Where is this screenshot from? Would like to read the larger article (if it's from one)
-
-
- 1 more reply
New conversation -
-
-
when is this from- has wave of cpu bug mitigations, requiring basically that the CPU get steam-cleaned between context switches affected this? also, the time from "syscall" starting decode to "sysret" retiring may be 50ns, but what perf penalty on followng insns b/c caches cool?
-
Doesn’t PCID fix all those issues?
- 2 more replies
New conversation -
-
-
ok but there's still a ton of business happening in the generic syscall handler (even before you even get to the code that does something specific for a particular syscall)
-
but what exactly does the author mean by "round trip"? (he can't mean "from the point of executing the syscall instruction in user code until control reaches the instruction that immediately follows the syscall", right?)
- 3 more replies
New conversation -
-
-
And yet the cheapest syscall took 500-1000 ns even before meltdown & spectre mitigations...
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
this isn’t the relevant metric to track for the expense of context switches though, right? the “fastest kernel in the west” sel4 is at a different order of magnitude (~100-200ns iirc) to round trip through the kernel.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
do we still need to flush TLB when context switching or has that been solved? cuz that might be quick but can have serious deferred perf penalties
-
Right? Isn't that the real cost you care about?
End of conversation
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.