Project

General

Profile

Patrick Mooney's activity

From 2020-07-01 to 2020-07-30

2020-07-30

03:50 PM illumos gate Bug #12999: MSR_AMD_DE_CFG is twice defined
I tested this via @wsdiff(1)@ and observed that beyond the normal DOF noise, there were no reported differences betwe... Patrick Mooney
03:41 PM illumos gate Bug #13000 (Closed): bhyve sources should reflect license details
As described in #12848, changes made to bhyve in illumos are dual licensed under both the upstream BSD license as wel... Patrick Mooney
02:22 AM illumos gate Bug #12998: OpenBSD needs DE_CFG MSR on AMD bhyve
With the proposed fix, an OpenBSD 6.7 guest is able to start on bhyve without panicking on the DE_CFG MSR access. A ... Patrick Mooney
01:26 AM illumos gate Bug #12936: bhyve vlapic needs ability to bypass isrvec_stk
It seems that few OSes use nested interrupt priorities these days. While illumos uses the TPR (unlike Linux) to elev... Patrick Mooney

2020-07-29

09:54 PM illumos gate Bug #12999 (Closed): MSR_AMD_DE_CFG is twice defined
While searching through our sources for references to MSR_AMD_DE_CFG (0xc0011029), I found that we've accidentally de... Patrick Mooney
08:40 PM illumos gate Bug #12998: OpenBSD needs DE_CFG MSR on AMD bhyve
Scraping some public AMD docs for how they've defined certain bits in this MSR for different families.
Family 10h:...
Patrick Mooney
08:02 PM illumos gate Bug #12998 (Closed): OpenBSD needs DE_CFG MSR on AMD bhyve
When attempting to boot openbsd 6.7 under bhyve on an AMD (Rome) machine, it attempts to unconditionally write to the... Patrick Mooney
03:58 PM illumos gate Bug #12997: clean up stale vdev driver in bhyve
As noted, this code is not reference by anything else in bhyve, and has been effectively dead since the FreeBSD 11 re... Patrick Mooney

2020-07-28

08:13 PM illumos gate Bug #12997 (Closed): clean up stale vdev driver in bhyve
When bhyve was imported from the Pluribus wad, it included a "vdev" driver in @uts/i86pc/io/vmm/io@. This has long s... Patrick Mooney
07:27 PM illumos gate Bug #12996 (Closed): bhyve kernel should be wscheck clean
When bhyve was imported into illumos, broad portions of it were added to the exception lists including the one for ws... Patrick Mooney
04:00 AM illumos gate Bug #12994 (New): CTASSERT should not clash on same line num
I was doing work on bhyve, which uses CTASSERT()s to enforce sizing and alignment for certain data structures. While... Patrick Mooney

2020-07-27

08:39 PM illumos gate Bug #12991 (Closed): bhyve vlapic should SIPI more carefully
Certain writes to the ICR LAPIC register can initiate transmission of a startup IPI (SIPI) to one of the other applic... Patrick Mooney
06:34 PM illumos gate Bug #12990 (Closed): improve interface boundary for bhyve ins/outs
While #12989 notes that in/out emulation is more straight forward than MMIO, due to the decoding acceleration present... Patrick Mooney
06:25 PM illumos gate Bug #12989 (Closed): improve interface boundary for bhyve MMIO
Doing instruction decoding and emulation for MMIO is a necessary task for bhyve. Unlike the in/out instructions, whi... Patrick Mooney

2020-07-24

09:46 PM illumos gate Bug #12981: vdev_hold prone to NULL dereference
Booting up a BE with this patch on the pool in question, I'm able to run a @zpool scrub@ without the machine panickin... Patrick Mooney
06:42 PM illumos gate Bug #12981 (Closed): vdev_hold prone to NULL dereference
I went to scrub my root pool which had been subject to testing as part of #12894. (It was a root pool that had underg... Patrick Mooney

2020-07-15

07:02 PM illumos gate Bug #12961 (In Progress): want support for AMD AVIC in bhyve
AMD's Advanced Virtual Interrupt Controller (AVIC) offers similar functionality to APICv on Intel CPUs. It has the p... Patrick Mooney
06:33 PM illumos gate Bug #12916: bhyve should be able to limit vmx capabilities
I fired this up on an older (and less capable) Westmere machine to make sure it properly detected and enabled the ava... Patrick Mooney
05:55 PM illumos gate Bug #12916: bhyve should be able to limit vmx capabilities
With this patch applied, I can see the VMX capabilities properly populated on an Ivy Bridge test system:... Patrick Mooney
05:43 PM illumos gate Bug #12917: bhyve should always use Unrestricted Guest
With this change applied, my bootrom-utilizing guests are still able to boot and run without issue. Without Unrestri... Patrick Mooney
05:49 AM illumos gate Bug #12960 (Closed): bhyve needs better SIPI notification
Today, when guest actions lead to a SIPI being issued by a vlapic to one of the not-yet-active vCPUs, this predicamen... Patrick Mooney

2020-07-14

03:18 PM illumos gate Bug #12939: avoid packing and bitfields in bhyve VMCB
Mike Zeller has also tested this in his environment, running VMs over the weekend on platform featuring the change. Patrick Mooney

2020-07-13

09:11 PM illumos gate Bug #12939: avoid packing and bitfields in bhyve VMCB
I fired up bhyve on a hardware machine and confirmed that my test instance booted and ran fine. Inspecting the VMCB ... Patrick Mooney
09:10 PM illumos gate Bug #12939: avoid packing and bitfields in bhyve VMCB
In a test compile, tsoome reported that bhyve is gcc9 clean with the proposed fix. Patrick Mooney
08:17 PM illumos gate Bug #12956: bhyve should set TMR on intr accept
See also: "OS-7622":https://smartos.org/bugview/OS-7622 bhyve vioapic writes can deadlock instance Patrick Mooney
08:15 PM illumos gate Bug #12956 (New): bhyve should set TMR on intr accept
Currently, bhyve manages the contents of trigger mode registers (TMRs) in the vlapic based on the state of interrupt ... Patrick Mooney

2020-07-10

03:00 PM illumos gate Bug #12922: bhyve vmm ops could have stronger guardrail
I've tested this on AMD and Intel systems to confirm that bhyve loads properly and is able to start VMs without issue. Patrick Mooney
03:38 AM illumos gate Bug #12945 (New): SMT exclusion should be by pid rather than zoneid
One part of #11048 was an API to restrict workloads from running on SMT pairs when vulnerable (or potentially aggress... Patrick Mooney

2020-07-09

11:59 PM illumos gate Bug #12511: vioblk should preallocate virtio chains
I have a test VM using a virtio-block device to back its root (and only) zpool. I booted it up with this change and ... Patrick Mooney
08:51 PM illumos gate Bug #12511 (In Progress): vioblk should preallocate virtio chains
Patrick Mooney
07:43 PM illumos gate Bug #12511: vioblk should preallocate virtio chains
I have a build VM (running OmniOSce 151034) which has been hit by this recently. An illumos-gate build will place th... Patrick Mooney

2020-07-08

08:17 PM illumos gate Bug #12939 (Closed): avoid packing and bitfields in bhyve VMCB
The control and state structs for the SVM VMCB in bhyve are defined as packed structs and make heavy use of C bitfiel... Patrick Mooney

2020-07-07

09:20 PM illumos gate Feature #12902: epoll should nominally support EPOLLEXCLUSIVE
It should be noted that there were no regressions in any of the other epoll tests when checking the proposed fix. Patrick Mooney
09:08 PM illumos gate Bug #12936 (Closed): bhyve vlapic needs ability to bypass isrvec_stk
The emulated APIC in bhyve (vlapic) contains a data structure @isrvec_stk@ which tracks the vectors of delivered inte... Patrick Mooney
04:09 PM illumos gate Bug #12910: EPOLLONESHOT should not remove fd
Prior to integration of the change, the oneshot epoll test fails, as expected:... Patrick Mooney
02:52 PM illumos gate Bug #12912: epoll mishandles excessive timeout negativity
Prior to integration of this change, a portion of the epoll test suite shows a problem relating to timeouts:... Patrick Mooney

2020-07-06

09:16 PM illumos gate Bug #12923: SMT exclusion should not balk at GZ
Mike Zeller confirmed that a mix of bhyve instances (both in the GZ and in NGZs) ran fine on a DEBUG kernel featuring... Patrick Mooney
08:54 PM illumos gate Bug #12909: epoll should better detect fd reassignment
I've re-run the epoll test suite to check for unexprected regressions and found none. The behavior in question was f... Patrick Mooney

2020-07-02

11:04 PM illumos gate Bug #12908: epoll should exclude normal files/directories
This has been in SmartOS (and OmniOSce) since 2017 Patrick Mooney
10:25 PM illumos gate Bug #12908: epoll should exclude normal files/directories
Without the fix, the epoll test suite notes problems for the file and dir tests:... Patrick Mooney
10:29 PM illumos gate Bug #12907: epoll_ctl can avoid EINTR entirely
I added an illumos-specific test to the epoll suite to verify the @/dev/poll@ write limit imposed as part of this pat... Patrick Mooney

2020-07-01

09:13 PM illumos gate Bug #12923: SMT exclusion should not balk at GZ
With the proposed patch, I was able to start a bhyve instance in the global zone on a DEBUG platform without an immed... Patrick Mooney
07:09 PM illumos gate Bug #12923 (Closed): SMT exclusion should not balk at GZ
The SMT exclusion, used to mitigate some micro-architectural leakage between threads on a single core, currently @ASS... Patrick Mooney
07:27 PM illumos gate Bug #12922: bhyve vmm ops could have stronger guardrail
This is just a minor clean-up, and considering it's already a well-protected error case, should not be at all visible... Patrick Mooney
06:23 PM illumos gate Bug #12922 (Closed): bhyve vmm ops could have stronger guardrail
When bhyve is initialized, it chooses a @vmm_ops@ set, based on the physical CPU (Intel or AMD) it's running on. The... Patrick Mooney
07:18 PM illumos gate Bug #12921: bhyve IPI statistics should not be a matrix
With the patch, the summarized stats are available:... Patrick Mooney
07:11 PM illumos gate Bug #12921: bhyve IPI statistics should not be a matrix
Prior to the change, querying statistics with @bhyvectl@ would show VM_MAXCPU count of IPI statistics per-vcpu:
<pre...
Patrick Mooney
06:21 PM illumos gate Bug #12921 (Closed): bhyve IPI statistics should not be a matrix
The statistics facility in bhyve currently stores counts of IPIs sent per-vcpu in a VM_MAXCPUs-sized array. While th... Patrick Mooney
 

Also available in: Atom