-
-
Notifications
You must be signed in to change notification settings - Fork 57
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Kvm syn #1408
base: devel
Are you sure you want to change the base?
Conversation
I finished the conversion. |
PVI cannot be enabled because then |
My only concern is that this raises the minimal kernel version, though I am not sure by how much and if it affects currently supported LTS distributions. I'll do a proper review when I have time (more likely tomorrow). |
Well, we only support bionic with |
torvalds/linux@01643c5#diff-57de5203cc2796c8dfb7cc4c308a2e3b01dde41795d495746034446d5d366c56 Bionic was released with 4.15 but 5.3 is available as well. Debian stable is at 4.19. So that seems ok, now "just" need to review the code. Indeed PVI doesn't help much, though I have thought about setting CPL=IOPL=3 for DPMI and maybe even vm86. It'd would change the port i/o handling to be for KVM (instead of a GPF) and some other things but still keeps enough protections (because CPL=3 still). |
Would be an interesting experiment, but |
Well if they fix KVM on bionic, then |
Seems hangs not fixed? |
Does it need a rebase? |
Even worse after rebase:
|
No, its not, its a bad rebase. :) |
Only many FS tests fail. |
Locally these tests pass. |
Moreover, the same FS tests that |
Is it possible with this branch that kvm just got a whole lot slower, as there seem to be many tests that are failing with timeout? |
Mm, yes, seems to have got twice |
Even more than that in some cases, the 'PP-DOS-GIT MFS DOSv3 share open set file attrs DOSv2' test took ~1 sec on GA (emulated) and ~12 secs on Travis(kvm) |
Can't confirm that weirdness:
... on kvm_sync and:
... on devel, so it only became slightly |
Interestingly when I run that test here I see a kvm error, just like those I saw a while back.
|
I'll retest on devel to see if the error is specific to kvm_syn |
test passes on current devel
|
Please post output of cpuid command |
Extract saved regs into sync_regs to continue immediately at guest.
Only exception frame left.
Even if ring0 is interrupted and exit reason is not hlt, we know its gotta be hlt as nothing else have remained.
This simplifies the code.
Andrew, could you please check |
Compare with old copies. This should speed up the things.
Seems no travis check any more? |
No Travis for anything but the weekly cron at the moment. I need to split out the kvm tests from the others and just run those. I'll try out the kvm_syn branch as soon as my devel test run is complete. |
So kvm_syn seems a lot better now than before (as far as i remember), as the kvm tests pass. I do see random GPF failures t1.zip with kvm_syn, and overall it's slower than devel.
|
What are the failures, are they Also good to know its not 30 times |
No, it's unrepeatable and seems random. I run the test again 5 times and it doesn't fail. In both the log files attached above the error seems similar, which to me looks like an attempt to execute zeroed memory. |
Not sure if there is still a reason |
What happened with the Core 2 Duo kernel problem, was there any agreement over the seemingly bad patch in 5.10 (and its backport)? |
Yes, absolutely. |
Thanks. I just read the kvm list and can see your back and forth with the kvm kernel devs, so it's still rumbling on... |
Its a minor details only. |
139e850
to
0e5522d
Compare
This converts kvm to sync_regs API.
But the remaining bit is to handle exceptions
w/o kvmmon.
Bart, would you like to review this, and maybe
complete the removal of kvmmon?