| Age | Commit message (Collapse) | Author |
|---|
|
This clock controller use the string comparison method to describe parent
relation between the clocks, which is not optimized.
Migrate to the new way by using .parent_hws where possible (ie. when
all clocks are local to the controller) and use .parent_data otherwise.
Signed-off-by: Alexandre Mergnat <amergnat@baylibre.com>
Signed-off-by: Jerome Brunet <jbrunet@baylibre.com>
|
|
This clock controller use the string comparison method to describe parent
relation between the clocks, which is not optimized.
Migrate to the new way by using .parent_hws where possible (ie. when
all clocks are local to the controller) and use .parent_data otherwise.
Signed-off-by: Alexandre Mergnat <amergnat@baylibre.com>
Reviewed-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
Tested-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
Signed-off-by: Jerome Brunet <jbrunet@baylibre.com>
|
|
Add the cts_i958 clock to control the clock source of the spdif output
block. It is used to select whether the clock source of the spdif output
is cts_amclk (when data are taken from i2s buffer) or the cts_mclk_i958
(when data are taken from the spdif buffer). The setup for this clock is
identical to GXBB, so this ports commit 7eaa44f6207fb6 ("clk: meson:
gxbb: add cts_i958 clock") to the Meson8/Meson8b/Meson8m2 clock driver.
Signed-off-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
Signed-off-by: Jerome Brunet <jbrunet@baylibre.com>
|
|
Add the SPDIF master clock also referred as cts_mclk_i958. The setup for
this clock is identical to GXBB, so this ports commit 3c277c247eabeb
("clk: meson: gxbb: add cts_mclk_i958") to the Meson8/Meson8b/Meson8m2
clock driver.
Signed-off-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
Signed-off-by: Jerome Brunet <jbrunet@baylibre.com>
|
|
Add the I2S master clock also referred as cts_amclk. The setup for this
clock is identical to GXBB, so this ports commit 4087bd4b21702d ("clk:
meson: gxbb: add cts_amclk") to the Meson8/Meson8b/Meson8m2 clock
driver.
Signed-off-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
Signed-off-by: Jerome Brunet <jbrunet@baylibre.com>
|
|
The variable which holds the parent names for the VPU clocks has a typo
in it. Fix this typo to make the variable naming in the driver
consistent. No functional changes.
Fixes: 41785ce562491d ("clk: meson: meson8b: add the VPU clock trees")
Signed-off-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
Signed-off-by: Jerome Brunet <jbrunet@baylibre.com>
|
|
This adds the four video decoder clock trees.
VDEC_1 is split into two paths on Meson8b and Meson8m2:
- input mux called "vdec_1_sel"
- two dividers ("vdec_1_1_div" and "vdec_1_2_div") and gates ("vdec_1_1"
and "vdec_1_2")
- and an output mux (probably glitch-free) called "vdec_1"
On Meson8 the VDEC_1 tree is simpler because there's only one path:
- input mux called "vdec_1_sel"
- divider ("vdec_1_1_div") and gate ("vdec_1_1")
- (the gate is used as output directly, there's no mux)
The VDEC_HCODEC and VDEC_2 clocks are simple composite clocks each
consisting of an input mux, divider and a gate.
The VDEC_HEVC clock seems to have two paths similar to the VDEC_1 clock.
However, the register offsets of the second clock path is not known.
Amlogic's 3.10 kernel (which is used as reference) sets
HHI_VDEC2_CLK_CNTL[31] to 1 before changing the VDEC_HEVC clock and back
to 0 afterwards. For now, leave a TODO comment and only add the first
path.
Signed-off-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
Reviewed-by: Neil Armstrong <narmstrong@baylibre.com>
Reviewed-by: Maxime Jourdan <mjourdan@baylibre.com>
Acked-by: Jerome Brunet <jbrunet@baylibre.com>
Signed-off-by: Neil Armstrong <narmstrong@baylibre.com>
Link: https://lkml.kernel.org/r/20190324151423.19063-3-martin.blumenstingl@googlemail.com
|
|
The VPU clock tree is slightly different on all three supported SoCs:
Meson8 only has an input mux (which chooses between "fclk_div4",
"fclk_div3", "fclk_div5" and "fclk_div7"), a divider and a gate.
Meson8b has two VPU clock trees, each with an input mux (using the same
parents as the input mux on Meson8), divider and a gates. The final VPU
clock is a glitch-free mux which chooses between VPU_1 and VPU_2.
Meson8m2 uses a similar clock tree as Meson8b but the last input clock
is different: instead of using "fclk_div7" as input Meson8m2 uses
"gp_pll". This was probably done in hardware to improve the accuracy of
the clock because fclk_div7 gives us 2550MHz / 7 = 364.286MHz while
GP_PLL can achieve 364.0MHz.
Signed-off-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
Reviewed-by: Neil Armstrong <narmstrong@baylibre.com>
Acked-by: Jerome Brunet <jbrunet@baylibre.com>
Signed-off-by: Neil Armstrong <narmstrong@baylibre.com>
Link: https://lkml.kernel.org/r/20190324151104.18397-5-martin.blumenstingl@googlemail.com
|
|
Meson8m2 has a GP_PLL clock (similar to GP0_PLL on GXBB/GXL/GXM) which
is used as input for the VPU clocks.
The only supported frequency (based on Amlogic's vendor kernel sources)
is 364MHz which is achieved using the following parameters:
- input: XTAL (24MHz)
- M = 182
- N = 3
- OD = 2 ^ 2
Signed-off-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
Reviewed-by: Neil Armstrong <narmstrong@baylibre.com>
Acked-by: Jerome Brunet <jbrunet@baylibre.com>
Signed-off-by: Neil Armstrong <narmstrong@baylibre.com>
Link: https://lkml.kernel.org/r/20190324151104.18397-4-martin.blumenstingl@googlemail.com
|
|
Meson8, Meson8b and Meson8m2 implement a similar clock controller.
However, there are a few differences between the three actual IP blocks.
One example where Meson8m2 differs from Meson8b is the VPU clock setup:
- the VPU input mux can choose between "fclk_div4", "fclk_div3",
"fclk_div5" and "fclk_div7" on Meson8b
- however, on Meson8m2 it can choose between "fclk_div4", "fclk_div3",
"fclk_div5" and "gp_pll" (GP_PLL only exists on Meson8m2, it's the
predecessor of the GP0_PLL clock on GXBB/GXL/GXM))
Add a separate clk_hw_onecell_data table for Meson8m2 so these
differences can be implemented in our clock controller driver. For now
meson8m2_hw_onecell_data is a clone of our existing
meson8b_hw_onecell_data.
Signed-off-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
Reviewed-by: Neil Armstrong <narmstrong@baylibre.com>
Acked-by: Jerome Brunet <jbrunet@baylibre.com>
Signed-off-by: Neil Armstrong <narmstrong@baylibre.com>
Link: https://lkml.kernel.org/r/20190324151104.18397-3-martin.blumenstingl@googlemail.com
|
|
Fix a typo in the APB clock names by renaming them from "abp" to "apb".
No functional changes.
Fixes: a7d19b05ce817d ("clk: meson: meson8b: add the CPU clock post divider clocks")
Signed-off-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
Signed-off-by: Neil Armstrong <narmstrong@baylibre.com>
Acked-by: Jerome Brunet <jbrunet@baylibre.com>
Link: https://lkml.kernel.org/r/20190210222603.6404-2-martin.blumenstingl@googlemail.com
|
|
Initially, the meson clock directory only hosted 2 controllers drivers,
for meson8 and gxbb. At the time, both used the same set of clock drivers
so managing the dependencies was not a big concern.
Since this ancient time, entropy did its job, controllers with different
requirement and specific clock drivers have been added. Unfortunately, we
did not do a great job at managing the dependencies between the
controllers and the different clock drivers. Some drivers, such as
clk-phase or vid-pll-div, are compiled even if they are useless on the
target (meson8). As we are adding new controllers, we need to be able to
pick a driver w/o pulling the whole thing.
The patch aims to clean things up by:
* providing a dedicated CONFIG_ for each clock drivers
* allowing clock drivers to be compiled as a modules, if possible
* stating explicitly which drivers are required by each controller.
Signed-off-by: Jerome Brunet <jbrunet@baylibre.com>
Signed-off-by: Neil Armstrong <narmstrong@baylibre.com>
Link: https://lkml.kernel.org/r/20190201125841.26785-5-jbrunet@baylibre.com
|
|
Add the GPU clock tree on Meson8, Meson8b and Meson8m2.
The GPU clock tree on Meson8b and Meson8m2 is almost identical to the
one one GXBB:
- there's a glitch-free mux at HHI_MALI_CLK_CNTL[31]
- there are two identical parents for this mux: mali_0 and mali_1, each
with a gate, divider and mux
- the parents of mali_0_sel and mali_1_sel are identical to GXBB except
there's no GP0_PLL on these 32-bit SoCs
Meson8 is different because it does not have the glitch-free mux.
Instead if only has the mali_0 clock tree. The parents of mali_0_sel are
identical to the ones on Meson8b and Meson8m2.
Signed-off-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
Reviewed-by: Neil Armstrong <narmstrong@baylibre.com>
Signed-off-by: Neil Armstrong <narmstrong@baylibre.com>
Link: https://lkml.kernel.org/r/20181208171247.22238-4-martin.blumenstingl@googlemail.com
|
|
The Meson8 SoC is slightly different compared to Meson8b and Meson8m2
because it does not have the glitch-free Mali GPU clock mux. For Meson8b
and Meson8m2 there are currently no known differences.
Add a separate clk_hw_onecell_data table for Meson8 so these differences
can be implemented. For now meson8_hw_onecell_data is a clone of our
existing meson8b_hw_onecell_data.
Signed-off-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
Reviewed-by: Neil Armstrong <narmstrong@baylibre.com>
Signed-off-by: Neil Armstrong <narmstrong@baylibre.com>
Link: https://lkml.kernel.org/r/20181208171247.22238-3-martin.blumenstingl@googlemail.com
|
|
Add all clocks to give us the final video clocks within the Meson8,
Meson8b and Meson8m2 SoCs. The final video clocks are:
- cts_enct
- cts_encl
- cts_encp
- cts_enci
- cts_vdac0
- hdmi_tx_pixel
- hdmi_sys
Add multiple clocks in between which are needed to implement these
clocks:
- Opposed to GXBB there is no pre-multiplier for the PLL input. The
assumption here is that the multiplier is required to achieve the HDMI
2.0 clock rates (which are up to twice the rate of the HDMI 1.4
rates).
- The main PLL is called "HDMI PLL" or "HPLL" in the datasheet. Rename
our existing "vid_pll_dco" to "hdmi_pll_dco". The actual VID_PLL clock
also exists further down the tree.
- Rename the existing "vid_pll" clock (which is the OD divider at
HHI_VID_PLL_CNTL[17:16]) to "hdmi_pll_lvds_out" to match the naming
from the datasheet.
- Add the second OD divider called "hdmi_pll_hdmi_out" at
HHI_VID_PLL_CNTL[19:18].
- Add the "vid_pll_in_sel" which can choose between "hdmi_pll_dco" and
another parent. However, the second parent is not use on Amlogic's
3.10 kernel for HDMI or CVBS output so just leave a TODO in the code.
- Add the "vid_pll_in_en" which is located after "vid_pll_in_sel"
according to the datasheet.
- Add "vid_pll_pre_div" which is used for divide-by-5 and divide-by-6 in
Amlogic's 3.10 kernel sources.
- Add "vid_pll_post_div" which divides the output of "vid_pll_pre_div"
further down. The Amlogic 3.10 kernel configures this as divide-by-2
with "vid_pll_pre_div" being configured as divide-by-5 to achieve a
total divider of 10.
- Add the real "vid_pll" clock which selects between "vid_pll_pre_div",
"vid_pll_post_div" and a third "vid_pll_pre_div_mult7_div2" (which is
"vid_pll_pre_div" divided by 3.5). The latter is not supported yet
because it's not used in Amlogic's 3.10 kernel. The "vid_pll" clock
rate can also be measured by clkmsr to check whether this
implementation is correct.
- Add "vid_pll_final_div" which is a post-divider for "vid_pll" and it's
used as input for "vclk" and "vclk2"
- Add the two symmetric "vclk" and "vclk" clock trees, each with a
divide-by-1, divide-by-2, divide-by-4, divide-by-6 and divide-by-12
clock and a divider for each clock.
- Add the "cts_enct", "cts_encp" and "hdmi_tx_pixel" clocks which each
have their own gate and can select between any of the five "vclk"
dividers.
- Add the "cts_encl" and "cts_vdac0" clocks which each have their own
gate and can select between any of the five "vclk2" dividers.
The "hdmi_sys" clock is a different than these video clocks. It takes
"xtal" as input (there are three more but unknown parents). Add this
clock as well as it's used by the HDMI controller. Amlogic's 3.10 kernel
always configures this as "xtal divided by 1", so we can ignore the
other parents for now.
This was tested on Meson8b and Meson8m2 boards by comparing the common
clock framework output with the clock measurer output. The following
video modes were first set in u-boot (by running "video dev open $mode")
before booting Linux:
4K2K30HZ (only supported by Meson8m2, not tested on Meson8b):
- vid_pll: 297000000Hz
- cts_encp: 297000000Hz
- hdmi_tx_pixel: 297000000Hz
1080P:
- vid_pll: 148500000Hz
- cts_encp: 148500000Hz
- hdmi_tx_pixel: 148500000Hz
720P:
- vid_pll: 148500000Hz
- cts_encp: 148500000Hz
- hdmi_tx_pixel: 74250000Hz
480P:
- vid_pll: 216000000Hz
- cts_encp: 54000000Hz
- hdmi_tx_pixel: 27000000Hz
Signed-off-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
Signed-off-by: Neil Armstrong <narmstrong@baylibre.com>
Link: https://lkml.kernel.org/r/20181202214220.7715-4-martin.blumenstingl@googlemail.com
|
|
This "vid_pll_dco" (which should be named HDMI_PLL or - as the datasheet
calls it - HPLL) has a 12-bit wide fractional parameter at
HHI_VID_PLL_CNTL2[11:0]. Add this so we correctly calculate the rate of
this PLL when u-boot is configured for a video mode which uses this
fractional parameter.
Signed-off-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
Signed-off-by: Neil Armstrong <narmstrong@baylibre.com>
Link: https://lkml.kernel.org/r/20181202214220.7715-3-martin.blumenstingl@googlemail.com
|
|
Unlike the other PLLs on Meson8b the N value "vid_pll_dco" (a better
name would be hdmi_pll_dco or - as the datasheet calls it - HPLL) is
located at HHI_VID_PLL_CNTL[14:10] instead of [13:9].
This results in an incorrect calculation of the rate of this PLL because
the value seen by the kernel is double the actual N (divider) value.
Update the offset of the N value to fix the calculation of the PLL rate.
Fixes: 28b9fcd016126e ("clk: meson8b: Add support for Meson8b clocks")
Reported-by: Jianxin Pan <jianxin.pan@amlogic.com>
Signed-off-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
Signed-off-by: Neil Armstrong <narmstrong@baylibre.com>
Link: https://lkml.kernel.org/r/20181202214220.7715-2-martin.blumenstingl@googlemail.com
|
|
There are four CPU clock post dividers:
- ABP
- PERIPH (used for the ARM global timer and ARM TWD timer)
- AXI
- L2 DRAM
Each of these clocks consists of two clocks:
- a mux to select between "cpu_clk" divided by 2, 3, 4, 5, 6, 7 or 8
- a "_clk_dis" gate. The public S805 datasheet states that this should
be set to 1 to disable the clock, the default value is 0. There is
also a hint that these are "just in case" bits which only exist in
case the corresponding mux implementation does not allow glitch-free
parent changes (the muxes are designed in a way that the clock can
stay enabled when changing the mux). It's still good practise to
describe this clock even if we're not supposed to modify it. Thus
this uses the read-only gate ops.
Signed-off-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
Acked-by: Jerome Brunet <jbrunet@baylibre.com>
Signed-off-by: Neil Armstrong <narmstrong@baylibre.com>
Link: https://lkml.kernel.org/r/20181122214017.25643-5-martin.blumenstingl@googlemail.com
|
|
The "cpu_div2" and "cpu_div3" take "cpu_in" as input and divide that by
2 or 3. The clock controller can also generate various CPU clock
post-dividers (2, 3, 4, 5, 6, 7, 8) which are derived from "cpu_clk".
When adding support for these post-dividers our clock naming could be
misleading as we have "cpu_div2" as well as "cpu_clk_div2".
Rename the existing "cpu_in" dividers so the name of the divider's
parent is part of the divider clock's name.
Signed-off-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
Acked-by: Jerome Brunet <jbrunet@baylibre.com>
Signed-off-by: Neil Armstrong <narmstrong@baylibre.com>
Link: https://lkml.kernel.org/r/20181122214017.25643-4-martin.blumenstingl@googlemail.com
|
|
Currently all clocks in the CPU clock tree are marked as read-only
(using the corresponding _ro_ clk_ops). This was correct since changing
the clock tree could cause the system to lock up.
Switch all clocks to their corresponding clk_ops variant which is not
read-only to allow changing the CPU clock tree since the bug which
locked up the system is now fixed (by switching the CPU clock temporary
to run off XTAL while changing the CPU clock tree).
Signed-off-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
Reviewed-by: Jerome Brunet <jbrunet@baylibre.com>
Signed-off-by: Neil Armstrong <narmstrong@baylibre.com>
Link: https://lkml.kernel.org/r/20181115224048.13511-7-martin.blumenstingl@googlemail.com
|
|
Changing the CPU clock requires changing various clocks including the
SYS PLL. The existing meson clk-pll and clk-regmap drivers can change
all of the relevant clocks already.
However, changing for exampe the SYS PLL is problematic because as long
as the CPU is running off a clock derived from SYS PLL changing the
latter results in a full system lockup.
Fix this system lockup by switching the CPU clock to run off the XTAL
while we are changing the any of the clocks in the CPU clock tree.
Signed-off-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
Reviewed-by: Jerome Brunet <jbrunet@baylibre.com>
Signed-off-by: Neil Armstrong <narmstrong@baylibre.com>
Link: https://lkml.kernel.org/r/20181115224048.13511-6-martin.blumenstingl@googlemail.com
|
|
The sys_pll on the EC-100 board is configured to 1584MHz at boot
(either by u-boot, firmware or chip defaults). This is achieved by using
M = 66, N = 1 (24MHz * 66 / 1).
At boot the CPU clock is running off sys_pll divided by 2 which results
in 792MHz. Thus M = 66 is considered to be a "safe" value for Meson8b.
To achieve 1608MHz (one of the CPU OPPs on Meson8 and Meson8m2) we need
M = 67, N = 1. I ran "stress --cpu 4" while infinitely cycling through
all available frequencies on my Meson8m2 board and could not spot any
issues with this setting (after ~12 hours of running this).
On Meson8, Meson8b and Meson8m2 we also want to be able to use 408MHz
and 816MHz CPU frequencies. These can be achieved by dividing sys_pll by
4 (for 408MHz) or 2 (for 816MHz). That means that sys_pll has to run at
1632MHz which can be generated using M = 68, N = 1.
Similarily we also want to be able to use 1008MHz as CPU frequency. This
means that sys_pll has to run either at 1008MHz or 2016MHz. The former
would result in an M value of 42, which is lower than the smallest value
used by the 3.10 GPL kernel sources from Amlogic (50 is the lower limit
there). Thus we need to run sys_pll at 2016MHz which can ge generated
using M = 84, N = 1.
I tested M = 68 and M = 84 on my Meson8b Odroid-C1 and my Meson8m2 board
by running "stress --cpu 4" while infinitely cycling thorugh all
available frequencies. I could not spot any issues after ~12 hours of
running this.
Amlogic's 3.10 GPL kernel sources have more M/N combinations. I did not
add them yet because M = 74 (to achieve close to 1800MHz on Meson8) and
M = 82 (to achieve close to 1992MHz on Meson8 as well) caused my
Meson8m2 board to hang randomly. It's not clear why this is (for example
because the board's voltage regulator design is bad, some missing bits
for these values in our clk-pll driver, etc.). Thus the following M
values from the Amlogic 3.10 GPL kernel sources are skipped as of now:
69, 70, 71, 72, 73, 74, 76, 78, 80, 82, 84, 86, 88, 90, 92, 94, 96, 98
Signed-off-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
Acked-by: Jerome Brunet <jbrunet@baylibre.com>
Signed-off-by: Neil Armstrong <narmstrong@baylibre.com>
Link: https://lkml.kernel.org/r/20181115224048.13511-5-martin.blumenstingl@googlemail.com
|
|
We don't want the common clock framework to disable the "cpu_clk" if
it's not used by any device. The cpufreq-dt driver does not enable the
CPU clocks. However, even if it would we would still want the CPU clock
to be enabled at all times because the CPU clock is also required even
if we disable CPU frequency scaling on a specific board.
The reason why we want the CPU clock to be enabled is a clock further up
in the tree:
Since commit 6f888e7bc7bd58 ("clk: meson: clk-pll: add enable bit") the
sys_pll can be disabled. However, since the CPU clock is derived from
sys_pll we don't want sys_pll to get disabled. The common clock
framework takes care of that for us by enabling all parent clocks of our
CPU clock when we mark the CPU clock with CLK_IS_CRITICAL.
Until now this is not a problem yet because all clocks in the CPU
clock's tree (including sys_pll) are read-only. However, once we allow
modifications to the clocks in that tree we will need this.
Signed-off-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
Acked-by: Jerome Brunet <jbrunet@baylibre.com>
Signed-off-by: Neil Armstrong <narmstrong@baylibre.com>
Link: https://lkml.kernel.org/r/20181115224048.13511-4-martin.blumenstingl@googlemail.com
|
|
The cpu_div3 clock (cpu_in divided by 3) generates a signal with a duty
cycle of 33%. The CPU clock however requires a clock signal with a duty
cycle of 50% to run stable.
cpu_div3 was observed to be problematic when cycling through all
available CPU frequencies (with additional patches on top of this one)
while running "stress --cpu 4" in the background. This caused sporadic
hangs where the whole system would fully lock up.
Amlogic's 3.10 kernel code also does not use the cpu_div3 clock either
when changing the CPU clock.
Signed-off-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
Reviewed-by: Jerome Brunet <jbrunet@baylibre.com>
Signed-off-by: Neil Armstrong <narmstrong@baylibre.com>
Link: https://lkml.kernel.org/r/20181115224048.13511-3-martin.blumenstingl@googlemail.com
|
|
According to the public S805 datasheet HHI_SYS_CPU_CLK_CNTL1[29:20] is
the register for the CPU scale_div clock. This matches the code in
Amlogic's 3.10 GPL kernel sources:
N = (aml_read_reg32(P_HHI_SYS_CPU_CLK_CNTL1) >> 20) & 0x3FF;
This means that the divider register is 10 bit wide instead of 9 bits.
So far this is not a problem since all u-boot versions I have seen are
not using the cpu_scale_div clock at all (instead they are configuring
the CPU clock to run off cpu_in_sel directly).
The fixes tag points to the latest rework of the CPU clocks. However,
even before the rework it was wrong. Commit 7a29a869434e8b ("clk: meson:
Add support for Meson clock controller") defines MESON_N_WIDTH as 9 (in
drivers/clk/meson/clk-cpu.c). But since the old clk-cpu implementation
this only carries the fixes tag for the CPU clock rewordk.
Fixes: 251b6fd38bcb9c ("clk: meson: rework meson8b cpu clock")
Signed-off-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
Signed-off-by: Neil Armstrong <narmstrong@baylibre.com>
Link: https://lkml.kernel.org/r/20180927085921.24627-3-martin.blumenstingl@googlemail.com
|
|
The public S805 datasheet only mentions that
HHI_SYS_CPU_CLK_CNTL1[20:29] contains a divider called "cpu_scale_div".
Unfortunately it does not mention how to use the register contents.
The Amlogic 3.10 GPL kernel sources are using the following code to
calculate the CPU clock based on that register (taken from
arch/arm/mach-meson8/clock.c in the 3.10 Amlogic kernel, shortened to
make it easier to read):
N = (aml_read_reg32(P_HHI_SYS_CPU_CLK_CNTL1) >> 20) & 0x3FF;
if (sel == 3) /* use cpu_scale_div */
div = 2 * N;
else
div = ... /* not relevant for this example */
cpu_clk = parent_clk / div;
This suggests that the formula is: parent_rate / 2 * register_value
However, running perf (which can measure the CPU clock rate thanks to
the ARM PMU) shows that this formula is not correct.
This can be reproduced with the following steps:
1. boot into u-boot
2. let the CPU clock run off the XTAL clock:
mw.l 0xC110419C 0x30 1
3. set the cpu_scale_div register:
to value 0x1: mw.l 0xC110415C 0x801016A2 1
to value 0x2: mw.l 0xC110415C 0x802016A2 1
to value 0x5: mw.l 0xC110415C 0x805016A2 1
4. let the CPU clock run off cpu_scale_div:
mw.l 0xC110419C 0xbd 1
5. boot Linux
6. run: perf stat -aB stress --cpu 4 --timeout 10
7. check the "cycles" value
I get the following results depending on the cpu_scale_div value:
- (cpu_in_sel - this is the input clock for cpu_scale_div - runs at
1.2GHz)
- 0x1 = 300MHz
- 0x2 = 200MHz
- 0x5 = 100MHz
This means that the actual formula to calculate the output of the
cpu_scale_div clock is: parent_rate / 2 * (register value + 1).
The register value 0x0 is reserved. When letting the CPU clock run off
the cpu_scale_div while the value is 0x0 the whole board hangs (even in
u-boot).
I also verified this with the TWD timer: when adding this to the .dts
without specifying it's clock it will auto-detect the PERIPH (which is
the input clock of the TWD) clock rate (and the result is shown in the
kernel log). On Meson8, Meson8b and Meson8m2 the PERIPH clock is CPUCLK
divided by 4. This also matched for all three test-cases from above (in
all cases the TWD timer clock rate was approx. one fourth of the CPU
clock rate).
A small note regarding the "fixes" tag: the original issue seems to
exist virtually since forever. Even commit 28b9fcd016126e ("clk:
meson8b: Add support for Meson8b clocks") seems to handle this wrong. I
still decided to use commit 251b6fd38bcb9c ("clk: meson: rework meson8b
cpu clock") because this is the first commit which gets the CPU hiearchy
correct and thus it's the first commit where the cpu_scale_div register
is used correctly (apart from the bug in the cpu_scale_table).
Fixes: 251b6fd38bcb9c ("clk: meson: rework meson8b cpu clock")
Signed-off-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
Signed-off-by: Neil Armstrong <narmstrong@baylibre.com>
Link: https://lkml.kernel.org/r/20180927085921.24627-2-martin.blumenstingl@googlemail.com
|
|
The clock controller is located in a register range (called "HHI") which
contains more than just registers for the clock controller. Known
consumers of the HHI register range are:
- the clock controller
- a reset controller
- temperature sensor calibration coefficient (TSC) (only on Meson8b and
Meson8m2)
- HDMI controller
The main reason for using a syscon is the "temperature sensor
calibration coefficient" which has to be set for the built-in temperature
sensor to work correctly. Four TSC bits are located in the SAR ADC's
register space. However on Meson8b and Meson8m2 there is a fifth TSC bit
which is unfortunately located in the HHI register space. To be more
precise, bit 9 of the HHI_DPLL_TOP_0 register (which sits right between
the HHI_SYS_PLL and HHI_VID_PLL registers).
Get the regmap from the parent (HHI syscon) node to support all
functionality of the HHI register range. Backwards compatibility with
old .dtbs is ensured by falling back to parsing the registers just like
before this change.
Signed-off-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
Acked-by: Neil Armstrong <narmstrong@baylibre.com>
Signed-off-by: Neil Armstrong <narmstrong@baylibre.com>
Link: https://lkml.kernel.org/r/20181028120859.5735-3-martin.blumenstingl@googlemail.com
|
|
For now the reset controller was using raw register access because the
early init did not initialize the regmap. However, now that clocks are
initialized early we can simply use the regmap also for the reset
controller.
No functional changes.
Signed-off-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
Signed-off-by: Jerome Brunet <jbrunet@baylibre.com>
|
|
Until now only the reset controller (part of the clock controller
register space) was registered early in the boot process, while the
clock controller itself was registered later on.
However, some parts of the SoC are initialized early in the boot process,
such as the SRAM and the TWD timer. The bootloader already enables these
clocks so we didn't see any issues so far.
Register the clock controller early so other drivers (such as the SRAM
and TWD timer) can use the clocks early in the boot process.
Signed-off-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
Signed-off-by: Jerome Brunet <jbrunet@baylibre.com>
|
|
Putting hard-coded rates inside the parameter tables assumes that
the parent is known and will never change. That's a big assumption
we should not make.
We have everything we need to recalculate the output rate using
the parent rate and the rest of the parameters. Let's do so and
drop the rates from the tables.
Acked-by: Neil Armstrong <narmstrong@baylibre.com>
Reviewed-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
Tested-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
Signed-off-by: Jerome Brunet <jbrunet@baylibre.com>
|
|
Remove od parameters from pll clocks and add post dividers clocks
instead. Some clock, especially the one which feature several ods,
may provide output between those ods. Also, some drivers, such
as the hdmi driver, may require a more detailed control of the
clock dividers, compared to what CCF would perform automatically.
One added benefit of removing ods is that it also greatly reduce the
size of the rate parameter tables.
In the future, we could possibly take the predivider 'n' out of this
driver as well. To do so, we will need to understand the constraints
for the PLL to lock and whether or not it depends on the input clock
rate.
Acked-by: Neil Armstrong <narmstrong@baylibre.com>
Reviewed-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
Tested-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
Signed-off-by: Jerome Brunet <jbrunet@baylibre.com>
|
|
CLK_GET_RATE_NOCACHE should only be necessary when the registers
controlling the rate of clock may change outside of CCF. On Amlogic,
it should only be the case for the hdmi pll which is directly controlled
by the display driver (WIP to fix this).
The other plls should not require this flag.
Reviewed-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
Tested-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
Signed-off-by: Jerome Brunet <jbrunet@baylibre.com>
|
|
Add the enable the bit of the pll clocks.
These pll clocks may be disabled but we can't model this as an external
gate since the pll needs to lock when enabled.
Adding this bit allows to drop the poke of the first register of PLL.
This will be useful to model the different components of the pll using
generic clocks elements
Acked-by: Neil Armstrong <narmstrong@baylibre.com>
Reviewed-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
Tested-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
Signed-off-by: Jerome Brunet <jbrunet@baylibre.com>
|
|
git://git.kernel.org/pub/scm/linux/kernel/git/clk/linux
Pull clk updates from Stephen Boyd:
"This time we have a good set of changes to the core framework that do
some general cleanups, but nothing too major. The majority of the diff
goes to two SoCs, Actions Semi and Qualcomm. A brand new driver is
introduced for Actions Semi so it takes up some lines to add all the
different types, and the Qualcomm diff is there because we add support
for two SoCs and it's quite a bit of data.
Otherwise the big driver updates are on TI Davinci and Amlogic
platforms. And then the long tail of driver updates for various fixes
and stuff follows after that.
Core:
- debugfs cleanups removing error checking and an unused provider API
- Removal of a clk init typedef that isn't used
- Usage of match_string() to simplify parent string name matching
- OF clk helpers moved to their own file (linux/of_clk.h)
- Make clk warnings more readable across kernel versions
New Drivers:
- Qualcomm SDM845 GCC and Video clk controllers
- Qualcomm MSM8998 GCC
- Actions Semi S900 SoC support
- Nuvoton npcm750 microcontroller clks
- Amlogic axg AO clock controller
Removed Drivers:
- Deprecated Rockchip clk-gate driver
Updates:
- debugfs functions stopped checking return values
- Support for the MSIOF module clocks on Rensas R-Car M3-N
- Support for the new Rensas RZ/G1C and R-Car E3 SoCs
- Qualcomm GDSC, RCG, and PLL updates for clk changes in new SoCs
- Berlin and Amlogic SPDX tagging
- Usage of of_clk_get_parent_count() in more places
- Proper implementation of the CDEV1/2 clocks on Tegra20
- Allwinner H6 PRCM clock support and R40 EMAC support
- Add critical flag to meson8b's fdiv2 as temporary fixup for ethernet
- Round closest support for meson's mpll driver
- Support for meson8b nand clocks and gxbb video decoder clocks
- Mediatek mali clks
- STM32MP1 fixes
- Uniphier LD11/LD20 stream demux system clock"
* tag 'clk-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/clk/linux: (134 commits)
clk: qcom: Export clk_fabia_pll_configure()
clk: bcm: Update and add Stingray clock entries
dt-bindings: clk: Update Stingray binding doc
clk-si544: Properly round requested frequency to nearest match
clk: ingenic: jz4770: Add 150us delay after enabling VPU clock
clk: ingenic: jz4770: Enable power of AHB1 bus after ungating VPU clock
clk: ingenic: jz4770: Modify C1CLK clock to disable CPU clock stop on idle
clk: ingenic: jz4770: Change OTG from custom to standard gated clock
clk: ingenic: Support specifying "wait for clock stable" delay
clk: ingenic: Add support for clocks whose gate bit is inverted
clk: use match_string() helper
clk: bcm2835: use match_string() helper
clk: Return void from debug_init op
clk: remove clk_debugfs_add_file()
clk: tegra: no need to check return value of debugfs_create functions
clk: davinci: no need to check return value of debugfs_create functions
clk: bcm2835: no need to check return value of debugfs_create functions
clk: no need to check return value of debugfs_create functions
clk: imx6: add EPIT clock support
clk: mvebu: use correct bit for 98DX3236 NAND
...
|
|
Until commit 05f814402d6174 ("clk: meson: add fdiv clock gates") we
relied on the bootloader to enable the fclk_div clock gates. It turns
out that our clock tree is incomplete at least on Meson8b (tested with
an Odroid-C1, which uses an RGMII PHY) because after the mentioned
commit Ethernet is not working anymore (no RX/TX activity can be seen).
At the same time Ethernet was still working on Meson8m2 with a RMII PHY.
Testing has shown that as soon as "fclk_div2" is disabled Ethernet stops
working on Odroid-C1. Unfortunately it's currently not clear what the
Ethernet controller IP block uses the fclk_div2 clock for. Mark the
clock as CLK_IS_CRITICAL to keep it enabled (as it's already enabled by
most bootloaders by default, which is why we didn't notice it before).
Fixes: 05f814402d6174 ("clk: meson: add fdiv clock gates")
Cc: stable@vger.kernel.org
Signed-off-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
Tested-by: Kevin Hilman <khilman@baylibre.com>
Signed-off-by: Jerome Brunet <jbrunet@baylibre.com>
|
|
Replace every license notices in drivers/clk/meson by SPDX license
identifiers, as described in license-rules.rst
Acked-by: Neil Armstrong <narmstrong@baylibre.com>
Acked-by: Stephen Boyd <sboyd@kernel.org>
Signed-off-by: Jerome Brunet <jbrunet@baylibre.com>
|
|
This adds the NAND clocks (from the HHI_NAND_CLK_CNTL register) to the
Meson8b clock driver. There are three NAND clocks: a gate which enables
or disables the NAND clock, a mux and a divider (which divides the mux
output).
Unfortunately the public S805 datasheet does not document the mux
parents. However, the vendor kernel has a few hints for us which allows
us to make an educated guess about the clock parents. To do this we need
to have a look at set_nand_core_clk() from the vendor's NAND driver (see
[0]):
- XTAL = (4<<9) | (1<<8) | 0
- 160MHz = (0<<9) | (1<<8) | 3)
- 182MHz = (3<<9) | (1<<8) | 1)
- 212MHz = (1<<9) | (1<<8) | 3)
- 255MHz = (2<<9) | (1<<8) | 1)
While there is a comment for the XTAL parent (which indicates that it
should only be used for debugging) we have to do a bit of math for the
other parents: target_freq * divider = rate of parent clock
Bit 8 above is the enable bit, so we can ignore it here. Bits 11:9 are
the mux index and bits 6:0 are the 0-based divider (so we need to add
1). This gives us:
- mux 0 (160MHz * 4) = fclk_div4 (actual rate = 637.5MHz, off by 2.5MHz)
- mux 1 (212MHz * 4) = fclk_div3 (actual rate = 850MHz, off by 2MHz)
- mux 2 (255MHz * 2) = fclk_div5 (matches exactly 510MHz)
- mux 3 (182MHz * 2) = fclk_div7 (actual rate = 346.3MHz, off by 0.3MHz)
[0] https://github.com/khadas/linux/blob/9587681285cb/drivers/amlogic/amlnf/dev/amlnf_ctrl.c#L314
Signed-off-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
Signed-off-by: Jerome Brunet <jbrunet@baylibre.com>
|
|
meson8b_cpu_clk has two parent clocks:
- meson8b_xtal
- meson8b_cpu_scale_out_sel
The name of the "xtal" clock parent is specified correctly. However,
there is a typo in the name of the second parent clock. The
meson8b_cpu_scale_out_sel definition uses the name "cpu_scale_out_sel"
(which matches the name from the datasheet). However, the mux parent
definition uses the name "cpu_out_sel" which does not match any existing
clock.
Fixes: 251b6fd38bcb9c ("clk: meson: rework meson8b cpu clock")
Signed-off-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
Signed-off-by: Jerome Brunet <jbrunet@baylibre.com>
|
|
The names of all fclk divider gate clocks follow the naming schema
"fclk_divN" and the name of all fclk fixed dividers follow the naming
schema "fclk_divN_div".
There's one exception to this rule: meson8b_fclk_div3_div's name is
"fclk_div_div3". It's child clock meson8b_fclk_div3 however references
it as "fclk_div3_div" (following the naming schema explained above).
Fix the naming of the meson8b_fclk_div3_div clock to follow the naming
schema. This also fixes serial console on my Meson8m2 board because
"clk81" uses fclk_div3 as parent. However, since the hierarchy stops at
meson8b_fclk_div3 there's no known parent clock and the rate of "clk81"
and all of it's children (UART clock, SDIO MMC controller clock, ...)
are all 0.
Fixes: 05f814402d6174 ("clk: meson: add fdiv clock gates")
Signed-off-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
Signed-off-by: Jerome Brunet <jbrunet@baylibre.com>
|
|
Fixes the following warnings:
drivers/clk/meson/meson8b.c:512:19: warning: symbol 'meson8b_mpeg_clk_div' was not declared. Should it be static?
drivers/clk/meson/meson8b.c:526:19: warning: symbol 'meson8b_clk81' was not declared. Should it be static?
drivers/clk/meson/meson8b.c:540:19: warning: symbol 'meson8b_cpu_in_sel' was not declared. Should it be static?
drivers/clk/meson/meson8b.c:591:19: warning: symbol 'meson8b_cpu_scale_div' was not declared. Should it be static?
drivers/clk/meson/meson8b.c:608:19: warning: symbol 'meson8b_cpu_scale_out_sel' was not declared. Should it be static?
drivers/clk/meson/meson8b.c:626:19: warning: symbol 'meson8b_cpu_clk' was not declared. Should it be static?
drivers/clk/meson/gxbb.c:392:27: warning: symbol 'gxbb_gp0_init_regs' was not declared. Should it be static?
drivers/clk/meson/gxbb.c:439:27: warning: symbol 'gxl_gp0_init_regs' was not declared. Should it be static?
drivers/clk/meson/axg.c:195:27: warning: symbol 'axg_gp0_init_regs' was not declared. Should it be static?
drivers/clk/meson/axg.c:248:27: warning: symbol 'axg_hifi_init_regs' was not declared. Should it be static?
drivers/clk/meson/meson8b.c: In function 'meson8b_clkc_probe':
drivers/clk/meson/meson8b.c:1052:14: warning: unused variable 'clk' [-Wunused-variable]
Signed-off-by: Stephen Boyd <sboyd@kernel.org>
|
|
clk81 is a composite clock which parents all the peripheral clocks of the
platform. It is a critical clock which is used as provided by the
bootloader. We don't want to change its rate or reparent it, ever.
Remove the CLK_IGNORE_UNUSED on the mux and divider. These clock can't
gate so the flag is useless, and the gate is already critical, so the
clock won't ever be unused.
Remove CLK_SET_RATE_NO_REPARENT from mux, it is useless since the mux is
read-only.
Remove CLK_SET_RATE_PARENT from the gate and divider and use ro_ops for
the divider. A peripheral clock should not try to change the rate of
clk81. Stopping the rate propagation is good way to make sure such request
would be ignored.
Signed-off-by: Jerome Brunet <jbrunet@baylibre.com>
Signed-off-by: Neil Armstrong <narmstrong@baylibre.com>
|
|
Fdiv fixed dividers clocks of the fixed_pll can actually gate
independently. We never had an issue so far because these clocks
were provided 'enabled' by the bootloader.
Add these gates to enable/disable the clocks when required.
Signed-off-by: Jerome Brunet <jbrunet@baylibre.com>
Signed-off-by: Neil Armstrong <narmstrong@baylibre.com>
|
|
mpll clocks parent can actually be divided by 1 or 2. So far, this
divider has always been set to 1, so the calculation was correct.
Now that we know it exists, model the tree correctly. If we ever get
a platform where the divider is different, we won't get into trouble
Signed-off-by: Jerome Brunet <jbrunet@baylibre.com>
Signed-off-by: Neil Armstrong <narmstrong@baylibre.com>
|
|
Add the missing frac parameter to the meson8b fixed_pll. It seems to be
always on this platform, so the rate remains unchanged
Signed-off-by: Jerome Brunet <jbrunet@baylibre.com>
Signed-off-by: Neil Armstrong <narmstrong@baylibre.com>
|
|
Instead of migrating meson cpu_clk to clk_regmap, like the other meson
clock drivers, we take advantage of the massive rework to get rid of it
completely, and solve (the first part) of the related FIXME notice.
As pointed out in the code comments, the cpu_clk should be modeled with
dividers and muxes it is made of, instead of one big composite clock.
The cpu_clk was not working correctly to enable dvfs on meson8b. It hangs
quite often when changing the cpu clock rate. This new implementation,
based on simple elements improves the situation but the platform will
still hang from time to time. This is not acceptable so, until we can
make the mechanism around the cpu clock stable, the cpu clock subtree
has been put in read-only mode, preventing any change of the cpu clock
The notifier and read-write operation will be added back when we have a
solution to the problem.
Cc: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
Signed-off-by: Jerome Brunet <jbrunet@baylibre.com>
Signed-off-by: Neil Armstrong <narmstrong@baylibre.com>
|
|
The mpll clock is a kind of fractional divider which can gate.
When the RW operation have been added, enable/disable ops have been
mistakenly inserted in this driver. These ops are essentially a
poor copy/paste of the generic gate ops.
This change removes the gate ops from the mpll driver and inserts a
generic gate clock on each mpll divider, simplifying the mpll
driver and reducing code duplication.
Signed-off-by: Jerome Brunet <jbrunet@baylibre.com>
Signed-off-by: Neil Armstrong <narmstrong@baylibre.com>
|
|
Rework meson pll driver to use clk_regmap and move meson8b, gxbb and
axg's clock using meson_clk_pll to clk_regmap.
This rework is not just about clk_regmap, there a serious clean-up of
the driver code:
* Add lock and reset field: Previously inferred from the n field.
* Simplify the reset logic: Code seemed to apply reset differently but
in fact it was always the same -> assert reset, apply params,
de-assert reset. The 2 lock checking loops have been kept for now, as
they seem to be necessary.
* Do the sequence of init register pokes only at .init() instead of in
.set_rate(). Redoing the init on every set_rate() is not necessary
Signed-off-by: Jerome Brunet <jbrunet@baylibre.com>
Signed-off-by: Neil Armstrong <narmstrong@baylibre.com>
|
|
Rework meson mpll driver to use clk_regmap and move meson8b, gxbb
and axg clocks using meson_clk_mpll to clk_regmap
Signed-off-by: Jerome Brunet <jbrunet@baylibre.com>
Signed-off-by: Neil Armstrong <narmstrong@baylibre.com>
|
|
Move meson8b, gxbb and axg clocks using clk_mux to clk_regmap
Also remove a few useless tables in the process
Signed-off-by: Jerome Brunet <jbrunet@baylibre.com>
Signed-off-by: Neil Armstrong <narmstrong@baylibre.com>
|
|
Move meson8b, gxbb and axg clocks using clk_divider to clk_regmap
Signed-off-by: Jerome Brunet <jbrunet@baylibre.com>
Signed-off-by: Neil Armstrong <narmstrong@baylibre.com>
|
