Uncore Manual

Reference Number: 329468-002 145
Uncore Performance Monitoring
Intel® QPI Link Layer Performance Monitoring
RxL0P_POWER_CYCLES
• Title: Cycles in L0p
• Category: POWER_RX Events
• Event Code: 0x10
• Max. Inc/Cyc:. 1, Register Restrictions: 0-3
• Definition: Number of QPI qfclk cycles spent in L0p power mode. L0p is a mode where we disable
1/2 of the QPI lanes, decreasing our bandwidth in order to save power. It increases snoop and
data transfer latencies and decreases overall bandwidth. This mode can be very useful in NUMA
optimized workloads that largely only utilize QPI for snoops and their responses. Use edge detect
to count the number of instances when the QPI link entered L0p. Link power states are per link
and per direction, so for example the Tx direction could be in one state while Rx was in another.
• NOTE: Using .edge_det to count transitions does not function if L1_POWER_CYCLES > 0.
RxL0_POWER_CYCLES
• Title: Cycles in L0
• Category: POWER_RX Events
• Event Code: 0x0f
• Max. Inc/Cyc:. 1, Register Restrictions: 0-3
• Definition: Number of QPI qfclk cycles spent in L0 power mode in the Link Layer. L0 is the default
mode which provides the highest performance with the most power. Use edge detect to count the
number of instances that the link entered L0. Link power states are per link and per direction, so
for example the Tx direction could be in one state while Rx was in another. The phy layer some-
times leaves L0 for training, which will not be captured by this event.
• NOTE: Includes L0p cycles. To get just L0, subtract RxL0P_POWER_CYCLES.
RxL_BYPASSED
• Title: Rx Flit Buffer Bypassed
• Category: RXQ Events
• Event Code: 0x09
• Max. Inc/Cyc:. 1, Register Restrictions: 0-3
• Definition: Counts the number of times that an incoming flit was able to bypass the flit buffer
and pass directly across the BGF and into the Egress. This is a latency optimization, and should
generally be the common case. If this value is less than the number of flits transferred, it implies
that there was queueing getting onto the ring, and thus the transactions saw higher latency.
RxL_CREDITS_CONSUMED_VN0
• Title: VN0 Credit Consumed
• Category: RX_CREDITS_CONSUMED Events
• Event Code: 0x1e
• Extra Select Bit: Y
• Max. Inc/Cyc:. 2, Register Restrictions: 0-3
• Definition: Counts the number of times that an RxQ VN0 credit was consumed (i.e. message
uses a VN0 credit for the Rx Buffer). This includes packets that went through the RxQ and those
that were bypasssed.
Table 2-140. Unit Masks for RxL_CREDITS_CONSUMED_VN0
Extension
umask
[15:8]
Description
DRS bxxxxxxx1 DRS
VN0 credit for the DRS message class.
NCB bxxxxxx1x NCB
VN0 credit for the NCB message class.