From: Noah Goldstein <goldstein.w.n@gmail.com>
To: unlisted-recipients:; (no To-header on input)
Cc: goldstein.w.n@gmail.com, tglx@linutronix.de, mingo@redhat.com,
bp@alien8.de, x86@kernel.org, hpa@zytor.com, luto@kernel.org,
linux-kernel@vger.kernel.org
Subject: [PATCH v3 2/2] x86/xstate: Make AVX512 status tracking more accurate
Date: Wed, 27 Oct 2021 11:26:15 -0500 [thread overview]
Message-ID: <20211027162615.1989004-2-goldstein.w.n@gmail.com> (raw)
In-Reply-To: <20211027162615.1989004-1-goldstein.w.n@gmail.com>
This patch splits the timestamps for tracking the AVX512 status into
'avx512_ZMM_Hi256_timestamp' and 'avx512_Hi16_ZMM_timestamp'. They are
used for tracking XFEATURE_ZMM_Hi256 and XFEATURE_Hi16_ZMM use
respectively.
The purpose of tracking the AVX512 status is to convey information
about possible frequency throttling. The current implementation has
false positives on XFEATURE_OPMASK use and any usage of the hi16 xmm
and ymm registers which are included in the XFEATURE_Hi16_ZMM set as
neither will cause frequency throttling.
This patches implementation avoids to add more clarity to the
output. The 'avx512_ZMM_Hi256_timestamp' will not have false positives
so its value will at least be indicative frequency throttling. Since
'avx512_Hi16_ZMM_timestamp' can still indicate frequency throttling
from zmm16...zmm31 use though had false positives it is separated.
Signed-off-by: Noah Goldstein <goldstein.w.n@gmail.com>
---
arch/x86/include/asm/fpu/types.h | 16 ++++++++--
arch/x86/kernel/fpu/xstate.c | 53 +++++++++++++++++++++++++++-----
2 files changed, 58 insertions(+), 11 deletions(-)
diff --git a/arch/x86/include/asm/fpu/types.h b/arch/x86/include/asm/fpu/types.h
index f5a38a5f3ae1..cb10909fa3da 100644
--- a/arch/x86/include/asm/fpu/types.h
+++ b/arch/x86/include/asm/fpu/types.h
@@ -330,11 +330,21 @@ struct fpu {
unsigned int last_cpu;
/*
- * @avx512_timestamp:
+ * @avx512_ZMM_Hi256_timestamp:
*
- * Records the timestamp of AVX512 use during last context switch.
+ * Records the timestamp of AVX512 use in the ZMM_Hi256 xfeature
+ * set. This include zmm0...zmm15.
*/
- unsigned long avx512_timestamp;
+ unsigned long avx512_ZMM_Hi256_timestamp;
+
+ /*
+ * @avx512_Hi16_ZMM_timestamp:
+ *
+ * Records the timestamp of AVX512 use in the Hi16_ZMM xfeature
+ * set. This includes usage of any of the hi16 xmm, ymm, or zmm
+ * registers.
+ */
+ unsigned long avx512_Hi16_ZMM_timestamp;
/*
* @state:
diff --git a/arch/x86/kernel/fpu/xstate.c b/arch/x86/kernel/fpu/xstate.c
index 00b495914be2..3bb1a425ce56 100644
--- a/arch/x86/kernel/fpu/xstate.c
+++ b/arch/x86/kernel/fpu/xstate.c
@@ -1246,7 +1246,7 @@ void xrstors(struct xregs_state *xstate, u64 mask)
}
/*
- * Track of the state of desired avx architecture features.
+ * Track of the state of desired avx related xfeatures.
*/
void fpu_update_avx_timestamp(struct fpu *fpu)
{
@@ -1254,18 +1254,28 @@ void fpu_update_avx_timestamp(struct fpu *fpu)
* AVX512 state is tracked here because its use is known to slow
* the max clock speed of the core.
*/
- if (fpu->state.xsave.header.xfeatures & XFEATURE_MASK_AVX512)
- fpu->avx512_timestamp = jiffies;
+
+ /*
+ * Store a separate state for ZMM_Hi256 and Hi16_ZMM xfeature use.
+ * If ZMM_Hi256 is used the machine has certainly used a zmm
+ * register. Hi16_ZMM, however, has false positives on usage of
+ * hi16 xmm and ymm registers.
+ */
+ if (fpu->state.xsave.header.xfeatures & XFEATURE_MASK_ZMM_Hi256)
+ fpu->avx512_ZMM_Hi256_timestamp = jiffies;
+ if (fpu->state.xsave.header.xfeatures & XFEATURE_MASK_Hi16_ZMM)
+ fpu->avx512_Hi16_ZMM_timestamp = jiffies;
}
+
#ifdef CONFIG_PROC_PID_ARCH_STATUS
+
/*
- * Report the amount of time elapsed in millisecond since last AVX512
- * use in the task.
+ * Helper function for computing proper output for avx512_status
+ * timestamp.
*/
-static void avx512_status(struct seq_file *m, struct task_struct *task)
+static long avx_status_compute_delta(unsigned long timestamp)
{
- unsigned long timestamp = READ_ONCE(task->thread.fpu.avx512_timestamp);
long delta;
if (!timestamp) {
@@ -1282,8 +1292,35 @@ static void avx512_status(struct seq_file *m, struct task_struct *task)
delta = LONG_MAX;
delta = jiffies_to_msecs(delta);
}
+ return delta;
+}
- seq_put_decimal_ll(m, "AVX512_elapsed_ms:\t", delta);
+/*
+ * Report the amount of time elapsed in millisecond since last AVX512
+ * use in the task.
+ */
+static void avx512_status(struct seq_file *m, struct task_struct *task)
+{
+ unsigned long timestamp;
+ long delta_ZMM_Hi256, delta_Hi16_ZMM, delta_unified;
+
+ timestamp = READ_ONCE(task->thread.fpu.avx512_ZMM_Hi256_timestamp);
+ delta_ZMM_Hi256 = avx_status_compute_delta(timestamp);
+
+ timestamp = READ_ONCE(task->thread.fpu.avx512_Hi16_ZMM_timestamp);
+ delta_Hi16_ZMM = avx_status_compute_delta(timestamp);
+
+ /*
+ * Report unified delta of most recent AVX512 usage from either
+ * Hi16_ZMM or ZMM_Hi256 xfeature sets.
+ */
+ delta_unified = timestamp ? delta_Hi16_ZMM : delta_ZMM_Hi256;
+
+ seq_put_decimal_ll(m, "AVX512_elapsed_ms:\t", delta_unified);
+ seq_putc(m, '\n');
+ seq_put_decimal_ll(m, "AVX512_ZMM_Hi256_elapsed_ms:\t", delta_ZMM_Hi256);
+ seq_putc(m, '\n');
+ seq_put_decimal_ll(m, "AVX512_Hi16_ZMM_elapsed_ms:\t", delta_Hi16_ZMM);
seq_putc(m, '\n');
}
--
2.25.1
next prev parent reply other threads:[~2021-10-27 16:26 UTC|newest]
Thread overview: 31+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-09-20 5:39 [PATCH v1] x86/fpu: Remove opmask state from avx512_timestamp check Noah Goldstein
2021-09-27 18:02 ` Noah Goldstein
2021-10-13 22:36 ` Noah Goldstein
2021-10-14 8:28 ` Borislav Petkov
2021-10-14 15:49 ` Noah Goldstein
2021-10-15 14:40 ` Borislav Petkov
2021-10-15 15:25 ` Dave Hansen
2021-10-15 17:30 ` Noah Goldstein
2021-10-15 20:47 ` [PATCH v2 1/2] x86/fpu: Add helper function for tracking AVX512 status Noah Goldstein
2021-10-26 23:15 ` Noah Goldstein
2021-10-15 20:47 ` [PATCH v2 2/2] x86/xstate: Make AVX512 status tracking more accurate Noah Goldstein
2021-10-27 11:07 ` Borislav Petkov
2021-10-27 16:28 ` Noah Goldstein
2021-10-27 16:58 ` Borislav Petkov
2021-10-27 17:18 ` Noah Goldstein
2021-10-27 16:26 ` [PATCH v3 1/2] x86/fpu: Add helper function for tracking AVX512 status Noah Goldstein
2021-10-27 16:26 ` Noah Goldstein [this message]
2021-10-27 17:11 ` [PATCH v3 2/2] x86/xstate: Make AVX512 status tracking more accurate Borislav Petkov
2021-10-27 17:37 ` Noah Goldstein
2021-10-27 17:45 ` Borislav Petkov
2021-10-27 18:21 ` Noah Goldstein
2021-11-03 20:22 ` Noah Goldstein
2021-11-03 20:48 ` Borislav Petkov
2021-11-03 20:56 ` Noah Goldstein
2021-10-27 17:17 ` [PATCH v4 1/2] x86/fpu: Add helper function for tracking AVX512 status Noah Goldstein
2021-10-27 17:17 ` [PATCH v4 2/2] x86/xstate: Make AVX512 status tracking more accurate Noah Goldstein
2021-10-27 18:21 ` [PATCH v5 1/2] x86/fpu: Add helper function for tracking AVX512 status Noah Goldstein
2021-10-27 18:21 ` [PATCH v5 2/2] x86/xstate: Make AVX512 status tracking more accurate Noah Goldstein
2021-11-16 16:21 ` [PATCH] x86/fpu: Correct AVX512 state tracking Borislav Petkov
2021-11-16 16:49 ` Noah Goldstein
2021-11-17 18:29 ` [tip: x86/fpu] " tip-bot2 for Noah Goldstein
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20211027162615.1989004-2-goldstein.w.n@gmail.com \
--to=goldstein.w.n@gmail.com \
--cc=bp@alien8.de \
--cc=hpa@zytor.com \
--cc=linux-kernel@vger.kernel.org \
--cc=luto@kernel.org \
--cc=mingo@redhat.com \
--cc=tglx@linutronix.de \
--cc=x86@kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox
all inboxes | Powered by JetHome®