k-mooney | ─────┴──────────── | |
| 14:12:57 | sean-k-mooney | the module approch forces you do reload the module to change it | |
| 14:13:08 | sean-k-mooney | and it does not allwo you to eaislly do this on a per device basis | |
| 14:13:21 | jgilaber | that's true | |
| 14:13:41 | chandankumar | sean-k-mooney: jgilaber I was trying to come out like this https://gistpreview.github.io/?9cf18df5754db0948330f0cc5110c652 but not yet implemented | |
| 14:14:43 | sean-k-mooney | skiming that that is mostly aligned on the sysfs interface to configure this | |
| 14:14:54 | sean-k-mooney | basiclly secitn 6 | |
| 14:15:31 | sean-k-mooney | we will need to document this in the contibutor docs for the module and cpature the logic as part of this work | |
| 14:15:49 | sean-k-mooney | jgilaber: the other option woudl be to have multiple modules that we load seperatly | |
| 14:15:53 | sean-k-mooney | with a shared core | |
| 14:15:59 | sean-k-mooney | i.e. one per fake device type | |
| 14:16:13 | sean-k-mooney | but i was kind fo hoping to only need one module for htis in general | |
| 14:16:47 | jgilaber | yeah, I'll try to rework the patch with this new approach | |
| 14:17:02 | jgilaber | and we can discuss again | |
| 14:17:17 | sean-k-mooney | so i think we need to spit this out into its won commit | |
| 14:17:36 | sean-k-mooney | i.e. one commit that intoduces the personatiy file reporting and configuration with just uart for now | |
| 14:17:49 | sean-k-mooney | the you can chandan can hopefully build on that in parallel | |
| 14:18:06 | sean-k-mooney | otherwise we will have to sequence both feature to not confilct | |
| 14:18:20 | jgilaber | yes, that is what I did already. The linked patch does not have the mdev bits | |
| 14:19:04 | sean-k-mooney | ah https://review.opendev.org/c/openstack/cyborg/+/998603 | |
| 14:19:10 | sean-k-mooney | ok so that is just the config interface | |
| 14:19:35 | sean-k-mooney | i woudl not be agaisn haveing module option to have a shorthand way to preallocate soem device fo a specifi tyep as well | |
| 14:19:44 | sean-k-mooney | i.e. UART=2 NVME=4 | |
| 14:19:55 | sean-k-mooney | buti still think the sysfs way is more flexible | |
| 14:20:32 | chandankumar | what about bringing both the approach and revisit in next meeting? | |
| 14:20:38 | jgilaber | ack, thanks for the feedback, I got what I need to continue | |
| 14:20:50 | sean-k-mooney | basiclly i woudl like to have a clean way to be able to do soemtihng like pci_sim_toplogy=uart:nova,nvme:cyborg,mdev:cyborg | |
| 14:20:50 | jgilaber | a note on timeline, I'll be on PTO starting Aug 5th | |
| 14:21:12 | sean-k-mooney | and have devstack create 3 pf assign 1 to nova and 2 to cyborg ectra | |
| 14:21:16 | jgilaber | so ideally it would be good to advance on this before the next meeting | |
| 14:21:37 | sean-k-mooney | i think we can contiue the fonvcertion here and in gerrit | |
| 14:21:45 | jgilaber | +1 | |
| 14:21:45 | sean-k-mooney | we dont need to wait for the meeting to make progress | |
| 14:21:52 | chandankumar | sounds good | |
| 14:22:20 | chandankumar | Any more questions on this topic? | |
| 14:22:56 | chandankumar | moving to next one | |
| 14:23:04 | chandankumar | #topic Fix for guest reboot test | |
| 14:23:24 | chandankumar | jgilaber is working on fixing guest reboot test | |
| 14:23:29 | chandankumar | below are the reviews | |
| 14:23:40 | chandankumar | #link 998283: Ensure guest is not reachable after reboot | https://review.opendev.org/c/openstack/cyborg-tempest-plugin/+/998283 | |
| 14:23:49 | jgilaber | yes, I see sean-k-mooney has already reviewed it, thanks | |
| 14:23:53 | chandankumar | #link Tested with pci and mdev tests in 997364: [WIP] Configure mdev driver lifecycle tests | https://review.opendev.org/c/openstack/cyborg/+/997364 | |
| 14:24:08 | jgilaber | it seems to fix the problem, at least I haven't seen any failure with it | |
| 14:24:15 | sean-k-mooney | ya i thinkt hat is ok ot go i was mainly waiting on ci on the followup | |
| 14:24:31 | sean-k-mooney | basiclly to get more executions of the wait for unablable logic | |
| 14:25:01 | sean-k-mooney | im going to do one more recheck but if that passes i think ill merge it later today | |
| 14:25:12 | jgilaber | works for me, thanks! | |
| 14:25:43 | chandankumar | thank you jgilaber for fixing the test. | |
| 14:25:50 | chandankumar | Moving to next one | |
| 14:25:57 | chandankumar | #topic Reviews | |
| 14:26:26 | chandankumar | Any reviews needs attension please added it to https://etherpad.opendev.org/p/cyborg-2026.2-status | |
| 14:26:53 | chandankumar | from etherpad I can see generic mdev driver is ready for review | |
| 14:27:08 | chandankumar | #link https://review.opendev.org/q/topic:%22generic-mdev-driver%22 | |
| 14:27:34 | sean-k-mooney | we need to review the nova part as well first right | |
| 14:27:48 | chandankumar | Do we want to add the nova piece to the etherpad? | |
| 14:27:50 | sean-k-mooney | at what point dose the depency start | |
| 14:28:33 | sean-k-mooney | i guess we can | |
| 14:28:37 | chandankumar | Here is the link to nova piece https://review.opendev.org/q/topic:%22cyborg-vgpu-support%22 | |
| 14:28:37 | jgilaber | technically I guess at https://review.opendev.org/c/openstack/cyborg/+/997050 | |
| 14:29:04 | melwitt | it's dependency that does not need or use Depends-On? | |
| 14:29:07 | sean-k-mooney | well that what i was debating | |
| 14:29:14 | jgilaber | but I added the depends-on at the last patch https://review.opendev.org/c/openstack/cyborg/+/997364/25 | |
| 14:29:34 | sean-k-mooney | melwitt: so we can fully enabel the cybrog side without techinally mdoifying nova | |
| 14:29:44 | sean-k-mooney | melwitt: but it wont actully work until we modify nova end ot end | |
| 14:29:51 | melwitt | oh ok it's on the last patch | |
| 14:29:51 | jgilaber | exactly | |
| 14:29:59 | sean-k-mooney | the reporting and device binding will on the cybrog cide | |
| 14:29:59 | melwitt | was just curious | |
| 14:30:02 | sean-k-mooney | *cide | |
| 14:30:12 | sean-k-mooney | ...side | |
| 14:30:12 | jgilaber | the last one enables the tests so it needs the nova patches to work | |