Date: Wed, 20 Jan 2021 12:18:44 +0200 From: Konstantin Belousov <kostikbel@gmail.com> To: Jakob Alvermark <jakob@alvermark.net> Cc: freebsd-current <freebsd-current@freebsd.org> Subject: Re: Waiting for bufdaemon Message-ID: <YAgDhJ9/YX29kH2a@kib.kiev.ua> In-Reply-To: <1d675d77-8bdb-c285-ffa2-28330b839734@alvermark.net> References: <369b3d82-98c5-b31e-6168-4003a042f174@FreeBSD.org> <YAGua8f8G57SbUMT@kib.kiev.ua> <556d40b8-92d7-303e-7d87-ea496d0ca733@FreeBSD.org> <YAG4g7dJbRWM7act@kib.kiev.ua> <ed52f8c1-5f43-b71e-1d56-625e2e064a7e@gwdg.de> <9ae3cc65-193a-39c3-4067-0d42e9f634b0@gwdg.de> <YAO9/cxnbGv5u3Bg@kib.kiev.ua> <adb9d42c-d855-1edf-a57c-9b16aad481b7@gwdg.de> <YAQINMoEi7ysBSD5@kib.kiev.ua> <1d675d77-8bdb-c285-ffa2-28330b839734@alvermark.net>
index | next in thread | previous in thread | raw e-mail
On Wed, Jan 20, 2021 at 11:02:21AM +0100, Jakob Alvermark wrote:
> This patch hides the problem for me. The system seems to work better now.
>
> No waiting on reboot, and the webcam works better.
I am curious what do you mean by the above reference to webcam.
Can you explain it with more details, even if only the impressions?
I probably going to commit the following patch in the next 24 hours.
commit 02505d07bca320a638c96918ac9076c6eece2fff
Author: Konstantin Belousov <kib@FreeBSD.org>
Date: Wed Jan 20 11:32:21 2021 +0200
AMD Zen CPUs: switch TSC timecounter to RDTSCP
Reported by: many
MFC after: 1 weel
Sponsored by: The FreeBSD Foundation
diff --git a/lib/libc/x86/sys/__vdso_gettc.c b/lib/libc/x86/sys/__vdso_gettc.c
index 7f224f8758cb..7a64f2a0b556 100644
--- a/lib/libc/x86/sys/__vdso_gettc.c
+++ b/lib/libc/x86/sys/__vdso_gettc.c
@@ -125,7 +125,7 @@ struct tsc_selector_tag {
};
static const struct tsc_selector_tag tsc_selector[] = {
- [0] = { /* Intel or AMD Zen+, LFENCE */
+ [0] = { /* Intel, LFENCE */
.ts_rdtsc32 = rdtsc32_mb_lfence,
.ts_rdtsc_low = rdtsc_low_mb_lfence,
},
@@ -164,9 +164,6 @@ tsc_selector_idx(u_int cpu_feature)
do_cpuid(1, p);
cpu_id = p[0];
- if (amd_cpu && CPUID_TO_FAMILY(cpu_id) >= 0x17)
- return (0);
-
if (cpu_feature != 0) {
do_cpuid(0x80000000, p);
cpu_exthigh = p[0];
diff --git a/sys/x86/x86/tsc.c b/sys/x86/x86/tsc.c
index 85924df98312..de0a1505c2f6 100644
--- a/sys/x86/x86/tsc.c
+++ b/sys/x86/x86/tsc.c
@@ -633,19 +633,12 @@ init_TSC_tc(void)
/*
* Timecounter implementation selection, top to bottom:
- * - For AMD Zens and newer, use LFENCE;RDTSC.
* - If RDTSCP is available, use RDTSCP.
* - If fence instructions are provided (SSE2), use LFENCE;RDTSC
* on Intel, and MFENCE;RDTSC on AMD.
* - For really old CPUs, just use RDTSC.
*/
- if ((cpu_vendor_id == CPU_VENDOR_AMD ||
- cpu_vendor_id == CPU_VENDOR_HYGON) &&
- CPUID_TO_FAMILY(cpu_id) >= 0x17) {
- tsc_timecounter.tc_get_timecount = shift > 0 ?
- tsc_get_timecount_low_lfence :
- tsc_get_timecount_lfence;
- } else if ((amd_feature & AMDID_RDTSCP) != 0) {
+ if ((amd_feature & AMDID_RDTSCP) != 0) {
tsc_timecounter.tc_get_timecount = shift > 0 ?
tscp_get_timecount_low : tscp_get_timecount;
} else if ((cpu_feature & CPUID_SSE2) != 0 && mp_ncpus > 1) {
help
Want to link to this message? Use this
URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?YAgDhJ9/YX29kH2a>