diff mbox series

[v4,24/24] x86/process: Reset hardware history in context switch

Message ID 20230613042422.5344-25-ricardo.neri-calderon@linux.intel.com
State New
Headers show
Series sched: Introduce classes of tasks for load balance | expand

Commit Message

Ricardo Neri June 13, 2023, 4:24 a.m. UTC 
Reset the classification history of the current task when switching to the
next task. Hardware will start the classification of the next task from
scratch.

Cc: Ben Segall <bsegall@google.com>
Cc: Daniel Bristot de Oliveira <bristot@redhat.com>
Cc: Dietmar Eggemann <dietmar.eggemann@arm.com>
Cc: Ionela Voinescu <ionela.voinescu@arm.com>
Cc: Joel Fernandes (Google) <joel@joelfernandes.org>
Cc: Len Brown <len.brown@intel.com>
Cc: Lukasz Luba <lukasz.luba@arm.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Perry Yuan <Perry.Yuan@amd.com>
Cc: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Cc: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Tim C. Chen <tim.c.chen@intel.com>
Cc: Valentin Schneider <vschneid@redhat.com>
Cc: Zhao Liu <zhao1.liu@linux.intel.com>
Cc: x86@kernel.org
Cc: linux-pm@vger.kernel.org
Cc: linux-kernel@vger.kernel.org
Signed-off-by: Ricardo Neri <ricardo.neri-calderon@linux.intel.com>
---
Changes since v3:
 * None

Changes since v2:
 * None

Changes since v1:
 * Measurements of the cost of the HRESET instruction

   Methodology:
   I created a tight loop with interrupts and preemption disabled. I
   recorded the value of the TSC counter before and after executing
   HRESET or RDTSC. I repeated the measurement 100,000 times.
   I performed the experiment using an Alder Lake S system. I set the
   frequency of the CPUs at a fixed value.

   The table below compares the cost of HRESET with RDTSC (expressed in
   the elapsed TSC count). The cost of the two instructions is
   comparable.

                              PCore      ECore
        Frequency (GHz)        5.0        3.8
        HRESET (avg)          28.5       44.7
        HRESET (stdev %)       3.6        2.3
        RDTSC  (avg)          25.2       35.7
        RDTSC  (stdev %)       3.9        2.6

 * Used an ALTERNATIVE macro instead of static_cpu_has() to execute HRESET
   when supported. (PeterZ)
---
 arch/x86/include/asm/hreset.h | 30 ++++++++++++++++++++++++++++++
 arch/x86/kernel/cpu/common.c  |  7 +++++++
 arch/x86/kernel/process_32.c  |  3 +++
 arch/x86/kernel/process_64.c  |  3 +++
 4 files changed, 43 insertions(+)
 create mode 100644 arch/x86/include/asm/hreset.h
diff mbox series

Patch

diff --git a/arch/x86/include/asm/hreset.h b/arch/x86/include/asm/hreset.h
new file mode 100644
index 000000000000..d68ca2fb8642
--- /dev/null
+++ b/arch/x86/include/asm/hreset.h
@@ -0,0 +1,30 @@ 
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef _ASM_X86_HRESET_H
+
+/**
+ * HRESET - History reset. Available since binutils v2.36.
+ *
+ * Request the processor to reset the history of task classification on the
+ * current logical processor. The history components to be
+ * reset are specified in %eax. Only bits specified in CPUID(0x20).EBX
+ * and enabled in the IA32_HRESET_ENABLE MSR can be selected.
+ *
+ * The assembly code looks like:
+ *
+ *	hreset %eax
+ *
+ * The corresponding machine code looks like:
+ *
+ *	F3 0F 3A F0 ModRM Imm
+ *
+ * The value of ModRM is 0xc0 to specify %eax register addressing.
+ * The ignored immediate operand is set to 0.
+ *
+ * The instruction is documented in the Intel SDM.
+ */
+
+#define __ASM_HRESET  ".byte 0xf3, 0xf, 0x3a, 0xf0, 0xc0, 0x0"
+
+void reset_hardware_history(void);
+
+#endif /* _ASM_X86_HRESET_H */
diff --git a/arch/x86/kernel/cpu/common.c b/arch/x86/kernel/cpu/common.c
index d47a442900ad..8cdd88dab7ed 100644
--- a/arch/x86/kernel/cpu/common.c
+++ b/arch/x86/kernel/cpu/common.c
@@ -53,6 +53,7 @@ 
 #include <asm/mce.h>
 #include <asm/msr.h>
 #include <asm/cacheinfo.h>
+#include <asm/hreset.h>
 #include <asm/memtype.h>
 #include <asm/microcode.h>
 #include <asm/microcode_intel.h>
@@ -398,6 +399,12 @@  static __always_inline void setup_umip(struct cpuinfo_x86 *c)
 
 static u32 hardware_history_features __ro_after_init;
 
+void reset_hardware_history(void)
+{
+	asm_inline volatile (ALTERNATIVE("", __ASM_HRESET, X86_FEATURE_HRESET)
+			     : : "a" (hardware_history_features) : "memory");
+}
+
 static __always_inline void setup_hreset(struct cpuinfo_x86 *c)
 {
 	if (!cpu_feature_enabled(X86_FEATURE_HRESET))
diff --git a/arch/x86/kernel/process_32.c b/arch/x86/kernel/process_32.c
index 708c87b88cc1..7353bb119e79 100644
--- a/arch/x86/kernel/process_32.c
+++ b/arch/x86/kernel/process_32.c
@@ -52,6 +52,7 @@ 
 #include <asm/switch_to.h>
 #include <asm/vm86.h>
 #include <asm/resctrl.h>
+#include <asm/hreset.h>
 #include <asm/proto.h>
 
 #include "process.h"
@@ -214,6 +215,8 @@  __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
 	/* Load the Intel cache allocation PQR MSR. */
 	resctrl_sched_in(next_p);
 
+	reset_hardware_history();
+
 	return prev_p;
 }
 
diff --git a/arch/x86/kernel/process_64.c b/arch/x86/kernel/process_64.c
index 3d181c16a2f6..fcf14cba2a9d 100644
--- a/arch/x86/kernel/process_64.c
+++ b/arch/x86/kernel/process_64.c
@@ -54,6 +54,7 @@ 
 #include <asm/xen/hypervisor.h>
 #include <asm/vdso.h>
 #include <asm/resctrl.h>
+#include <asm/hreset.h>
 #include <asm/unistd.h>
 #include <asm/fsgsbase.h>
 #ifdef CONFIG_IA32_EMULATION
@@ -659,6 +660,8 @@  __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
 	/* Load the Intel cache allocation PQR MSR. */
 	resctrl_sched_in(next_p);
 
+	reset_hardware_history();
+
 	return prev_p;
 }