As GDSCs are turned on and off some associated clocks are momentarily
enabled for house keeping purposes. For this, and similar, purposes the
"shared RCGs" will park the RCG on a source clock which is known to be
available.
When the RCG is parked, a safe clock source will be selected and
committed, then the original source would be written back and upon enable
the change back to the unparked source would be committed.
But starting with SM8350 this fails, as the value in CFG is committed by
the GDSC handshake and without a ticking parent the GDSC enablement will
time out.
This becomes a concrete problem if the runtime supended state of a
device includes disabling such rcg's parent clock. As the device
attempts to power up the domain again the rcg will fail to enable and
hence the GDSC enablement will fail, preventing the device from
returning from the suspended state.
This can be seen in e.g. the display stack during probe on SM8350.
To avoid this problem, the software needs to ensure that the RCG is
configured to a active parent clock while it is disabled. This is done
by caching the CFG register content while the shared RCG is parked on
this safe source.
Writes to M, N and D registers are committed as they are requested. New
helpers for get_parent() and recalc_rate() are extracted from their
previous implementations and __clk_rcg2_configure() is modified to allow
it to operate on the cached value.
Fixes: 7ef6f11887 ("clk: qcom: Configure the RCGs to a safe source as needed")
Signed-off-by: Bjorn Andersson <bjorn.andersson@linaro.org>
Reviewed-by: Stephen Boyd <sboyd@kernel.org>
Link: https://lore.kernel.org/r/20220426212136.1543984-1-bjorn.andersson@linaro.org
Add clk_rcg_floor_ops for clock that can't provide a stable freq and
require to use a floor freq to provide the requested frequency.
Signed-off-by: Ansuel Smith <ansuelsmth@gmail.com>
Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org>
Reviewed-by: Stephen Boyd <sboyd@kernel.org>
Tested-by: Jonathan McDowell <noodles@earth.li>
Signed-off-by: Bjorn Andersson <bjorn.andersson@linaro.org>
Link: https://lore.kernel.org/r/20220226135235.10051-10-ansuelsmth@gmail.com
The function clk_gfx3d_determine_rate is selecting different PLLs
to manage the GFX3D clock source in a special way: this one needs
to be ping-pong'ed on different PLLs to ensure stability during
frequency switching (set a PLL rate, let it stabilize, switch the
RCG to the new PLL) and fast frequency transitions.
This technique is currently being used in the MSM8996 SoC and the
function was assuming that the parents were always at a specific
index in the parents list, which is TRUE, if we use this only on
the MSM8996 MMCC.
Unfortunately, MSM8996 is not the only SoC that needs to ping-pong
the graphics RCG, so choices are:
1. Make new special ops just to hardcode *again* other indexes,
creating code duplication for (imo) no reason; or
2. Generalize this function, so that it becomes usable for a range
of SoCs with slightly different ping-pong configuration.
In this commit, the second road was taken: define a new "special"
struct clk_rcg2_gfx3d, containing the ordered list of parents to
ping-pong the graphics clock on, and the "regular" rcg2 clock
structure in order to generalize the clk_gfx3d_determine_rate
function and make it working for other SoCs.
As for the function itself it is left with the assumption that we
need to ping-pong over three parents. The reasons for this are:
1. The initial model was MSM8996, which has 3 parents for the
graphics clock pingpong;
2. The other example that was taken into consideration is the
SDM630/636/660 SoC gpu clock controller, which is ping-ponging
over two dynamic clocked and one fixed clock PLL.
Signed-off-by: AngeloGioacchino Del Regno <angelogioacchino.delregno@somainline.org>
Link: https://lore.kernel.org/r/20210113183817.447866-6-angelogioacchino.delregno@somainline.org
[sboyd@kernel.org: Grow some local variables, drop do_div() usage in
favor of plain division, we're not dealing with a u64 here]
Signed-off-by: Stephen Boyd <sboyd@kernel.org>
Update the init data name for each of the dynamic frequency switch
controlled clock associated with the RCG clock name, so that it can be
generated as per the hardware plan. Thus update the macro accordingly.
Signed-off-by: Taniya Das <tdas@codeaurora.org>
Link: https://lkml.kernel.org/r/20191014102308.27441-2-tdas@codeaurora.org
Signed-off-by: Stephen Boyd <sboyd@kernel.org>
The RCG CFG/M/N/D register base could be at a different offset than
the CMD register, so introduce a cfg_offset to identify the offset
with respect to the CMD RCGR register.
Signed-off-by: Taniya Das <tdas@codeaurora.org>
Signed-off-by: Anu Ramanathan <anur@codeaurora.org>
Signed-off-by: Shawn Guo <shawn.guo@linaro.org>
Signed-off-by: Vinod Koul <vkoul@kernel.org>
Signed-off-by: Stephen Boyd <sboyd@kernel.org>
Dynamic Frequency switch is a feature of clock controller by which request
from peripherals allows automatic switching frequency of input clock
without SW intervention. There are various performance levels associated
with a root clock. When the input performance state changes, the source
clocks and division ratios of the new performance state are loaded on to
RCG via HW and the RCG switches to new clock frequency when the RCG is in
DFS HW enabled mode.
Register the root clock generators(RCG) to switch to use the dfs clock ops
in the cases where DFS is enabled. The clk_round_rate() called by the clock
consumer would invoke the dfs determine clock ops and would read the DFS
performance level registers to identify all the frequencies supported and
update the frequency table. The DFS clock consumers would maintain these
frequency mapping and request the desired performance levels.
Signed-off-by: Taniya Das <tdas@codeaurora.org>
[sboyd@kernel.org: Rework registration logic to stop copying, change
recalc_rate() to index directly into the table if possible and fallback
to calculating on the fly with an assumed correct parent]
Signed-off-by: Stephen Boyd <sboyd@kernel.org>
Frequency table macro is used by multiple clock drivers, move frequency
table macro to common header file.
Signed-off-by: Taniya Das <tdas@codeaurora.org>
Signed-off-by: Stephen Boyd <sboyd@kernel.org>
For some root clock generators, there could be child branches which are
controlled by an entity other than application processor subsystem. For
such RCGs, as per application processor subsystem clock driver, all of
its downstream clocks are disabled and RCG is in disabled state but in
reality downstream clocks can be left enabled before.
So in this scenario, when RCG is disabled as per clock driver's point of
view and when rate scaling request comes before downstream clock enable
request, then RCG fails to update its configuration because in reality
RCG is on and it expects its new source to already be in enable state but
in reality new source is off. In order to avoid having the RCG to go into
an invalid state, add support to update the CFG, M, N and D registers
during set_rate() without configuration update and defer the actual RCG
configuration update to be done during clk_enable() as at this point of
time, both its new parent and safe source will be already enabled and RCG
can safely switch to new parent.
During clk_disable() request, configure it to safe source as both its
parents, safe source and current parent will be enabled and RCG can
safely execute a switch.
Signed-off-by: Taniya Das <tdas@codeaurora.org>
Signed-off-by: Amit Nischal <anischal@codeaurora.org>
Signed-off-by: Stephen Boyd <sboyd@kernel.org>
Currently the driver assumes the register configuration value
is identical to its index in the parent map. This patch adds
the parent map field in regmap mux clock node which contains
the mapping of parent index with actual register configuration
value. If regmap node contains this parent map then the
configuration value will be taken from this
parent map instead of simply writing the index value.
Signed-off-by: Abhishek Sahu <absahu@codeaurora.org>
Signed-off-by: Stephen Boyd <sboyd@codeaurora.org>
The RCGs ops for shared branches are not used now, so remove it.
Signed-off-by: Georgi Djakov <georgi.djakov@linaro.org>
Signed-off-by: Stephen Boyd <sboyd@codeaurora.org>
The default behaviour with clk_rcg2_ops is for the
clk_round_rate()/clk_set_rate() to return/set a ceil clock
rate closest to the requested rate by looking up the corresponding
frequency table.
However, we do have some instances (mainly sdcc on various platforms)
of clients expecting a clk_set_rate() to set a floor value instead.
Add a new clk_rcg2_floor_ops to handle this for such specific
rcg instances
Signed-off-by: Rajendra Nayak <rnayak@codeaurora.org>
Signed-off-by: Ritesh Harjani <riteshh@codeaurora.org>
Signed-off-by: Stephen Boyd <sboyd@codeaurora.org>
The GPU clocks on msm8996 have three dedicated PLLs, MMPLL2,
MMPLL8, and MMPLL9. We leave MMPLL9 at the maximum speed (624
MHz), and we use MMPLL2 and MMPLL8 for the other frequencies. To
make switching frequencies faster, we ping-pong between MMPLL2
and MMPLL8 when we're switching between frequencies that aren't
the maximum. Implement custom rcg clk ops for this type of
frequency switching.
Signed-off-by: Stephen Boyd <sboyd@codeaurora.org>
DSI specific RCG clocks required customized clk_ops. There are
a total of 4 RCGs per DSI block: DSI, BYTE, ESC and PIXEL.
There are a total of 2 clocks coming from the DSI PLL, which serve as
inputs to these RCGs. The BYTE and ESC RCGs are fed by one of the
post dividers of DSI1 or DSI2 PLLs, and the DSI and PIXEL RCGs are fed by
another divider of the PLL.
In each of the 2 groups above, only one of the clocks sets its parent.
These are BYTE RCG and DSI RCG for each of the groups respectively, as
shown in the diagram below.
The DSI and BYTE RCGs serve as bypass clocks. We create a new set of ops
clk_rcg_bypass2_ops, which are like the regular bypass ops, but don't
take in a freq table, since the DSI driver using these clocks is
parent-able.
The PIXEL RCG needs to derive the required pixel clock using dsixpll.
It parses a m/n frac table to retrieve the correct clock.
The ESC RCG doesn't have a frac M/N block, it can just apply a pre-
divider. Its ops simply check if the required clock rate can be
achieved by the pre-divider.
+-------------------+
| |---dsixpllbyte---o---> To byte RCG
| | | (sets parent rate)
| | |
| | |
| DSI 1/2 PLL | |
| | o---> To esc RCG
| | (doesn't set parent rate)
| |
| |----dsixpll-----o---> To dsi RCG
+-------------------+ | (sets parent rate)
( x = 1, 2 ) |
|
o---> To pixel rcg
(doesn't set parent rate)
Signed-off-by: Archit Taneja <architt@codeaurora.org>
Signed-off-by: Stephen Boyd <sboyd@codeaurora.org>
Some root clock generators may have child branches that are controlled
by different CPUs. These RCGs require some special operations:
- some enable bits have to be toggled when we set the rate;
- if RCG is disabled we only cache the rate and set it later when enabled;
- when the RCG is disabled, the mux is set to the safe source;
Signed-off-by: Georgi Djakov <georgi.djakov@linaro.org>
[sboyd@codeaurora.org: Simplify recalc_rate implementation]
Signed-off-by: Stephen Boyd <sboyd@codeaurora.org>
Sometimes the display driver may want to change the parent PLL of
the display clocks (byte and pixel clocks) depending on the
use-case. Currently the parent is fixed by means of having a
frequency table with one entry that chooses a particular parent.
Remove this restriction and use the parent the clock is
configured for in the hardware during clk_set_rate(). This
requires consumers to rely on the default parent or to configure
the parent with clk_set_parent()/assigned-clock-parents on the
clocks before calling clk_set_rate().
Tested-by: Archit Taneja <architt@codeaurora.org>
Cc: Hai Li <hali@codeaurora.org>
Signed-off-by: Stephen Boyd <sboyd@codeaurora.org>
In the current parent mapping code, we can get duplicate or inconsistent
indexes, which leads to discrepancy between the number of elements in the
array and the number of parents. Until now, this was solved with some
reordering but this is not always possible.
This patch introduces index tables that are used to define the relations
between the PLL source and the hardware mux configuration value.
To accomplish this, here we do the following:
- Define a parent_map struct to map the relations between PLL source index
and register configuration value.
- Add a qcom_find_src_index() function for finding the index of a clock
matching the specific PLL configuration.
- Update the {set,get}_parent RCG functions use the newly introduced
parent_map struct.
- Convert all existing drivers to the new parent_map tables.
Signed-off-by: Georgi Djakov <georgi.djakov@linaro.org>
Signed-off-by: Stephen Boyd <sboyd@codeaurora.org>
The ahbix clock can never be turned off in practice. To change the
rates we need to switch the mux off the M/N counter to an always on
source (XO), reprogram the M/N counter to get the rate we want and
finally switch back to the M/N counter. Add a new ops structure
for this type of clock so that we can set the rate properly.
Fixes: c99e515a92 "clk: qcom: Add IPQ806X LPASS clock controller (LCC) driver"
Tested-by: Kenneth Westfield <kwestfie@codeaurora.org>
Signed-off-by: Stephen Boyd <sboyd@codeaurora.org>
The banked MD RCGs in global clock control have a different
register layout than the ones implemented in multimedia clock
control. Add support for these types of clocks so we can change
the rates of the UBI32 clocks.
Signed-off-by: Stephen Boyd <sboyd@codeaurora.org>
In the case of HDMI clocks, we want to bypass the RCG's ability
to divide the output clock and pass through the parent HDMI PLL
rate. Add a simple set of clk_ops to configure the RCG to do
this. This removes the need to keep adding more frequency entries
to the tv_src clock whenever we want to support a new rate.
Tested-by: Rob Clark <robdclark@gmail.com>
Signed-off-by: Stephen Boyd <sboyd@codeaurora.org>
Add support for the DSI/EDP/HDMI RCG clocks. With the proper
display driver in place this should allow us to support display
clocks on msm8974 based devices.
Signed-off-by: Stephen Boyd <sboyd@codeaurora.org>
Signed-off-by: Mike Turquette <mturquette@linaro.org>
Add support for the root clock generators on Qualcomm devices.
RCGs are highly customizable mux/divider/counter clocks that can
be used to generate almost any rate desired given some input
source that is faster than the desired rate.
Signed-off-by: Stephen Boyd <sboyd@codeaurora.org>
Signed-off-by: Mike Turquette <mturquette@linaro.org>
Bjorn Andersson