| Message ID | 20190520104708.11980-4-nsaenzjulienne@suse.de (mailing list archive) |
|---|---|
| State | RFC |
| Headers | show |
| Series | cpufreq support for the Raspberry Pi | expand |
Hi Nicolas, the following comments applies only in case Eric is fine with the whole approach. On 20.05.19 12:47, Nicolas Saenz Julienne wrote: > Raspberry Pi's firmware, which runs in a dedicated processor, keeps maybe we should clarify that the firmware is running in the VPU > track of the board's temperature and voltage. It's resposible for > scaling the CPU frequency whenever it deems the device reached an unsafe > state. On top of that the firmware provides an interface which allows > Linux to to query the clock's state or change it's frequency. I think this requires a separate update of the devicetree binding. > > Being the sole user of the bcm2835 clock driver, this integrates the > firmware interface into the clock driver and adds a first user: the CPU > pll, also known as 'pllb'. Please verify that the kernel still works (and this clock driver probe) under the following conditions: - CONFIG_RASPBERRYPI_FIRMWARE=n - CONFIG_RASPBERRYPI_FIRMWARE=m - older DTBs without patch #1 > > Signed-off-by: Nicolas Saenz Julienne <nsaenzjulienne@suse.de> > --- > drivers/clk/bcm/clk-bcm2835.c | 274 ++++++++++++++++++++++++++++++++-- > 1 file changed, 259 insertions(+), 15 deletions(-) > > diff --git a/drivers/clk/bcm/clk-bcm2835.c b/drivers/clk/bcm/clk-bcm2835.c > index 5aea110672f3..ce5b16f3704e 100644 > --- a/drivers/clk/bcm/clk-bcm2835.c > +++ b/drivers/clk/bcm/clk-bcm2835.c > @@ -35,6 +35,7 @@ > #include <linux/platform_device.h> > #include <linux/slab.h> > #include <dt-bindings/clock/bcm2835.h> > +#include <soc/bcm2835/raspberrypi-firmware.h> > > #define CM_PASSWORD 0x5a000000 > > @@ -289,6 +290,30 @@ > #define LOCK_TIMEOUT_NS 100000000 > #define BCM2835_MAX_FB_RATE 1750000000u > > +#define RPI_FIRMWARE_ARM_CLK_ID 0x000000003 > +#define RPI_FIRMWARE_STATE_ENABLE_BIT 0x1 > +#define RPI_FIRMWARE_STATE_WAIT_BIT 0x2 > + > +/* > + * Structure of the message passed to Raspberry Pi's firmware in order to > + * change clock rates. The 'disable_turbo' option is only available to the ARM > + * clock (pllb) which we enable by default as turbo mode will alter multiple > + * clocks at once. > + * > + * Even though we're able to access the clock registers directly we're bound to > + * use the firmware interface as the firmware ultimately takes care of > + * mitigating overheating/undervoltage situations and we would be changing > + * frequencies behind his back. > + * > + * For more information on the firmware interface check: > + * https://github.com/raspberrypi/firmware/wiki/Mailbox-property-interface > + */ > +struct bcm2835_firmware_prop { > + u32 id; > + u32 val; > + u32 disable_turbo; > +} __packed; > + > /* > * Names of clocks used within the driver that need to be replaced > * with an external parent's name. This array is in the order that > @@ -316,6 +341,8 @@ struct bcm2835_cprman { > */ > const char *real_parent_names[ARRAY_SIZE(cprman_parent_names)]; > > + struct rpi_firmware *firmware; > + > /* Must be last */ > struct clk_hw_onecell_data onecell; > }; > @@ -424,6 +451,23 @@ struct bcm2835_pll_data { > unsigned long max_fb_rate; > }; > > +struct bcm2835_fw_pll_data { > + const char *name; > + int firmware_clk_id; > + u32 flags; > + > + unsigned long min_rate; > + unsigned long max_rate; > + > + /* > + * The rate provided to the firmware interface might not always match > + * the clock subsystem's. For instance, in the case of the ARM cpu > + * clock, even though the firmware ultimately edits 'pllb' it takes the > + * expected 'pllb_arm' rate as it's input. > + */ > + unsigned int rate_rescale; > +}; > + > struct bcm2835_pll_ana_bits { > u32 mask0; > u32 set0; > @@ -505,6 +549,7 @@ struct bcm2835_pll { > struct clk_hw hw; > struct bcm2835_cprman *cprman; > const struct bcm2835_pll_data *data; > + const struct bcm2835_fw_pll_data *fw_data; > }; > > static int bcm2835_pll_is_on(struct clk_hw *hw) > @@ -517,6 +562,41 @@ static int bcm2835_pll_is_on(struct clk_hw *hw) > A2W_PLL_CTRL_PRST_DISABLE; > } > > +static int raspberrypi_clock_property(struct rpi_firmware *firmware, u32 tag, > + u32 clk, u32 *val) > +{ > + int ret; > + struct bcm2835_firmware_prop msg = { > + .id = clk, > + .val = *val, > + .disable_turbo = 1, > + }; > + > + ret = rpi_firmware_property(firmware, tag, &msg, sizeof(msg)); > + if (ret) > + return ret; > + > + *val = msg.val; > + > + return 0; > +} > + > +static int bcm2835_fw_pll_is_on(struct clk_hw *hw) > +{ > + struct bcm2835_pll *pll = container_of(hw, struct bcm2835_pll, hw); > + struct bcm2835_cprman *cprman = pll->cprman; > + u32 val = 0; > + int ret; > + > + ret = raspberrypi_clock_property(cprman->firmware, > + RPI_FIRMWARE_GET_CLOCK_STATE, > + pll->fw_data->firmware_clk_id, &val); > + if (ret) > + return 0; > + > + return !!(val & RPI_FIRMWARE_STATE_ENABLE_BIT); > +} > + > static void bcm2835_pll_choose_ndiv_and_fdiv(unsigned long rate, > unsigned long parent_rate, > u32 *ndiv, u32 *fdiv) > @@ -547,16 +627,37 @@ static long bcm2835_pll_round_rate(struct clk_hw *hw, unsigned long rate, > unsigned long *parent_rate) > { > struct bcm2835_pll *pll = container_of(hw, struct bcm2835_pll, hw); > - const struct bcm2835_pll_data *data = pll->data; > u32 ndiv, fdiv; > > - rate = clamp(rate, data->min_rate, data->max_rate); > + if (pll->data) > + rate = clamp(rate, pll->data->min_rate, pll->data->max_rate); > + else > + rate = clamp(rate, pll->fw_data->min_rate, > + pll->fw_data->max_rate); > > bcm2835_pll_choose_ndiv_and_fdiv(rate, *parent_rate, &ndiv, &fdiv); > > return bcm2835_pll_rate_from_divisors(*parent_rate, ndiv, fdiv, 1); > } > > +static unsigned long bcm2835_fw_pll_get_rate(struct clk_hw *hw, > + unsigned long parent_rate) > +{ > + struct bcm2835_pll *pll = container_of(hw, struct bcm2835_pll, hw); > + struct bcm2835_cprman *cprman = pll->cprman; > + u32 val = 0; > + int ret; > + > + ret = raspberrypi_clock_property(cprman->firmware, > + RPI_FIRMWARE_GET_CLOCK_RATE, > + pll->fw_data->firmware_clk_id, > + &val); > + if (ret) > + return ret; > + > + return val * pll->fw_data->rate_rescale; > +} > + > static unsigned long bcm2835_pll_get_rate(struct clk_hw *hw, > unsigned long parent_rate) > { > @@ -584,6 +685,35 @@ static unsigned long bcm2835_pll_get_rate(struct clk_hw *hw, > return bcm2835_pll_rate_from_divisors(parent_rate, ndiv, fdiv, pdiv); > } > > +static int bcm2835_fw_pll_on(struct clk_hw *hw) > +{ > + struct bcm2835_pll *pll = container_of(hw, struct bcm2835_pll, hw); > + struct bcm2835_cprman *cprman = pll->cprman; > + u32 val; > + int ret; > + > + val = RPI_FIRMWARE_STATE_ENABLE_BIT | RPI_FIRMWARE_STATE_WAIT_BIT; > + > + ret = raspberrypi_clock_property(cprman->firmware, > + RPI_FIRMWARE_SET_CLOCK_STATE, > + pll->fw_data->firmware_clk_id, &val); > + if (ret) > + return ret; > + > + return 0; > +} > + > +static void bcm2835_fw_pll_off(struct clk_hw *hw) > +{ > + struct bcm2835_pll *pll = container_of(hw, struct bcm2835_pll, hw); > + struct bcm2835_cprman *cprman = pll->cprman; > + u32 val = RPI_FIRMWARE_STATE_WAIT_BIT; > + > + raspberrypi_clock_property(cprman->firmware, > + RPI_FIRMWARE_SET_CLOCK_STATE, > + pll->fw_data->firmware_clk_id, &val); > +} > + > static void bcm2835_pll_off(struct clk_hw *hw) > { > struct bcm2835_pll *pll = container_of(hw, struct bcm2835_pll, hw); > @@ -651,6 +781,25 @@ bcm2835_pll_write_ana(struct bcm2835_cprman *cprman, u32 ana_reg_base, u32 *ana) > cprman_write(cprman, ana_reg_base + i * 4, ana[i]); > } > > +static int bcm2835_fw_pll_set_rate(struct clk_hw *hw, unsigned long rate, > + unsigned long parent_rate) > +{ > + struct bcm2835_pll *pll = container_of(hw, struct bcm2835_pll, hw); > + u32 new_rate = rate / pll->fw_data->rate_rescale; > + struct bcm2835_cprman *cprman = pll->cprman; > + int ret; > + > + ret = raspberrypi_clock_property(cprman->firmware, > + RPI_FIRMWARE_SET_CLOCK_RATE, > + pll->fw_data->firmware_clk_id, > + &new_rate); > + if (ret) > + dev_err(cprman->dev, "Failed to change %s frequency: %d", > + clk_hw_get_name(hw), ret); I think we should print this error only once or at least rate limited. > + > + return ret; > +} > + > static int bcm2835_pll_set_rate(struct clk_hw *hw, > unsigned long rate, unsigned long parent_rate) > { > @@ -759,6 +908,15 @@ static const struct clk_ops bcm2835_pll_clk_ops = { > .debug_init = bcm2835_pll_debug_init, > }; > > +static const struct clk_ops bcm2835_firmware_pll_clk_ops = { > + .is_prepared = bcm2835_fw_pll_is_on, > + .prepare = bcm2835_fw_pll_on, > + .unprepare = bcm2835_fw_pll_off, > + .recalc_rate = bcm2835_fw_pll_get_rate, > + .set_rate = bcm2835_fw_pll_set_rate, > + .round_rate = bcm2835_pll_round_rate, > +}; > + > struct bcm2835_pll_divider { > struct clk_divider div; > struct bcm2835_cprman *cprman; > @@ -1287,6 +1445,83 @@ static const struct clk_ops bcm2835_vpu_clock_clk_ops = { > .debug_init = bcm2835_clock_debug_init, > }; > > +static int bcm2835_firmware_get_min_max_rates(struct bcm2835_cprman *cprman, > + struct bcm2835_fw_pll_data *data) > +{ > + u32 min_rate = 0, max_rate = 0; > + int ret; > + > + ret = raspberrypi_clock_property(cprman->firmware, > + RPI_FIRMWARE_GET_MIN_CLOCK_RATE, > + data->firmware_clk_id, > + &min_rate); > + if (ret) { > + dev_err(cprman->dev, "Failed to get %s min freq: %d\n", > + data->name, ret); > + return ret; > + } > + > + ret = raspberrypi_clock_property(cprman->firmware, > + RPI_FIRMWARE_GET_MAX_CLOCK_RATE, > + data->firmware_clk_id, > + &max_rate); > + if (ret) { > + dev_err(cprman->dev, "Failed to get %s max freq: %d\n", > + data->name, ret); > + return ret; > + } > + > + data->min_rate = min_rate * data->rate_rescale; > + data->max_rate = max_rate * data->rate_rescale; > + > + dev_info(cprman->dev, "min %lu, max %lu\n", data->min_rate, data->max_rate); > + return 0; > +} > + > +static struct clk_hw *bcm2835_register_fw_pll(struct bcm2835_cprman *cprman, > + const struct bcm2835_fw_pll_data *data) > +{ > + struct bcm2835_fw_pll_data *fw_data; > + struct clk_init_data init; > + struct bcm2835_pll *pll; > + int ret; > + > + memset(&init, 0, sizeof(init)); > + > + /* All of the PLLs derive from the external oscillator. */ > + init.parent_names = &cprman->real_parent_names[0]; > + init.num_parents = 1; > + init.name = data->name; > + init.ops = &bcm2835_firmware_pll_clk_ops; > + init.flags = data->flags | CLK_IGNORE_UNUSED; > + > + pll = kzalloc(sizeof(*pll), GFP_KERNEL); > + if (!pll) > + return NULL; > + > + /* > + * As the max and min rate values are firmware dependent we need a non > + * constant version of struct bcm2835_fw_pll_data. > + */ > + fw_data = devm_kmemdup(cprman->dev, data, sizeof(*data), > + GFP_KERNEL); > + if (!fw_data) > + return NULL; > + > + ret = bcm2835_firmware_get_min_max_rates(cprman, fw_data); > + if (ret) > + return NULL; > + > + pll->cprman = cprman; > + pll->hw.init = &init; > + pll->fw_data = fw_data; > + > + ret = devm_clk_hw_register(cprman->dev, &pll->hw); > + if (ret) > + return NULL; > + return &pll->hw; > +} > + > static struct clk_hw *bcm2835_register_pll(struct bcm2835_cprman *cprman, > const struct bcm2835_pll_data *data) > { > @@ -1462,6 +1697,9 @@ struct bcm2835_clk_desc { > #define REGISTER_PLL(...) _REGISTER(&bcm2835_register_pll, \ > &(struct bcm2835_pll_data) \ > {__VA_ARGS__}) > +#define REGISTER_FW_PLL(...) _REGISTER(&bcm2835_register_fw_pll, \ > + &(struct bcm2835_fw_pll_data) \ > + {__VA_ARGS__}) > #define REGISTER_PLL_DIV(...) _REGISTER(&bcm2835_register_pll_divider, \ > &(struct bcm2835_pll_divider_data) \ > {__VA_ARGS__}) > @@ -1654,21 +1892,11 @@ static const struct bcm2835_clk_desc clk_desc_array[] = { > .flags = CLK_SET_RATE_PARENT), > > /* PLLB is used for the ARM's clock. */ > - [BCM2835_PLLB] = REGISTER_PLL( > + [BCM2835_PLLB] = REGISTER_FW_PLL( > .name = "pllb", > - .cm_ctrl_reg = CM_PLLB, > - .a2w_ctrl_reg = A2W_PLLB_CTRL, > - .frac_reg = A2W_PLLB_FRAC, > - .ana_reg_base = A2W_PLLB_ANA0, > - .reference_enable_mask = A2W_XOSC_CTRL_PLLB_ENABLE, > - .lock_mask = CM_LOCK_FLOCKB, > .flags = CLK_GET_RATE_NOCACHE, > - > - .ana = &bcm2835_ana_default, > - > - .min_rate = 600000000u, > - .max_rate = 3000000000u, > - .max_fb_rate = BCM2835_MAX_FB_RATE), > + .rate_rescale = 2, > + .firmware_clk_id = RPI_FIRMWARE_ARM_CLK_ID), > [BCM2835_PLLB_ARM] = REGISTER_PLL_DIV( > .name = "pllb_arm", > .source_pll = "pllb", > @@ -2133,9 +2361,24 @@ static int bcm2835_clk_probe(struct platform_device *pdev) > struct resource *res; > const struct bcm2835_clk_desc *desc; > const size_t asize = ARRAY_SIZE(clk_desc_array); > + struct device_node *firmware_node; > + struct rpi_firmware *firmware; > size_t i; > int ret; > > + firmware_node = of_find_node_by_name(NULL, "firmware"); I prefer of_find_compatible_node() which is more specific. > + if (!firmware_node) { > + dev_err(dev, "Missing firmware node\n"); > + return -ENOENT; > + } The firmware node should be optional for the driver. > + > + firmware = rpi_firmware_get(firmware_node); > + of_node_put(firmware_node); > + if (!firmware) { > + dev_err(dev, "Can't get raspberrypi's firmware\n"); For in case the driver for the Raspberry Pifirmware is enabled, we shouldn't spam with errors here. Stefan
On 20.05.19 14:11, Stefan Wahren wrote: > > Please verify that the kernel still works (and this clock driver probe) > under the following conditions: > > - CONFIG_RASPBERRYPI_FIRMWARE=n > - CONFIG_RASPBERRYPI_FIRMWARE=m > - older DTBs without patch #1 Sorry, i meant patch #4
On Mo, 2019-05-20 at 12:47 +0200, Nicolas Saenz Julienne wrote: > + * For more information on the firmware interface check: > + * https://github.com/raspberrypi/firmware/wiki/Mailbox-property-interface > + */ > +struct bcm2835_firmware_prop { > + u32 id; > + u32 val; > + u32 disable_turbo; > +} __packed; Hi, technically we are not in arch and those fields have a defined endianness. Regards Oliver
Hi Oliver, thanks for the review. On Mon, 2019-05-20 at 14:43 +0200, Oliver Neukum wrote: > On Mo, 2019-05-20 at 12:47 +0200, Nicolas Saenz Julienne wrote: > > + * For more information on the firmware interface check: > > + * https://github.com/raspberrypi/firmware/wiki/Mailbox-property-interface > > + */ > > +struct bcm2835_firmware_prop { > > + u32 id; > > + u32 val; > > + u32 disable_turbo; > > +} __packed; > > Hi, > > technically we are not in arch and those fields have a defined > endianness. > Well I set it as packed since it's 'sent' through a memory mapped firmware interface. Hence the need for the structure format to be fixed. So I guessed we're safer with it, as I'm not 100% sure what the different compilers are going to do with it (although it's very likely it'll stay the same). BTW this will be built both for arm & arm64. Regards, Nicolas
On Tue, 21 May 2019 13:39:31 +0200 Nicolas Saenz Julienne <nsaenzjulienne@suse.de> wrote: > Hi Oliver, thanks for the review. > > On Mon, 2019-05-20 at 14:43 +0200, Oliver Neukum wrote: > > On Mo, 2019-05-20 at 12:47 +0200, Nicolas Saenz Julienne wrote: > > > + * For more information on the firmware interface check: > > > + * https://github.com/raspberrypi/firmware/wiki/Mailbox-property-interface > > > + */ > > > +struct bcm2835_firmware_prop { > > > + u32 id; > > > + u32 val; > > > + u32 disable_turbo; > > > +} __packed; > > > > Hi, > > > > technically we are not in arch and those fields have a defined > > endianness. > > > > Well I set it as packed since it's 'sent' through a memory mapped firmware > interface. Hence the need for the structure format to be fixed. So I guessed > we're safer with it, as I'm not 100% sure what the different compilers are > going to do with it (although it's very likely it'll stay the same). BTW this > will be built both for arm & arm64. I believe that's not the point Oliver was trying to make. You should use __le32 instead of u32. That's because u32 means "host byte order" and this code is not located under arch/, so host endianness is unknown, but the mailbox interface requires little-endian. It's nit-picking, and that's why Oliver writes 'technically'; there is probably no way this firmware interface could be used on a big-endian CPU... Petr T
On Tue, 2019-05-21 at 14:14 +0200, Petr Tesarik wrote: > On Tue, 21 May 2019 13:39:31 +0200 > Nicolas Saenz Julienne <nsaenzjulienne@suse.de> wrote: > > > Hi Oliver, thanks for the review. > > > > On Mon, 2019-05-20 at 14:43 +0200, Oliver Neukum wrote: > > > On Mo, 2019-05-20 at 12:47 +0200, Nicolas Saenz Julienne wrote: > > > > + * For more information on the firmware interface check: > > > > + * > > > > https://github.com/raspberrypi/firmware/wiki/Mailbox-property-interface > > > > + */ > > > > +struct bcm2835_firmware_prop { > > > > + u32 id; > > > > + u32 val; > > > > + u32 disable_turbo; > > > > +} __packed; > > > > > > Hi, > > > > > > technically we are not in arch and those fields have a defined > > > endianness. > > > > > > > Well I set it as packed since it's 'sent' through a memory mapped firmware > > interface. Hence the need for the structure format to be fixed. So I guessed > > we're safer with it, as I'm not 100% sure what the different compilers are > > going to do with it (although it's very likely it'll stay the same). BTW > > this > > will be built both for arm & arm64. > > I believe that's not the point Oliver was trying to make. You should > use __le32 instead of u32. > > That's because u32 means "host byte order" and this code is not located > under arch/, so host endianness is unknown, but the mailbox interface > requires little-endian. > > It's nit-picking, and that's why Oliver writes 'technically'; there is > probably no way this firmware interface could be used on a big-endian > CPU... Understood, thanks for the clarification. Regards, Nicolas
Hi Nicolas, On 20.05.19 14:11, Stefan Wahren wrote: > Hi Nicolas, > > the following comments applies only in case Eric is fine with the whole > approach. > > On 20.05.19 12:47, Nicolas Saenz Julienne wrote: >> Raspberry Pi's firmware, which runs in a dedicated processor, keeps > maybe we should clarify that the firmware is running in the VPU >> track of the board's temperature and voltage. It's resposible for >> scaling the CPU frequency whenever it deems the device reached an unsafe >> state. On top of that the firmware provides an interface which allows >> Linux to to query the clock's state or change it's frequency. > I think this requires a separate update of the devicetree binding. >> Being the sole user of the bcm2835 clock driver, this integrates the >> firmware interface into the clock driver and adds a first user: the CPU >> pll, also known as 'pllb'. > Please verify that the kernel still works (and this clock driver probe) > under the following conditions: > > - CONFIG_RASPBERRYPI_FIRMWARE=n > - CONFIG_RASPBERRYPI_FIRMWARE=m > - older DTBs without patch #1 i thought about this and the case this driver would return -EPROBE_DEFER. The clock driver is too essential for doing such a thing. So i think the best solution would be to move these changes into a separate driver which should be register by the clock driver (similiar to vchiq). This also avoid the need of a new device tree binding.
Hi Stefan, thanks for your comments! On Tue, 2019-05-21 at 14:40 +0200, Stefan Wahren wrote: > Hi Nicolas, > > On 20.05.19 14:11, Stefan Wahren wrote: > > Hi Nicolas, > > > > the following comments applies only in case Eric is fine with the whole > > approach. > > > > On 20.05.19 12:47, Nicolas Saenz Julienne wrote: > > > Raspberry Pi's firmware, which runs in a dedicated processor, keeps > > maybe we should clarify that the firmware is running in the VPU > > > track of the board's temperature and voltage. It's resposible for > > > scaling the CPU frequency whenever it deems the device reached an unsafe > > > state. On top of that the firmware provides an interface which allows > > > Linux to to query the clock's state or change it's frequency. > > I think this requires a separate update of the devicetree binding. > > > Being the sole user of the bcm2835 clock driver, this integrates the > > > firmware interface into the clock driver and adds a first user: the CPU > > > pll, also known as 'pllb'. > > Please verify that the kernel still works (and this clock driver probe) > > under the following conditions: > > > > - CONFIG_RASPBERRYPI_FIRMWARE=n > > - CONFIG_RASPBERRYPI_FIRMWARE=m > > - older DTBs without patch #1 > i thought about this and the case this driver would return > -EPROBE_DEFER. The clock driver is too essential for doing such a thing. > So i think the best solution would be to move these changes into a > separate driver which should be register by the clock driver (similiar > to vchiq). This also avoid the need of a new device tree binding. I understand your concerns. Wouldn't you prefer registering the device trough the device tree? I'd go with the same approach as the firmware touchscreen driver, which is registered after the firmware's probe trough dt's 'simple-bus'. That said, it's not a strongly held opinion, I'm happy with whatever solution as long as it works. I get from your comments that you'd like the register based version of 'pllb' and 'pllb_arm' to be loaded if for some reason the firmware isn't available. Is that right? The main problem I see with this is the duplication of 'pllb' and 'pllb_arm'. Both drivers will create the same clock device through different interfaces. Any suggestions on how to deal with that? If not I can simply remove 'pllb' and 'pllb_arm' from clk-bcm2835.c. Regards, Nicolas
> Nicolas Saenz Julienne <nsaenzjulienne@suse.de> hat am 21. Mai 2019 um 17:47 geschrieben: > > > Hi Stefan, thanks for your comments! > > On Tue, 2019-05-21 at 14:40 +0200, Stefan Wahren wrote: > > Hi Nicolas, > > > > On 20.05.19 14:11, Stefan Wahren wrote: > > > Hi Nicolas, > > > > > > the following comments applies only in case Eric is fine with the whole > > > approach. > > > > > > On 20.05.19 12:47, Nicolas Saenz Julienne wrote: > > > > Raspberry Pi's firmware, which runs in a dedicated processor, keeps > > > maybe we should clarify that the firmware is running in the VPU > > > > track of the board's temperature and voltage. It's resposible for > > > > scaling the CPU frequency whenever it deems the device reached an unsafe > > > > state. On top of that the firmware provides an interface which allows > > > > Linux to to query the clock's state or change it's frequency. > > > I think this requires a separate update of the devicetree binding. > > > > Being the sole user of the bcm2835 clock driver, this integrates the > > > > firmware interface into the clock driver and adds a first user: the CPU > > > > pll, also known as 'pllb'. > > > Please verify that the kernel still works (and this clock driver probe) > > > under the following conditions: > > > > > > - CONFIG_RASPBERRYPI_FIRMWARE=n > > > - CONFIG_RASPBERRYPI_FIRMWARE=m > > > - older DTBs without patch #1 > > i thought about this and the case this driver would return > > -EPROBE_DEFER. The clock driver is too essential for doing such a thing. > > So i think the best solution would be to move these changes into a > > separate driver which should be register by the clock driver (similiar > > to vchiq). This also avoid the need of a new device tree binding. > > I understand your concerns. > > Wouldn't you prefer registering the device trough the device tree? I'd go with > the same approach as the firmware touchscreen driver, which is registered after > the firmware's probe trough dt's 'simple-bus'. That said, it's not a strongly > held opinion, I'm happy with whatever solution as long as it works. A devicetree binding always introduce some kind of inflexibility. In case someone finds a better solution later things can get really messy. A recent example is the clock handling for i2c-bcm2835. > > I get from your comments that you'd like the register based version of 'pllb' > and 'pllb_arm' to be loaded if for some reason the firmware isn't available. Is > that right? This wasn't my intention. I would prefer a simple approch here (no handover). > The main problem I see with this is the duplication of 'pllb' and > 'pllb_arm'. Both drivers will create the same clock device through different > interfaces. Any suggestions on how to deal with that? If not I can simply > remove 'pllb' and 'pllb_arm' from clk-bcm2835.c. Yes. So even if this driver is disabled, there shouldn't be a regression. Or did i miss something? > > Regards, > Nicolas >
On Tue, 2019-05-21 at 23:43 +0200, Stefan Wahren wrote: > > Nicolas Saenz Julienne <nsaenzjulienne@suse.de> hat am 21. Mai 2019 um 17:47 > > geschrieben: > > > > > > Hi Stefan, thanks for your comments! > > > > On Tue, 2019-05-21 at 14:40 +0200, Stefan Wahren wrote: > > > Hi Nicolas, > > > > > > On 20.05.19 14:11, Stefan Wahren wrote: > > > > Hi Nicolas, > > > > > > > > the following comments applies only in case Eric is fine with the whole > > > > approach. > > > > > > > > On 20.05.19 12:47, Nicolas Saenz Julienne wrote: > > > > > Raspberry Pi's firmware, which runs in a dedicated processor, keeps > > > > maybe we should clarify that the firmware is running in the VPU > > > > > track of the board's temperature and voltage. It's resposible for > > > > > scaling the CPU frequency whenever it deems the device reached an > > > > > unsafe > > > > > state. On top of that the firmware provides an interface which allows > > > > > Linux to to query the clock's state or change it's frequency. > > > > I think this requires a separate update of the devicetree binding. > > > > > Being the sole user of the bcm2835 clock driver, this integrates the > > > > > firmware interface into the clock driver and adds a first user: the > > > > > CPU > > > > > pll, also known as 'pllb'. > > > > Please verify that the kernel still works (and this clock driver probe) > > > > under the following conditions: > > > > > > > > - CONFIG_RASPBERRYPI_FIRMWARE=n > > > > - CONFIG_RASPBERRYPI_FIRMWARE=m > > > > - older DTBs without patch #1 > > > i thought about this and the case this driver would return > > > -EPROBE_DEFER. The clock driver is too essential for doing such a thing. > > > So i think the best solution would be to move these changes into a > > > separate driver which should be register by the clock driver (similiar > > > to vchiq). This also avoid the need of a new device tree binding. > > > > I understand your concerns. > > > > Wouldn't you prefer registering the device trough the device tree? I'd go > > with > > the same approach as the firmware touchscreen driver, which is registered > > after > > the firmware's probe trough dt's 'simple-bus'. That said, it's not a > > strongly > > held opinion, I'm happy with whatever solution as long as it works. > > A devicetree binding always introduce some kind of inflexibility. In case > someone finds a better solution later things can get really messy. A recent > example is the clock handling for i2c-bcm2835. Fair enough. > > I get from your comments that you'd like the register based version of > > 'pllb' > > and 'pllb_arm' to be loaded if for some reason the firmware isn't available. > > Is > > that right? > > This wasn't my intention. I would prefer a simple approch here (no handover). > > > The main problem I see with this is the duplication of 'pllb' and > > 'pllb_arm'. Both drivers will create the same clock device through different > > interfaces. Any suggestions on how to deal with that? If not I can simply > > remove 'pllb' and 'pllb_arm' from clk-bcm2835.c. > > Yes. So even if this driver is disabled, there shouldn't be a regression. Or > did i miss something? No, there shoudn't be any regressions as these clocks are not being used at the moment. I'll send a follow-up series soon :) Regrads, Nicolas
diff --git a/drivers/clk/bcm/clk-bcm2835.c b/drivers/clk/bcm/clk-bcm2835.c index 5aea110672f3..ce5b16f3704e 100644 --- a/drivers/clk/bcm/clk-bcm2835.c +++ b/drivers/clk/bcm/clk-bcm2835.c @@ -35,6 +35,7 @@ #include <linux/platform_device.h> #include <linux/slab.h> #include <dt-bindings/clock/bcm2835.h> +#include <soc/bcm2835/raspberrypi-firmware.h> #define CM_PASSWORD 0x5a000000 @@ -289,6 +290,30 @@ #define LOCK_TIMEOUT_NS 100000000 #define BCM2835_MAX_FB_RATE 1750000000u +#define RPI_FIRMWARE_ARM_CLK_ID 0x000000003 +#define RPI_FIRMWARE_STATE_ENABLE_BIT 0x1 +#define RPI_FIRMWARE_STATE_WAIT_BIT 0x2 + +/* + * Structure of the message passed to Raspberry Pi's firmware in order to + * change clock rates. The 'disable_turbo' option is only available to the ARM + * clock (pllb) which we enable by default as turbo mode will alter multiple + * clocks at once. + * + * Even though we're able to access the clock registers directly we're bound to + * use the firmware interface as the firmware ultimately takes care of + * mitigating overheating/undervoltage situations and we would be changing + * frequencies behind his back. + * + * For more information on the firmware interface check: + * https://github.com/raspberrypi/firmware/wiki/Mailbox-property-interface + */ +struct bcm2835_firmware_prop { + u32 id; + u32 val; + u32 disable_turbo; +} __packed; + /* * Names of clocks used within the driver that need to be replaced * with an external parent's name. This array is in the order that @@ -316,6 +341,8 @@ struct bcm2835_cprman { */ const char *real_parent_names[ARRAY_SIZE(cprman_parent_names)]; + struct rpi_firmware *firmware; + /* Must be last */ struct clk_hw_onecell_data onecell; }; @@ -424,6 +451,23 @@ struct bcm2835_pll_data { unsigned long max_fb_rate; }; +struct bcm2835_fw_pll_data { + const char *name; + int firmware_clk_id; + u32 flags; + + unsigned long min_rate; + unsigned long max_rate; + + /* + * The rate provided to the firmware interface might not always match + * the clock subsystem's. For instance, in the case of the ARM cpu + * clock, even though the firmware ultimately edits 'pllb' it takes the + * expected 'pllb_arm' rate as it's input. + */ + unsigned int rate_rescale; +}; + struct bcm2835_pll_ana_bits { u32 mask0; u32 set0; @@ -505,6 +549,7 @@ struct bcm2835_pll { struct clk_hw hw; struct bcm2835_cprman *cprman; const struct bcm2835_pll_data *data; + const struct bcm2835_fw_pll_data *fw_data; }; static int bcm2835_pll_is_on(struct clk_hw *hw) @@ -517,6 +562,41 @@ static int bcm2835_pll_is_on(struct clk_hw *hw) A2W_PLL_CTRL_PRST_DISABLE; } +static int raspberrypi_clock_property(struct rpi_firmware *firmware, u32 tag, + u32 clk, u32 *val) +{ + int ret; + struct bcm2835_firmware_prop msg = { + .id = clk, + .val = *val, + .disable_turbo = 1, + }; + + ret = rpi_firmware_property(firmware, tag, &msg, sizeof(msg)); + if (ret) + return ret; + + *val = msg.val; + + return 0; +} + +static int bcm2835_fw_pll_is_on(struct clk_hw *hw) +{ + struct bcm2835_pll *pll = container_of(hw, struct bcm2835_pll, hw); + struct bcm2835_cprman *cprman = pll->cprman; + u32 val = 0; + int ret; + + ret = raspberrypi_clock_property(cprman->firmware, + RPI_FIRMWARE_GET_CLOCK_STATE, + pll->fw_data->firmware_clk_id, &val); + if (ret) + return 0; + + return !!(val & RPI_FIRMWARE_STATE_ENABLE_BIT); +} + static void bcm2835_pll_choose_ndiv_and_fdiv(unsigned long rate, unsigned long parent_rate, u32 *ndiv, u32 *fdiv) @@ -547,16 +627,37 @@ static long bcm2835_pll_round_rate(struct clk_hw *hw, unsigned long rate, unsigned long *parent_rate) { struct bcm2835_pll *pll = container_of(hw, struct bcm2835_pll, hw); - const struct bcm2835_pll_data *data = pll->data; u32 ndiv, fdiv; - rate = clamp(rate, data->min_rate, data->max_rate); + if (pll->data) + rate = clamp(rate, pll->data->min_rate, pll->data->max_rate); + else + rate = clamp(rate, pll->fw_data->min_rate, + pll->fw_data->max_rate); bcm2835_pll_choose_ndiv_and_fdiv(rate, *parent_rate, &ndiv, &fdiv); return bcm2835_pll_rate_from_divisors(*parent_rate, ndiv, fdiv, 1); } +static unsigned long bcm2835_fw_pll_get_rate(struct clk_hw *hw, + unsigned long parent_rate) +{ + struct bcm2835_pll *pll = container_of(hw, struct bcm2835_pll, hw); + struct bcm2835_cprman *cprman = pll->cprman; + u32 val = 0; + int ret; + + ret = raspberrypi_clock_property(cprman->firmware, + RPI_FIRMWARE_GET_CLOCK_RATE, + pll->fw_data->firmware_clk_id, + &val); + if (ret) + return ret; + + return val * pll->fw_data->rate_rescale; +} + static unsigned long bcm2835_pll_get_rate(struct clk_hw *hw, unsigned long parent_rate) { @@ -584,6 +685,35 @@ static unsigned long bcm2835_pll_get_rate(struct clk_hw *hw, return bcm2835_pll_rate_from_divisors(parent_rate, ndiv, fdiv, pdiv); } +static int bcm2835_fw_pll_on(struct clk_hw *hw) +{ + struct bcm2835_pll *pll = container_of(hw, struct bcm2835_pll, hw); + struct bcm2835_cprman *cprman = pll->cprman; + u32 val; + int ret; + + val = RPI_FIRMWARE_STATE_ENABLE_BIT | RPI_FIRMWARE_STATE_WAIT_BIT; + + ret = raspberrypi_clock_property(cprman->firmware, + RPI_FIRMWARE_SET_CLOCK_STATE, + pll->fw_data->firmware_clk_id, &val); + if (ret) + return ret; + + return 0; +} + +static void bcm2835_fw_pll_off(struct clk_hw *hw) +{ + struct bcm2835_pll *pll = container_of(hw, struct bcm2835_pll, hw); + struct bcm2835_cprman *cprman = pll->cprman; + u32 val = RPI_FIRMWARE_STATE_WAIT_BIT; + + raspberrypi_clock_property(cprman->firmware, + RPI_FIRMWARE_SET_CLOCK_STATE, + pll->fw_data->firmware_clk_id, &val); +} + static void bcm2835_pll_off(struct clk_hw *hw) { struct bcm2835_pll *pll = container_of(hw, struct bcm2835_pll, hw); @@ -651,6 +781,25 @@ bcm2835_pll_write_ana(struct bcm2835_cprman *cprman, u32 ana_reg_base, u32 *ana) cprman_write(cprman, ana_reg_base + i * 4, ana[i]); } +static int bcm2835_fw_pll_set_rate(struct clk_hw *hw, unsigned long rate, + unsigned long parent_rate) +{ + struct bcm2835_pll *pll = container_of(hw, struct bcm2835_pll, hw); + u32 new_rate = rate / pll->fw_data->rate_rescale; + struct bcm2835_cprman *cprman = pll->cprman; + int ret; + + ret = raspberrypi_clock_property(cprman->firmware, + RPI_FIRMWARE_SET_CLOCK_RATE, + pll->fw_data->firmware_clk_id, + &new_rate); + if (ret) + dev_err(cprman->dev, "Failed to change %s frequency: %d", + clk_hw_get_name(hw), ret); + + return ret; +} + static int bcm2835_pll_set_rate(struct clk_hw *hw, unsigned long rate, unsigned long parent_rate) { @@ -759,6 +908,15 @@ static const struct clk_ops bcm2835_pll_clk_ops = { .debug_init = bcm2835_pll_debug_init, }; +static const struct clk_ops bcm2835_firmware_pll_clk_ops = { + .is_prepared = bcm2835_fw_pll_is_on, + .prepare = bcm2835_fw_pll_on, + .unprepare = bcm2835_fw_pll_off, + .recalc_rate = bcm2835_fw_pll_get_rate, + .set_rate = bcm2835_fw_pll_set_rate, + .round_rate = bcm2835_pll_round_rate, +}; + struct bcm2835_pll_divider { struct clk_divider div; struct bcm2835_cprman *cprman; @@ -1287,6 +1445,83 @@ static const struct clk_ops bcm2835_vpu_clock_clk_ops = { .debug_init = bcm2835_clock_debug_init, }; +static int bcm2835_firmware_get_min_max_rates(struct bcm2835_cprman *cprman, + struct bcm2835_fw_pll_data *data) +{ + u32 min_rate = 0, max_rate = 0; + int ret; + + ret = raspberrypi_clock_property(cprman->firmware, + RPI_FIRMWARE_GET_MIN_CLOCK_RATE, + data->firmware_clk_id, + &min_rate); + if (ret) { + dev_err(cprman->dev, "Failed to get %s min freq: %d\n", + data->name, ret); + return ret; + } + + ret = raspberrypi_clock_property(cprman->firmware, + RPI_FIRMWARE_GET_MAX_CLOCK_RATE, + data->firmware_clk_id, + &max_rate); + if (ret) { + dev_err(cprman->dev, "Failed to get %s max freq: %d\n", + data->name, ret); + return ret; + } + + data->min_rate = min_rate * data->rate_rescale; + data->max_rate = max_rate * data->rate_rescale; + + dev_info(cprman->dev, "min %lu, max %lu\n", data->min_rate, data->max_rate); + return 0; +} + +static struct clk_hw *bcm2835_register_fw_pll(struct bcm2835_cprman *cprman, + const struct bcm2835_fw_pll_data *data) +{ + struct bcm2835_fw_pll_data *fw_data; + struct clk_init_data init; + struct bcm2835_pll *pll; + int ret; + + memset(&init, 0, sizeof(init)); + + /* All of the PLLs derive from the external oscillator. */ + init.parent_names = &cprman->real_parent_names[0]; + init.num_parents = 1; + init.name = data->name; + init.ops = &bcm2835_firmware_pll_clk_ops; + init.flags = data->flags | CLK_IGNORE_UNUSED; + + pll = kzalloc(sizeof(*pll), GFP_KERNEL); + if (!pll) + return NULL; + + /* + * As the max and min rate values are firmware dependent we need a non + * constant version of struct bcm2835_fw_pll_data. + */ + fw_data = devm_kmemdup(cprman->dev, data, sizeof(*data), + GFP_KERNEL); + if (!fw_data) + return NULL; + + ret = bcm2835_firmware_get_min_max_rates(cprman, fw_data); + if (ret) + return NULL; + + pll->cprman = cprman; + pll->hw.init = &init; + pll->fw_data = fw_data; + + ret = devm_clk_hw_register(cprman->dev, &pll->hw); + if (ret) + return NULL; + return &pll->hw; +} + static struct clk_hw *bcm2835_register_pll(struct bcm2835_cprman *cprman, const struct bcm2835_pll_data *data) { @@ -1462,6 +1697,9 @@ struct bcm2835_clk_desc { #define REGISTER_PLL(...) _REGISTER(&bcm2835_register_pll, \ &(struct bcm2835_pll_data) \ {__VA_ARGS__}) +#define REGISTER_FW_PLL(...) _REGISTER(&bcm2835_register_fw_pll, \ + &(struct bcm2835_fw_pll_data) \ + {__VA_ARGS__}) #define REGISTER_PLL_DIV(...) _REGISTER(&bcm2835_register_pll_divider, \ &(struct bcm2835_pll_divider_data) \ {__VA_ARGS__}) @@ -1654,21 +1892,11 @@ static const struct bcm2835_clk_desc clk_desc_array[] = { .flags = CLK_SET_RATE_PARENT), /* PLLB is used for the ARM's clock. */ - [BCM2835_PLLB] = REGISTER_PLL( + [BCM2835_PLLB] = REGISTER_FW_PLL( .name = "pllb", - .cm_ctrl_reg = CM_PLLB, - .a2w_ctrl_reg = A2W_PLLB_CTRL, - .frac_reg = A2W_PLLB_FRAC, - .ana_reg_base = A2W_PLLB_ANA0, - .reference_enable_mask = A2W_XOSC_CTRL_PLLB_ENABLE, - .lock_mask = CM_LOCK_FLOCKB, .flags = CLK_GET_RATE_NOCACHE, - - .ana = &bcm2835_ana_default, - - .min_rate = 600000000u, - .max_rate = 3000000000u, - .max_fb_rate = BCM2835_MAX_FB_RATE), + .rate_rescale = 2, + .firmware_clk_id = RPI_FIRMWARE_ARM_CLK_ID), [BCM2835_PLLB_ARM] = REGISTER_PLL_DIV( .name = "pllb_arm", .source_pll = "pllb", @@ -2133,9 +2361,24 @@ static int bcm2835_clk_probe(struct platform_device *pdev) struct resource *res; const struct bcm2835_clk_desc *desc; const size_t asize = ARRAY_SIZE(clk_desc_array); + struct device_node *firmware_node; + struct rpi_firmware *firmware; size_t i; int ret; + firmware_node = of_find_node_by_name(NULL, "firmware"); + if (!firmware_node) { + dev_err(dev, "Missing firmware node\n"); + return -ENOENT; + } + + firmware = rpi_firmware_get(firmware_node); + of_node_put(firmware_node); + if (!firmware) { + dev_err(dev, "Can't get raspberrypi's firmware\n"); + return -EPROBE_DEFER; + } + cprman = devm_kzalloc(dev, struct_size(cprman, onecell.hws, asize), GFP_KERNEL); @@ -2144,6 +2387,7 @@ static int bcm2835_clk_probe(struct platform_device *pdev) spin_lock_init(&cprman->regs_lock); cprman->dev = dev; + cprman->firmware = firmware; res = platform_get_resource(pdev, IORESOURCE_MEM, 0); cprman->regs = devm_ioremap_resource(dev, res); if (IS_ERR(cprman->regs))
Raspberry Pi's firmware, which runs in a dedicated processor, keeps track of the board's temperature and voltage. It's resposible for scaling the CPU frequency whenever it deems the device reached an unsafe state. On top of that the firmware provides an interface which allows Linux to to query the clock's state or change it's frequency. Being the sole user of the bcm2835 clock driver, this integrates the firmware interface into the clock driver and adds a first user: the CPU pll, also known as 'pllb'. Signed-off-by: Nicolas Saenz Julienne <nsaenzjulienne@suse.de> --- drivers/clk/bcm/clk-bcm2835.c | 274 ++++++++++++++++++++++++++++++++-- 1 file changed, 259 insertions(+), 15 deletions(-)