Project

General

Profile

Actions

Bug #14886

open

vmm_drv_test needs to be less IPS-dependent

Added by Dan McDonald about 2 months ago. Updated about 2 months ago.

Status:
New
Priority:
Normal
Assignee:
-
Category:
-
Start date:
Due date:
% Done:

0%

Estimated time:
Difficulty:
Medium
Tags:
Gerrit CR:
External Bug:

Description

During upstream merge of #14864 I discovered that the vmm_drv_test driver is depending on the IPS manifest, and strictly speaking it shouldn't.

I filed SmartOS bug OS-8401 (https://smartos.org/bugview/OS-8401) and committed https://github.com/TritonDataCenter/illumos-joyent/commit/cef7020412fbcaba82b6bbbc172e0f614dde1e80 to our downstream illumos. It appears to work for us, and since it leverages the already-in-gate techniques used in os-test's ksensor_test driver (#12830), we should make sure it is in -gate as well.


Related issues

Related to illumos gate - Bug #14864: want device for testing vmm_drv interfaceClosedPatrick Mooney

Actions
Related to illumos gate - Feature #12830: Want centralized ksensor frameworkClosedRobert Mustacchi

Actions
Actions #1

Updated by Dan McDonald about 2 months ago

  • Related to Bug #14864: want device for testing vmm_drv interface added
Actions #2

Updated by Dan McDonald about 2 months ago

Actions #3

Updated by Dan McDonald about 2 months ago

  • Subject changed from vmm_drv_test needs to be less IPS-dependent to vmm_drv_test needs to be less IPS-dependent
Actions #4

Updated by Joshua M. Clulow about 2 months ago

I don't really think it's open and shut that it shouldn't use the infrastructure we have for installing drivers?

Actions #5

Updated by Joshua M. Clulow about 2 months ago

To be clear, SmartOS doesn't have to use IPS, but should (and this is generally true) be using the manifests wholesale to construct the correct driver configuration across the board -- which also covers this. I don't think we want to avoid using the first class facilities we have for expressing driver metadata because of one distribution.

Actions #6

Updated by Dan McDonald about 2 months ago

I mentioned the `ksensor_test` driver earlier, which does not have an IPS driver entry in its delivery to system-test-ostest.mf. Because it and vmm_drv_test are drivers purpose-built for testing, I thought maybe they should be consistent in their delivery.

Further inspection for "driver" and "drv" in $SRC/pkg/manifests/*test* yield a 50-50 split between in-IPS driver delivery, vs. module-only driver delivery. (Note that "driver" in the crypto tests is a dependency.)

nowhere(~/ws/illumos-gate)[1]% grep -l /drv/ usr/src/pkg/manifests/*test*
usr/src/pkg/manifests/system-bhyve-tests.p5m
usr/src/pkg/manifests/system-file-system-zfs-tests.p5m
usr/src/pkg/manifests/system-io-tests.p5m
usr/src/pkg/manifests/system-test-ostest.p5m
nowhere(~/ws/illumos-gate)[0]% grep -l ^driver usr/src/pkg/manifests/*test*
usr/src/pkg/manifests/system-bhyve-tests.p5m
usr/src/pkg/manifests/system-file-system-zfs-tests.p5m
nowhere(~/ws/illumos-gate)[0]% 

Clearly there's in-gate inconsistency here. Consider this issue my vote for (in the case of tests) module-only delivery. I can be convinced otherwise, however.

Actions

Also available in: Atom PDF