| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-Id: <1438044416-15588-4-git-send-email-christopher.s.hall@intel.com>
Date: Mon, 27 Jul 2015 17:46:54 -0700
From: Christopher Hall <christopher.s.hall@...el.com>
To: john.stultz@...aro.org, tglx@...utronix.de,
richardcochran@...il.com, mingo@...hat.com,
jeffrey.t.kirsher@...el.com, john.ronciak@...el.com, hpa@...or.com,
x86@...nel.org
Cc: linux-kernel@...r.kernel.org, netdev@...r.kernel.org,
Christopher Hall <christopher.s.hall@...el.com>
Subject: [PATCH 3/5] Add calls to translate Always Running Timer (ART) to system time
* art_to_mono64
* art_to_rawmono64
* art_to_realtime64
Intel audio and PCH ethernet devices use the Always Running Timer (ART) to
relate their device clock to system time
Signed-off-by: Christopher Hall <christopher.s.hall@...el.com>
---
arch/x86/Kconfig | 12 ++++
arch/x86/include/asm/art.h | 42 ++++++++++++++
arch/x86/kernel/Makefile | 1 +
arch/x86/kernel/art.c | 134 +++++++++++++++++++++++++++++++++++++++++++++
arch/x86/kernel/tsc.c | 4 ++
5 files changed, 193 insertions(+)
create mode 100644 arch/x86/include/asm/art.h
create mode 100644 arch/x86/kernel/art.c
diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
index b3a1a5d..1ef9985 100644
--- a/arch/x86/Kconfig
+++ b/arch/x86/Kconfig
@@ -1175,6 +1175,18 @@ config X86_CPUID
with major 203 and minors 0 to 31 for /dev/cpu/0/cpuid to
/dev/cpu/31/cpuid.
+config X86_ART
+ bool "Always Running Timer"
+ default y
+ depends on X86_TSC
+ ---help---
+ This option provides functionality to drivers and devices that use
+ the always-running-timer (ART) to correlate their device clock
+ counter with the system clock counter. The TSC is *exactly* related
+ to the ART by a ratio m/n specified by CPUID leaf 0x15
+ (n=EAX,m=EBX). If ART is unused or unavailable there isn't any
+ performance impact. It's safe to say Y.
+
choice
prompt "High Memory Support"
default HIGHMEM4G
diff --git a/arch/x86/include/asm/art.h b/arch/x86/include/asm/art.h
new file mode 100644
index 0000000..da58ce4
--- /dev/null
+++ b/arch/x86/include/asm/art.h
@@ -0,0 +1,42 @@
+/*
+ * x86 ART related functions
+ */
+#ifndef _ASM_X86_ART_H
+#define _ASM_X86_ART_H
+
+#ifndef CONFIG_X86_ART
+
+static inline int setup_art(void)
+{
+ return 0;
+}
+
+static inline bool has_art(void)
+{
+ return false;
+}
+
+static inline int art_to_rawmono64(struct timespec64 *rawmono, cycle_t art)
+{
+ return -ENXIO;
+}
+static inline int art_to_realtime64(struct timespec64 *realtime, cycle_t art)
+{
+ return -ENXIO;
+}
+static inline int art_to_mono64(struct timespec64 *mono, cycle_t art)
+{
+ return -ENXIO;
+}
+
+#else
+
+extern int setup_art(void);
+extern bool has_art(void);
+extern int art_to_rawmono64(struct timespec64 *rawmono, cycle_t art);
+extern int art_to_realtime64(struct timespec64 *realtime, cycle_t art);
+extern int art_to_mono64(struct timespec64 *mono, cycle_t art);
+
+#endif
+
+#endif/*_ASM_X86_ART_H*/
diff --git a/arch/x86/kernel/Makefile b/arch/x86/kernel/Makefile
index 0f15af4..0908311 100644
--- a/arch/x86/kernel/Makefile
+++ b/arch/x86/kernel/Makefile
@@ -109,6 +109,7 @@ obj-$(CONFIG_PERF_EVENTS) += perf_regs.o
obj-$(CONFIG_TRACING) += tracepoint.o
obj-$(CONFIG_IOSF_MBI) += iosf_mbi.o
obj-$(CONFIG_PMC_ATOM) += pmc_atom.o
+obj-$(CONFIG_X86_ART) += art.o
###
# 64 bit specific files
diff --git a/arch/x86/kernel/art.c b/arch/x86/kernel/art.c
new file mode 100644
index 0000000..1906cf0
--- /dev/null
+++ b/arch/x86/kernel/art.c
@@ -0,0 +1,134 @@
+#include <asm/tsc.h>
+#include <asm/cpufeature.h>
+#include <asm/processor.h>
+#include <linux/spinlock.h>
+#include <linux/seqlock.h>
+
+#define CPUID_ART_LEAF 0x15
+
+static bool art_present;
+
+static struct art_state {
+ seqcount_t seq;
+ u32 art_ratio_numerator;
+ u32 art_ratio_denominator;
+
+ cycle_t prev_art;
+ cycle_t prev_tsc_corr_art; /*This is the TSC value corresponding to
+ prev_art */
+ u32 tsc_remainder;
+} art_state ____cacheline_aligned;
+
+static DEFINE_RAW_SPINLOCK(art_lock);
+
+#define MIN_DENOMINATOR 2
+int setup_art(void)
+{
+ if (boot_cpu_data.cpuid_level < CPUID_ART_LEAF)
+ return 0;
+ art_state.art_ratio_denominator = cpuid_eax(CPUID_ART_LEAF);
+ if (art_state.art_ratio_denominator < MIN_DENOMINATOR)
+ return 0;
+ art_state.art_ratio_numerator = cpuid_ebx(CPUID_ART_LEAF);
+
+ art_present = true;
+ return 0;
+}
+
+static bool has_art(void)
+{
+ return art_present;
+}
+EXPORT_SYMBOL(has_art);
+
+#define ROLLOVER_THRESHOLD (2ULL << 23)
+
+static u32 art_scale(struct art_state *art_state, cycle_t *art)
+{
+ u32 rem;
+
+ *art *= art_state->art_ratio_numerator;
+
+ switch (art_state->art_ratio_denominator) {
+ default:
+ rem = do_div(*art, art_state->art_ratio_denominator);
+ case 2:
+ rem = *art & 0x1;
+ *art >>= 1;
+ }
+ return rem + art_state->tsc_remainder;
+}
+
+static cycle_t art_to_tsc(cycle_t art)
+{
+ unsigned seq;
+ cycle_t tsc_next;
+ u32 rem_next;
+ bool backward = false;
+ unsigned long flags;
+
+ do {
+ seq = read_seqcount_begin(&art_state.seq);
+
+ if (art < art_state.prev_art &&
+ art_state.prev_art - art < ROLLOVER_THRESHOLD) {
+ tsc_next = (art_state.prev_art-art);
+ art_scale(&art_state, &tsc_next);
+ tsc_next = art_state.prev_tsc_corr_art - tsc_next;
+ backward = true;
+ } else {
+ tsc_next = art - art_state.prev_art;
+ rem_next = art_scale(&art_state, &tsc_next);
+ tsc_next += art_state.prev_tsc_corr_art;
+
+ tsc_next += rem_next /
+ art_state.art_ratio_denominator;
+ rem_next %= art_state.art_ratio_denominator;
+ }
+ } while (read_seqcount_retry(&art_state.seq, seq));
+
+ /* There's no need to update after every read, if an update is
+ already in progress by someone else just exit */
+ if (!backward && raw_spin_trylock_irqsave(&art_lock, flags)) {
+ write_seqcount_begin(&art_state.seq);
+ art_state.prev_art = art;
+ art_state.prev_tsc_corr_art = tsc_next;
+ art_state.tsc_remainder = rem_next;
+ write_seqcount_end(&art_state.seq);
+ raw_spin_unlock_irqrestore(&art_lock, flags);
+ }
+
+ return tsc_next;
+}
+
+static bool checked_art_to_tsc(cycle_t *tsc)
+{
+ if (!has_art())
+ return false;
+ *tsc = art_to_tsc(*tsc);
+ return true;
+}
+
+static int art_to_rawmono64(struct timespec64 *rawmono, cycle_t art)
+{
+ if (!checked_art_to_tsc(&art))
+ return -ENXIO;
+ return tsc_to_rawmono64(rawmono, art);
+}
+EXPORT_SYMBOL(art_to_rawmono64);
+
+static int art_to_realtime64(struct timespec64 *realtime, cycle_t art)
+{
+ if (!checked_art_to_tsc(&art))
+ return -ENXIO;
+ return tsc_to_realtime64(realtime, art);
+}
+EXPORT_SYMBOL(art_to_realtime64);
+
+static int art_to_mono64(struct timespec64 *mono, cycle_t art)
+{
+ if (!checked_art_to_tsc(&art))
+ return -ENXIO;
+ return tsc_to_mono64(mono, art);
+}
+EXPORT_SYMBOL(art_to_mono64);
diff --git a/arch/x86/kernel/tsc.c b/arch/x86/kernel/tsc.c
index a192271..828c4b3 100644
--- a/arch/x86/kernel/tsc.c
+++ b/arch/x86/kernel/tsc.c
@@ -22,6 +22,8 @@
#include <asm/nmi.h>
#include <asm/x86_init.h>
+#include <asm/art.h>
+
unsigned int __read_mostly cpu_khz; /* TSC clocks / usec, not used here */
EXPORT_SYMBOL(cpu_khz);
@@ -1177,6 +1179,8 @@ static int __init init_tsc_clocksource(void)
return 0;
}
+ setup_art();
+
schedule_delayed_work(&tsc_irqwork, 0);
return 0;
}
--
1.9.1
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Powered by blists - more mailing lists