Re: vmm(4): swap in log(9) for printf(9) [vmx 3/3]

2021-11-28 Thread Mark Kettenis
> From: Dave Voutila > Date: Sun, 28 Nov 2021 22:51:59 -0500 > > The last vmm diff I'll be sending tonight...promise! This swaps out > usage of printf(9) outside the autoconf(4) functions. > > The reason for this change is printf(9) could acquire a sleepable > lock. Huh? /* * printf: print a

Re: vmm(4): bump vmclear spinout [vmx 1/3]

2021-11-28 Thread Mike Larkin
On Sun, Nov 28, 2021 at 10:32:47PM -0500, Dave Voutila wrote: > Smallest of the VMX/VMCS stability diffs. This bumps the spinout to be > the same number of ticks used by the mplock debug. This is needed on > older/slower hosts. > > ok? > > -dv > > diff e8c587551f20ba6fdaa0f483ea768aade9f66f7d >

vmm(4): swap in log(9) for printf(9) [vmx 3/3]

2021-11-28 Thread Dave Voutila
The last vmm diff I'll be sending tonight...promise! This swaps out usage of printf(9) outside the autoconf(4) functions. The reason for this change is printf(9) could acquire a sleepable lock. For VMX-based hosts, this can corrupt the active VMCS as the scheduler may decide to migrate the

vmm(4): restore vmcs after sleep points [vmx 2/3]

2021-11-28 Thread Dave Voutila
This diff removes instability from VMX-based hosts by either removing the possibility of the process sleeping while the VMCS is active or reloading it if we had no choice. A mutex is added to help guard the VMCS state so testing with witness has helped verify the diff. The rwlock on the cpu

vmm(4): bump vmclear spinout [vmx 1/3]

2021-11-28 Thread Dave Voutila
Smallest of the VMX/VMCS stability diffs. This bumps the spinout to be the same number of ticks used by the mplock debug. This is needed on older/slower hosts. ok? -dv diff e8c587551f20ba6fdaa0f483ea768aade9f66f7d 981a8cfd4e1dfe412e9c72fb5b47e7e46813bfbb blob -

mpsafe dwge(4)

2021-11-28 Thread Jonathan Matthew
This applies our normal strategies for making network drivers mpsafe, and also writes to GMAC_TX_POLL_DEMAND once per call to dwge_start() rather than once per packet, and returns rx slots once per interrupt rather than once per packet. I've tested this on a rockpro64, where it makes tcpbench

Re: vmm(4) testers wanted: VMX stability fixes

2021-11-28 Thread Dave Voutila
Dave Voutila writes: > Greetings tech@, > > I'm looking for testers of the following diff which addresses spurious > VMCS corruption on Intel VMX hosts. If since upgrading to 7.0 or > following -current you've had guests die and errors about failure to > vmresume, please try this diff. I've

Re: slaacd(8): prevent crash when interface disappears

2021-11-28 Thread Florian Obser
On 2021-11-28 05:13 UTC, Klemens Nanni wrote: > On Thu, Nov 18, 2021 at 09:02:00AM +0100, Florian Obser wrote: >> This might be the crash kn@ was seeing once in a blue moon. > > I somewhat doubt it, since slaacd crashed on my notebook using trunk(4) > over em(4) and athn(4), none of these

Re: slaacd(8): prevent crash when interface disappears

2021-11-28 Thread Klemens Nanni
On Thu, Nov 18, 2021 at 09:02:00AM +0100, Florian Obser wrote: > This is split in two for easier review and I also intend to commit it > like this. > > The first diff shuffles setting of if_index around so that it's > available in all switch cases and uses it consistently instead of >