From: Robert Richter <robert.richter@amd.com>
To: Ingo Molnar <mingo@elte.hu>
Cc: LKML <linux-kernel@vger.kernel.org>,
oprofile-list <oprofile-list@lists.sourceforge.net>,
Robert Richter <robert.richter@amd.com>
Subject: [PATCH 07/15] oprofile/x86: implement randomization for IBS periodic op counter
Date: Fri, 26 Feb 2010 18:29:59 +0100 [thread overview]
Message-ID: <1267205407-6523-8-git-send-email-robert.richter@amd.com> (raw)
In-Reply-To: <1267205407-6523-1-git-send-email-robert.richter@amd.com>
IBS selects an op (execution operation) for sampling by counting
either cycles or dispatched ops. Better statistical samples can be
produced by adding a software generated random offset to the periodic
op counter value with each sample.
This patch adds software randomization to the IBS periodic op
counter. The lower 12 bits of the 20 bit counter are
randomized. IbsOpCurCnt is initialized with a 12 bit random value.
There is a work around if the hw can not write to IbsOpCurCnt. Then
the lower 8 bits of the 16 bit IbsOpMaxCnt [15:0] value are randomized
in the range of -128 to +127 by adding/subtracting an offset to the
maximum count (IbsOpMaxCnt).
The linear feedback shift register (LFSR) algorithm is used for
pseudo-random number generation to have low impact to the memory
system.
Signed-off-by: Robert Richter <robert.richter@amd.com>
---
arch/x86/oprofile/op_model_amd.c | 69 ++++++++++++++++++++++++++++++++++---
1 files changed, 63 insertions(+), 6 deletions(-)
diff --git a/arch/x86/oprofile/op_model_amd.c b/arch/x86/oprofile/op_model_amd.c
index 97c84eb..a9d1947 100644
--- a/arch/x86/oprofile/op_model_amd.c
+++ b/arch/x86/oprofile/op_model_amd.c
@@ -52,7 +52,7 @@ static unsigned long reset_value[NUM_VIRT_COUNTERS];
#define IBS_FETCH_ENABLE (1ULL<<48)
#define IBS_FETCH_CNT_MASK 0xFFFF0000ULL
-/*IbsOpCtl bits */
+/* IbsOpCtl bits */
#define IBS_OP_CNT_CTL (1ULL<<19)
#define IBS_OP_VAL (1ULL<<18)
#define IBS_OP_ENABLE (1ULL<<17)
@@ -72,6 +72,7 @@ struct op_ibs_config {
};
static struct op_ibs_config ibs_config;
+static u64 ibs_op_ctl;
/*
* IBS cpuid feature detection
@@ -84,8 +85,16 @@ static struct op_ibs_config ibs_config;
* bit 0 is used to indicate the existence of IBS.
*/
#define IBS_CAPS_AVAIL (1LL<<0)
+#define IBS_CAPS_RDWROPCNT (1LL<<3)
#define IBS_CAPS_OPCNT (1LL<<4)
+/*
+ * IBS randomization macros
+ */
+#define IBS_RANDOM_BITS 12
+#define IBS_RANDOM_MASK ((1ULL << IBS_RANDOM_BITS) - 1)
+#define IBS_RANDOM_MAXCNT_OFFSET (1ULL << (IBS_RANDOM_BITS - 5))
+
static u32 get_ibs_caps(void)
{
u32 ibs_caps;
@@ -241,6 +250,38 @@ static unsigned int lfsr_random(void)
return lfsr_value;
}
+/*
+ * IBS software randomization
+ *
+ * The IBS periodic op counter is randomized in software. The lower 12
+ * bits of the 20 bit counter are randomized. IbsOpCurCnt is
+ * initialized with a 12 bit random value.
+ */
+static inline u64 op_amd_randomize_ibs_op(u64 val)
+{
+ unsigned int random = lfsr_random();
+
+ if (!(ibs_caps & IBS_CAPS_RDWROPCNT))
+ /*
+ * Work around if the hw can not write to IbsOpCurCnt
+ *
+ * Randomize the lower 8 bits of the 16 bit
+ * IbsOpMaxCnt [15:0] value in the range of -128 to
+ * +127 by adding/subtracting an offset to the
+ * maximum count (IbsOpMaxCnt).
+ *
+ * To avoid over or underflows and protect upper bits
+ * starting at bit 16, the initial value for
+ * IbsOpMaxCnt must fit in the range from 0x0081 to
+ * 0xff80.
+ */
+ val += (s8)(random >> 4);
+ else
+ val |= (u64)(random & IBS_RANDOM_MASK) << 32;
+
+ return val;
+}
+
static inline void
op_amd_handle_ibs(struct pt_regs * const regs,
struct op_msrs const * const msrs)
@@ -290,8 +331,7 @@ op_amd_handle_ibs(struct pt_regs * const regs,
oprofile_write_commit(&entry);
/* reenable the IRQ */
- ctl &= ~IBS_OP_VAL & 0xFFFFFFFF;
- ctl |= IBS_OP_ENABLE;
+ ctl = op_amd_randomize_ibs_op(ibs_op_ctl);
wrmsrl(MSR_AMD64_IBSOPCTL, ctl);
}
}
@@ -312,10 +352,27 @@ static inline void op_amd_start_ibs(void)
}
if (ibs_config.op_enabled) {
- val = (ibs_config.max_cnt_op >> 4) & 0xFFFF;
+ ibs_op_ctl = ibs_config.max_cnt_op >> 4;
+ if (!(ibs_caps & IBS_CAPS_RDWROPCNT)) {
+ /*
+ * IbsOpCurCnt not supported. See
+ * op_amd_randomize_ibs_op() for details.
+ */
+ ibs_op_ctl = clamp(ibs_op_ctl, 0x0081ULL, 0xFF80ULL);
+ } else {
+ /*
+ * The start value is randomized with a
+ * positive offset, we need to compensate it
+ * with the half of the randomized range. Also
+ * avoid underflows.
+ */
+ ibs_op_ctl = min(ibs_op_ctl + IBS_RANDOM_MAXCNT_OFFSET,
+ 0xFFFFULL);
+ }
if (ibs_caps & IBS_CAPS_OPCNT && ibs_config.dispatched_ops)
- val |= IBS_OP_CNT_CTL;
- val |= IBS_OP_ENABLE;
+ ibs_op_ctl |= IBS_OP_CNT_CTL;
+ ibs_op_ctl |= IBS_OP_ENABLE;
+ val = op_amd_randomize_ibs_op(ibs_op_ctl);
wrmsrl(MSR_AMD64_IBSOPCTL, val);
}
}
--
1.6.6
next prev parent reply other threads:[~2010-02-26 17:37 UTC|newest]
Thread overview: 19+ messages / expand[flat|nested] mbox.gz Atom feed top
2010-02-26 17:29 [PATCH 00/15] oprofile fixes and updates for v2.6.34 Robert Richter
2010-02-26 17:29 ` [PATCH 01/15] oprofile: remove tracing build dependency Robert Richter
2010-02-26 17:29 ` [PATCH 02/15] oprofile: remove EXPERIMENTAL from the config option description Robert Richter
2010-02-26 17:29 ` [PATCH 03/15] oprofile/x86: remove OPROFILE_IBS config option Robert Richter
2010-02-26 17:29 ` [PATCH 04/15] oprofile/x86: remove node check in AMD IBS initialization Robert Richter
2010-02-26 17:29 ` [PATCH 05/15] oprofile/x86: implement IBS cpuid feature detection Robert Richter
2010-02-26 17:29 ` [PATCH 06/15] oprofile/x86: implement lsfr pseudo-random number generator for IBS Robert Richter
2010-02-26 17:29 ` Robert Richter [this message]
2010-02-26 17:30 ` [PATCH 08/15] oprofile/x86: warn user if a counter is already active Robert Richter
2010-02-26 17:30 ` [PATCH 09/15] oprofile/x86: add comment to counter-in-use warning Robert Richter
2010-02-26 17:30 ` [PATCH 10/15] oprofile/x86: fix perfctr nmi reservation for mulitplexing Robert Richter
2010-02-26 17:30 ` [PATCH 11/15] oprofile/x86: use kzalloc() instead of kmalloc() Robert Richter
2010-02-26 17:30 ` [PATCH 12/15] oprofile/x86: fix msr access to reserved counters Robert Richter
2010-02-26 17:30 ` [PATCH 13/15] perf, x86: make IBS macros available in perf_event.h Robert Richter
2010-02-26 17:30 ` [PATCH 14/15] perf, x86: add some IBS macros to perf_event.h Robert Richter
2010-02-26 17:30 ` [PATCH 15/15] perf, x86: rename macro in ARCH_PERFMON_EVENTSEL_ENABLE Robert Richter
2010-02-27 9:03 ` [PATCH 00/15] oprofile fixes and updates for v2.6.34 Ingo Molnar
2010-03-01 14:39 ` Robert Richter
2010-03-02 10:01 ` Ingo Molnar
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=1267205407-6523-8-git-send-email-robert.richter@amd.com \
--to=robert.richter@amd.com \
--cc=linux-kernel@vger.kernel.org \
--cc=mingo@elte.hu \
--cc=oprofile-list@lists.sourceforge.net \
/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
Powered by JetHome