[dpdk-dev] [PATCH v2] timer: add lfence before TSC read

[dpdk-dev] [PATCH v2] timer: add lfence before TSC read

[Didier Pallard]

According to Intel Developer's Manual:

"The RDTSC instruction is not a serializing instruction. It does not necessarily wait
 until all previous instructions have been executed before reading the counter. Simi-
 larly, subsequent instructions may begin execution before the read operation is
 performed. If software requires RDTSC to be executed only after all previous instruc-
 tions have completed locally, it can either use RDTSCP (if the processor supports that
 instruction) or execute the sequence LFENCE;RDTSC."

So add a rte_rdtsc_precise function that do a lfence instruction before rdtsc to
synchronize read operations and ensure that the TSC read is done at the expected
place.

Signed-off-by: Didier Pallard <didier.pallard@6wind.com>
---

Introduce a rte_rdtsc_precise function that adds lfence before reading tsc counter.

 lib/librte_eal/common/include/rte_cycles.h |   37 ++++++++++++++++++++++++++++
 1 file changed, 37 insertions(+)

diff --git a/lib/librte_eal/common/include/rte_cycles.h b/lib/librte_eal/common/include/rte_cycles.h
index cc6fe71..662f62d 100644
--- a/lib/librte_eal/common/include/rte_cycles.h
+++ b/lib/librte_eal/common/include/rte_cycles.h
@@ -128,6 +128,43 @@ rte_rdtsc(void)
 }
 
 /**
+ * Read the TSC register precisely where function is called.
+ *
+ * @return
+ *   The TSC for this lcore.
+ */
+static inline uint64_t
+rte_rdtsc_precise(void)
+{
+	union {
+		uint64_t tsc_64;
+		struct {
+			uint32_t lo_32;
+			uint32_t hi_32;
+		};
+	} tsc;
+
+	/* serialize previous load instructions in pipe */
+	asm volatile("lfence");
+
+#ifdef RTE_LIBRTE_EAL_VMWARE_TSC_MAP_SUPPORT
+	if (unlikely(rte_cycles_vmware_tsc_map)) {
+		/* ecx = 0x10000 corresponds to the physical TSC for VMware */
+		asm volatile("rdpmc" :
+		             "=a" (tsc.lo_32),
+		             "=d" (tsc.hi_32) :
+		             "c"(0x10000));
+		return tsc.tsc_64;
+	}
+#endif
+
+	asm volatile("rdtsc" :
+		     "=a" (tsc.lo_32),
+		     "=d" (tsc.hi_32));
+	return tsc.tsc_64;
+}
+
+/**
  * Get the measured frequency of the RDTSC counter
  *
  * @return
-- 
1.7.10.4

[dpdk-dev] [PATCH v3] timer: add new rte_rdtsc_precise function

[Didier Pallard]

According to Intel Developer's Manual:

"The RDTSC instruction is not a serializing instruction. It does not necessarily wait
 until all previous instructions have been executed before reading the counter. Simi-
 larly, subsequent instructions may begin execution before the read operation is
 performed. If software requires RDTSC to be executed only after all previous instruc-
 tions have completed locally, it can either use RDTSCP (if the processor supports that
 instruction) or execute the sequence LFENCE;RDTSC."

So add a rte_rdtsc_precise function that do a memory barrier before rdtsc to
synchronize operations and ensure that the TSC read is done at the expected
place.

Signed-off-by: Didier Pallard <didier.pallard@6wind.com>
---

Call rte_mb() and rte_rdtsc() rather than duplicating rte_rdtsc function.
Use r/w memory barrier instead of lfence to serialize both load and stores.

 lib/librte_eal/common/include/rte_cycles.h |   14 ++++++++++++++
 1 file changed, 14 insertions(+)

diff --git a/lib/librte_eal/common/include/rte_cycles.h b/lib/librte_eal/common/include/rte_cycles.h
index cc6fe71..e91edf8 100644
--- a/lib/librte_eal/common/include/rte_cycles.h
+++ b/lib/librte_eal/common/include/rte_cycles.h
@@ -76,6 +76,7 @@ extern "C" {
 
 #include <stdint.h>
 #include <rte_debug.h>
+#include <rte_atomic.h>
 
 #ifdef RTE_LIBRTE_EAL_VMWARE_TSC_MAP_SUPPORT
 /** Global switch to use VMWARE mapping of TSC instead of RDTSC */
@@ -128,6 +129,19 @@ rte_rdtsc(void)
 }
 
 /**
+ * Read the TSC register precisely where function is called.
+ *
+ * @return
+ *   The TSC for this lcore.
+ */
+static inline uint64_t
+rte_rdtsc_precise(void)
+{
+	rte_mb();
+	return(rte_rdtsc());
+}
+
+/**
  * Get the measured frequency of the RDTSC counter
  *
  * @return
-- 
1.7.10.4

Re: [dpdk-dev] [PATCH v3] timer: add new rte_rdtsc_precise function

[Thomas Monjalon]

19/02/2014 17:46, Didier Pallard:
> According to Intel Developer's Manual:
> 
> "The RDTSC instruction is not a serializing instruction. It does not
> necessarily wait until all previous instructions have been executed before
> reading the counter. Simi- larly, subsequent instructions may begin
> execution before the read operation is performed. If software requires
> RDTSC to be executed only after all previous instruc- tions have completed
> locally, it can either use RDTSCP (if the processor supports that
> instruction) or execute the sequence LFENCE;RDTSC."
> 
> So add a rte_rdtsc_precise function that do a memory barrier before rdtsc
> to synchronize operations and ensure that the TSC read is done at the
> expected place.
> 
> Signed-off-by: Didier Pallard <didier.pallard@6wind.com>
> ---
> 
> Call rte_mb() and rte_rdtsc() rather than duplicating rte_rdtsc function.
> Use r/w memory barrier instead of lfence to serialize both load and stores.

Acked and applied.

Thanks to all
-- 
Thomas