From: Hans Rosenfeld <hans.rosenfeld@amd.com>
To: <mingo@elte.hu>
Cc: <hpa@zytor.com>, <tglx@linutronix.de>,
<suresh.b.siddha@intel.com>, <eranian@google.com>,
<brgerst@gmail.com>, <robert.richter@amd.com>,
<Andreas.Herrmann3@amd.com>, <x86@kernel.org>,
<linux-kernel@vger.kernel.org>, <bebl@mageta.org>,
Benjamin Block <benjamin.block@amd.com>,
Hans Rosenfeld <hans.rosenfeld@amd.com>
Subject: [RFC 1/5] x86, perf: Implement software-activation of lwp
Date: Fri, 16 Dec 2011 17:12:19 +0100 [thread overview]
Message-ID: <1324051943-21112-1-git-send-email-hans.rosenfeld@amd.com> (raw)
In-Reply-To: <20111216160757.GL665@escobedo.osrc.amd.com>
From: Benjamin Block <benjamin.block@amd.com>
Adds activation of lwp for ring3-software. With this patch and
scheduler-support (xsave-patches done by Hans Rosenfeld) a
userspace-application can use the lwp-instructions (llwpcb,
slwpcb, ..) to activate lwp.
Support for the threshold-interrupt is not given and thus the
corresponding bits are not set. It is therefore unavailable for
userspace-applications.
Signed-off-by: Benjamin Block <benjamin.block@amd.com>
Signed-off-by: Hans Rosenfeld <hans.rosenfeld@amd.com>
---
arch/x86/include/asm/msr-index.h | 1 +
arch/x86/kernel/cpu/Makefile | 2 +-
arch/x86/kernel/cpu/perf_event_amd_lwp.c | 138 ++++++++++++++++++++++++++++++
3 files changed, 140 insertions(+), 1 deletions(-)
create mode 100644 arch/x86/kernel/cpu/perf_event_amd_lwp.c
diff --git a/arch/x86/include/asm/msr-index.h b/arch/x86/include/asm/msr-index.h
index 2d9cf3c..557a6fd 100644
--- a/arch/x86/include/asm/msr-index.h
+++ b/arch/x86/include/asm/msr-index.h
@@ -136,6 +136,7 @@
#define MSR_AMD64_IBSDCPHYSAD 0xc0011039
#define MSR_AMD64_IBSCTL 0xc001103a
#define MSR_AMD64_IBSBRTARGET 0xc001103b
+#define MSR_AMD64_LWP_CFG 0xc0000105
#define MSR_AMD64_LWP_CBADDR 0xc0000106
/* Fam 15h MSRs */
diff --git a/arch/x86/kernel/cpu/Makefile b/arch/x86/kernel/cpu/Makefile
index 6042981..9973465 100644
--- a/arch/x86/kernel/cpu/Makefile
+++ b/arch/x86/kernel/cpu/Makefile
@@ -20,7 +20,7 @@ obj-$(CONFIG_X86_32) += bugs.o
obj-$(CONFIG_X86_64) += bugs_64.o
obj-$(CONFIG_CPU_SUP_INTEL) += intel.o
-obj-$(CONFIG_CPU_SUP_AMD) += amd.o
+obj-$(CONFIG_CPU_SUP_AMD) += amd.o perf_event_amd_lwp.o
obj-$(CONFIG_CPU_SUP_CYRIX_32) += cyrix.o
obj-$(CONFIG_CPU_SUP_CENTAUR) += centaur.o
obj-$(CONFIG_CPU_SUP_TRANSMETA_32) += transmeta.o
diff --git a/arch/x86/kernel/cpu/perf_event_amd_lwp.c b/arch/x86/kernel/cpu/perf_event_amd_lwp.c
new file mode 100644
index 0000000..9aa9a91
--- /dev/null
+++ b/arch/x86/kernel/cpu/perf_event_amd_lwp.c
@@ -0,0 +1,138 @@
+#include <linux/perf_event.h>
+#include <linux/module.h>
+
+#include <asm/cpufeature.h>
+#include <asm/processor.h>
+
+/* masks only the events as of spec r3.08 (lwp v1) */
+#define LWP_EVENT_MASK 0x7E
+
+struct lwp_capabilities {
+#define LWP_CAPS_LWP 0
+#define LWP_CAPS_THRESHOLD 31
+ unsigned long supported_events;
+ unsigned long available_events;
+
+ u8 size_lwpcb;
+ u8 size_event;
+ u8 size_max_event_id;
+ u8 size_event_offset;
+
+#define LWP_CAPS_FILTER_BRANCH 28
+#define LWP_CAPS_FILTER_IP 29
+#define LWP_CAPS_FILTER_CACHE_LVL 30
+#define LWP_CAPS_FILTER_CACHE_LAT 31
+ unsigned long features;
+};
+
+union lwp_cfg_msr {
+ struct {
+ u32 allowed_events;
+ u8 core_id;
+ u8 interrupt_vector;
+ u16 reserved;
+ } cfg;
+ u64 msr_value;
+};
+
+static struct lwp_capabilities lwp_caps;
+
+static void get_lwp_caps(struct lwp_capabilities *caps)
+{
+ u32 sizes;
+
+ memset(caps, 0, sizeof(*caps));
+ cpuid(0x8000001C, (u32 *) &caps->available_events, &sizes,
+ (u32 *) &caps->features,
+ (u32 *) &caps->supported_events);
+
+ caps->size_lwpcb = (u8) sizes;
+ caps->size_event = (u8) (sizes >> 0x8);
+ caps->size_max_event_id = (u8) (sizes >> 0x10);
+ caps->size_event_offset = (u8) (sizes >> 0x18);
+}
+
+static void lwp_start_cpu(void *c)
+{
+ struct lwp_capabilities *caps = (struct lwp_capabilities *) c;
+ union lwp_cfg_msr msr;
+
+ msr.msr_value = 0;
+ /* allow supported events of lwpv1 [1..6] */
+ msr.cfg.allowed_events |= ((u32) caps->supported_events) &
+ LWP_EVENT_MASK;
+ /*
+ * The value showing up in the core-id field of a event-record.
+ * I currently only 8 bits wide.
+ */
+ msr.cfg.core_id = (u8) smp_processor_id();
+
+ /*
+ * We currently do not support the threshold-interrupt so
+ * bit 31 and [40..47] of msr.msr_value keep 0
+ *
+ * msr.cfg.allowed_events |= (1U << 31);
+ * msr.cfg.interrupt_vector = xxx;
+ */
+
+ wrmsrl(MSR_AMD64_LWP_CFG, msr.msr_value);
+}
+
+static int __cpuinit
+lwp_cpu_notifier(struct notifier_block *self, unsigned long action,
+ void *hcpu)
+{
+ switch (action & ~CPU_TASKS_FROZEN) {
+ case CPU_STARTING:
+ lwp_start_cpu(&lwp_caps);
+ break;
+ default:
+ return NOTIFY_DONE;
+ }
+
+ return NOTIFY_OK;
+}
+
+static __init int amd_lwp_init(void)
+{
+ if (!static_cpu_has(X86_FEATURE_LWP))
+ return -ENODEV;
+
+ /* read the _supported_ events */
+ get_lwp_caps(&lwp_caps);
+
+ /* we currently only support lwp v1; spec r3.08 */
+ if (!test_bit(LWP_CAPS_LWP, &lwp_caps.supported_events) ||
+ (((lwp_caps.features >> 9) & 0x7F) != 1))
+ return -ENODEV;
+
+ if (!test_bit(LWP_CAPS_THRESHOLD, &lwp_caps.supported_events))
+ return -ENODEV;
+
+ get_online_cpus();
+
+ barrier();
+
+ perf_cpu_notifier(lwp_cpu_notifier);
+ smp_call_function(lwp_start_cpu, &lwp_caps, 1);
+
+ put_online_cpus();
+
+ /*
+ * The values returned by cpuid are corresponding to the values in
+ * MSR_AMD64_LWP_CFG and determine what events are available. As we
+ * have just changed MSR_AMD64_LWP_CFG, we have to re-read the lwp_caps.
+ */
+ get_lwp_caps(&lwp_caps);
+
+ printk(KERN_INFO "perf: AMD LWP caps: "
+ "[%#lx],[%#hhx|%#hhx|%#hhx|%#hhx],[%#lx],[%#lx]",
+ lwp_caps.available_events, lwp_caps.size_lwpcb,
+ lwp_caps.size_event, lwp_caps.size_max_event_id,
+ lwp_caps.size_event_offset, lwp_caps.features,
+ lwp_caps.supported_events);
+
+ return 0;
+}
+
+device_initcall(amd_lwp_init);
--
1.7.7
next prev parent reply other threads:[~2011-12-16 16:12 UTC|newest]
Thread overview: 54+ messages / expand[flat|nested] mbox.gz Atom feed top
2011-11-29 12:41 [PATCH 0/9] rework of extended state handling, LWP support Hans Rosenfeld
2011-11-29 12:41 ` [PATCH 1/9] x86, xsave: warn on #NM exceptions caused by the kernel Hans Rosenfeld
2011-11-29 12:41 ` [PATCH 2/9] x86, xsave: cleanup fpu/xsave support Hans Rosenfeld
2011-11-29 12:41 ` [PATCH 3/9] x86, xsave: cleanup fpu/xsave signal frame setup Hans Rosenfeld
2011-11-29 12:41 ` [PATCH 4/9] x86, xsave: rework fpu/xsave support Hans Rosenfeld
2011-11-29 12:41 ` [PATCH 5/9] x86, xsave: remove unused code Hans Rosenfeld
2011-11-29 12:41 ` [PATCH 6/9] x86, xsave: more cleanups Hans Rosenfeld
2011-11-29 12:41 ` [PATCH 7/9] x86, xsave: remove lazy allocation of xstate area Hans Rosenfeld
2011-11-29 12:41 ` [PATCH 8/9] x86, xsave: add support for non-lazy xstates Hans Rosenfeld
2011-11-29 12:41 ` [PATCH 9/9] x86, xsave: add kernel support for AMDs Lightweight Profiling (LWP) Hans Rosenfeld
2011-11-29 21:31 ` [PATCH 0/9] rework of extended state handling, LWP support Andi Kleen
2011-11-30 17:37 ` Hans Rosenfeld
2011-11-30 21:52 ` Andi Kleen
2011-12-01 20:36 ` Hans Rosenfeld
2011-12-02 2:01 ` H. Peter Anvin
2011-12-02 11:20 ` Hans Rosenfeld
2011-12-07 19:57 ` Hans Rosenfeld
2011-12-07 20:00 ` [PATCH 7/8] x86, xsave: add support for non-lazy xstates Hans Rosenfeld
2011-12-07 20:00 ` [PATCH 8/8] x86, xsave: add kernel support for AMDs Lightweight Profiling (LWP) Hans Rosenfeld
2011-12-05 10:22 ` [PATCH 0/9] rework of extended state handling, LWP support Ingo Molnar
2011-12-16 16:07 ` Hans Rosenfeld
2011-12-16 16:12 ` Hans Rosenfeld [this message]
2011-12-16 16:12 ` [RFC 2/5] perf: adds prototype for a new perf-context-type Hans Rosenfeld
2011-12-16 16:12 ` [RFC 3/5] perf: adds a new pmu-initialization-call Hans Rosenfeld
2011-12-16 16:12 ` [RFC 4/5] x86, perf: implements lwp-perf-integration (rc1) Hans Rosenfeld
2011-12-18 8:04 ` Ingo Molnar
2011-12-18 15:22 ` Benjamin Block
2011-12-18 23:43 ` Ingo Molnar
2011-12-19 9:09 ` Robert Richter
2011-12-19 10:54 ` Ingo Molnar
2011-12-19 11:12 ` Avi Kivity
2011-12-19 11:40 ` Ingo Molnar
2011-12-19 11:58 ` Avi Kivity
2011-12-19 18:13 ` Benjamin
2011-12-20 8:56 ` Ingo Molnar
2011-12-20 9:15 ` Ingo Molnar
2011-12-20 9:47 ` Avi Kivity
2011-12-20 10:09 ` Ingo Molnar
2011-12-20 15:27 ` Joerg Roedel
2011-12-20 18:40 ` Ingo Molnar
2011-12-21 0:07 ` Joerg Roedel
2011-12-21 12:34 ` Ingo Molnar
2011-12-21 12:44 ` Avi Kivity
2011-12-21 13:22 ` Ingo Molnar
2011-12-21 22:49 ` Joerg Roedel
2011-12-23 10:53 ` Ingo Molnar
2011-12-21 11:46 ` Gleb Natapov
2011-12-23 10:56 ` Ingo Molnar
2011-12-20 15:48 ` Vince Weaver
2011-12-20 18:27 ` Ingo Molnar
2011-12-20 22:47 ` Vince Weaver
2011-12-21 12:00 ` Ingo Molnar
2011-12-21 13:55 ` Vince Weaver
2011-12-16 16:12 ` [RFC 5/5] x86, perf: adds support for the LWP threshold-int Hans Rosenfeld
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=1324051943-21112-1-git-send-email-hans.rosenfeld@amd.com \
--to=hans.rosenfeld@amd.com \
--cc=Andreas.Herrmann3@amd.com \
--cc=bebl@mageta.org \
--cc=benjamin.block@amd.com \
--cc=brgerst@gmail.com \
--cc=eranian@google.com \
--cc=hpa@zytor.com \
--cc=linux-kernel@vger.kernel.org \
--cc=mingo@elte.hu \
--cc=robert.richter@amd.com \
--cc=suresh.b.siddha@intel.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®