HN Gopher Feed (2017-08-04) - page 1 of 10
Ryzen CPU HyperThreads break if 100% busy and interrupted to Top of
31 points by soundshttps://svnweb.freebsd.org/base?view=revision&revision=321899
StillBored - 13 minutes ago
Its like we have all collectively forgotten that the first release
of a hardware or software project is just an expensive beta. I
don't buy 1st gen microarches because i'm not interested in paying
top dollar to be a tester. What I find odd, is that apparently both
intel and AMD have also forgotten this, as Intel seems to be moving
toward making their enterprise customers the beta testers, while
AMD seems so desperate for marketshare as to have released zen as a
volume product before releasing it as a high end one. Meaning that
if they have to do a recall, they are both losing their entire
margin, as well as having to replace a large number of devices.
loeg - 42 minutes ago
Note that we still don't really know what's going on; just that
this change seems to alleviate the symptoms.
copx - 1 hours ago
I wonder if the Ryzen bugs can really be fixed with a microcode
nly - 1 hours ago
They will probably just issue an update to turn off
lewisj489 - 50 minutes ago
Ryzen CPU's aren't made by Intel.
kartD - 1 hours ago
Even with SMT disabled it causes the faults
strmpnk - 50 minutes ago
I don't think that is the case for this particular issue.
There may be other reported SMT bugs (still some instability)
but here it has to be a pair of HTs: if one hyperthread is
in a cpu-bound loop of any kind (can be in user mode),
and the other hyperthread is returning from an interrupt
via IRETQ ...
sqeaky - 28 minutes ago
Does this affect only BSDs then?I have a Ryzen machine I bought and
put into a Jenkins cluster. It does builds on a bunch of VMs
occasionally and pegs the CPUs. I have had no issue so far.
shmerl - 19 minutes ago
It affects Linux as well.
0x0 - 28 minutes ago
Not the first bug reported for Ryzen. Wasn't there a couple of
others too, one with linux locking up and another triggered by the
ocaml compiler emitting opcodes refering AH/BH/CH/DH registers in a
moogly - 20 minutes ago
The OCaml thing was Skylake. But at least that one was fixed back
bloaf - 18 minutes ago
The ocaml compiler found a bug in Intel's Skylake/Kaby Lake. If
they found one for Ryzen too, I haven't heard about
cma - 1 hours ago
> When a cpu-thread stalls in this manner it appears to stall
INSIDE the microcode for IRETQ. It doesn't make it to the return
pc, and the cpu thread cannot take any IPIs or other hardware
interrupts while in this state.So maybe fixable with a microcode
db48x - 8 minutes ago
Most cpu errata are. (Most of the rest are ignorable).
shmerl - 22 minutes ago
See this thread: https://community.amd.com/message/2815893