I replied[1] to this patch series[2] and described how I think interconnect
bandwidth voting should be captured in DT and how it should work.
So sending out a patch series implementing that. This patch series does the
following:
- Allow required-opps to point to any device's OPP
- Add support to devfreq passive governor to look at the "parent"
devfreq's required-opps to decide the "slave" device's frequency
- Adds Bandwidth OPP table support
- Adds a devfreq library for interconnect paths
Interconnects and interconnect paths quantify they performance levels in
terms of bandwidth. So similar to how we have frequency based OPP tables
in DT and in the OPP framework, this patch series adds bandwidth OPP
table support in the OPP framework and in DT.
To simplify voting for interconnects, this patch series adds helper
functions to create devfreq devices out of interconnect paths. This
allows drivers to add a single line of code to add interconnect voting
capability.
To add devfreq device for the "gpu-mem" interconnect path and use
required OPPs to do decide gpu-mem bandwidth:
icc_create_devfreq(dev, "gpu-mem", gpu_devfreq);
To add devfreq device for the "gpu-mem" interconnect path and use some
other governor to device gpu-mem bandwidth:
icc_create_devfreq(dev, "gpu-mem", NULL);
So with the DT bindings added in this patch series, the DT for a GPU
that does bandwidth voting from GPU to Cache and GPU to DDR would look
something like this:
gpu_cache_opp_table: gpu_cache_opp_table {
compatible = "operating-points-v2";
gpu_cache_3000: opp-3000 {
opp-peak-KBps = <3000>;
opp-avg-KBps = <1000>;
};
gpu_cache_6000: opp-6000 {
opp-peak-KBps = <6000>;
opp-avg-KBps = <2000>;
};
gpu_cache_9000: opp-9000 {
opp-peak-KBps = <9000>;
opp-avg-KBps = <9000>;
};
};
gpu_ddr_opp_table: gpu_ddr_opp_table {
compatible = "operating-points-v2";
gpu_ddr_1525: opp-1525 {
opp-peak-KBps = <1525>;
opp-avg-KBps = <452>;
};
gpu_ddr_3051: opp-3051 {
opp-peak-KBps = <3051>;
opp-avg-KBps = <915>;
};
gpu_ddr_7500: opp-7500 {
opp-peak-KBps = <7500>;
opp-avg-KBps = <3000>;
};
};
gpu_opp_table: gpu_opp_table {
compatible = "operating-points-v2";
opp-shared;
opp-200000000 {
opp-hz = /bits/ 64 <200000000>;
required-opps = <&gpu_cache_3000>, <&gpu_ddr_1525>;
};
opp-400000000 {
opp-hz = /bits/ 64 <400000000>;
required-opps = <&gpu_cache_6000>, <&gpu_ddr_3051>;
};
};
gpu@7864000 {
...
operating-points-v2 = <&gpu_opp_table>, <&gpu_cache_opp_table>, <&gpu_ddr_opp_table>;
interconnects = <&mmnoc MASTER_GPU_1 &bimc SLAVE_SYSTEL_CACHE>,
<&mmnoc MASTER_GPU_1 &bimc SLAVE_DDR>;
interconnect-names = "gpu-cache", "gpu-mem";
interconnect-opp-table = <&gpu_cache_opp_table>, <&gpu_ddr_opp_table>
};
Cheers,
Saravana
[1] - https://lore.kernel.org/lkml/20190601021228.210574-1-saravanak@xxxxxxxxxx/
[2] - https://lore.kernel.org/lkml/20190423132823.7915-1-georgi.djakov@xxxxxxxxxx/
Saravana Kannan (11):
OPP: Allow required-opps even if the device doesn't have power-domains
OPP: Add function to look up required OPP's for a given OPP
PM / devfreq: Add required OPPs support to passive governor
dt-bindings: opp: Introduce opp-peak-KBps and opp-avg-KBps bindings
OPP: Add support for bandwidth OPP tables
OPP: Add helper function for bandwidth OPP tables
OPP: Add API to find an OPP table from its DT node
dt-bindings: interconnect: Add interconnect-opp-table property
interconnect: Add OPP table support for interconnects
OPP: Allow copying OPPs tables between devices
interconnect: Add devfreq support
.../bindings/interconnect/interconnect.txt | 8 +
Documentation/devicetree/bindings/opp/opp.txt | 15 +-
drivers/devfreq/governor_passive.c | 25 ++-
drivers/interconnect/Makefile | 2 +-
drivers/interconnect/core.c | 27 ++-
drivers/interconnect/icc-devfreq.c | 156 ++++++++++++++++++
drivers/opp/core.c | 115 ++++++++++++-
drivers/opp/of.c | 90 +++++++---
drivers/opp/opp.h | 4 +-
include/linux/interconnect.h | 21 +++
include/linux/pm_opp.h | 44 +++++
11 files changed, 474 insertions(+), 33 deletions(-)
create mode 100644 drivers/interconnect/icc-devfreq.c
--
2.22.0.rc2.383.gf4fbbf30c2-goog
