DragonFly’s tcp keepalive was changed from milliseconds to seconds. This happened in both DragonFly-current and in the 5.6 release, and it changes the networking API, which means a dports rebuild is needed… or a pkg upgrade, for which happily all packages have been rebuilt. So, on your next update of the system, be sure to update packages too.
The module formerly known as ‘radeonkms’ is now just plain ‘radeon’. There have been changes in other commits, but this is the only usage change.
Matthew Dillon has made some changes to DragonFly’s scheduling system His further tests show an improvement in basic forking.
There’s several bug fixes that have gone into DragonFly over the past few days, in an attempt to track down an odd bug. They’ve been committed to 5.6, too, so you can pick them up if you update.
I imagine this will turn into a 5.6.2 release, but not until we find the cause of the error mentioned in that link.
You’ll all be happy to know ACPI errors are less noisy now. (And it was updated to 20190509, before the 5.6 release.)
Matthew Dillon’s made a change to the DragonFly kernel that could be disruptive, but will help make sure chromium runs. If you update after this point, make sure to update your dports, too, just to be sure everything is in sync. This applies to 5.6 and 5.7.
Because of some changes Matthew Dillon made to maxvnodes calculation in DragonFly, you may find yourself using 5%-10% less RAM. If you’ve upgraded to 5.6, you already have this benefit.
DragonFly now has retpoline turned on (stats included in that link) as a side effect of having gcc-8 as default, and SMAP/SMEP are also supported. I enjoy just saying these words out loud. SMEP SMEP SMEP SMEPSMEPSMEPSMEP.
I’m still backlogged, so here’s a May 14th mitigation in DragonFly for MDS attacks possible with Intel CPUs from 2011 onward. It’s in the current release.
A happy note for the end of the week: I like seeing cross-BSD work. (and POSIX)
A last-minute drm change in DragonFly 5.6 turned out to cause a reproducible lockup, so there’s changes in place for it. This means 5.6.1 will need to be rolled, which I will do in a day or two. If you want to update now, the normal buildworld/buildkernel process will get you this change.
This will turn into a real 5.6 release probably by weekend if no problems are found. See the tag commit message for a list of the commits since 5.4.
The next release of DragonFly should be smaller; Sascha Wildner and Rimvydas Jasinskas have removed or substituted enough packages on the installer image to drop the package disk usage 50%.
Rimvydas Jasinskas has done the tedious but useful work to update a number of utilities in DragonFly to newer vendor versions: LibreSSL 2.9.1, libarchive 3.3.3, xz 5.2.4, ldns 1.7.0, and OpenSSH 8.0p1.
opie(4) is no longer an option in several places in DragonFly. It’s also known as S/Key, and I’d be mildly surprised if you’ve used it.
Sascha Wildner has committed mandoc(1) to DragonFly to use for man, whatis, apropos, and other functions. One less GNU utility, and also means groff can come out of the base system. That is almost the last C++ code in base… I am not sure what remains.
Because of the ongoing pmap work from Matthew Dillon, building vkernels may not work for a short period in DragonFly-master. “A short period” usually means a few days for this sort of thing.
Update: all better now.
I’m jumping ahead in my very full queue of DragonFly items to post to something new: Matthew Dillon has committed extensive work to the virtual memory system in DragonFly. He has a message to users@ that sums it up.
If that’s not enough reading for you, I’ll point at commits where he reports speed improvements, changes to systat(1), or just 2-decade-old copyright items.
DragonFly 5.4.3 is out. My users@ post describes upgrading, as do the 5.4 release notes. This release has a fix for an Intel floating-point bug.
Images are available for download at various mirrors, too. If you’ve recently upgraded to 5.4, it’s the normal build process. There’s a brand new complete build of all packages uploaded, too, so plan on a ‘pkg update’.
This may never ever matter if you manage to avoid fdisk your whole life. But if you don’t pull that off, here’s the reminder: label your DragonFly slices with 108.
(Yes, I do in fact have a backlog of two months with DragonFly material. It’s been that constant.)