From d296065b477378cf17d503f1c5430ce235122f30 Mon Sep 17 00:00:00 2001 From: Kevin Hui Date: Thu, 18 Apr 2024 12:09:11 -0700 Subject: [PATCH] Bump SKID_MARGIN_RCBS from 60 to 70 Summary: For AMD hosts I'm seeing issues with trying to run `mkfs.ext3` at large sizes using Antlir. We can raise it slightly to deal with this issue. If it becomes too much of a problem for Intel hosts, we can always break out compilation for the two processor types Reviewed By: jasonwhite Differential Revision: D56271985 fbshipit-source-id: 1af3a007363a241cfa1c3e12585d6f837b274d19 --- reverie-ptrace/src/timer.rs | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/reverie-ptrace/src/timer.rs b/reverie-ptrace/src/timer.rs index 88f3aa2..7b2a11e 100644 --- a/reverie-ptrace/src/timer.rs +++ b/reverie-ptrace/src/timer.rs @@ -403,8 +403,8 @@ impl EventStatus { /// `rr` uses a value of 100 for almost all platforms, but with precise_ip = 0. /// Enabling Intel PEBS via precise_ip > 0 seems to reduce observed skid by 1/2, /// in synthetic benchmarks, though it makes counter _values_ incorrect. As a -/// result, we choose 60. -const SKID_MARGIN_RCBS: u64 = 60; +/// result, we choose 70. +const SKID_MARGIN_RCBS: u64 = 70; /// We refuse to schedule a "perf timeout" for this or fewer RCBs, instead /// choosing to directly single step. This is because I am somewhat paranoid