Konstantin Belousov

@zoneminder Unlocking in non-owning thread is UB.

Today we continued to try to figure out the problem with FreeBSD. It appears that c++ std::unique_locks are just not working. We lock in one thread, unlock in another the unlock doesn't appear to be happening. Anyone got insight on this? Doesn't affect Linux.

It's National Dog Day! Join us in celebrating by sharing some photos of your #dogsofFreeBSD.

@disruptnhandlr cmovcc of course

@disruptnhandlr Do movcc to unmapped address, for instance. AFAIR traps in the transactional region are suppressed, but transactions are aborted.

Question to x86 experts: what’s the best way to conditionally cause an abort of a TSX transaction in a single instruction? xabort_cc would be it but there’s no such instruction.

@cperciva @kaevans91 @markjdb2 @FrameworkPuter Ah right it is cninit() to bring up console early, for boot -d.

@kostikbel2 @kaevans91 @markjdb2 @FrameworkPuter It's brought up before we exit hammer_time. Something something debugging early kernel boot.

FreeBSD boot time -= 1971 ms (my laptop) FreeBSD boot time -= 1997 ms (c5.xlarge EC2 instance) cgit.freebsd.org/src/commit/?id… The "Reviewed by" is a hint to how long I've been working on this patch: I emailed out the first version for review in August 2017.

@kaevans91 @markjdb2 @cperciva @FrameworkPuter It is very strange, kbd controller should not be touched that early.

@markjdb2 @cperciva @kostikbel2 @FrameworkPuter bah, hint.atkbd.0.disabled

@kaevans91 @cperciva @FrameworkPuter How much memory do you have? If a lot, it is most likely vm_page_array[] initialization.

@cperciva Does this happen before SI_SUB_COPYRIGHT? i.e. was this the sucker introducing a three or even four second delay before the kernel started writing to the fb on my @FrameworkPuter?

@DarkainMX @freebsd Well, you can run even 1.0 binaries on amd64. And you do not need a jail for that, in fact.

FYI yet another reason why I absolutely LOVE @FreeBSD! Wanna run really old binaries? Sure, go for it. It "just works" out of the box thanks to these options. :) This means you can create a FreeBSD 4 jail on a FreeBSD 14 host.

@awesomekling ARMv8 8.7 List of LD* instructions goes from C6.2.98 to C6.2.177. Then ST* are in C6.2.248 to C6.2.311. Total 144.

Esp. if you start counting from the first change (~10 years ago) when trampolines were moved from stack to dedicated shared pages (to allow stacks be nx).

Signal trampolines on FreeBSD/amd64 (both 64bit and 32bit) are now proper shared libs. This is one of the most complicated NOPs I ever implemented.

@klarainc @vermaden # cc -v gcc version 2.7.2.1

Tell us when you started using #FreeBSD without telling us the version or the year!

@__phantomderp @bgianf Each return from kernel mode to userspace checks for pending signals and deliver them. Not only runtime calls, but also returns from hw interrupts (think timers). Look for e.g. setitimer(2) to get periodic nudge.

@bsdimp Can we have this as devfs aliases, e.g. symlinks under /dev/UEFI/ ?

Prior to this hint, this specific NVMe drive was nvme2: nvme2: <Generic NVMe Device> mem 0xe6b00000-0xe6b03fff irq 150 at device 0.0 numa-domain 2 on pci10 Serial(X) will also wire to uart unit to a serial with _UID X Next stop: Adding USB support (PCI and ACPI work now)

Working on new hints: hint.nvme.77.at="UEFI:PcieRoot(2)/Pci(0x1,0x1)/Pci(0x0,0x0)" creates nvme77: <Generic NVMe Device> mem 0xe6b00000-0xe6b03fff irq 150 at device 0.0 numa-domain 2 on pci10 verify with # devctl getpath UEFI nvme77 PcieRoot(2)/Pci(0x1,0x1)/Pci(0x0,0x0)

@ed_maste only SIGKILL, ^C wouldn't work

@joshbloch Is it something about class init procedure? The class of the lambda is inner, so it waits for init lock of Oddity, while rules allow it continue if lambda class is nestingly inited in the same thread that Oddity.

Puzzler: What's it print? Why? public class Oddity { private static int[] o = ThreadLocalRandom.current().ints() .limit(1000000) .parallel() .filter(i -> (i&1)==1).toArray(); public static void main(String[] args) { System.out.println(o.length); } }

I was pleasantly surprised that this took less than a week from user report to actual fix cgit.freebsd.org/src/commit/?id…

software.intel.com/content/dam/de… remote TLB shootdown

@MOS_8502 0f 45 ... is cmovne, which is pentiumpro and later. K6-2 did not have that instructions. FreeBSD 13 (actually clang) compiles i386 code for ppro.

@MengTangmu @thatcks You are mixing vnode locks and range locks. In fact, on 13, if all pages to be read are resident, UFS does not take vnode lock for read at all (so not affected by e.g. fsync). But range lock is taken to guarantee atomicity.

@kostikbel2 @thatcks Yeah. But that is not a VFS limitation, it could set MNTK_SHARED_WRITES (like ZFS). Then it would have new synchronisation problems to solve, of course, but it would be very nice if fsync() and write() didn't block read() (longer version of this tweet: wiki.postgresql.org/wiki/FreeBSD/A…)

Concurrent reads and writes to the same region of a file. postgresql.org/message-id/fla…