10. Performance Monitors¶
Currently, the RISC-V privilege spec (v1.11) describes a basic hardware performance facility at the hart (core) level. Three counters for dedicated functions have been defined by the spec:
Address |
Name |
Description |
---|---|---|
0xB01 |
|
counts the number of cycles executed by the hart starting from an arbitrary point of time. |
0xB02 |
|
counts the number of instructions executed by the hart starting from an arbitrary point of time. |
0xC03 |
|
this is a read-only CSR which reads the memory mapped value of the platforms real-time counter. |
Each of the above are 64-bit counters. Shadow CSRs of the above also exist in the user-space.
Apart from the above, RISC-V also provides provision to instantiate additional 29 64-bit event
counters: mhpmcounter3
- mhpmcounter31
. The event selectors for these counters are also
defined: mhpmevent3
- mhpmevent31
. The meaning of these events is defined by the platform
and can be customized for each platform.
This instance of the design instantiates 2
mhpmcounters
counters.
In addition, a single 32-bit counter-enable register : mcounteren
is
available.
Each bit in this register corresponds to each of the 32 event-counters described above. This register
controls only the accessibility of the counter registers and has no effect on the underlying counters,
which will continue to increment irrespective of the settings of the mcounteren
fields.
Clearing a bit in the mcounteren
only indicates that the event-counters cannot be accessed by
lower level privilege modes.
10.1. Need for Daisy Chaining¶
Each event-counter is mapped to a read/write CSR address. Thus, each 64-bit counter will have an additional 12-bit decoder to select that counter in case of a read/write CSR op.
Since all CSRs are accessed in the write-back stage of the core, the 12-bit address from this unit, fans-out to all CSRs. Since the event-counters are implemented as 64-bit adders, the fan-out load is further increased as they become part of the CSR read/write op.
Further more, suppose there are 30 events defined by the core/platform and each event-counter is configurable to choose any of the 30 events to track. This leads to an additional 30 is 1 demux on each event-counter.
All the three factors defined above can cause the event-counters to become critical in terms of area and frequency closure.
To address the issues 1 and 2
listed above, it is possible to implement the
CSRs as a daisy-chain. Here the CSRs are grouped based on their functionality and accesses to
CSRs can thus take variable number of cycles. For eg, less frequently accessed CSRs like
fcssr
or *scratch
or debug registers
can be placed in GRP-2 or GRP-3.
Performance counters and status registers can be placed in GRP-1 to enable quick and
fast access. Such daisy chaining will reduce the comparator fan-out while performing CSR read/write ops.
10.2. List of Events for¶
The core will support capturing the following 26 events:
Event number |
Description |
1 |
Number of misprediction |
2 |
Number of exceptions |
3 |
Number of interrupts |
4 |
Number of csr operations |
5 |
Number of jumps |
6 |
Number of branches |
7 |
Reserved |
8 |
Number of mul/div operations |
9 |
Number of RAW stalls |
10 |
Number of Execute Unit stalls |
11 |
Number of I$ accesses |
12 |
Number of I$ miss |
13 |
Number of I$ fill-buffer hits |
14 |
Number of I$ non cacheable accesses |
15 |
Number of I$ fill-buffer releases |
16 |
Number of D$ read accesses |
17 |
Number of D$ write accesses |
18 |
Number of D$ atomic accesses |
19 |
Number of D$ Non-cached read access |
20 |
Number of D$ Non-cached write acces |
21 |
Number of D$ read misses |
22 |
Number of D$ write misses |
23 |
Number of D$ atomic misses |
24 |
Number of D$ read fill-buffer hits |
25 |
Number of D$ write fill-buffer hits |
26 |
Number of D$ atomic fill-buffer hits |
27 |
Number of D$ fill-buffer releases |
28 |
Number of D$ line evictions |
29 |
Reserved |
30 |
Reserved |
10.3. Interrupts from Counters¶
There is a need to raise an interrupt when a particular counter has observed delta
number of counts.
This feature is however, not part of the current RISC-V ISA, since it does not mandate how the
counters are interpreted and neither specify on which direction should the counters move (up or down).
We use a custom CSR :mhpmcounters
to generate interrupts when a counter crosses a threshold.
The encoding for this CSR is the same as that of mcounteren/mcountinhibit.
When a particular bit is set, it indicates that the corresponding counter will generate an
interrupt when the value reaches 0 and the counter is enabled (mhpmevent != 0). The interrupt
can be disabled by writing a 0 to the corresponding mhpmevent
register
(equivalent to disabling the counter)
Following is an example of how such a framework can be used:
> csrw mhpminterrupten, 0x4 # enable interrupt for mhpmcounter3
> addi x31, x0, -delta # note the negative delta
> csrw mhpmcounter3, x31
> csrw mhpmevent3, 0x9 # enable mhpmcounter3 to track event-code-9
> ...
> interrupt is generated jump to isr!
> ...
>
ISR Routine
> csrw mhpmevent3, x0 # disable mhphmcounter3 will also disable the interrupt.