rpmi: performance: Add memory region service #81
Conversation
yeongjoshua
force-pushed
the
perf-memory-region
branch
from
2a28a69 to
f04b957
Compare
yeongjoshua
force-pushed
the
perf-memory-region
branch
from
f04b957 to
0d2c4c5
Compare
yeongjoshua
force-pushed
the
perf-memory-region
branch
from
0d2c4c5 to
4689f53
Compare
yeongjoshua
force-pushed
the
perf-memory-region
branch
from
4689f53 to
7536240
Compare
yeongjoshua
force-pushed
the
perf-memory-region
branch
from
995b1d6 to
992d6ee
Compare
| factor of `1000`. | ||
| the metric is entirely platform-dependent. Values on this scale are represented | ||
| with `performance level index`, and the platform has complete control over | ||
| mapping these performance operating points to performance states, which are |
There was a problem hiding this comment.
So, performance operating point == performance level index ??
There was a problem hiding this comment.
Why can't we simply say it Performance Level instead of Performance Level Index.
Reason being, index term is always used to retrieve some other item which is not available directly.
There was a problem hiding this comment.
We also need to mention property of performance scale that when the performance scales up or scales down if we move on the performance scale.
For example
if [perflvl1, perflvl2, perflvl3, perflvl4, perflvl5, ...., perflvlN] is the performance scale where perflvl1 represents the first level and perflvlN represents the last level. Then, if the AP moves from perflvl1 -> perflvl2 the performance scales up. and vice-versa, it scales down.
|
@pathakraul would this one is more preferred instead? |
yeongjoshua
force-pushed
the
perf-memory-region
branch
2 times, most recently
from
865aace to
fd069f8
Compare
|
Apart from above i dont have any more comments. There are some minor formatting required which i will handle to make it similar to rest of the spec. |
|
Also please rebase this PR with latest main branch |
Clarify the performance level represented by the performance index which links to its attributes through PERF_GET_SUPPORTED_LEVELS. Segregate the RPMI performance fast shared region to ease PMP/SPMP on the fast channel memory region block. Signed-off-by: Joshua Yeong <[email protected]>
yeongjoshua
force-pushed
the
perf-memory-region
branch
from
fd069f8 to
06c5cb7
Compare
yeongjoshua
force-pushed
the
perf-memory-region
branch
from
06c5cb7 to
6abb148
Compare
Add explaination to multi-linear voltage format. Signed-off-by: Joshua Yeong <[email protected]>
yeongjoshua
force-pushed
the
perf-memory-region
branch
from
6abb148 to
d579bb4
Compare
|
|
||
| | 3 | ||
| | STEP | ||
| | Step size in microvolts (uV). | ||
| |=== | ||
|
|
||
| ---- | ||
| For example, a system with voltage profile of |
| (a) 900mV to 1650mV, step size of 50mV | ||
| (b) 1800mV to 3300mV, step size of 100mV | ||
|
|
||
| Multi-linear format structure: |
There was a problem hiding this comment.
Still its not clear. Please explain how VOLTAGE_MIN_SEL and VOLTAGE_MAX_SEL are derived and how they are used in the example
There was a problem hiding this comment.
From my reading, the selectors are not used by any service (VOLT_SET_LEVEL and VOLT_GET_LEVEL use microvolts directly). They are purely an implementation detail of how the PuC programs the underlying hardware. Both the Simple-Linear and Multi-Linear formats support NUM_LEVELS > 1, so from a client perspective they appear to be interchangeable, and the Multi-Linear format is redundant.
|
Performance group is fine. only one comment in the voltage group above |
Add 'PERF_GET_FAST_CHANNEL_REGION' service to get a continuous block of physical memory region for fast channel access.