Skip to contentSkip to conversation

x86, mwaitt: introduce AMD mwaitt support

PATCH In review Huang Rui 4 participants 13 messages Aug 5, 2015 – Aug 22, 2015

Subsystems (2)
lore
Share & tools
lore

Huang Rui

author
#0 lore raw

v6 cover x86, mwaitt: introduce AMD mwaitt support

Hi,

This patch set introduces a new instruction support on AMD Carrizo (Family 15h, Model 60h-6fh). It adds mwaitx delay function with a configurable timer.

Andy and Boris provide a suggestion which uses mwaitx on delay method.

As Peter's suggestion of last version (v5), the serial of patch set provides a test result.
http://marc.info/?l=linux-kernel&m=143436586513713&w=2

Some discussions of the background, please see:
http://marc.info/?l=linux-kernel&m=143202042530498&w=2
http://marc.info/?l=linux-kernel&m=143161327003541&w=2
http://marc.info/?l=linux-kernel&m=143222815331016&w=2

Patch set is rebased on tip/master.

Changes from v1 -> v2
- Remove mwaitx idle implementation since some disputes without power improvement.
- Add a patch which implement another use case on delay.
- Introduce a kernel parameter (delay) to make delay method configurable.

Changes from v2 -> v3
- Add compared data on commit message
- Remove kernel parameter
- Add hint to avoid to access deep state in future
- Update mwaitx delay method as Petter's suggestion

Changes from v3 -> v4
- Put the MONITORX/MWAITX description into comments

Changes from v4 -> v5
- Remove mwaitx function
- Use mwaitx_delay at init_amd
- Use cpu_tts as montioring address scope

Changes from v5 -> v6
- Move definitions into patch 1
- Completed the power consumption testing both with MWAITX and without MWAITX
- Use mwaitx_delay at bsp_init_amd

In MWAITX delay, the CPU core will be quiesced in a waiting phase, diminishing its power consumption.

Run a simple test to measure power consumption:

cat /sys/bus/pci/devices/0000\:00\:18.4/hwmon/hwmon0/power1_acc; sleep 10000s;
cat /sys/bus/pci/devices/0000\:00\:18.4/hwmon/hwmon0/power1_acc;

* TSC-based default delay: 485115 uWatts average power
* MWAITX-based delay: 252738 uWatts average power

Thus, that's about 240 milliWatts less power consumption. The test method relies on the support of AMD CPU accumulated power algorithm in fam15_power for which patches are forthcoming.

Thanks,
Rui

Huang Rui (2):
x86, mwaitt: add monitorx and mwaitx instruction x86, mwaitt: introduce mwaitx delay with a configurable timer

 arch/x86/include/asm/cpufeature.h |  1 +
 arch/x86/include/asm/delay.h      |  1 +
 arch/x86/include/asm/mwait.h      | 43 +++++++++++++++++++++++++++++++++++
 arch/x86/kernel/cpu/amd.c         |  4 ++++
 arch/x86/lib/delay.c              | 48 ++++++++++++++++++++++++++++++++++++++-
 5 files changed, 96 insertions(+), 1 deletion(-)
Signature · 1 line
1.9.1
author
#1 lore raw

PATCH v6 1/2 x86, mwaitt: add monitorx and mwaitx instruction

On AMD Carrizo processors (Family 15h, Model 60h-6fh), there is a new feature called MWAITT (MWAIT with a timer) as an extension of MONITOR/MWAIT.

MWAITT, another name is MWAITX (MWAIT with extensions), has a configurable timer that causes MWAITX to exit on expiration.

Compared with MONITOR/MWAIT, there are minor differences in opcode and input parameters.

MWAITX ECX[1]: enable timer if set
MWAITX EBX[31:0]: max wait time expressed in SW P0 clocks

MWAIT MWAITX
opcode 0f 01 c9 | 0f 01 fb
ECX[0] value of RFLAGS.IF seen by instruction
ECX[1] unused/#GP if set | enable timer if set
ECX[31:2] unused/#GP if set
EAX unused (reserve for hint)
EBX[31:0] unused | max wait time (loops)

MONITOR MONITORX
opcode 0f 01 c8 | 0f 01 fa
EAX (logical) address to monitor
ECX #GP if not zero

The software P0 frequency is the same as the TSC frequency.

Max timeout = EBX/(TSC frequency)

Cc Borislav Petkov <[email protected]>
Cc Andy Lutomirski <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc Andreas Herrmann <[email protected]>
---
 arch/x86/include/asm/cpufeature.h |  1 +
 arch/x86/include/asm/mwait.h      | 43 +++++++++++++++++++++++++++++++++++++++
 2 files changed, 44 insertions(+)
arch/x86/include/asm/cpufeature.h +1 -0
index 4b11974..9978a98 100644
@@ -177,6 +177,7 @@
 #define X86_FEATURE_PERFCTR_NB  ( 6*32+24) /* NB performance counter extensions */
 #define X86_FEATURE_BPEXT	(6*32+26) /* data breakpoint extension */
 #define X86_FEATURE_PERFCTR_L2	( 6*32+28) /* L2 performance counter extensions */
+#define X86_FEATURE_MWAITT	( 6*32+29) /* MWAIT extension (MONITORX/MWAITX) */
 
 /*
  * Auxiliary flags: Linux defined - For features scattered in various
arch/x86/include/asm/mwait.h +43 -0

61 lines not on this page · raw message

Signature · 1 line
1.9.1
author
#2 lore raw

PATCH v6 2/2 x86, mwaitt: introduce mwaitx delay with a configurable timer

In reply to Huang Rui

MWAITX can enable a timer and a corresponding timer value specified in SW P0 clocks. The SW P0 frequency is the same as TSC. The timer provides an upper bound on how long the instruction waits before exiting.

The implementation of delay function in kernel can leverage the timer of MWAITX. This patch provides a new method (delay_mwaitx) to measure delay time.

In MWAITX delay, the CPU core will be quiesced in a waiting phase, diminishing its power consumption.

Run a simple test to measure power consumption:

cat /sys/bus/pci/devices/0000\:00\:18.4/hwmon/hwmon0/power1_acc; sleep 10000s;
cat /sys/bus/pci/devices/0000\:00\:18.4/hwmon/hwmon0/power1_acc;

* TSC-based default delay: 485115 uWatts average power
* MWAITX-based delay: 252738 uWatts average power

Thus, that's about 240 milliWatts less power consumption. The test method relies on the support of AMD CPU accumulated power algorithm in fam15_power for which patches are forthcoming.

---
 arch/x86/include/asm/delay.h |  1 +
 arch/x86/kernel/cpu/amd.c    |  4 ++++
 arch/x86/lib/delay.c         | 48 +++++++++++++++++++++++++++++++++++++++++++-
 3 files changed, 52 insertions(+), 1 deletion(-)
arch/x86/include/asm/delay.h +1 -0
index 9b3b4f2..36a760b 100644
@@ -4,5 +4,6 @@
 #include <asm-generic/delay.h>
 
 void use_tsc_delay(void);
+void use_mwaitx_delay(void);
 
 #endif /* _ASM_X86_DELAY_H */
arch/x86/kernel/cpu/amd.c +4 -0
index 51ad2af..730e620 100644
@@ -11,6 +11,7 @@
 #include <asm/cpu.h>
 #include <asm/smp.h>
 #include <asm/pci-direct.h>
+#include <asm/delay.h>
 
 #ifdef CONFIG_X86_64
 # include <asm/mmconfig.h>
@@ -506,6 +507,9 @@ static void bsp_init_amd(struct cpuinfo_x86 *c)
 		/* A random value per boot for bit slice [12:upper_bit) */
 		va_align.bits = get_random_int() & va_align.mask;
 	}
+
+	if (cpu_has(c, X86_FEATURE_MWAITT))
+		use_mwaitx_delay();
 }
 
 static void early_init_amd(struct cpuinfo_x86 *c)
arch/x86/lib/delay.c +47 -1

66 lines not on this page · raw message

Signature · 1 line
1.9.1
#3 lore raw

Re: [PATCH v6 2/2] x86, mwaitt: introduce mwaitx delay with a configurable timer

Quoted 8 lines · Huang Rui

MWAITX can enable a timer and a corresponding timer value specified in SW P0 clocks. The SW P0 frequency is the same as TSC. The timer provides an upper bound on how long the instruction waits before exiting.

The implementation of delay function in kernel can leverage the timer of MWAITX. This patch provides a new method (delay_mwaitx) to measure delay time.

...

Review

Quoted 4 lines
(quoted diff) +4 -0
+static void delay_mwaitx(unsigned long __loops)
+{
+	u32 delay, loops = __loops;
+	u64 end, start;

Hmm, this truncates __loops in case someone wants to delay for more than (u32)-1 TSC clocks. I guess the right thing to do is to do the calculation with u64s and MWAITX_MAX_LOOPS will keep us within bounds.

Here's what I did:

---
From: Huang Rui <[email protected]>
Date: Wed, 5 Aug 2015 11:18:52 +0800
Subject: [PATCH] x86/asm: Introduce an MWAITX-based delay with a configurable
timer
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

MWAITX can enable a timer and a corresponding timer value specified in SW P0 clocks. The SW P0 frequency is the same as TSC. The timer provides an upper bound on how long the instruction waits before exiting.

This way, a delay function in the kernel can leverage that MWAITX timer of MWAITX.

When a CPU core executes MWAITX, it will be quiesced in a waiting phase, diminishing its power consumption. This way, we can save power in comparison to our default TSC-based delays.

A simple test shows that:

$ cat /sys/bus/pci/devices/0000\:00\:18.4/hwmon/hwmon0/power1_acc
$ sleep 10000s
$ cat /sys/bus/pci/devices/0000\:00\:18.4/hwmon/hwmon0/power1_acc

Results:

* TSC-based default delay: 485115 uWatts average power
* MWAITX-based delay: 252738 uWatts average power

Thus, that's about 240 milliWatts less power consumption. The test method relies on the support of AMD CPU accumulated power algorithm in fam15h_power for which patches are forthcoming.

Cc Aaron Lu <[email protected]>
Cc Andreas Herrmann <[email protected]>
Cc Andy Lutomirski <[email protected]>
Cc Aravind Gopalakrishnan <[email protected]>
Cc Fengguang Wu <[email protected]>
Cc Frédéric Weisbecker <[email protected]>
Cc Hector Marco-Gisbert <[email protected]>
Cc "H. Peter Anvin" <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Jacob Shin <[email protected]>
Cc Jiri Olsa <[email protected]>
Cc John Stultz <[email protected]>
Cc Len Brown <[email protected]>
Cc Paolo Bonzini <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc "Rafael J. Wysocki" <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc Tony Li <[email protected]>
Cc x86-ml <[email protected]>
[ Fix delay truncation. ]
---
 arch/x86/include/asm/delay.h |  1 +
 arch/x86/kernel/cpu/amd.c    |  4 ++++
 arch/x86/lib/delay.c         | 47 +++++++++++++++++++++++++++++++++++++++++++-
 3 files changed, 51 insertions(+), 1 deletion(-)
arch/x86/include/asm/delay.h +1 -0
index 9b3b4f2754c7..36a760bda462 100644
@@ -4,5 +4,6 @@
 #include <asm-generic/delay.h>
 
 void use_tsc_delay(void);
+void use_mwaitx_delay(void);
 
 #endif /* _ASM_X86_DELAY_H */
arch/x86/kernel/cpu/amd.c +4 -0
index 51ad2af84a72..4a70fc6d400a 100644
@@ -11,6 +11,7 @@
 #include <asm/cpu.h>
 #include <asm/smp.h>
 #include <asm/pci-direct.h>
+#include <asm/delay.h>
 
 #ifdef CONFIG_X86_64
 # include <asm/mmconfig.h>
@@ -506,6 +507,9 @@ static void bsp_init_amd(struct cpuinfo_x86 *c)
 		/* A random value per boot for bit slice [12:upper_bit) */
 		va_align.bits = get_random_int() & va_align.mask;
 	}
+
+	if (cpu_has(c, X86_FEATURE_MWAITX))
+		use_mwaitx_delay();
 }
 
 static void early_init_amd(struct cpuinfo_x86 *c)
arch/x86/lib/delay.c +46 -1

65 lines not on this page · raw message

Signature · 8 lines
2.5.0.rc2.28.g6003e7f

-- 
Regards/Gruss,
    Boris.

ECO tip #101: Trim your mails when you reply.
--
author
#4 lore raw

Re: [PATCH v6 2/2] x86, mwaitt: introduce mwaitx delay with a configurable timer

Quoted 10 lines · Borislav Petkov

arch/x86/lib/delay.c static void delay_tsc(unsigned long __loops)

Quoted 4 lines · Huang Rui
arch/x86/lib/delay.c +4 -0
+static void delay_mwaitx(unsigned long __loops)
+{
+	u32 delay, loops = __loops;
+	u64 end, start;

Hmm, this truncates __loops in case someone wants to delay for more than (u32)-1 TSC clocks. I guess the right thing to do is to do the calculation with u64s and MWAITX_MAX_LOOPS will keep us within bounds.

Yes, you're right. Thanks to update. :)

Thanks,
Rui

maintainer
#5 lore raw
In reply to Huang Rui
Quoted 10 lines · Huang Rui

cat /sys/bus/pci/devices/0000\:00\:18.4/hwmon/hwmon0/power1_acc; sleep 10000s;
cat /sys/bus/pci/devices/0000\:00\:18.4/hwmon/hwmon0/power1_acc;

* TSC-based default delay: 485115 uWatts average power
* MWAITX-based delay: 252738 uWatts average power

Thus, that's about 240 milliWatts less power consumption. The test method relies on the support of AMD CPU accumulated power algorithm in fam15_power for which patches are forthcoming.

Cool power consumption drop is actually even measureable.

Also, I think implementing it as a loop, as Peter suggested, was the right thing to do due to this statement in MWAITX's definition in the APM:

"There is no indication after exiting MWAITX of why the processor exited or if the timer expired. It is up to software to check whether the awaiting store has occurred, and if not, determining how much time has elapsed if it wants to re-establish the MONITORX with a new timer value."

So all in all, those patches are starting to shape up nicely. One small nit I have is using "MWAITT" (with a T) together with MWAITX while the APM calls it only MWAITX. But I can fix that when applying and drop all MWAITT occurrences.

Thanks.

Signature · 5 lines
Regards/Gruss,
    Boris.

ECO tip #101: Trim your mails when you reply.
--
maintainer
#6 lore raw

cover tip-queue 2015-08-10

From: Borislav Petkov <[email protected]>

Hi Ingo,

here's a little stuff which got done baking. More specifically, the use of this new MWAITX insn on AMD as a more power-optimal delay function. Patch 3's commit message has details as to how exactly it was measured. The power measurement interface will be part of fam15h_power soon too.

Please queue for 4.3.

Btw, patch 3 uses rdtsc_ordered() which is in tip/x86/asm... Patch 1 goes to tip/x86/microcode, of course.

Thanks.

Andrzej Hajda (1):
x86/microcode: Use kmemdup() rather than duplicating its
implementation

Huang Rui (2):
x86/asm: Add MONITORX/MWAITX insns support
x86/asm: Introduce an MWAITX-based delay with a configurable timer

maintainer
#7 lore raw

PATCH 1/3 x86/microcode: Use kmemdup() rather than duplicating its implementation

From: Andrzej Hajda <[email protected]>

The patch was generated using fixed coccinelle semantic patch scripts/coccinelle/api/memdup.cocci.

Cc Andrzej Hajda <[email protected]>
Cc Bartlomiej Zolnierkiewicz <[email protected]>
Cc "H. Peter Anvin" <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Marek Szyprowski <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc x86-ml <[email protected]>
---
 arch/x86/kernel/cpu/microcode/amd.c         | 4 +---
 arch/x86/kernel/cpu/microcode/intel_early.c | 4 +---
 2 files changed, 2 insertions(+), 6 deletions(-)
arch/x86/kernel/cpu/microcode/amd.c +1 -3
index c7d2415b8a24..be37f101ce5a 100644
@@ -366,15 +366,13 @@ static int verify_and_add_patch(u8 family, u8 *fw, unsigned int leftover)
 		return -EINVAL;
 	}
 
-	patch->data = kzalloc(patch_size, GFP_KERNEL);
+	patch->data = kmemdup(fw + SECTION_HDR_SIZE, patch_size, GFP_KERNEL);
 	if (!patch->data) {
 		pr_err("Patch data allocation failure.\n");
 		kfree(patch);
 		return -EINVAL;
 	}
 
-	/* All looks ok, copy patch... */
-	memcpy(patch->data, fw + SECTION_HDR_SIZE, patch_size);
 	INIT_LIST_HEAD(&patch->plist);
 	patch->patch_id  = mc_hdr->patch_id;
 	patch->equiv_cpu = proc_id;
arch/x86/kernel/cpu/microcode/intel_early.c +1 -3
index 8187b7247d1c..101f0ac5f6e1 100644
@@ -207,13 +207,11 @@ save_microcode(struct mc_saved_data *mc_saved_data,
 		mc_hdr = &mc->hdr;
 		size   = get_totalsize(mc_hdr);
 
-		saved_ptr[i] = kmalloc(size, GFP_KERNEL);
+		saved_ptr[i] = kmemdup(mc, size, GFP_KERNEL);
 		if (!saved_ptr[i]) {
 			ret = -ENOMEM;
 			goto err;
 		}
-
-		memcpy(saved_ptr[i], mc, size);
 	}
 
 	/*
Signature · 1 line
2.5.0.rc2.28.g6003e7f
#8 lore raw

[tip:x86/microcode] x86/microcode: Use kmemdup() rather than duplicating its implementation

x86/microcode: Use kmemdup() rather than duplicating its implementation
Commit-ID:  d4e963644768b33aa3db7f470c35d74ed78d8354
Gitweb:     http://git.kernel.org/tip/d4e963644768b33aa3db7f470c35d74ed78d8354
Author:     Andrzej Hajda <[email protected]>
AuthorDate: Mon, 10 Aug 2015 12:19:52 +0200
Committer:  Ingo Molnar <[email protected]>
CommitDate: Sat, 22 Aug 2015 14:49:35 +0200

x86/microcode: Use kmemdup() rather than duplicating its implementation

The patch was generated using fixed coccinelle semantic patch scripts/coccinelle/api/memdup.cocci.

Cc Bartlomiej Zolnierkiewicz <[email protected]>
Cc H. Peter Anvin <[email protected]>
Cc Linus Torvalds <[email protected]>
Cc Marek Szyprowski <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Thomas Gleixner <[email protected]>
---
 arch/x86/kernel/cpu/microcode/amd.c         | 4 +---
 arch/x86/kernel/cpu/microcode/intel_early.c | 4 +---
 2 files changed, 2 insertions(+), 6 deletions(-)
arch/x86/kernel/cpu/microcode/amd.c +1 -3
index c7d2415..be37f10 100644
@@ -366,15 +366,13 @@ static int verify_and_add_patch(u8 family, u8 *fw, unsigned int leftover)
 		return -EINVAL;
 	}
 
-	patch->data = kzalloc(patch_size, GFP_KERNEL);
+	patch->data = kmemdup(fw + SECTION_HDR_SIZE, patch_size, GFP_KERNEL);
 	if (!patch->data) {
 		pr_err("Patch data allocation failure.\n");
 		kfree(patch);
 		return -EINVAL;
 	}
 
-	/* All looks ok, copy patch... */
-	memcpy(patch->data, fw + SECTION_HDR_SIZE, patch_size);
 	INIT_LIST_HEAD(&patch->plist);
 	patch->patch_id  = mc_hdr->patch_id;
 	patch->equiv_cpu = proc_id;
arch/x86/kernel/cpu/microcode/intel_early.c +1 -3
index 8187b72..101f0ac 100644
@@ -207,13 +207,11 @@ save_microcode(struct mc_saved_data *mc_saved_data,
 		mc_hdr = &mc->hdr;
 		size   = get_totalsize(mc_hdr);
 
-		saved_ptr[i] = kmalloc(size, GFP_KERNEL);
+		saved_ptr[i] = kmemdup(mc, size, GFP_KERNEL);
 		if (!saved_ptr[i]) {
 			ret = -ENOMEM;
 			goto err;
 		}
-
-		memcpy(saved_ptr[i], mc, size);
 	}
 
 	/*
maintainer
#9 lore raw

PATCH 2/3 x86/asm: Add MONITORX/MWAITX insns support

In reply to Borislav Petkov

From: Huang Rui <[email protected]>

AMD Carrizo processors (Family 15h, Models 60h-6fh) add a new feature called MWAITX (MWAIT with extensions) as an extension to MONITOR/MWAIT.

This new instruction controls a configurable timer which causes the core to exit wait state on timer expiration, in addition to "normal" MWAIT condition of reading from a monitored VA.

Compared to MONITOR/MWAIT, there are minor differences in opcode and input parameters:

MWAITX ECX[1]: enable timer if set
MWAITX EBX[31:0]: max wait time expressed in SW P0 clocks == TSC. The software P0 frequency is the same as the TSC frequency.

MWAIT MWAITX
opcode 0f 01 c9 | 0f 01 fb
ECX[0] value of RFLAGS.IF seen by instruction
ECX[1] unused/#GP if set | enable timer if set
ECX[31:2] unused/#GP if set
EAX unused (reserve for hint)
EBX[31:0] unused | max wait time (SW P0 == TSC)

MONITOR MONITORX
opcode 0f 01 c8 | 0f 01 fa
EAX (logical) address to monitor
ECX #GP if not zero

Max timeout = EBX/(TSC frequency)

Cc Aaron Lu <[email protected]>
Cc Alexander Shishkin <[email protected]>
Cc Andreas Herrmann <[email protected]>
Cc Andy Lutomirski <[email protected]>
Cc Dave Hansen <[email protected]>
Cc Dirk Brandewie <[email protected]>
Cc Fengguang Wu <[email protected]>
Cc Frédéric Weisbecker <[email protected]>
Cc "H. Peter Anvin" <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc John Stultz <[email protected]>
Cc Josh Triplett <[email protected]>
Cc Len Brown <[email protected]>
Cc Mike Galbraith <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc "Rafael J. Wysocki" <[email protected]>
Cc Ross Zwisler <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc Tony Li <[email protected]>
Cc x86-ml <[email protected]>
---
 arch/x86/include/asm/cpufeature.h |  1 +
 arch/x86/include/asm/mwait.h      | 45 +++++++++++++++++++++++++++++++++++++++
 2 files changed, 46 insertions(+)
arch/x86/include/asm/cpufeature.h +1 -0
index 3d6606fb97d0..a39e5708209b 100644
@@ -176,6 +176,7 @@
 #define X86_FEATURE_PERFCTR_NB  ( 6*32+24) /* NB performance counter extensions */
 #define X86_FEATURE_BPEXT	(6*32+26) /* data breakpoint extension */
 #define X86_FEATURE_PERFCTR_L2	( 6*32+28) /* L2 performance counter extensions */
+#define X86_FEATURE_MWAITX	( 6*32+29) /* MWAIT extension (MONITORX/MWAITX) */
 
 /*
  * Auxiliary flags: Linux defined - For features scattered in various
arch/x86/include/asm/mwait.h +45 -0

63 lines not on this page · raw message

Signature · 1 line
2.5.0.rc2.28.g6003e7f
#10 lore raw

[tip:x86/asm] x86/asm: Add MONITORX/MWAITX instruction support

x86/asm: Add MONITORX/MWAITX instruction support
Commit-ID:  f96756746c7909de37db3d03ac5fd5cfb2757f38
Gitweb:     http://git.kernel.org/tip/f96756746c7909de37db3d03ac5fd5cfb2757f38
Author:     Huang Rui <[email protected]>
AuthorDate: Mon, 10 Aug 2015 12:19:53 +0200
Committer:  Ingo Molnar <[email protected]>
CommitDate: Sat, 22 Aug 2015 14:52:16 +0200

x86/asm: Add MONITORX/MWAITX instruction support

AMD Carrizo processors (Family 15h, Models 60h-6fh) added a new feature called MWAITX (MWAIT with extensions) as an extension to MONITOR/MWAIT.

This new instruction controls a configurable timer which causes the core to exit wait state on timer expiration, in addition to "normal" MWAIT condition of reading from a monitored VA.

Compared to MONITOR/MWAIT, there are minor differences in opcode and input parameters:

MWAITX ECX[1]: enable timer if set
MWAITX EBX[31:0]: max wait time expressed in SW P0 clocks == TSC. The software P0 frequency is the same as the TSC frequency.

MWAIT MWAITX
opcode 0f 01 c9 | 0f 01 fb
ECX[0] value of RFLAGS.IF seen by instruction
ECX[1] unused/#GP if set | enable timer if set
ECX[31:2] unused/#GP if set
EAX unused (reserve for hint)
EBX[31:0] unused | max wait time (SW P0 == TSC)

MONITOR MONITORX
opcode 0f 01 c8 | 0f 01 fa
EAX (logical) address to monitor
ECX #GP if not zero

Max timeout = EBX/(TSC frequency)

Cc Aaron Lu <[email protected]>
Cc Alexander Shishkin <[email protected]>
Cc Andreas Herrmann <[email protected]>
Cc Andy Lutomirski <[email protected]>
Cc Dave Hansen <[email protected]>
Cc Dirk Brandewie <[email protected]>
Cc Fengguang Wu <[email protected]>
Cc Frédéric Weisbecker <[email protected]>
Cc H. Peter Anvin <[email protected]>
Cc John Stultz <[email protected]>
Cc Josh Triplett <[email protected]>
Cc Len Brown <[email protected]>
Cc Linus Torvalds <[email protected]>
Cc Mike Galbraith <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Rafael J. Wysocki <[email protected]>
Cc Ross Zwisler <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc Tony Li <[email protected]>
---
 arch/x86/include/asm/cpufeature.h |  1 +
 arch/x86/include/asm/mwait.h      | 45 +++++++++++++++++++++++++++++++++++++++
 2 files changed, 46 insertions(+)
arch/x86/include/asm/cpufeature.h +1 -0
index 3d6606f..a39e570 100644
@@ -176,6 +176,7 @@
 #define X86_FEATURE_PERFCTR_NB  ( 6*32+24) /* NB performance counter extensions */
 #define X86_FEATURE_BPEXT	(6*32+26) /* data breakpoint extension */
 #define X86_FEATURE_PERFCTR_L2	( 6*32+28) /* L2 performance counter extensions */
+#define X86_FEATURE_MWAITX	( 6*32+29) /* MWAIT extension (MONITORX/MWAITX) */
 
 /*
  * Auxiliary flags: Linux defined - For features scattered in various
arch/x86/include/asm/mwait.h +45 -0

63 lines not on this page · raw message

maintainer
#11 lore raw

PATCH 3/3 x86/asm: Introduce an MWAITX-based delay with a configurable timer

In reply to Borislav Petkov

From: Huang Rui <[email protected]>

MWAITX can enable a timer and a corresponding timer value specified in SW P0 clocks. The SW P0 frequency is the same as TSC. The timer provides an upper bound on how long the instruction waits before exiting.

This way, a delay function in the kernel can leverage that MWAITX timer of MWAITX.

When a CPU core executes MWAITX, it will be quiesced in a waiting phase, diminishing its power consumption. This way, we can save power in comparison to our default TSC-based delays.

A simple test shows that:

$ cat /sys/bus/pci/devices/0000\:00\:18.4/hwmon/hwmon0/power1_acc
$ sleep 10000s
$ cat /sys/bus/pci/devices/0000\:00\:18.4/hwmon/hwmon0/power1_acc

Results:

* TSC-based default delay: 485115 uWatts average power
* MWAITX-based delay: 252738 uWatts average power

Thus, that's about 240 milliWatts less power consumption. The test method relies on the support of AMD CPU accumulated power algorithm in fam15h_power for which patches are forthcoming.

Cc Aaron Lu <[email protected]>
Cc Andreas Herrmann <[email protected]>
Cc Andy Lutomirski <[email protected]>
Cc Aravind Gopalakrishnan <[email protected]>
Cc Fengguang Wu <[email protected]>
Cc Frédéric Weisbecker <[email protected]>
Cc Hector Marco-Gisbert <[email protected]>
Cc "H. Peter Anvin" <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Jacob Shin <[email protected]>
Cc Jiri Olsa <[email protected]>
Cc John Stultz <[email protected]>
Cc Len Brown <[email protected]>
Cc Paolo Bonzini <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc "Rafael J. Wysocki" <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc Tony Li <[email protected]>
Cc x86-ml <[email protected]>
[ Fix delay truncation. ]
---
 arch/x86/include/asm/delay.h |  1 +
 arch/x86/kernel/cpu/amd.c    |  4 ++++
 arch/x86/lib/delay.c         | 47 +++++++++++++++++++++++++++++++++++++++++++-
 3 files changed, 51 insertions(+), 1 deletion(-)
arch/x86/include/asm/delay.h +1 -0
index 9b3b4f2754c7..36a760bda462 100644
@@ -4,5 +4,6 @@
 #include <asm-generic/delay.h>
 
 void use_tsc_delay(void);
+void use_mwaitx_delay(void);
 
 #endif /* _ASM_X86_DELAY_H */
arch/x86/kernel/cpu/amd.c +4 -0
index 51ad2af84a72..4a70fc6d400a 100644
@@ -11,6 +11,7 @@
 #include <asm/cpu.h>
 #include <asm/smp.h>
 #include <asm/pci-direct.h>
+#include <asm/delay.h>
 
 #ifdef CONFIG_X86_64
 # include <asm/mmconfig.h>
@@ -506,6 +507,9 @@ static void bsp_init_amd(struct cpuinfo_x86 *c)
 		/* A random value per boot for bit slice [12:upper_bit) */
 		va_align.bits = get_random_int() & va_align.mask;
 	}
+
+	if (cpu_has(c, X86_FEATURE_MWAITX))
+		use_mwaitx_delay();
 }
 
 static void early_init_amd(struct cpuinfo_x86 *c)
arch/x86/lib/delay.c +46 -1

65 lines not on this page · raw message

Signature · 1 line
2.5.0.rc2.28.g6003e7f
#12 lore raw

[tip:x86/asm] x86/asm/delay: Introduce an MWAITX-based delay with a configurable timer

x86/asm/delay: Introduce an MWAITX-based delay with a configurable timer
Commit-ID:  b466bdb614823aaaa7188e85516177d2850f4782
Gitweb:     http://git.kernel.org/tip/b466bdb614823aaaa7188e85516177d2850f4782
Author:     Huang Rui <[email protected]>
AuthorDate: Mon, 10 Aug 2015 12:19:54 +0200
Committer:  Ingo Molnar <[email protected]>
CommitDate: Sat, 22 Aug 2015 14:52:16 +0200

x86/asm/delay: Introduce an MWAITX-based delay with a configurable timer

MWAITX can enable a timer and a corresponding timer value specified in SW P0 clocks. The SW P0 frequency is the same as TSC. The timer provides an upper bound on how long the instruction waits before exiting.

This way, a delay function in the kernel can leverage that MWAITX timer of MWAITX.

When a CPU core executes MWAITX, it will be quiesced in a waiting phase, diminishing its power consumption. This way, we can save power in comparison to our default TSC-based delays.

A simple test shows that:

	$ cat /sys/bus/pci/devices/0000\:00\:18.4/hwmon/hwmon0/power1_acc
	$ sleep 10000s
	$ cat /sys/bus/pci/devices/0000\:00\:18.4/hwmon/hwmon0/power1_acc

Results:

* TSC-based default delay: 485115 uWatts average power
* MWAITX-based delay: 252738 uWatts average power

Thus, that's about 240 milliWatts less power consumption. The test method relies on the support of AMD CPU accumulated power algorithm in fam15h_power for which patches are forthcoming.

[ Fix delay truncation. ]
Cc Aaron Lu <[email protected]>
Cc Andreas Herrmann <[email protected]>
Cc Aravind Gopalakrishnan <[email protected]>
Cc Fengguang Wu <[email protected]>
Cc Frédéric Weisbecker <[email protected]>
Cc H. Peter Anvin <[email protected]>
Cc Hector Marco-Gisbert <[email protected]>
Cc Jacob Shin <[email protected]>
Cc Jiri Olsa <[email protected]>
Cc John Stultz <[email protected]>
Cc Len Brown <[email protected]>
Cc Linus Torvalds <[email protected]>
Cc Paolo Bonzini <[email protected]>
Cc Rafael J. Wysocki <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc Tony Li <[email protected]>
---
 arch/x86/include/asm/delay.h |  1 +
 arch/x86/kernel/cpu/amd.c    |  4 ++++
 arch/x86/lib/delay.c         | 47 +++++++++++++++++++++++++++++++++++++++++++-
 3 files changed, 51 insertions(+), 1 deletion(-)
arch/x86/include/asm/delay.h +1 -0
index 9b3b4f2..36a760b 100644
@@ -4,5 +4,6 @@
 #include <asm-generic/delay.h>
 
 void use_tsc_delay(void);
+void use_mwaitx_delay(void);
 
 #endif /* _ASM_X86_DELAY_H */
arch/x86/kernel/cpu/amd.c +4 -0
index 51ad2af..4a70fc6 100644
@@ -11,6 +11,7 @@
 #include <asm/cpu.h>
 #include <asm/smp.h>
 #include <asm/pci-direct.h>
+#include <asm/delay.h>
 
 #ifdef CONFIG_X86_64
 # include <asm/mmconfig.h>
@@ -506,6 +507,9 @@ static void bsp_init_amd(struct cpuinfo_x86 *c)
 		/* A random value per boot for bit slice [12:upper_bit) */
 		va_align.bits = get_random_int() & va_align.mask;
 	}
+
+	if (cpu_has(c, X86_FEATURE_MWAITX))
+		use_mwaitx_delay();
 }
 
 static void early_init_amd(struct cpuinfo_x86 *c)
arch/x86/lib/delay.c +46 -1

65 lines not on this page · raw message