| Message ID | 1431978219-14226-3-git-send-email-eric@anholt.net |
|---|---|
| State | New |
| Headers | show |
Hi Eric,
On Mon, May 18, 2015 at 12:43:34PM -0700, Eric Anholt wrote:
> +DEFINE_MUTEX(delayed_clock_init);
Static?
baruch
Stephen Warren <swarren@wwwdotorg.org> writes: > On 05/18/2015 01:43 PM, Eric Anholt wrote: >> + init.flags = CLK_IS_ROOT; > > Is it possible to add clock parent information to the driver, so the > clocks are all hooked together into the correct tree, rather than all > looking like root clocks? > > One of the many reasons I didn't do anything FW-wise for the kernel was > the hope that such information would be forthcoming, and hence we could > have complete kernel drivers. As far as I can tell, none of these clocks are the parent of any other. There's a layer of PLLs, then clockman almost always just dividing off of those. In response to a clockman request (which is what this property tag quickly maps to), only PLLH ever gets changed as a result of RPI_CLOCK_PIXEL. PLLH feeds the HDMI and VEC (SDTV). We don't have an SDTV clock for us to control through these interfaces, so there are no conflicts that I can see. Only two clockman clocks I see divide off of on other clockman clocks, and we don't have access to those.
Stephen Boyd <sboyd@codeaurora.org> writes: > On 05/28, Stephen Warren wrote: >> On 05/18/2015 01:43 PM, Eric Anholt wrote: >> > Unfortunately, the clock manager's registers are not accessible by the >> > ARM, so we have to request that the firmware modify our clocks for us. >> > >> > This driver only registers the clocks at the point they are requested >> > by a client driver. This is partially to support returning >> > -EPROBE_DEFER when the firmware driver isn't supported yet, but it >> > also avoids issues with disabling "unused" clocks due to them not yet >> > being connected to their consumers in the DT. >> >> It looks like you forgot to CC the clock subsystem maintainers: >> >> M: Mike Turquette <mturquette@linaro.org> >> M: Stephen Boyd <sboyd@codeaurora.org> >> > > Thanks, except I don't have the full patch context here to review > the patch. > >> The description above sounds like a neat solution, but has the >> disadvantage that the clocks without a client won't show up in debugfs. >> I wonder if the clock maintainers know of a better way? > > Can you mark them as CLK_IGNORE_UNUSED? The probe defer problem > has a solution in sight (see more below). >> > + init.flags = CLK_IS_ROOT; >> >> Is it possible to add clock parent information to the driver, so the >> clocks are all hooked together into the correct tree, rather than all >> looking like root clocks? >> >> One of the many reasons I didn't do anything FW-wise for the kernel was >> the hope that such information would be forthcoming, and hence we could >> have complete kernel drivers. >> >> > +void __init rpi_firmware_init_clock_provider(struct device_node *node) >> > +{ >> > + /* We delay construction of our struct clks until get time, >> > + * because we need to be able to return -EPROBE_DEFER if the >> > + * firmware driver isn't up yet. clk core doesn't support >> > + * re-probing on -EPROBE_DEFER, but callers of clk_get can. >> >> Oh, that's nasty. What would it take to fix the clock core to support >> deferred probe? It really should. > > There are patches to support probe defer from clk_get() but they > stalled because sunxi is needs to keep clocks on from their > providing driver (termed "critical clocks"). If we can resolve > the "critical clocks" thing then we should be able to support > probe defer, unless we find other users of orphaned clk > pointers. Great! I'm certainly happy to switch to a normal registration of all my clocks once -EPROBE_DEFER works from the clock provider init. If those patches aren't landing this release, that also gives us a release to wire up all the clock consumers in the DT before we get hit by stable DT ABI, so we'll be able to give a nice limited set of CLOCK_IGNORE_UNUSED in the flags when we transition.
Stephen Warren <swarren@wwwdotorg.org> writes: > On 05/18/2015 01:43 PM, Eric Anholt wrote: >> obj-$(CONFIG_ARCH_BCM2835) += clk-bcm2835.o >> +obj-$(CONFIG_ARCH_BCM2835) += clk-raspberrypi.o > > Shouldn't this replace the old legacy code in clk-bcm2835.c? I don't think we can, because of DT backwards compat (Sigh). > > I wonder if a separate Kconfig symbol is warranted for the clock driver. > Looking at the context in the patch, it's common not to do that though. I've moved it under RASPBERRYPI_FIRMWARE, since it needs that to build. >> diff --git a/drivers/clk/clk-raspberrypi.c b/drivers/clk/clk-raspberrypi.c > >> +struct rpi_firmware_clock { >> + /* Clock definitions in our static struct. */ >> + int clock_id; > > Why not just use the raw HW IDs (from the mailbox protocol) as the ID in > DT and the driver? The only thing that would need to change is to add a > error check for "clkspec->args[0] == 0" in > rpi_firmware_delayed_get_clk(). The advantage would be that > include/dt-bindings/clk/raspberrypi.h would exactly match the firmware > protocol, and hence be easily correlated with the documentation. Done. >> +static int rpi_clk_set_rate(struct clk_hw *hw, >> + unsigned long rate, unsigned long parent_rate) >> +{ >> + struct rpi_firmware_clock *rpi_clk = >> + container_of(hw, struct rpi_firmware_clock, hw); >> + u32 packet[2]; >> + int ret; >> + >> + packet[0] = rpi_clk->clock_id; >> + packet[1] = rate; >> + ret = rpi_firmware_property(rpi_clk->firmware_node, > > The docs indicate there's a third word here "skip setting turbo". Is > that optional? > > https://github.com/raspberrypi/firmware/wiki/Mailbox-property-interface Yeah, it's optional based on the buffer size specified in the packet. >> +static long rpi_clk_round_rate(struct clk_hw *hw, unsigned long rate, >> + unsigned long *parent_rate) >> +{ >> + /* >> + * The firmware will end up rounding our rate to something, >> + * but we don't have an interface for it. Just return the >> + * requested value, and it'll get updated after the clock gets >> + * set. >> + */ >> + return rate; >> +} > > Hmm. I wonder if that will confuse things; I thought the purpose of > round_rate() was so code could know exactly what would happen ahead of time? Well, we don't have the ability. Things seem to work, anyway. Certainly better than just living with fixed clocks for everything like we are today. >> +static struct clk *rpi_firmware_delayed_get_clk(struct >> of_phandle_args *clkspec, + void *_data) > >> + rpi_clk = &rpi_clocks[clkspec->args[0]]; >> + >> + firmware_node = of_parse_phandle(of_node, "firmware", 0); >> + if (!firmware_node) { >> + dev_err(dev, "%s: Missing firmware node\n", rpi_clk->name); >> + return ERR_PTR(-ENODEV); >> + } >> + >> + /* Try a no-op transaction to see if the driver is loaded yet. */ >> + ret = rpi_firmware_property_list(firmware_node, NULL, 0); >> + if (ret) >> + return ERR_PTR(ret); > > I would move all that into this driver's probe(). We can't move all this into the driver's probe, because this is where we're returning -EPROBE_DEFER. We could potentially do just the phandle parse up front and allocate some memory to pass it and our own device node to this function through the _data arg, but I don't see much point. >> +CLK_OF_DECLARE(rpi_firmware_clocks, "raspberrypi,firmware-clocks", >> + rpi_firmware_init_clock_provider); > > Oh, I guess the same comment as for the firmware node applies re: the > over-genericity of the DT compatible value applies here too. Perhaps > raspberrypi,bcm2835-firmware-clocks to match Lee's proposal for the > firmware node? Done.
On 05/29/2015 03:02 PM, Eric Anholt wrote: > Stephen Warren <swarren@wwwdotorg.org> writes: > >> On 05/18/2015 01:43 PM, Eric Anholt wrote: >>> +static struct clk *rpi_firmware_delayed_get_clk(struct >>> of_phandle_args *clkspec, + void *_data) >> >>> + rpi_clk = &rpi_clocks[clkspec->args[0]]; + + firmware_node = >>> of_parse_phandle(of_node, "firmware", 0); + if (!firmware_node) >>> { + dev_err(dev, "%s: Missing firmware node\n", >>> rpi_clk->name); + return ERR_PTR(-ENODEV); + } + + /* Try a >>> no-op transaction to see if the driver is loaded yet. */ + ret >>> = rpi_firmware_property_list(firmware_node, NULL, 0); + if >>> (ret) + return ERR_PTR(ret); >> >> I would move all that into this driver's probe(). > > We can't move all this into the driver's probe, because this is > where we're returning -EPROBE_DEFER. We could potentially do just > the phandle parse up front and allocate some memory to pass it and > our own device node to this function through the _data arg, but I > don't see much point. Well, once the clock core correctly supports deferred probe, that can be moved. Aside from that, I think all your other replies to my replies in this thread/series make sense.
diff --git a/drivers/clk/Makefile b/drivers/clk/Makefile index 3d00c25..6a5dafa 100644 --- a/drivers/clk/Makefile +++ b/drivers/clk/Makefile @@ -20,6 +20,7 @@ obj-$(CONFIG_MACH_ASM9260) += clk-asm9260.o obj-$(CONFIG_COMMON_CLK_AXI_CLKGEN) += clk-axi-clkgen.o obj-$(CONFIG_ARCH_AXXIA) += clk-axm5516.o obj-$(CONFIG_ARCH_BCM2835) += clk-bcm2835.o +obj-$(CONFIG_ARCH_BCM2835) += clk-raspberrypi.o obj-$(CONFIG_COMMON_CLK_CDCE706) += clk-cdce706.o obj-$(CONFIG_ARCH_CLPS711X) += clk-clps711x.o obj-$(CONFIG_ARCH_EFM32) += clk-efm32gg.o diff --git a/drivers/clk/clk-raspberrypi.c b/drivers/clk/clk-raspberrypi.c new file mode 100644 index 0000000..5745875 --- /dev/null +++ b/drivers/clk/clk-raspberrypi.c @@ -0,0 +1,241 @@ +/* + * Copyright © 2015 Broadcom + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + */ + +/* + * Implements a clock provider for the clocks controlled by the + * firmware on Raspberry Pi. + * + * These clocks are controlled by the CLOCKMAN peripheral in the + * hardware, but the ARM doesn't have access to the registers for + * them. As a result, we have to call into the firmware to get it to + * enable, disable, and set their frequencies. + * + * We don't have an interface for getting the set of frequencies + * available from the hardware. We can request a min/max, but other + * than that we have to request a frequency and take what it gives us. + */ + +#include <dt-bindings/clk/raspberrypi.h> +#include <linux/clk-provider.h> +#include <soc/bcm2835/raspberrypi-firmware-property.h> + +struct rpi_firmware_clock { + /* Clock definitions in our static struct. */ + int clock_id; + const char *name; + + /* The rest are filled in at init time. */ + struct clk_hw hw; + struct device *dev; + struct device_node *firmware_node; +}; + +static struct rpi_firmware_clock rpi_clocks[] = { + [RPI_CLOCK_EMMC] = { 1, "emmc" }, + [RPI_CLOCK_UART0] = { 2, "uart0" }, + [RPI_CLOCK_ARM] = { 3, "arm" }, + [RPI_CLOCK_CORE] = { 4, "core" }, + [RPI_CLOCK_V3D] = { 5, "v3d" }, + [RPI_CLOCK_H264] = { 6, "h264" }, + [RPI_CLOCK_ISP] = { 7, "isp" }, + [RPI_CLOCK_SDRAM] = { 8, "sdram" }, + [RPI_CLOCK_PIXEL] = { 9, "pixel" }, + [RPI_CLOCK_PWM] = { 10, "pwm" }, +}; + +static int rpi_clk_is_on(struct clk_hw *hw) +{ + struct rpi_firmware_clock *rpi_clk = + container_of(hw, struct rpi_firmware_clock, hw); + u32 packet[2]; + int ret; + + packet[0] = rpi_clk->clock_id; + packet[1] = 0; + ret = rpi_firmware_property(rpi_clk->firmware_node, + RPI_FIRMWARE_GET_CLOCK_STATE, + &packet, sizeof(packet)); + if (ret) { + dev_err(rpi_clk->dev, "Failed to get clock state\n"); + return 0; + } + + return packet[1] != 0; +} + +static int rpi_clk_set_state(struct rpi_firmware_clock *rpi_clk, bool on) +{ + u32 packet[2]; + int ret; + + packet[0] = rpi_clk->clock_id; + packet[1] = on; + ret = rpi_firmware_property(rpi_clk->firmware_node, + RPI_FIRMWARE_SET_CLOCK_STATE, + &packet, sizeof(packet)); + if (ret || (packet[1] & (1 << 1))) { + dev_err(rpi_clk->dev, "Failed to set clock state\n"); + return -EINVAL; + } + + return 0; +} + +static int rpi_clk_on(struct clk_hw *hw) +{ + struct rpi_firmware_clock *rpi_clk = + container_of(hw, struct rpi_firmware_clock, hw); + + return rpi_clk_set_state(rpi_clk, true); +} + +static void rpi_clk_off(struct clk_hw *hw) +{ + struct rpi_firmware_clock *rpi_clk = + container_of(hw, struct rpi_firmware_clock, hw); + + rpi_clk_set_state(rpi_clk, false); +} + +static unsigned long rpi_clk_get_rate(struct clk_hw *hw, + unsigned long parent_rate) +{ + struct rpi_firmware_clock *rpi_clk = + container_of(hw, struct rpi_firmware_clock, hw); + u32 packet[2]; + int ret; + + packet[0] = rpi_clk->clock_id; + packet[1] = 0; + ret = rpi_firmware_property(rpi_clk->firmware_node, + RPI_FIRMWARE_GET_CLOCK_RATE, + &packet, sizeof(packet)); + if (ret) { + dev_err(rpi_clk->dev, "Failed to get clock rate\n"); + return 0; + } + + return packet[1]; +} + +static int rpi_clk_set_rate(struct clk_hw *hw, + unsigned long rate, unsigned long parent_rate) +{ + struct rpi_firmware_clock *rpi_clk = + container_of(hw, struct rpi_firmware_clock, hw); + u32 packet[2]; + int ret; + + packet[0] = rpi_clk->clock_id; + packet[1] = rate; + ret = rpi_firmware_property(rpi_clk->firmware_node, + RPI_FIRMWARE_SET_CLOCK_RATE, + &packet, sizeof(packet)); + if (ret) { + dev_err(rpi_clk->dev, "Failed to set clock rate\n"); + return ret; + } + + return 0; +} + +static long rpi_clk_round_rate(struct clk_hw *hw, unsigned long rate, + unsigned long *parent_rate) +{ + /* + * The firmware will end up rounding our rate to something, + * but we don't have an interface for it. Just return the + * requested value, and it'll get updated after the clock gets + * set. + */ + return rate; +} + +static struct clk_ops rpi_clk_ops = { + .is_prepared = rpi_clk_is_on, + .prepare = rpi_clk_on, + .unprepare = rpi_clk_off, + .recalc_rate = rpi_clk_get_rate, + .set_rate = rpi_clk_set_rate, + .round_rate = rpi_clk_round_rate, +}; + +DEFINE_MUTEX(delayed_clock_init); +static struct clk *rpi_firmware_delayed_get_clk(struct of_phandle_args *clkspec, + void *_data) +{ + struct device_node *of_node = _data; + struct platform_device *pdev = of_find_device_by_node(of_node); + struct device *dev = &pdev->dev; + struct device_node *firmware_node; + struct clk_init_data init; + struct rpi_firmware_clock *rpi_clk; + struct clk *ret_clk; + int ret; + + if (clkspec->args_count != 1) { + dev_err(dev, "clock phandle should have 1 argument\n"); + return ERR_PTR(-ENODEV); + } + + if (clkspec->args[0] >= ARRAY_SIZE(rpi_clocks)) { + dev_err(dev, "clock phandle index %d too large\n", + clkspec->args[0]); + return ERR_PTR(-ENODEV); + } + + rpi_clk = &rpi_clocks[clkspec->args[0]]; + + firmware_node = of_parse_phandle(of_node, "firmware", 0); + if (!firmware_node) { + dev_err(dev, "%s: Missing firmware node\n", rpi_clk->name); + return ERR_PTR(-ENODEV); + } + + /* Try a no-op transaction to see if the driver is loaded yet. */ + ret = rpi_firmware_property_list(firmware_node, NULL, 0); + if (ret) + return ERR_PTR(ret); + + mutex_lock(&delayed_clock_init); + if (rpi_clk->hw.clk) { + mutex_unlock(&delayed_clock_init); + return rpi_clk->hw.clk; + } + memset(&init, 0, sizeof(init)); + init.ops = &rpi_clk_ops; + + rpi_clk->firmware_node = firmware_node; + rpi_clk->dev = dev; + rpi_clk->hw.init = &init; + init.name = rpi_clk->name; + init.flags = CLK_IS_ROOT; + + ret_clk = clk_register(dev, &rpi_clk->hw); + mutex_unlock(&delayed_clock_init); + if (!IS_ERR(ret_clk)) + dev_info(dev, "registered %s clock\n", rpi_clk->name); + else { + dev_err(dev, "%s clock failed to init: %ld\n", rpi_clk->name, + PTR_ERR(ret_clk)); + } + return ret_clk; +} + +void __init rpi_firmware_init_clock_provider(struct device_node *node) +{ + /* We delay construction of our struct clks until get time, + * because we need to be able to return -EPROBE_DEFER if the + * firmware driver isn't up yet. clk core doesn't support + * re-probing on -EPROBE_DEFER, but callers of clk_get can. + */ + of_clk_add_provider(node, rpi_firmware_delayed_get_clk, node); +} + +CLK_OF_DECLARE(rpi_firmware_clocks, "raspberrypi,firmware-clocks", + rpi_firmware_init_clock_provider); diff --git a/include/dt-bindings/clk/raspberrypi.h b/include/dt-bindings/clk/raspberrypi.h new file mode 100644 index 0000000..c9fa85c --- /dev/null +++ b/include/dt-bindings/clk/raspberrypi.h @@ -0,0 +1,23 @@ +#/* + * Copyright © 2015 Broadcom + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + */ + +#ifndef _DT_BINDINGS_CLK_RASPBERRYPI_H +#define _DT_BINDINGS_CLK_RASPBERRYPI_H + +#define RPI_CLOCK_EMMC 0 +#define RPI_CLOCK_UART0 1 +#define RPI_CLOCK_ARM 2 +#define RPI_CLOCK_CORE 3 +#define RPI_CLOCK_V3D 4 +#define RPI_CLOCK_H264 5 +#define RPI_CLOCK_ISP 6 +#define RPI_CLOCK_SDRAM 7 +#define RPI_CLOCK_PIXEL 8 +#define RPI_CLOCK_PWM 9 + +#endif
Unfortunately, the clock manager's registers are not accessible by the ARM, so we have to request that the firmware modify our clocks for us. This driver only registers the clocks at the point they are requested by a client driver. This is partially to support returning -EPROBE_DEFER when the firmware driver isn't supported yet, but it also avoids issues with disabling "unused" clocks due to them not yet being connected to their consumers in the DT. Signed-off-by: Eric Anholt <eric@anholt.net> --- drivers/clk/Makefile | 1 + drivers/clk/clk-raspberrypi.c | 241 ++++++++++++++++++++++++++++++++++ include/dt-bindings/clk/raspberrypi.h | 23 ++++ 3 files changed, 265 insertions(+) create mode 100644 drivers/clk/clk-raspberrypi.c create mode 100644 include/dt-bindings/clk/raspberrypi.h