Add interconnect support to QSPI and QUP drivers
Subsystems (9)
v3 cover Add interconnect support to QSPI and QUP drivers
dt-binding patch for QUP drivers.
- https://patchwork.kernel.org/patch/11436621/ [Convert QUP bindings to YAML and add ICC, pin swap doc]
dt-binding patch for QSPI.
- https://patchwork.kernel.org/patch/11436719/ [Convert QSPI binding to YAML and add interconnect doc]
High level design:
- QUP wrapper/common driver.
Vote for QUP core on behalf of earlycon from probe. Remove BW vote during earlycon exit call
- SERIAL driver.
Vote only for CPU/CORE path because driver is in FIFO mode only Vote/unvote from qcom_geni_serial_pm func.
Bump up the CPU vote from set_termios call based on real time need
- I2C driver.
Vote for CORE/CPU/DDR path
Vote/unvote from runtime resume/suspend callback
As bus speed for I2C is fixed from probe itself no need for bump up.
- SPI QUP driver.
Vote only for CPU/CORE path because driver is in FIFO mode only Vote/unvote from runtime resume/suspend callback Bump up CPU vote based on real time need per transfer.
- QSPI driver.
Vote only for CPU path
Vote/unvote from runtime resume/suspend callback Bump up CPU vote based on real time need per transfer.
Changes in V2:
- Add devm_of_icc_get() API interconnect core.
- Add ICC support to common driver to fix earlyconsole crash.
Changes in V3:
- Define common ICC APIs in geni-se driver and use it across geni based I2C,SPI and UART driver.
Akash Asthana (8):
interconnect: Add devm_of_icc_get() as exported API for users
soc: qcom: geni: Support for ICC voting
soc: qcom-geni-se: Add interconnect support to fix earlycon crash
i2c: i2c-qcom-geni: Add interconnect support
spi: spi-geni-qcom: Add interconnect support
tty: serial: qcom_geni_serial: Add interconnect support
spi: spi-qcom-qspi: Add interconnect support
arm64: dts: sc7180: Add interconnect for QUP and QSPI
arch/arm64/boot/dts/qcom/sc7180.dtsi | 127 +++++++++++++++++++++++++++++ drivers/i2c/busses/i2c-qcom-geni.c | 30 ++++++- drivers/interconnect/core.c | 25 ++++++ drivers/soc/qcom/qcom-geni-se.c | 149 ++++++++++++++++++++++++++++++++++ drivers/spi/spi-geni-qcom.c | 31 ++++++- drivers/spi/spi-qcom-qspi.c | 46 ++++++++++- drivers/tty/serial/qcom_geni_serial.c | 35 +++++++- include/linux/interconnect.h | 7 ++ include/linux/qcom-geni-se.h | 38 +++++++++ 9 files changed, 482 insertions(+), 6 deletions(-)
Signature · 1 line
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,\na Linux Foundation Collaborative Project
PATCH v3 1/8 interconnect: Add devm_of_icc_get() as exported API for users
Users can use devm version of of_icc_get() to benefit from automatic resource release.
Reviewed by: Matthias Kaehlcke <[email protected]>
--- drivers/interconnect/core.c | 25 +++++++++++++++++++++++++ include/linux/interconnect.h | 7 +++++++ 2 files changed, 32 insertions(+)
drivers/interconnect/core.c +25 -0
@@ -350,6 +350,31 @@ static struct icc_node *of_icc_get_from_provider(struct of_phandle_args *spec)return node; } +static void devm_icc_release (struct device *dev,void *res) +{ +icc_put (*(struct icc_path **)res); +} + +struct icc_path *devm_of_icc_get (struct device *dev,const char *name) +{ +struct icc_path **ptr, *path; + + ptr =devres_alloc (devm_icc_release,sizeof (**ptr),GFP_KERNEL ); +if (!ptr) +return ERR_PTR (-ENOMEM ); + + path =of_icc_get (dev, name); +if (!IS_ERR (path)) { + *ptr = path; +devres_add (dev, ptr); + }else { +devres_free (ptr); + } + +return path; +} +EXPORT_SYMBOL_GPL (devm_of_icc_get); +/** * of_icc_get() - get a path handle from a DT node based on name * @dev: device pointer for the consumer device
include/linux/interconnect.h +7 -0
@@ -28,6 +28,7 @@ struct device;struct icc_path *icc_get (struct device *dev,const int src_id,const int dst_id);struct icc_path *of_icc_get (struct device *dev,const char *name); +struct icc_path *devm_of_icc_get (struct device *dev,const char *name);void icc_put (struct icc_path *path);int icc_set_bw (struct icc_path *path,u32 avg_bw,u32 peak_bw);void icc_set_tag (struct icc_path *path,u32 tag);
@@ -46,6 +47,12 @@ static inline struct icc_path *of_icc_get(struct device *dev,return NULL ; } +static inline struct icc_path *devm_of_icc_get (struct device *dev, +const char *name) +{ +return NULL ; +} +static inline void icc_put (struct icc_path *path) { }
Signature · 1 line
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,\na Linux Foundation Collaborative Project
PATCH v3 2/8 soc: qcom: geni: Support for ICC voting
In reply to Akash AsthanaAdd necessary macros and structure variables to support ICC BW voting from individual SE drivers.
---
Changes in V2:
- As per Bjorn's comment dropped enums for ICC paths, given the three paths individual members
Changes in V3:
- Add geni_icc_get, geni_icc_vote_on and geni_icc_vote_off as helper API.
- Add geni_icc_path structure in common header
drivers/soc/qcom/qcom-geni-se.c | 98 +++++++++++++++++++++++++++++++++++++++++ include/linux/qcom-geni-se.h | 36 +++++++++++++++ 2 files changed, 134 insertions(+)
drivers/soc/qcom/qcom-geni-se.c +98 -0
@@ -720,6 +720,104 @@ void geni_se_rx_dma_unprep(struct geni_se *se, dma_addr_t iova, size_t len) }EXPORT_SYMBOL (geni_se_rx_dma_unprep); +int geni_icc_get (struct geni_se *se,const char *icc_core,const char *icc_cpu, +const char *icc_ddr) +{ +if (icc_core) { + se->to_core.path =devm_of_icc_get (se->dev,"qup-core" ); +if (IS_ERR (se->to_core.path)) +return PTR_ERR (se->to_core.path); + } + +if (icc_cpu) { + se->from_cpu.path =devm_of_icc_get (se->dev,"qup-config" ); +if (IS_ERR (se->from_cpu.path)) +return PTR_ERR (se->from_cpu.path); + } + +if (icc_ddr) { + se->to_ddr.path =devm_of_icc_get (se->dev,"qup-memory" ); +if (IS_ERR (se->to_ddr.path)) +return PTR_ERR (se->to_ddr.path); + } + +return 0 ; +} +EXPORT_SYMBOL (geni_icc_get); + +int geni_icc_vote_on (struct geni_se *se) +{ +int ret; + +if (se->to_core.path) { + ret =icc_set_bw (se->to_core.path, se->to_core.avg_bw, + se->to_core.peak_bw); +if (ret) { +dev_err_ratelimited (se->dev,"%s: ICC BW voting failed for core\n" , + __func__); +return ret; + } + } + +if (se->from_cpu.path) { + ret =icc_set_bw (se->from_cpu.path, se->from_cpu.avg_bw, + se->from_cpu.peak_bw); +if (ret) { +dev_err_ratelimited (se->dev,"%s: ICC BW voting failed for cpu\n" , + __func__); +return ret; + } + } + +if (se->to_ddr.path) { + ret =icc_set_bw (se->to_ddr.path, se->to_ddr.avg_bw, + se->to_ddr.peak_bw); +if (ret) { +dev_err_ratelimited (se->dev,"%s: ICC BW voting failed for ddr\n" , + __func__); +return ret; + } + } + +return 0 ; +} +EXPORT_SYMBOL (geni_icc_vote_on); + +int geni_icc_vote_off (struct geni_se *se) +{ +int ret; + +if (se->to_core.path) { + ret =icc_set_bw (se->to_core.path,0 ,0 ); +if (ret) { +dev_err_ratelimited (se->dev,"%s: ICC BW remove failed for core\n" , + __func__); +return ret; + } + } + +if (se->from_cpu.path) { + ret =icc_set_bw (se->from_cpu.path,0 ,0 ); +if (ret) { +dev_err_ratelimited (se->dev,"%s: ICC BW remove failed for cpu\n" , + __func__); +return ret; + } + } + +if (se->to_ddr.path) { + ret =icc_set_bw (se->to_ddr.path,0 ,0 ); +if (ret) { +dev_err_ratelimited (se->dev,"%s: ICC BW remove failed for ddr\n" , + __func__); +return ret; + } + } + +return 0 ; +} +EXPORT_SYMBOL (geni_icc_vote_off); +static int geni_se_probe (struct platform_device *pdev) {struct device *dev = &pdev->dev;
include/linux/qcom-geni-se.h +36 -0
@@ -6,6 +6,8 @@#ifndef _LINUX_QCOM_GENI_SE#define _LINUX_QCOM_GENI_SE +#include <linux/interconnect.h> +/* Transfer mode supported by GENI Serial Engines */ enum geni_se_xfer_mode { GENI_SE_INVALID,
@@ -25,6 +27,12 @@ enum geni_se_protocol_type {struct geni_wrapper ;struct clk ; +struct geni_icc_path { +struct icc_path *path; +unsigned int avg_bw; +unsigned int peak_bw; +}; +/** * struct geni_se - GENI Serial Engine * @base: Base Address of the Serial Engine's register block
@@ -33,6 +41,9 @@ struct clk;* @clk: Handle to the core serial engine clock * @num_clk_levels: Number of valid clock levels in clk_perf_tbl * @clk_perf_tbl: Table of clock frequency input to serial engine clock +* @to_core: ICC path structure for geni to core +* @from_cpu: ICC path structure for cpu to geni +* @to_ddr: ICC path structure for geni to ddr */ struct geni_se {void __iomem *base;
@@ -41,6 +52,9 @@ struct geni_se {struct clk *clk;unsigned int num_clk_levels;unsigned long *clk_perf_tbl; +struct geni_icc_path to_core; +struct geni_icc_path from_cpu; +struct geni_icc_path to_ddr; };/* Common SE registers */
@@ -229,6 +243,21 @@ struct geni_se {#define GENI_SE_VERSION_MINOR (ver) ((ver & HW_VER_MINOR_MASK) >> HW_VER_MINOR_SHFT)#define GENI_SE_VERSION_STEP (ver) (ver & HW_VER_STEP_MASK) +/* +* Define bandwidth thresholds that cause the underlying Core 2X interconnect +* clock to run at the named frequency. These baseline values are recommended +* by the hardware team, and are not dynamically scaled with GENI bandwidth +* beyond basic on/off. +*/ +#define CORE_2X_19_2_MHZ 960 +#define CORE_2X_50_MHZ 2500 +#define CORE_2X_100_MHZ 5000 +#define CORE_2X_150_MHZ 7500 +#define CORE_2X_200_MHZ 10000 +#define CORE_2X_236_MHZ 16383 + +#define GENI_DEFAULT_BW Bps_to_icc(1000) +#if IS_ENABLED (CONFIG_QCOM_GENI_SE )u32 geni_se_get_qup_hw_version (struct geni_se *se);
@@ -416,5 +445,12 @@ int geni_se_rx_dma_prep(struct geni_se *se, void *buf, size_t len,void geni_se_tx_dma_unprep (struct geni_se *se,dma_addr_t iova,size_t len);void geni_se_rx_dma_unprep (struct geni_se *se,dma_addr_t iova,size_t len); + +int geni_icc_get (struct geni_se *se,const char *icc_core,const char *icc_cpu, +const char *icc_ddr); + +int geni_icc_vote_on (struct geni_se *se); + +int geni_icc_vote_off (struct geni_se *se);#endif #endif
Signature · 1 line
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,\na Linux Foundation Collaborative Project
PATCH v3 4/8 i2c: i2c-qcom-geni: Add interconnect support
In reply to Akash AsthanaGet the interconnect paths for I2C based Serial Engine device and vote according to the bus speed of the driver.
---
Changes in V2:
- As per Bjorn's comment, removed se == NULL check from geni_i2c_icc_get
- As per Bjorn's comment, removed code to set se->icc_path* to NULL in failure
- As per Bjorn's comment, introduced and using devm_of_icc_get API for getting path handle
- As per Matthias comment, added error handling for icc_set_bw call
Changes in V3:
- As per Matthias comment, use common library APIs defined in geni-se driver for ICC functionality.
drivers/i2c/busses/i2c-qcom-geni.c | 30 +++++++++++++++++++++++++++++- 1 file changed, 29 insertions(+), 1 deletion(-)
drivers/i2c/busses/i2c-qcom-geni.c +29 -1
@@ -557,6 +557,26 @@ static int geni_i2c_probe(struct platform_device *pdev) gi2c->adap.dev.of_node = dev->of_node;strlcpy (gi2c->adap.name,"Geni-I2C" ,sizeof (gi2c->adap.name)); + ret =geni_icc_get (&gi2c->se,"qup-core" ,"qup-config" ,"qup-memory" ); +if (ret) +return ret; +/* +* Set the bus quota for core and cpu to a reasonable value for +* register access. +* Set quota for DDR based on bus speed, assume peak requirement +* as twice of avg bw. +*/ + gi2c->se.to_core.avg_bw =GENI_DEFAULT_BW ; + gi2c->se.to_core.peak_bw =Bps_to_icc (CORE_2X_100_MHZ ); + gi2c->se.from_cpu.avg_bw =GENI_DEFAULT_BW ; + gi2c->se.from_cpu.peak_bw =GENI_DEFAULT_BW ; + gi2c->se.to_ddr.avg_bw =Bps_to_icc (gi2c->clk_freq_out); + gi2c->se.to_ddr.peak_bw =Bps_to_icc (2 * gi2c->clk_freq_out); + + ret =geni_icc_vote_on (&gi2c->se); +if (ret) +return ret; + ret =geni_se_resources_on (&gi2c->se);if (ret) {dev_err (dev,"Error turning on resources %d\n" , ret);
@@ -579,6 +599,10 @@ static int geni_i2c_probe(struct platform_device *pdev)return ret; } + ret =geni_icc_vote_off (&gi2c->se); +if (ret) +return ret; +dev_dbg (dev,"i2c fifo/se-dma mode. fifo depth:%d\n" , tx_depth); gi2c->suspended =1 ;
@@ -623,7 +647,7 @@ static int __maybe_unused geni_i2c_runtime_suspend(struct device *dev) gi2c->suspended =1 ; } -return 0 ; +return geni_icc_vote_off (&gi2c->se); }static int __maybe_unused geni_i2c_runtime_resume (struct device *dev)
@@ -631,6 +655,10 @@ static int __maybe_unused geni_i2c_runtime_resume(struct device *dev)int ret;struct geni_i2c_dev *gi2c =dev_get_drvdata (dev); + ret =geni_icc_vote_on (&gi2c->se); +if (ret) +return ret; + ret =geni_se_resources_on (&gi2c->se);if (ret)return ret;
Signature · 1 line
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,\na Linux Foundation Collaborative Project
PATCH v3 6/8 tty: serial: qcom_geni_serial: Add interconnect support
In reply to Akash AsthanaGet the interconnect paths for Uart based Serial Engine device and vote according to the baud rate requirement of the driver.
---
Changes in V2:
- As per Bjorn's comment, removed se == NULL check from geni_serial_icc_get
- As per Bjorn's comment, removed code to set se->icc_path* to NULL in failure
- As per Bjorn's comment, introduced and using devm_of_icc_get API for getting path handle
- As per Matthias comment, added error handling for icc_set_bw call
Changes in V3:
- As per Matthias comment, use common library APIs defined in geni-se driver for ICC functionality.
drivers/tty/serial/qcom_geni_serial.c | 28 +++++++++++++++++++++++++--- 1 file changed, 25 insertions(+), 3 deletions(-)
drivers/tty/serial/qcom_geni_serial.c +25 -3
@@ -965,6 +965,14 @@ static void qcom_geni_serial_set_termios(struct uart_port *uport, ser_clk_cfg =SER_CLK_EN ; ser_clk_cfg |= clk_div <<CLK_DIV_SHFT ; +/* +* Bump up BW vote on CPU path as driver supports FIFO mode only. +* Assume peak_bw as twice of avg_bw. +*/ + port->se.from_cpu.avg_bw =Bps_to_icc (baud); + port->se.from_cpu.peak_bw =Bps_to_icc (2 * baud); +geni_icc_vote_on (&port->se); +/* parity */ tx_trans_cfg =readl (uport->membase +SE_UART_TX_TRANS_CFG ); tx_parity_cfg =readl (uport->membase +SE_UART_TX_PARITY_CFG );
@@ -1202,11 +1210,14 @@ static void qcom_geni_serial_pm(struct uart_port *uport,if (old_state ==UART_PM_STATE_UNDEFINED ) old_state =UART_PM_STATE_OFF ; -if (new_state ==UART_PM_STATE_ON && old_state ==UART_PM_STATE_OFF ) +if (new_state ==UART_PM_STATE_ON && old_state ==UART_PM_STATE_OFF ) { +geni_icc_vote_on (&port->se);geni_se_resources_on (&port->se); -else if (new_state ==UART_PM_STATE_OFF && - old_state ==UART_PM_STATE_ON ) + }else if (new_state ==UART_PM_STATE_OFF && + old_state ==UART_PM_STATE_ON ) {geni_se_resources_off (&port->se); +geni_icc_vote_off (&port->se); + } }static const struct uart_ops qcom_geni_console_pops = {
@@ -1304,6 +1315,17 @@ static int qcom_geni_serial_probe(struct platform_device *pdev)return -ENOMEM ; } + ret =geni_icc_get (&port->se,"qup-core" ,"qup-config" ,NULL ); +if (ret) +return ret; +/* Set the bus quota to a reasonable value */ + port->se.to_core.avg_bw = console ?GENI_DEFAULT_BW : +Bps_to_icc (CORE_2X_50_MHZ ); + port->se.to_core.peak_bw = console ?GENI_DEFAULT_BW : +Bps_to_icc (CORE_2X_100_MHZ ); + port->se.from_cpu.avg_bw =GENI_DEFAULT_BW ; + port->se.from_cpu.peak_bw =GENI_DEFAULT_BW ; + port->name =devm_kasprintf (uport->dev,GFP_KERNEL ,"qcom_geni_serial_%s%d" ,uart_console (uport) ?"console" :"uart" , uport->line);
Signature · 1 line
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,\na Linux Foundation Collaborative Project
PATCH v3 8/8 arm64: dts: sc7180: Add interconnect for QUP and QSPI
In reply to Akash AsthanaAdd interconnect ports for GENI QUPs and QSPI to set bus capabilities.
---
Changes in V2:
- As per Bjorn's comment, ignoring 80 char limit in defining interconnects paths.
Changes in V3:
- No change.
Note: This won't compile without [email protected]/patch/11430403
[arm64: dts: sc7180: Include interconnect definitions]
arch/arm64/boot/dts/qcom/sc7180.dtsi | 127 +++++++++++++++++++++++++++++++++++ 1 file changed, 127 insertions(+)
arch/arm64/boot/dts/qcom/sc7180.dtsi +127 -0
@@ -427,6 +427,8 @@ #size-cells = <2>; ranges; iommus = <&apps_smmu 0x43 0x0>; + interconnects = <&qup_virt MASTER_QUP_CORE_0 &qup_virt SLAVE_QUP_CORE_0>; + interconnect-names = "qup-core"; status = "disabled"; i2c0: i2c@880000 {
@@ -439,6 +441,11 @@ interrupts = <GIC_SPI 601 IRQ_TYPE_LEVEL_HIGH>; #address-cells = <1>; #size-cells = <0>; + interconnects = <&qup_virt MASTER_QUP_CORE_0 &qup_virt SLAVE_QUP_CORE_0>, + <&gem_noc MASTER_APPSS_PROC &config_noc SLAVE_QUP_0>, + <&aggre1_noc MASTER_QUP_0 &mc_virt SLAVE_EBI1>; + interconnect-names = "qup-core", "qup-config", + "qup-memory"; status = "disabled"; };
@@ -452,6 +459,9 @@ interrupts = <GIC_SPI 601 IRQ_TYPE_LEVEL_HIGH>; #address-cells = <1>; #size-cells = <0>; + interconnects = <&qup_virt MASTER_QUP_CORE_0 &qup_virt SLAVE_QUP_CORE_0>, + <&gem_noc MASTER_APPSS_PROC &config_noc SLAVE_QUP_0>; + interconnect-names = "qup-core", "qup-config"; status = "disabled"; };
@@ -463,6 +473,9 @@ pinctrl-names = "default"; pinctrl-0 = <&qup_uart0_default>; interrupts = <GIC_SPI 601 IRQ_TYPE_LEVEL_HIGH>; + interconnects = <&qup_virt MASTER_QUP_CORE_0 &qup_virt SLAVE_QUP_CORE_0>, + <&gem_noc MASTER_APPSS_PROC &config_noc SLAVE_QUP_0>; + interconnect-names = "qup-core", "qup-config"; status = "disabled"; };
@@ -476,6 +489,11 @@ interrupts = <GIC_SPI 602 IRQ_TYPE_LEVEL_HIGH>; #address-cells = <1>; #size-cells = <0>; + interconnects = <&qup_virt MASTER_QUP_CORE_0 &qup_virt SLAVE_QUP_CORE_0>, + <&gem_noc MASTER_APPSS_PROC &config_noc SLAVE_QUP_0>, + <&aggre1_noc MASTER_QUP_0 &mc_virt SLAVE_EBI1>; + interconnect-names = "qup-core", "qup-config", + "qup-memory"; status = "disabled"; };
@@ -489,6 +507,9 @@ interrupts = <GIC_SPI 602 IRQ_TYPE_LEVEL_HIGH>; #address-cells = <1>; #size-cells = <0>; + interconnects = <&qup_virt MASTER_QUP_CORE_0 &qup_virt SLAVE_QUP_CORE_0>, + <&gem_noc MASTER_APPSS_PROC &config_noc SLAVE_QUP_0>; + interconnect-names = "qup-core", "qup-config"; status = "disabled"; };
@@ -500,6 +521,9 @@ pinctrl-names = "default"; pinctrl-0 = <&qup_uart1_default>; interrupts = <GIC_SPI 602 IRQ_TYPE_LEVEL_HIGH>; + interconnects = <&qup_virt MASTER_QUP_CORE_0 &qup_virt SLAVE_QUP_CORE_0>, + <&gem_noc MASTER_APPSS_PROC &config_noc SLAVE_QUP_0>; + interconnect-names = "qup-core", "qup-config"; status = "disabled"; };
@@ -513,6 +537,11 @@ interrupts = <GIC_SPI 603 IRQ_TYPE_LEVEL_HIGH>; #address-cells = <1>; #size-cells = <0>; + interconnects = <&qup_virt MASTER_QUP_CORE_0 &qup_virt SLAVE_QUP_CORE_0>, + <&gem_noc MASTER_APPSS_PROC &config_noc SLAVE_QUP_0>, + <&aggre1_noc MASTER_QUP_0 &mc_virt SLAVE_EBI1>; + interconnect-names = "qup-core", "qup-config", + "qup-memory"; status = "disabled"; };
@@ -524,6 +553,9 @@ pinctrl-names = "default"; pinctrl-0 = <&qup_uart2_default>; interrupts = <GIC_SPI 603 IRQ_TYPE_LEVEL_HIGH>; + interconnects = <&qup_virt MASTER_QUP_CORE_0 &qup_virt SLAVE_QUP_CORE_0>, + <&gem_noc MASTER_APPSS_PROC &config_noc SLAVE_QUP_0>; + interconnect-names = "qup-core", "qup-config"; status = "disabled"; };
@@ -537,6 +569,11 @@ interrupts = <GIC_SPI 604 IRQ_TYPE_LEVEL_HIGH>; #address-cells = <1>; #size-cells = <0>; + interconnects = <&qup_virt MASTER_QUP_CORE_0 &qup_virt SLAVE_QUP_CORE_0>, + <&gem_noc MASTER_APPSS_PROC &config_noc SLAVE_QUP_0>, + <&aggre1_noc MASTER_QUP_0 &mc_virt SLAVE_EBI1>; + interconnect-names = "qup-core", "qup-config", + "qup-memory"; status = "disabled"; };
@@ -550,6 +587,9 @@ interrupts = <GIC_SPI 604 IRQ_TYPE_LEVEL_HIGH>; #address-cells = <1>; #size-cells = <0>; + interconnects = <&qup_virt MASTER_QUP_CORE_0 &qup_virt SLAVE_QUP_CORE_0>, + <&gem_noc MASTER_APPSS_PROC &config_noc SLAVE_QUP_0>; + interconnect-names = "qup-core", "qup-config"; status = "disabled"; };
@@ -561,6 +601,9 @@ pinctrl-names = "default"; pinctrl-0 = <&qup_uart3_default>; interrupts = <GIC_SPI 604 IRQ_TYPE_LEVEL_HIGH>; + interconnects = <&qup_virt MASTER_QUP_CORE_0 &qup_virt SLAVE_QUP_CORE_0>, + <&gem_noc MASTER_APPSS_PROC &config_noc SLAVE_QUP_0>; + interconnect-names = "qup-core", "qup-config"; status = "disabled"; };
@@ -574,6 +617,11 @@ interrupts = <GIC_SPI 605 IRQ_TYPE_LEVEL_HIGH>; #address-cells = <1>; #size-cells = <0>; + interconnects = <&qup_virt MASTER_QUP_CORE_0 &qup_virt SLAVE_QUP_CORE_0>, + <&gem_noc MASTER_APPSS_PROC &config_noc SLAVE_QUP_0>, + <&aggre1_noc MASTER_QUP_0 &mc_virt SLAVE_EBI1>; + interconnect-names = "qup-core", "qup-config", + "qup-memory"; status = "disabled"; };
@@ -585,6 +633,9 @@ pinctrl-names = "default"; pinctrl-0 = <&qup_uart4_default>; interrupts = <GIC_SPI 605 IRQ_TYPE_LEVEL_HIGH>; + interconnects = <&qup_virt MASTER_QUP_CORE_0 &qup_virt SLAVE_QUP_CORE_0>, + <&gem_noc MASTER_APPSS_PROC &config_noc SLAVE_QUP_0>; + interconnect-names = "qup-core", "qup-config"; status = "disabled"; };
@@ -598,6 +649,11 @@ interrupts = <GIC_SPI 606 IRQ_TYPE_LEVEL_HIGH>; #address-cells = <1>; #size-cells = <0>; + interconnects = <&qup_virt MASTER_QUP_CORE_0 &qup_virt SLAVE_QUP_CORE_0>, + <&gem_noc MASTER_APPSS_PROC &config_noc SLAVE_QUP_0>, + <&aggre1_noc MASTER_QUP_0 &mc_virt SLAVE_EBI1>; + interconnect-names = "qup-core", "qup-config", + "qup-memory"; status = "disabled"; };
@@ -611,6 +667,9 @@ interrupts = <GIC_SPI 606 IRQ_TYPE_LEVEL_HIGH>; #address-cells = <1>; #size-cells = <0>; + interconnects = <&qup_virt MASTER_QUP_CORE_0 &qup_virt SLAVE_QUP_CORE_0>, + <&gem_noc MASTER_APPSS_PROC &config_noc SLAVE_QUP_0>; + interconnect-names = "qup-core", "qup-config"; status = "disabled"; };
@@ -622,6 +681,9 @@ pinctrl-names = "default"; pinctrl-0 = <&qup_uart5_default>; interrupts = <GIC_SPI 606 IRQ_TYPE_LEVEL_HIGH>; + interconnects = <&qup_virt MASTER_QUP_CORE_0 &qup_virt SLAVE_QUP_CORE_0>, + <&gem_noc MASTER_APPSS_PROC &config_noc SLAVE_QUP_0>; + interconnect-names = "qup-core", "qup-config"; status = "disabled"; }; };
@@ -636,6 +698,8 @@ #size-cells = <2>; ranges; iommus = <&apps_smmu 0x4c3 0x0>; + interconnects = <&qup_virt MASTER_QUP_CORE_1 &qup_virt SLAVE_QUP_CORE_1>; + interconnect-names = "qup-core"; status = "disabled"; i2c6: i2c@a80000 {
@@ -648,6 +712,11 @@ interrupts = <GIC_SPI 353 IRQ_TYPE_LEVEL_HIGH>; #address-cells = <1>; #size-cells = <0>; + interconnects = <&qup_virt MASTER_QUP_CORE_1 &qup_virt SLAVE_QUP_CORE_1>, + <&gem_noc MASTER_APPSS_PROC &config_noc SLAVE_QUP_1>, + <&aggre2_noc MASTER_QUP_1 &mc_virt SLAVE_EBI1>; + interconnect-names = "qup-core", "qup-config", + "qup-memory"; status = "disabled"; };
@@ -661,6 +730,9 @@ interrupts = <GIC_SPI 353 IRQ_TYPE_LEVEL_HIGH>; #address-cells = <1>; #size-cells = <0>; + interconnects = <&qup_virt MASTER_QUP_CORE_1 &qup_virt SLAVE_QUP_CORE_1>, + <&gem_noc MASTER_APPSS_PROC &config_noc SLAVE_QUP_1>; + interconnect-names = "qup-core", "qup-config"; status = "disabled"; };
@@ -672,6 +744,9 @@ pinctrl-names = "default"; pinctrl-0 = <&qup_uart6_default>; interrupts = <GIC_SPI 353 IRQ_TYPE_LEVEL_HIGH>; + interconnects = <&qup_virt MASTER_QUP_CORE_1 &qup_virt SLAVE_QUP_CORE_1>, + <&gem_noc MASTER_APPSS_PROC &config_noc SLAVE_QUP_1>; + interconnect-names = "qup-core", "qup-config"; status = "disabled"; };
@@ -685,6 +760,11 @@ interrupts = <GIC_SPI 354 IRQ_TYPE_LEVEL_HIGH>; #address-cells = <1>; #size-cells = <0>; + interconnects = <&qup_virt MASTER_QUP_CORE_1 &qup_virt SLAVE_QUP_CORE_1>, + <&gem_noc MASTER_APPSS_PROC &config_noc SLAVE_QUP_1>, + <&aggre2_noc MASTER_QUP_1 &mc_virt SLAVE_EBI1>; + interconnect-names = "qup-core", "qup-config", + "qup-memory"; status = "disabled"; };
@@ -696,6 +776,9 @@ pinctrl-names = "default"; pinctrl-0 = <&qup_uart7_default>; interrupts = <GIC_SPI 354 IRQ_TYPE_LEVEL_HIGH>; + interconnects = <&qup_virt MASTER_QUP_CORE_1 &qup_virt SLAVE_QUP_CORE_1>, + <&gem_noc MASTER_APPSS_PROC &config_noc SLAVE_QUP_1>; + interconnect-names = "qup-core", "qup-config"; status = "disabled"; };
@@ -709,6 +792,11 @@ interrupts = <GIC_SPI 355 IRQ_TYPE_LEVEL_HIGH>; #address-cells = <1>; #size-cells = <0>; + interconnects = <&qup_virt MASTER_QUP_CORE_1 &qup_virt SLAVE_QUP_CORE_1>, + <&gem_noc MASTER_APPSS_PROC &config_noc SLAVE_QUP_1>, + <&aggre2_noc MASTER_QUP_1 &mc_virt SLAVE_EBI1>; + interconnect-names = "qup-core", "qup-config", + "qup-memory"; status = "disabled"; };
@@ -722,6 +810,9 @@ interrupts = <GIC_SPI 355 IRQ_TYPE_LEVEL_HIGH>; #address-cells = <1>; #size-cells = <0>; + interconnects = <&qup_virt MASTER_QUP_CORE_1 &qup_virt SLAVE_QUP_CORE_1>, + <&gem_noc MASTER_APPSS_PROC &config_noc SLAVE_QUP_1>; + interconnect-names = "qup-core", "qup-config"; status = "disabled"; };
@@ -733,6 +824,9 @@ pinctrl-names = "default"; pinctrl-0 = <&qup_uart8_default>; interrupts = <GIC_SPI 355 IRQ_TYPE_LEVEL_HIGH>; + interconnects = <&qup_virt MASTER_QUP_CORE_1 &qup_virt SLAVE_QUP_CORE_1>, + <&gem_noc MASTER_APPSS_PROC &config_noc SLAVE_QUP_1>; + interconnect-names = "qup-core", "qup-config"; status = "disabled"; };
@@ -746,6 +840,11 @@ interrupts = <GIC_SPI 356 IRQ_TYPE_LEVEL_HIGH>; #address-cells = <1>; #size-cells = <0>; + interconnects = <&qup_virt MASTER_QUP_CORE_1 &qup_virt SLAVE_QUP_CORE_1>, + <&gem_noc MASTER_APPSS_PROC &config_noc SLAVE_QUP_1>, + <&aggre2_noc MASTER_QUP_1 &mc_virt SLAVE_EBI1>; + interconnect-names = "qup-core", "qup-config", + "qup-memory"; status = "disabled"; };
@@ -757,6 +856,9 @@ pinctrl-names = "default"; pinctrl-0 = <&qup_uart9_default>; interrupts = <GIC_SPI 356 IRQ_TYPE_LEVEL_HIGH>; + interconnects = <&qup_virt MASTER_QUP_CORE_1 &qup_virt SLAVE_QUP_CORE_1>, + <&gem_noc MASTER_APPSS_PROC &config_noc SLAVE_QUP_1>; + interconnect-names = "qup-core", "qup-config"; status = "disabled"; };
@@ -770,6 +872,11 @@ interrupts = <GIC_SPI 357 IRQ_TYPE_LEVEL_HIGH>; #address-cells = <1>; #size-cells = <0>; + interconnects = <&qup_virt MASTER_QUP_CORE_1 &qup_virt SLAVE_QUP_CORE_1>, + <&gem_noc MASTER_APPSS_PROC &config_noc SLAVE_QUP_1>, + <&aggre2_noc MASTER_QUP_1 &mc_virt SLAVE_EBI1>; + interconnect-names = "qup-core", "qup-config", + "qup-memory"; status = "disabled"; };
@@ -783,6 +890,9 @@ interrupts = <GIC_SPI 357 IRQ_TYPE_LEVEL_HIGH>; #address-cells = <1>; #size-cells = <0>; + interconnects = <&qup_virt MASTER_QUP_CORE_1 &qup_virt SLAVE_QUP_CORE_1>, + <&gem_noc MASTER_APPSS_PROC &config_noc SLAVE_QUP_1>; + interconnect-names = "qup-core", "qup-config"; status = "disabled"; };
@@ -794,6 +904,9 @@ pinctrl-names = "default"; pinctrl-0 = <&qup_uart10_default>; interrupts = <GIC_SPI 357 IRQ_TYPE_LEVEL_HIGH>; + interconnects = <&qup_virt MASTER_QUP_CORE_1 &qup_virt SLAVE_QUP_CORE_1>, + <&gem_noc MASTER_APPSS_PROC &config_noc SLAVE_QUP_1>; + interconnect-names = "qup-core", "qup-config"; status = "disabled"; };
@@ -807,6 +920,11 @@ interrupts = <GIC_SPI 358 IRQ_TYPE_LEVEL_HIGH>; #address-cells = <1>; #size-cells = <0>; + interconnects = <&qup_virt MASTER_QUP_CORE_1 &qup_virt SLAVE_QUP_CORE_1>, + <&gem_noc MASTER_APPSS_PROC &config_noc SLAVE_QUP_1>, + <&aggre2_noc MASTER_QUP_1 &mc_virt SLAVE_EBI1>; + interconnect-names = "qup-core", "qup-config", + "qup-memory"; status = "disabled"; };
@@ -820,6 +938,9 @@ interrupts = <GIC_SPI 358 IRQ_TYPE_LEVEL_HIGH>; #address-cells = <1>; #size-cells = <0>; + interconnects = <&qup_virt MASTER_QUP_CORE_1 &qup_virt SLAVE_QUP_CORE_1>, + <&gem_noc MASTER_APPSS_PROC &config_noc SLAVE_QUP_1>; + interconnect-names = "qup-core", "qup-config"; status = "disabled"; };
@@ -831,6 +952,9 @@ pinctrl-names = "default"; pinctrl-0 = <&qup_uart11_default>; interrupts = <GIC_SPI 358 IRQ_TYPE_LEVEL_HIGH>; + interconnects = <&qup_virt MASTER_QUP_CORE_1 &qup_virt SLAVE_QUP_CORE_1>, + <&gem_noc MASTER_APPSS_PROC &config_noc SLAVE_QUP_1>; + interconnect-names = "qup-core", "qup-config"; status = "disabled"; }; };
@@ -1335,6 +1459,9 @@ clocks = <&gcc GCC_QSPI_CNOC_PERIPH_AHB_CLK>, <&gcc GCC_QSPI_CORE_CLK>; clock-names = "iface", "core"; + interconnects = <&gem_noc MASTER_APPSS_PROC + &config_noc SLAVE_QSPI_0>; + interconnect-names = "qspi-config"; status = "disabled"; };
Signature · 1 line
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,\na Linux Foundation Collaborative Project
PATCH v3 3/8 soc: qcom-geni-se: Add interconnect support to fix earlycon crash
In reply to Akash AsthanaQUP core clock is shared among all the SE drivers present on particular QUP wrapper, the system will reset(unclocked access) if earlycon used after QUP core clock is put to 0 from other SE drivers before real console comes up.
As earlycon can't vote for it's QUP core need, to fix this add ICC support to common/QUP wrapper driver and put vote for QUP core from probe on behalf of earlycon and remove vote during earlycon exit call.
---
Change is V3:
- Add geni_remove_earlycon_icc_vote API that will be used by earlycon exit function to remove ICC vote for earlyconsole.
- Remove suspend/resume hook for geni-se driver as we are no longer removing earlyconsole ICC vote from system suspend, we are removing from earlycon exit.
drivers/soc/qcom/qcom-geni-se.c | 51 +++++++++++++++++++++++++++++++++++ drivers/tty/serial/qcom_geni_serial.c | 7 +++++ include/linux/qcom-geni-se.h | 2 ++ 3 files changed, 60 insertions(+)
drivers/soc/qcom/qcom-geni-se.c +51 -0
@@ -90,8 +90,11 @@ struct geni_wrapper {struct device *dev;void __iomem *base;struct clk_bulk_data ahb_clks[NUM_AHB_CLKS ]; +struct geni_icc_path to_core; }; +struct geni_wrapper *earlycon_wrapper; +#define QUP_HW_VER_REG 0x4/* Common SE registers */
@@ -818,6 +821,26 @@ int geni_icc_vote_off(struct geni_se *se) }EXPORT_SYMBOL (geni_icc_vote_off); +void geni_remove_earlycon_icc_vote (void ) +{ +struct geni_wrapper *wrapper = earlycon_wrapper; +struct device_node *parent =of_get_next_parent (wrapper->dev->of_node); +struct device_node *child; + +for_each_child_of_node (parent, child) { +if (of_device_is_compatible (child,"qcom,geni-se-qup" )) { + wrapper =platform_get_drvdata (of_find_device_by_node ( + child)); +icc_put (wrapper->to_core.path); + wrapper->to_core.path =NULL ; + } + } +of_node_put (parent); + + earlycon_wrapper =NULL ; +} +EXPORT_SYMBOL (geni_remove_earlycon_icc_vote); +static int geni_se_probe (struct platform_device *pdev) {struct device *dev = &pdev->dev;
@@ -845,6 +868,34 @@ static int geni_se_probe(struct platform_device *pdev) } } +#ifdef CONFIG_SERIAL_EARLYCON + wrapper->to_core.path =devm_of_icc_get (dev,"qup-core" ); +if (IS_ERR (wrapper->to_core.path)) +return PTR_ERR (wrapper->to_core.path); +/* +* Put minmal BW request on core clocks on behalf of early console. +* The vote will be removed earlycon exit function. +* +* Note: We are putting vote on each QUP wrapper instead only to which +* earlycon is connected because QUP core clock of different wrapper +* share same voltage domain. If core1 is put to 0, then core2 will +* also run at 0, if not voted. Default ICC vote will be removed ASA +* we touch any of the core clock. +* core1 = core2 = max(core1, core2) +*/ + ret =icc_set_bw (wrapper->to_core.path,GENI_DEFAULT_BW , +GENI_DEFAULT_BW ); +if (ret) { +dev_err (&pdev->dev,"%s: ICC BW voting failed for core\n" , + __func__); +return ret; + } + +if (of_get_compatible_child (pdev->dev.of_node,"qcom,geni-debug-uart" )) + earlycon_wrapper = wrapper; +of_node_put (pdev->dev.of_node); +#endif +dev_set_drvdata (dev, wrapper);dev_dbg (dev,"GENI SE Driver probed\n" );return devm_of_platform_populate (dev);
drivers/tty/serial/qcom_geni_serial.c +7 -0
@@ -1090,6 +1090,12 @@ static void qcom_geni_serial_earlycon_write(struct console *con,__qcom_geni_serial_console_write (&dev->port, s, n); } +static int qcom_geni_serial_earlycon_exit (struct console *con) +{ +geni_remove_earlycon_icc_vote (); +return 0 ; +} +static int __init qcom_geni_serial_earlycon_setup (struct earlycon_device *dev,const char *opt) {
@@ -1135,6 +1141,7 @@ static int __init qcom_geni_serial_earlycon_setup(struct earlycon_device *dev,writel (stop_bit_len, uport->membase +SE_UART_TX_STOP_BIT_LEN ); dev->con->write = qcom_geni_serial_earlycon_write; + dev->con->exit = qcom_geni_serial_earlycon_exit; dev->con->setup =NULL ;return 0 ; }
include/linux/qcom-geni-se.h +2 -0
@@ -452,5 +452,7 @@ int geni_icc_get(struct geni_se *se, const char *icc_core, const char *icc_cpu,int geni_icc_vote_on (struct geni_se *se);int geni_icc_vote_off (struct geni_se *se); + +void geni_remove_earlycon_icc_vote (void );#endif #endif
Signature · 1 line
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,\na Linux Foundation Collaborative Project
PATCH v3 5/8 spi: spi-geni-qcom: Add interconnect support
In reply to Akash AsthanaGet the interconnect paths for SPI based Serial Engine device and vote according to the current bus speed of the driver.
---
Changes in V2:
- As per Bjorn's comment, removed se == NULL check from geni_spi_icc_get
- As per Bjorn's comment, removed code to set se->icc_path* to NULL in failure
- As per Bjorn's comment, introduced and using devm_of_icc_get API for getting path handle
- As per Matthias comment, added error handling for icc_set_bw call
Changes in V3:
- As per Matthias's comment, use helper ICC function from geni-se driver.
drivers/spi/spi-geni-qcom.c | 31 ++++++++++++++++++++++++++++++- 1 file changed, 30 insertions(+), 1 deletion(-)
drivers/spi/spi-geni-qcom.c +30 -1
@@ -234,6 +234,16 @@ static int setup_fifo_params(struct spi_device *spi_slv,return ret; } +/* +* Set BW quota for CPU as driver supports FIFO mode only. +* Assume peak bw as twice of avg bw. +*/ + se->from_cpu.avg_bw =Bps_to_icc (mas->cur_speed_hz); + se->from_cpu.peak_bw =Bps_to_icc (2 * mas->cur_speed_hz); + ret =geni_icc_vote_on (se); +if (ret) +return ret; + clk_sel = idx &CLK_SEL_MSK ; m_clk_cfg = (div <<CLK_DIV_SHFT ) |SER_CLK_EN ;spi_setup_word_len (mas, spi_slv->mode, spi_slv->bits_per_word);
@@ -578,6 +588,15 @@ static int spi_geni_probe(struct platform_device *pdev)spin_lock_init (&mas->lock);pm_runtime_enable (dev); + ret =geni_icc_get (&mas->se,"qup-core" ,"qup-config" ,NULL ); +if (ret) + goto spi_geni_probe_runtime_disable; +/* Set the bus quota to a reasonable value for register access */ + mas->se.to_core.avg_bw =Bps_to_icc (CORE_2X_50_MHZ ); + mas->se.to_core.peak_bw =Bps_to_icc (CORE_2X_100_MHZ ); + mas->se.from_cpu.avg_bw =GENI_DEFAULT_BW ; + mas->se.from_cpu.peak_bw =GENI_DEFAULT_BW ; + ret =spi_geni_init (mas);if (ret) goto spi_geni_probe_runtime_disable;
@@ -616,14 +635,24 @@ static int __maybe_unused spi_geni_runtime_suspend(struct device *dev) {struct spi_master *spi =dev_get_drvdata (dev);struct spi_geni_master *mas =spi_master_get_devdata (spi); +int ret; + + ret =geni_se_resources_off (&mas->se); +if (ret) +return ret; -return geni_se_resources_off (&mas->se); +return geni_icc_vote_off (&mas->se); }static int __maybe_unused spi_geni_runtime_resume (struct device *dev) {struct spi_master *spi =dev_get_drvdata (dev);struct spi_geni_master *mas =spi_master_get_devdata (spi); +int ret; + + ret =geni_icc_vote_on (&mas->se); +if (ret) +return ret;return geni_se_resources_on (&mas->se); }
Signature · 1 line
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,\na Linux Foundation Collaborative Project
PATCH v3 7/8 spi: spi-qcom-qspi: Add interconnect support
In reply to Akash AsthanaGet the interconnect paths for QSPI device and vote according to the current bus speed of the driver.
---
Changes in V2:
- As per Bjorn's comment, introduced and using devm_of_icc_get API for getting path handle
- As per Matthias comment, added error handling for icc_set_bw call
Changes in V3:
- No Change.
drivers/spi/spi-qcom-qspi.c | 46 ++++++++++++++++++++++++++++++++++++++++++++- 1 file changed, 45 insertions(+), 1 deletion(-)
drivers/spi/spi-qcom-qspi.c +45 -1
@@ -2,6 +2,7 @@// Copyright (c) 2017-2018, The Linux foundation. All rights reserved. #include <linux/clk.h> +#include <linux/interconnect.h> #include <linux/interrupt.h> #include <linux/io.h> #include <linux/module.h>
@@ -139,7 +140,10 @@ struct qcom_qspi {struct device *dev;struct clk_bulk_data *clks;struct qspi_xfer xfer; -/* Lock to protect xfer and IRQ accessed registers */ +struct icc_path *icc_path_cpu_to_qspi; +unsigned int avg_bw_cpu; +unsigned int peak_bw_cpu; +/* Lock to protect data accessed by IRQs */ spinlock_t lock; };
@@ -241,6 +245,20 @@ static int qcom_qspi_transfer_one(struct spi_master *master,return ret; } +/* +* Set BW quota for CPU as driver supports FIFO mode only. +* Assume peak bw as twice of avg bw. +*/ + ctrl->avg_bw_cpu =Bps_to_icc (speed_hz); + ctrl->peak_bw_cpu =Bps_to_icc (2 * speed_hz); + ret =icc_set_bw (ctrl->icc_path_cpu_to_qspi, ctrl->avg_bw_cpu, + ctrl->peak_bw_cpu); +if (ret) { +dev_err (ctrl->dev,"%s: ICC BW voting failed for cpu\n" , + __func__); +return ret; + } +spin_lock_irqsave (&ctrl->lock, flags);/* We are half duplex, so either rx or tx will be set */
@@ -458,6 +476,15 @@ static int qcom_qspi_probe(struct platform_device *pdev)if (ret) goto exit_probe_master_put; + ctrl->icc_path_cpu_to_qspi =devm_of_icc_get (dev,"qspi-config" ); +if (IS_ERR (ctrl->icc_path_cpu_to_qspi)) { + ret =PTR_ERR (ctrl->icc_path_cpu_to_qspi); + goto exit_probe_master_put; + } +/* Put BW vote on CPU path for register access */ + ctrl->avg_bw_cpu =Bps_to_icc (1000 ); + ctrl->peak_bw_cpu =Bps_to_icc (1000 ); + ret =platform_get_irq (pdev,0 );if (ret <0 ) goto exit_probe_master_put;
@@ -511,9 +538,17 @@ static int __maybe_unused qcom_qspi_runtime_suspend(struct device *dev) {struct spi_master *master =dev_get_drvdata (dev);struct qcom_qspi *ctrl =spi_master_get_devdata (master); +int ret;clk_bulk_disable_unprepare (QSPI_NUM_CLKS , ctrl->clks); + ret =icc_set_bw (ctrl->icc_path_cpu_to_qspi,0 ,0 ); +if (ret) { +dev_err_ratelimited (ctrl->dev,"%s: ICC BW remove failed for cpu\n" , + __func__); +return ret; + } +return 0 ; }
@@ -521,6 +556,15 @@ static int __maybe_unused qcom_qspi_runtime_resume(struct device *dev) {struct spi_master *master =dev_get_drvdata (dev);struct qcom_qspi *ctrl =spi_master_get_devdata (master); +int ret; + + ret =icc_set_bw (ctrl->icc_path_cpu_to_qspi, ctrl->avg_bw_cpu, + ctrl->peak_bw_cpu); +if (ret) { +dev_err_ratelimited (ctrl->dev,"%s: ICC BW voting failed for cpu\n" , + __func__); +return ret; + }return clk_bulk_prepare_enable (QSPI_NUM_CLKS , ctrl->clks); }
Signature · 1 line
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,\na Linux Foundation Collaborative Project
Mark Brown
maintainerRe: [PATCH V3 7/8] spi: spi-qcom-qspi: Add interconnect support
drivers/spi/spi-qcom-qspi.c static int qcom_qspi_transfer_one(struct spi_master *master, · line 253
Quoted 6 lines · Akash Asthana
drivers/spi/spi-qcom-qspi.c +6
-0+/* +* Set BW quota for CPU as driver supports FIFO mode only. +* Assume peak bw as twice of avg bw. +*/ + ctrl->avg_bw_cpu =Bps_to_icc (speed_hz); + ctrl->peak_bw_cpu =Bps_to_icc (2 * speed_hz);
I thought you were going to factor this best guess handling of peak bandwidth out into the core?
Re: [PATCH V3 2/8] soc: qcom: geni: Support for ICC voting
In reply to Akash AsthanaHi Akash,
Quoted 73 lines · Akash Asthana
Add necessary macros and structure variables to support ICC BW voting from individual SE drivers.
Signed-off-by Akash Asthana <[email protected]>---
Changes in V2:
- As per Bjorn's comment dropped enums for ICC paths, given the three paths individual membersChanges in V3:
- Addgeni_icc_get,geni_icc_vote_onandgeni_icc_vote_offas helper API.
- Addgeni_icc_pathstructure in common headerdrivers/soc/qcom/qcom-geni-se.c | 98 +++++++++++++++++++++++++++++++++++++++++ include/linux/qcom-geni-se.h | 36 +++++++++++++++ 2 files changed, 134 insertions(+)drivers/soc/qcom/qcom-geni-se.c +48
-0@@ -720,6 +720,104 @@ void geni_se_rx_dma_unprep(struct geni_se *se, dma_addr_t iova, size_t len) }EXPORT_SYMBOL (geni_se_rx_dma_unprep); +int geni_icc_get (struct geni_se *se,const char *icc_core,const char *icc_cpu, +const char *icc_ddr) +{ +if (icc_core) { + se->to_core.path =devm_of_icc_get (se->dev,"qup-core" ); +if (IS_ERR (se->to_core.path)) +return PTR_ERR (se->to_core.path); + } + +if (icc_cpu) { + se->from_cpu.path =devm_of_icc_get (se->dev,"qup-config" ); +if (IS_ERR (se->from_cpu.path)) +return PTR_ERR (se->from_cpu.path); + } + +if (icc_ddr) { + se->to_ddr.path =devm_of_icc_get (se->dev,"qup-memory" ); +if (IS_ERR (se->to_ddr.path)) +return PTR_ERR (se->to_ddr.path); + } + +return 0 ; +} +EXPORT_SYMBOL (geni_icc_get); + +int geni_icc_vote_on (struct geni_se *se) +{ +int ret; + +if (se->to_core.path) { + ret =icc_set_bw (se->to_core.path, se->to_core.avg_bw, + se->to_core.peak_bw); +if (ret) { +dev_err_ratelimited (se->dev,"%s: ICC BW voting failed for core\n" , + __func__); +return ret; + } + } + +if (se->from_cpu.path) { + ret =icc_set_bw (se->from_cpu.path, se->from_cpu.avg_bw, + se->from_cpu.peak_bw); +if (ret) { +dev_err_ratelimited (se->dev,"%s: ICC BW voting failed for cpu\n" , + __func__); +return ret; + } + }
drivers/soc/qcom/qcom-geni-se.c void geni_se_rx_dma_unprep(struct geni_se *se, dma_addr_t iova, size_t len) · line 780
Quoted 10 lines
drivers/soc/qcom/qcom-geni-se.c +10
-0+ +if (se->to_ddr.path) { + ret =icc_set_bw (se->to_ddr.path, se->to_ddr.avg_bw, + se->to_ddr.peak_bw); +if (ret) { +dev_err_ratelimited (se->dev,"%s: ICC BW voting failed for ddr\n" , + __func__); +return ret; + } + }
With an array of 'struct geni_icc_path' pointers the above could be reduced to:
for (i = 0; i < ARRAY_SIZE(se->icc_paths); i++) {
if (!se->icc_paths[i])
continue;
ret = icc_set_bw(se->icc_paths[i]->path, se->icc_paths[i]->avg_bw,
se->icc_paths[i]->peak_bw);
if (ret) {
dev_err_ratelimited(se->dev, "%s: ICC BW voting failed\n",
__func__);
return ret;
}
}similar for geni_icc_vote_off()
It's just a suggestion, looks also good to me as is.
Re: [PATCH V3 3/8] soc: qcom-geni-se: Add interconnect support to fix earlycon crash
In reply to Akash AsthanaHi Akash,
Quoted 29 lines · Akash Asthana
QUP core clock is shared among all the SE drivers present on particular QUP wrapper, the system will reset(unclocked access) if earlycon used after QUP core clock is put to 0 from other SE drivers before real console comes up.
As earlycon can't vote for it's QUP core need, to fix this add ICC support to common/QUP wrapper driver and put vote for QUP core from probe on behalf of earlycon and remove vote during earlycon exit call.
Signed-off-by Akash Asthana <[email protected]>Reported-by Matthias Kaehlcke <[email protected]>---
Change is V3:
- Addgeni_remove_earlycon_icc_voteAPI that will be used by earlycon exit function to remove ICC vote for earlyconsole.
- Remove suspend/resume hook for geni-se driver as we are no longer removing earlyconsole ICC vote from system suspend, we are removing from earlycon exit.drivers/soc/qcom/qcom-geni-se.c | 51 +++++++++++++++++++++++++++++++++++ drivers/tty/serial/qcom_geni_serial.c | 7 +++++ include/linux/qcom-geni-se.h | 2 ++ 3 files changed, 60 insertions(+)drivers/soc/qcom/qcom-geni-se.c +0
-0
drivers/soc/qcom/qcom-geni-se.c struct geni_wrapper { · line 96
Quoted 7 lines
drivers/soc/qcom/qcom-geni-se.c +2
-0@@ -90,8 +90,11 @@ struct geni_wrapper {struct device *dev;void __iomem *base;struct clk_bulk_data ahb_clks[NUM_AHB_CLKS ]; +struct geni_icc_path to_core; }; +struct geni_wrapper *earlycon_wrapper;
should be static
Quoted 17 lines
drivers/soc/qcom/qcom-geni-se.c +10
-0+#define QUP_HW_VER_REG 0x4/* Common SE registers */ @@ -818,6 +821,26 @@ int geni_icc_vote_off(struct geni_se *se) }EXPORT_SYMBOL (geni_icc_vote_off); +void geni_remove_earlycon_icc_vote (void ) +{ +struct geni_wrapper *wrapper = earlycon_wrapper; +struct device_node *parent =of_get_next_parent (wrapper->dev->of_node); +struct device_node *child; + +for_each_child_of_node (parent, child) { +if (of_device_is_compatible (child,"qcom,geni-se-qup" )) { + wrapper = platform_get_drvdata(of_find_device_by_node(
drivers/soc/qcom/qcom-geni-se.c int geni_icc_vote_off(struct geni_se *se) · line 842
Quoted 10 lines
drivers/soc/qcom/qcom-geni-se.c +10
-0+ child)); +icc_put (wrapper->to_core.path); + wrapper->to_core.path =NULL ; + } + } +of_node_put (parent); + + earlycon_wrapper =NULL ; +} +EXPORT_SYMBOL (geni_remove_earlycon_icc_vote);
I didn't know that consoles have an exit handler, this is way nicer than the miscellaneous triggers we discussed earlier :)
Quoted 13 lines
drivers/soc/qcom/qcom-geni-se.c +6
-0+static int geni_se_probe (struct platform_device *pdev) {struct device *dev = &pdev->dev;@@ -845,6 +868,34 @@ static int geni_se_probe(struct platform_device *pdev) } } +#ifdef CONFIG_SERIAL_EARLYCON + wrapper->to_core.path =devm_of_icc_get (dev,"qup-core" ); +if (IS_ERR (wrapper->to_core.path)) +return PTR_ERR (wrapper->to_core.path); +/*
drivers/soc/qcom/qcom-geni-se.c static int geni_se_probe(struct platform_device *pdev) · line 885
Quoted 10 lines
drivers/soc/qcom/qcom-geni-se.c +10
-0+* Put minmal BW request on core clocks on behalf of early console. +* The vote will be removed earlycon exit function. +* +* Note: We are putting vote on each QUP wrapper instead only to which +* earlycon is connected because QUP core clock of different wrapper +* share same voltage domain. If core1 is put to 0, then core2 will +* also run at 0, if not voted. Default ICC vote will be removed ASA +* we touch any of the core clock. +* core1 = core2 = max(core1, core2) +*/
I don't really understand this part. According to the comment if we vote (let's say) for core2 but not for core1 then:
core1: 0
core2: GENI_DEFAULT_BW
core1 = core2 = max(core1, core2)
or
core1 = core2 = max(0, GENI_DEFAULT_BW)
hence
core1 = core2 = GENI_DEFAULT_BW
What am I missing, why is it necessary to vote for both/all?
Re: [PATCH V3 4/8] i2c: i2c-qcom-geni: Add interconnect support
In reply to Akash AsthanaHi Akash,
Quoted 40 lines · Akash Asthana
Get the interconnect paths for I2C based Serial Engine device and vote according to the bus speed of the driver.
Signed-off-by Akash Asthana <[email protected]>---
Changes in V2:
- As per Bjorn's comment, removed se == NULL check fromgeni_i2c_icc_get
- As per Bjorn's comment, removed code to set se->icc_path* to NULL in failure
- As per Bjorn's comment, introduced and usingdevm_of_icc_getAPI for getting path handle
- As per Matthias comment, added error handling foricc_set_bwcallChanges in V3:
- As per Matthias comment, use common library APIs defined in geni-se driver for ICC functionality.drivers/i2c/busses/i2c-qcom-geni.c | 30 +++++++++++++++++++++++++++++- 1 file changed, 29 insertions(+), 1 deletion(-)drivers/i2c/busses/i2c-qcom-geni.c +13
-0@@ -557,6 +557,26 @@ static int geni_i2c_probe(struct platform_device *pdev) gi2c->adap.dev.of_node = dev->of_node;strlcpy (gi2c->adap.name,"Geni-I2C" ,sizeof (gi2c->adap.name)); + ret =geni_icc_get (&gi2c->se,"qup-core" ,"qup-config" ,"qup-memory" ); +if (ret) +return ret; +/* +* Set the bus quota for core and cpu to a reasonable value for +* register access. +* Set quota for DDR based on bus speed, assume peak requirement +* as twice of avg bw. +*/ + gi2c->se.to_core.avg_bw =GENI_DEFAULT_BW ; + gi2c->se.to_core.peak_bw =Bps_to_icc (CORE_2X_100_MHZ ); + gi2c->se.from_cpu.avg_bw =GENI_DEFAULT_BW ; + gi2c->se.from_cpu.peak_bw =GENI_DEFAULT_BW ;
drivers/i2c/busses/i2c-qcom-geni.c static int geni_i2c_probe(struct platform_device *pdev) · line 582
Quoted 10 lines
drivers/i2c/busses/i2c-qcom-geni.c +7
-0+ gi2c->se.to_ddr.avg_bw =Bps_to_icc (gi2c->clk_freq_out); + gi2c->se.to_ddr.peak_bw =Bps_to_icc (2 * gi2c->clk_freq_out); + + ret =geni_icc_vote_on (&gi2c->se); +if (ret) +return ret; + ret =geni_se_resources_on (&gi2c->se);if (ret) {dev_err (dev,"Error turning on resources %d\n" , ret);
I think you need to call geni_icc_vote_off() here and in other error paths.
Re: [PATCH V3 5/8] spi: spi-geni-qcom: Add interconnect support
In reply to Akash AsthanaQuoted 40 lines · Akash Asthana
Get the interconnect paths for SPI based Serial Engine device and vote according to the current bus speed of the driver.
Signed-off-by Akash Asthana <[email protected]>---
Changes in V2:
- As per Bjorn's comment, removed se == NULL check fromgeni_spi_icc_get
- As per Bjorn's comment, removed code to set se->icc_path* to NULL in failure
- As per Bjorn's comment, introduced and usingdevm_of_icc_getAPI for getting path handle
- As per Matthias comment, added error handling foricc_set_bwcallChanges in V3:
- As per Matthias's comment, use helper ICC function from geni-se driver.drivers/spi/spi-geni-qcom.c | 31 ++++++++++++++++++++++++++++++- 1 file changed, 30 insertions(+), 1 deletion(-)drivers/spi/spi-geni-qcom.c +10
-0@@ -234,6 +234,16 @@ static int setup_fifo_params(struct spi_device *spi_slv,return ret; } +/* +* Set BW quota for CPU as driver supports FIFO mode only. +* Assume peak bw as twice of avg bw. +*/ + se->from_cpu.avg_bw =Bps_to_icc (mas->cur_speed_hz); + se->from_cpu.peak_bw =Bps_to_icc (2 * mas->cur_speed_hz); + ret =geni_icc_vote_on (se); +if (ret) +return ret; + clk_sel = idx &CLK_SEL_MSK ; m_clk_cfg = (div <<CLK_DIV_SHFT ) |SER_CLK_EN ;spi_setup_word_len (mas, spi_slv->mode, spi_slv->bits_per_word);
drivers/spi/spi-geni-qcom.c static int spi_geni_probe(struct platform_device *pdev) · line 593
Quoted 6 lines
drivers/spi/spi-geni-qcom.c +3
-0@@ -578,6 +588,15 @@ static int spi_geni_probe(struct platform_device *pdev)spin_lock_init (&mas->lock);pm_runtime_enable (dev); + ret =geni_icc_get (&mas->se,"qup-core" ,"qup-config" ,NULL ); +if (ret) + goto spi_geni_probe_runtime_disable;
This fails without providing any hints why, besides the error code. It might be worth to add error logging to geni_icc_get().
Quoted 25 lines
drivers/spi/spi-geni-qcom.c +12
-1+/* Set the bus quota to a reasonable value for register access */ + mas->se.to_core.avg_bw =Bps_to_icc (CORE_2X_50_MHZ ); + mas->se.to_core.peak_bw =Bps_to_icc (CORE_2X_100_MHZ ); + mas->se.from_cpu.avg_bw =GENI_DEFAULT_BW ; + mas->se.from_cpu.peak_bw =GENI_DEFAULT_BW ; + ret =spi_geni_init (mas);if (ret) goto spi_geni_probe_runtime_disable;@@ -616,14 +635,24 @@ static int __maybe_unused spi_geni_runtime_suspend(struct device *dev) {struct spi_master *spi =dev_get_drvdata (dev);struct spi_geni_master *mas =spi_master_get_devdata (spi); +int ret; + + ret =geni_se_resources_off (&mas->se); +if (ret) +return ret; -return geni_se_resources_off (&mas->se); +return geni_icc_vote_off (&mas->se); }static int __maybe_unused spi_geni_runtime_resume (struct device *dev) {
drivers/spi/spi-geni-qcom.c static int __maybe_unused spi_geni_runtime_suspend(struct device *dev) · line 658
Quoted 10 lines
drivers/spi/spi-geni-qcom.c +5
-0struct spi_master *spi =dev_get_drvdata (dev);struct spi_geni_master *mas =spi_master_get_devdata (spi); +int ret; + + ret =geni_icc_vote_on (&mas->se); +if (ret) +return ret;return geni_se_resources_on (&mas->se); }
Re: [PATCH V3 6/8] tty: serial: qcom_geni_serial: Add interconnect support
In reply to Akash AsthanaHi Akash,
Quoted 58 lines · Akash Asthana
Get the interconnect paths for Uart based Serial Engine device and vote according to the baud rate requirement of the driver.
Signed-off-by Akash Asthana <[email protected]>---
Changes in V2:
- As per Bjorn's comment, removed se == NULL check fromgeni_serial_icc_get
- As per Bjorn's comment, removed code to set se->icc_path* to NULL in failure
- As per Bjorn's comment, introduced and usingdevm_of_icc_getAPI for getting path handle
- As per Matthias comment, added error handling foricc_set_bwcallChanges in V3:
- As per Matthias comment, use common library APIs defined in geni-se driver for ICC functionality.drivers/tty/serial/qcom_geni_serial.c | 28 +++++++++++++++++++++++++--- 1 file changed, 25 insertions(+), 3 deletions(-)drivers/tty/serial/qcom_geni_serial.c +14
-3@@ -965,6 +965,14 @@ static void qcom_geni_serial_set_termios(struct uart_port *uport, ser_clk_cfg =SER_CLK_EN ; ser_clk_cfg |= clk_div <<CLK_DIV_SHFT ; +/* +* Bump up BW vote on CPU path as driver supports FIFO mode only. +* Assume peak_bw as twice of avg_bw. +*/ + port->se.from_cpu.avg_bw =Bps_to_icc (baud); + port->se.from_cpu.peak_bw =Bps_to_icc (2 * baud); +geni_icc_vote_on (&port->se); +/* parity */ tx_trans_cfg =readl (uport->membase +SE_UART_TX_TRANS_CFG ); tx_parity_cfg =readl (uport->membase +SE_UART_TX_PARITY_CFG );@@ -1202,11 +1210,14 @@ static void qcom_geni_serial_pm(struct uart_port *uport,if (old_state ==UART_PM_STATE_UNDEFINED ) old_state =UART_PM_STATE_OFF ; -if (new_state ==UART_PM_STATE_ON && old_state ==UART_PM_STATE_OFF ) +if (new_state ==UART_PM_STATE_ON && old_state ==UART_PM_STATE_OFF ) { +geni_icc_vote_on (&port->se);geni_se_resources_on (&port->se); -else if (new_state ==UART_PM_STATE_OFF && - old_state ==UART_PM_STATE_ON ) + }else if (new_state ==UART_PM_STATE_OFF && + old_state ==UART_PM_STATE_ON ) {geni_se_resources_off (&port->se); +geni_icc_vote_off (&port->se); + } }static const struct uart_ops qcom_geni_console_pops = {@@ -1304,6 +1315,17 @@ static int qcom_geni_serial_probe(struct platform_device *pdev)return -ENOMEM ;
drivers/tty/serial/qcom_geni_serial.c static int qcom_geni_serial_probe(struct platform_device *pdev) · line 1325
Quoted 10 lines
drivers/tty/serial/qcom_geni_serial.c +8
-0} + ret =geni_icc_get (&port->se,"qup-core" ,"qup-config" ,NULL ); +if (ret) +return ret; +/* Set the bus quota to a reasonable value */ + port->se.to_core.avg_bw = console ?GENI_DEFAULT_BW : +Bps_to_icc (CORE_2X_50_MHZ ); + port->se.to_core.peak_bw = console ?GENI_DEFAULT_BW : +Bps_to_icc (CORE_2X_100_MHZ );
I'm still unconvinced about the setting of the core bandwidth based on whether the port is used as console or not. It could possibly break consoles working at speeds > 115kbs and reserve more bandwidth than necessary for ports with 'slow' devices.
Why not scale the core bandwidth dynamically? You said earlier that there is no clear/linear translation of port speed to bandwidth, but you could use the same logic that is implicitly used here:
if (baudrate <= 115200) {
avg_bw = GENI_DEFAULT_BW;
peak_bw = GENI_DEFAULT_BW;
} else {
avg_bw = Bps_to_icc(CORE_2X_50_MHZ);
peak_bw = Bps_to_icc(CORE_2X_100_MHZ);
}This would be more robust, power efficient and future readers of the code don't have to wonder "why is the console special?" when our discussions on this will be long forgotten.
Re: [PATCH V3 7/8] spi: spi-qcom-qspi: Add interconnect support
In reply to Akash AsthanaQuoted 101 lines · Akash Asthana
Get the interconnect paths for QSPI device and vote according to the current bus speed of the driver.
Signed-off-by Akash Asthana <[email protected]>---
Changes in V2:
- As per Bjorn's comment, introduced and usingdevm_of_icc_getAPI for getting path handle
- As per Matthias comment, added error handling foricc_set_bwcallChanges in V3:
- No Change.drivers/spi/spi-qcom-qspi.c | 46 ++++++++++++++++++++++++++++++++++++++++++++- 1 file changed, 45 insertions(+), 1 deletion(-)drivers/spi/spi-qcom-qspi.c +38
-1@@ -2,6 +2,7 @@// Copyright (c) 2017-2018, The Linux foundation. All rights reserved. #include <linux/clk.h> +#include <linux/interconnect.h> #include <linux/interrupt.h> #include <linux/io.h> #include <linux/module.h> @@ -139,7 +140,10 @@ struct qcom_qspi {struct device *dev;struct clk_bulk_data *clks;struct qspi_xfer xfer; -/* Lock to protect xfer and IRQ accessed registers */ +struct icc_path *icc_path_cpu_to_qspi; +unsigned int avg_bw_cpu; +unsigned int peak_bw_cpu; +/* Lock to protect data accessed by IRQs */ spinlock_t lock; };@@ -241,6 +245,20 @@ static int qcom_qspi_transfer_one(struct spi_master *master,return ret; } +/* +* Set BW quota for CPU as driver supports FIFO mode only. +* Assume peak bw as twice of avg bw. +*/ + ctrl->avg_bw_cpu =Bps_to_icc (speed_hz); + ctrl->peak_bw_cpu =Bps_to_icc (2 * speed_hz); + ret =icc_set_bw (ctrl->icc_path_cpu_to_qspi, ctrl->avg_bw_cpu, + ctrl->peak_bw_cpu); +if (ret) { +dev_err (ctrl->dev,"%s: ICC BW voting failed for cpu\n" , + __func__); +return ret; + } +spin_lock_irqsave (&ctrl->lock, flags);/* We are half duplex, so either rx or tx will be set */ @@ -458,6 +476,15 @@ static int qcom_qspi_probe(struct platform_device *pdev)if (ret) goto exit_probe_master_put; + ctrl->icc_path_cpu_to_qspi =devm_of_icc_get (dev,"qspi-config" ); +if (IS_ERR (ctrl->icc_path_cpu_to_qspi)) { + ret =PTR_ERR (ctrl->icc_path_cpu_to_qspi); + goto exit_probe_master_put; + } +/* Put BW vote on CPU path for register access */ + ctrl->avg_bw_cpu =Bps_to_icc (1000 ); + ctrl->peak_bw_cpu =Bps_to_icc (1000 ); + ret =platform_get_irq (pdev,0 );if (ret <0 ) goto exit_probe_master_put;@@ -511,9 +538,17 @@ static int __maybe_unused qcom_qspi_runtime_suspend(struct device *dev) {struct spi_master *master =dev_get_drvdata (dev);struct qcom_qspi *ctrl =spi_master_get_devdata (master); +int ret;clk_bulk_disable_unprepare (QSPI_NUM_CLKS , ctrl->clks); + ret =icc_set_bw (ctrl->icc_path_cpu_to_qspi,0 ,0 ); +if (ret) { +dev_err_ratelimited (ctrl->dev,"%s: ICC BW remove failed for cpu\n" , + __func__); +return ret; + } +return 0 ; }@@ -521,6 +556,15 @@ static int __maybe_unused qcom_qspi_runtime_resume(struct device *dev) {struct spi_master *master =dev_get_drvdata (dev);struct qcom_qspi *ctrl =spi_master_get_devdata (master); +int ret; +
drivers/spi/spi-qcom-qspi.c static int __maybe_unused qcom_qspi_runtime_resume(struct device *dev) · line 570
Quoted 10 lines
drivers/spi/spi-qcom-qspi.c +7
-0+ ret =icc_set_bw (ctrl->icc_path_cpu_to_qspi, ctrl->avg_bw_cpu, + ctrl->peak_bw_cpu); +if (ret) { +dev_err_ratelimited (ctrl->dev,"%s: ICC BW voting failed for cpu\n" , + __func__); +return ret; + }return clk_bulk_prepare_enable (QSPI_NUM_CLKS , ctrl->clks); }
Looks good to me besides Mark's concern about the bandwith calculation logic.
Re: [PATCH V3 2/8] soc: qcom: geni: Support for ICC voting
In reply to Akash AsthanaQuoted 22 lines · Akash Asthana
Add necessary macros and structure variables to support ICC BW voting from individual SE drivers.
Signed-off-by Akash Asthana <[email protected]>---
Changes in V2:
- As per Bjorn's comment dropped enums for ICC paths, given the three paths individual membersChanges in V3:
- Addgeni_icc_get,geni_icc_vote_onandgeni_icc_vote_offas helper API.
- Addgeni_icc_pathstructure in common headerdrivers/soc/qcom/qcom-geni-se.c | 98 +++++++++++++++++++++++++++++++++++++++++ include/linux/qcom-geni-se.h | 36 +++++++++++++++ 2 files changed, 134 insertions(+)drivers/soc/qcom/qcom-geni-se.c +0
-0
drivers/soc/qcom/qcom-geni-se.c void geni_se_rx_dma_unprep(struct geni_se *se, dma_addr_t iova, size_t len) · line 726
Quoted 7 lines
drivers/soc/qcom/qcom-geni-se.c +4
-0@@ -720,6 +720,104 @@ void geni_se_rx_dma_unprep(struct geni_se *se, dma_addr_t iova, size_t len) }EXPORT_SYMBOL (geni_se_rx_dma_unprep); +int geni_icc_get (struct geni_se *se,const char *icc_core,const char *icc_cpu, +const char *icc_ddr) +{ +if (icc_core) {
Afaict it's only this that might be passed as NULL, so please drop these conditionals (keep the last one).
drivers/soc/qcom/qcom-geni-se.c void geni_se_rx_dma_unprep(struct geni_se *se, dma_addr_t iova, size_t len) · line 728
Quoted 2 lines
drivers/soc/qcom/qcom-geni-se.c +2
-0+ se->to_core.path =devm_of_icc_get (se->dev,"qup-core" ); +if (IS_ERR (se->to_core.path))
It would be useful to print an error message here (if PTR_ERR(path) != -EPROBE_DEFER).
Quoted 14 lines
drivers/soc/qcom/qcom-geni-se.c +14
-0+return PTR_ERR (se->to_core.path); + } + +if (icc_cpu) { + se->from_cpu.path =devm_of_icc_get (se->dev,"qup-config" ); +if (IS_ERR (se->from_cpu.path)) +return PTR_ERR (se->from_cpu.path); + } + +if (icc_ddr) { + se->to_ddr.path =devm_of_icc_get (se->dev,"qup-memory" ); +if (IS_ERR (se->to_ddr.path)) +return PTR_ERR (se->to_ddr.path); + }
drivers/soc/qcom/qcom-geni-se.c void geni_se_rx_dma_unprep(struct geni_se *se, dma_addr_t iova, size_t len) · line 752
Quoted 10 lines
drivers/soc/qcom/qcom-geni-se.c +10
-0+ +return 0 ; +} +EXPORT_SYMBOL (geni_icc_get); + +int geni_icc_vote_on (struct geni_se *se) +{ +int ret; + +if (se->to_core.path) {
icc_set_bw(NULL, ...) is valid and will return 0, so these checks doesn't add any value.
drivers/soc/qcom/qcom-geni-se.c void geni_se_rx_dma_unprep(struct geni_se *se, dma_addr_t iova, size_t len) · line 757
Quoted 5 lines
drivers/soc/qcom/qcom-geni-se.c +5
-0+ ret =icc_set_bw (se->to_core.path, se->to_core.avg_bw, + se->to_core.peak_bw); +if (ret) { +dev_err_ratelimited (se->dev,"%s: ICC BW voting failed for core\n" , + __func__);
Please drop the __func__, the message is specific enough.
Quoted 52 lines
drivers/soc/qcom/qcom-geni-se.c +52
-0+return ret; + } + } + +if (se->from_cpu.path) { + ret =icc_set_bw (se->from_cpu.path, se->from_cpu.avg_bw, + se->from_cpu.peak_bw); +if (ret) { +dev_err_ratelimited (se->dev,"%s: ICC BW voting failed for cpu\n" , + __func__); +return ret; + } + } + +if (se->to_ddr.path) { + ret =icc_set_bw (se->to_ddr.path, se->to_ddr.avg_bw, + se->to_ddr.peak_bw); +if (ret) { +dev_err_ratelimited (se->dev,"%s: ICC BW voting failed for ddr\n" , + __func__); +return ret; + } + } + +return 0 ; +} +EXPORT_SYMBOL (geni_icc_vote_on); + +int geni_icc_vote_off (struct geni_se *se) +{ +int ret; + +if (se->to_core.path) { + ret =icc_set_bw (se->to_core.path,0 ,0 ); +if (ret) { +dev_err_ratelimited (se->dev,"%s: ICC BW remove failed for core\n" , + __func__); +return ret; + } + } + +if (se->from_cpu.path) { + ret =icc_set_bw (se->from_cpu.path,0 ,0 ); +if (ret) { +dev_err_ratelimited (se->dev,"%s: ICC BW remove failed for cpu\n" , + __func__); +return ret; + } + } + +if (se->to_ddr.path) { + ret =icc_set_bw (se->to_ddr.path,0 ,0 );
drivers/soc/qcom/qcom-geni-se.c void geni_se_rx_dma_unprep(struct geni_se *se, dma_addr_t iova, size_t len) · line 819
Quoted 10 lines
drivers/soc/qcom/qcom-geni-se.c +10
-0+if (ret) { +dev_err_ratelimited (se->dev,"%s: ICC BW remove failed for ddr\n" , + __func__); +return ret; + } + } + +return 0 ; +} +EXPORT_SYMBOL (geni_icc_vote_off);
Given that these two functions only switch the bandwidth request between some value and 0, I really think we should carry a "bool enabled" on the path and replace these two functions with icc_bulk_enable()/icc_bulk_disable().
The added benefit of this would be that you call icc_set_bw() instead of changing the geni_icc_path->{avg_bw,peak_bw} and don't need to keep track of them here.
Regards,
Bjorn
Quoted 87 lines
drivers/soc/qcom/qcom-geni-se.c +1
-0+static int geni_se_probe (struct platform_device *pdev) {struct device *dev = &pdev->dev;include/linux/qcom-geni-se.h +36
-0@@ -6,6 +6,8 @@#ifndef _LINUX_QCOM_GENI_SE#define _LINUX_QCOM_GENI_SE +#include <linux/interconnect.h> +/* Transfer mode supported by GENI Serial Engines */ enum geni_se_xfer_mode { GENI_SE_INVALID,@@ -25,6 +27,12 @@ enum geni_se_protocol_type {struct geni_wrapper ;struct clk ; +struct geni_icc_path { +struct icc_path *path; +unsigned int avg_bw; +unsigned int peak_bw; +}; +/** * struct geni_se - GENI Serial Engine * @base: Base Address of the Serial Engine's register block @@ -33,6 +41,9 @@ struct clk;* @clk: Handle to the core serial engine clock * @num_clk_levels: Number of valid clock levels in clk_perf_tbl * @clk_perf_tbl: Table of clock frequency input to serial engine clock +* @to_core: ICC path structure for geni to core +* @from_cpu: ICC path structure for cpu to geni +* @to_ddr: ICC path structure for geni to ddr */ struct geni_se {void __iomem *base;@@ -41,6 +52,9 @@ struct geni_se {struct clk *clk;unsigned int num_clk_levels;unsigned long *clk_perf_tbl; +struct geni_icc_path to_core; +struct geni_icc_path from_cpu; +struct geni_icc_path to_ddr; };/* Common SE registers */ @@ -229,6 +243,21 @@ struct geni_se {#define GENI_SE_VERSION_MINOR (ver) ((ver & HW_VER_MINOR_MASK) >> HW_VER_MINOR_SHFT)#define GENI_SE_VERSION_STEP (ver) (ver & HW_VER_STEP_MASK) +/* +* Define bandwidth thresholds that cause the underlying Core 2X interconnect +* clock to run at the named frequency. These baseline values are recommended +* by the hardware team, and are not dynamically scaled with GENI bandwidth +* beyond basic on/off. +*/ +#define CORE_2X_19_2_MHZ 960 +#define CORE_2X_50_MHZ 2500 +#define CORE_2X_100_MHZ 5000 +#define CORE_2X_150_MHZ 7500 +#define CORE_2X_200_MHZ 10000 +#define CORE_2X_236_MHZ 16383 + +#define GENI_DEFAULT_BW Bps_to_icc(1000) +#if IS_ENABLED (CONFIG_QCOM_GENI_SE )u32 geni_se_get_qup_hw_version (struct geni_se *se);@@ -416,5 +445,12 @@ int geni_se_rx_dma_prep(struct geni_se *se, void *buf, size_t len,void geni_se_tx_dma_unprep (struct geni_se *se,dma_addr_t iova,size_t len);void geni_se_rx_dma_unprep (struct geni_se *se,dma_addr_t iova,size_t len); + +int geni_icc_get (struct geni_se *se,const char *icc_core,const char *icc_cpu, +const char *icc_ddr); + +int geni_icc_vote_on (struct geni_se *se); + +int geni_icc_vote_off (struct geni_se *se);#endif #endif Signature · 1 line
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,\na Linux Foundation Collaborative Project
Re: [PATCH V3 2/8] soc: qcom: geni: Support for ICC voting
Quoted 146 lines · Bjorn Andersson
Quoted 22 lines · Akash Asthana
Add necessary macros and structure variables to support ICC BW voting from individual SE drivers.
Signed-off-by Akash Asthana <[email protected]>---
Changes in V2:
- As per Bjorn's comment dropped enums for ICC paths, given the three paths individual membersChanges in V3:
- Addgeni_icc_get,geni_icc_vote_onandgeni_icc_vote_offas helper API.
- Addgeni_icc_pathstructure in common headerdrivers/soc/qcom/qcom-geni-se.c | 98 +++++++++++++++++++++++++++++++++++++++++ include/linux/qcom-geni-se.h | 36 +++++++++++++++ 2 files changed, 134 insertions(+)drivers/soc/qcom/qcom-geni-se.c +0
-0drivers/soc/qcom/qcom-geni-se.c void geni_se_rx_dma_unprep(struct geni_se *se, dma_addr_t iova, size_t len) · line 726
Quoted 7 lines
drivers/soc/qcom/qcom-geni-se.c +4
-0@@ -720,6 +720,104 @@ void geni_se_rx_dma_unprep(struct geni_se *se, dma_addr_t iova, size_t len) }EXPORT_SYMBOL (geni_se_rx_dma_unprep); +int geni_icc_get (struct geni_se *se,const char *icc_core,const char *icc_cpu, +const char *icc_ddr) +{ +if (icc_core) {Afaict it's only this that might be passed as NULL, so please drop these conditionals (keep the last one).
drivers/soc/qcom/qcom-geni-se.c void geni_se_rx_dma_unprep(struct geni_se *se, dma_addr_t iova, size_t len) · line 728
Quoted 2 lines
drivers/soc/qcom/qcom-geni-se.c +2
-0+ se->to_core.path =devm_of_icc_get (se->dev,"qup-core" ); +if (IS_ERR (se->to_core.path))It would be useful to print an error message here (if
PTR_ERR(path)!= -EPROBE_DEFER).Quoted 14 lines
drivers/soc/qcom/qcom-geni-se.c +14
-0+return PTR_ERR (se->to_core.path); + } + +if (icc_cpu) { + se->from_cpu.path =devm_of_icc_get (se->dev,"qup-config" ); +if (IS_ERR (se->from_cpu.path)) +return PTR_ERR (se->from_cpu.path); + } + +if (icc_ddr) { + se->to_ddr.path =devm_of_icc_get (se->dev,"qup-memory" ); +if (IS_ERR (se->to_ddr.path)) +return PTR_ERR (se->to_ddr.path); + }drivers/soc/qcom/qcom-geni-se.c void geni_se_rx_dma_unprep(struct geni_se *se, dma_addr_t iova, size_t len) · line 752
Quoted 10 lines
drivers/soc/qcom/qcom-geni-se.c +10
-0+ +return 0 ; +} +EXPORT_SYMBOL (geni_icc_get); + +int geni_icc_vote_on (struct geni_se *se) +{ +int ret; + +if (se->to_core.path) {
icc_set_bw(NULL, ...)is valid and will return 0, so these checks doesn't add any value.drivers/soc/qcom/qcom-geni-se.c void geni_se_rx_dma_unprep(struct geni_se *se, dma_addr_t iova, size_t len) · line 757
Quoted 5 lines
drivers/soc/qcom/qcom-geni-se.c +5
-0+ ret =icc_set_bw (se->to_core.path, se->to_core.avg_bw, + se->to_core.peak_bw); +if (ret) { +dev_err_ratelimited (se->dev,"%s: ICC BW voting failed for core\n" , + __func__);Please drop the
__func__, the message is specific enough.Quoted 52 lines
drivers/soc/qcom/qcom-geni-se.c +52
-0+return ret; + } + } + +if (se->from_cpu.path) { + ret =icc_set_bw (se->from_cpu.path, se->from_cpu.avg_bw, + se->from_cpu.peak_bw); +if (ret) { +dev_err_ratelimited (se->dev,"%s: ICC BW voting failed for cpu\n" , + __func__); +return ret; + } + } + +if (se->to_ddr.path) { + ret =icc_set_bw (se->to_ddr.path, se->to_ddr.avg_bw, + se->to_ddr.peak_bw); +if (ret) { +dev_err_ratelimited (se->dev,"%s: ICC BW voting failed for ddr\n" , + __func__); +return ret; + } + } + +return 0 ; +} +EXPORT_SYMBOL (geni_icc_vote_on); + +int geni_icc_vote_off (struct geni_se *se) +{ +int ret; + +if (se->to_core.path) { + ret =icc_set_bw (se->to_core.path,0 ,0 ); +if (ret) { +dev_err_ratelimited (se->dev,"%s: ICC BW remove failed for core\n" , + __func__); +return ret; + } + } + +if (se->from_cpu.path) { + ret =icc_set_bw (se->from_cpu.path,0 ,0 ); +if (ret) { +dev_err_ratelimited (se->dev,"%s: ICC BW remove failed for cpu\n" , + __func__); +return ret; + } + } + +if (se->to_ddr.path) { + ret =icc_set_bw (se->to_ddr.path,0 ,0 );drivers/soc/qcom/qcom-geni-se.c void geni_se_rx_dma_unprep(struct geni_se *se, dma_addr_t iova, size_t len) · line 819
Quoted 10 lines
drivers/soc/qcom/qcom-geni-se.c +10
-0+if (ret) { +dev_err_ratelimited (se->dev,"%s: ICC BW remove failed for ddr\n" , + __func__); +return ret; + } + } + +return 0 ; +} +EXPORT_SYMBOL (geni_icc_vote_off);Given that these two functions only switch the bandwidth request between some value and 0, I really think we should carry a "bool enabled" on the path and replace these two functions with icc_bulk_enable()/icc_bulk_disable().
Quoted 3 lines
The added benefit of this would be that you call
icc_set_bw()instead of changing the geni_icc_path->{avg_bw,peak_bw} and don't need to keep track of them here.
Yes yes! I had the same thought here [1].
Georgi, what do you think?
-Evan
Re: [PATCH V3 3/8] soc: qcom-geni-se: Add interconnect support to fix earlycon crash
In reply to Matthias KaehlckeQuoted 64 lines · Matthias Kaehlcke
Hi Akash,
Quoted 29 lines · Akash Asthana
QUP core clock is shared among all the SE drivers present on particular QUP wrapper, the system will reset(unclocked access) if earlycon used after QUP core clock is put to 0 from other SE drivers before real console comes up.
As earlycon can't vote for it's QUP core need, to fix this add ICC support to common/QUP wrapper driver and put vote for QUP core from probe on behalf of earlycon and remove vote during earlycon exit call.
Signed-off-by Akash Asthana <[email protected]>Reported-by Matthias Kaehlcke <[email protected]>---
Change is V3:
- Addgeni_remove_earlycon_icc_voteAPI that will be used by earlycon exit function to remove ICC vote for earlyconsole.
- Remove suspend/resume hook for geni-se driver as we are no longer removing earlyconsole ICC vote from system suspend, we are removing from earlycon exit.drivers/soc/qcom/qcom-geni-se.c | 51 +++++++++++++++++++++++++++++++++++ drivers/tty/serial/qcom_geni_serial.c | 7 +++++ include/linux/qcom-geni-se.h | 2 ++ 3 files changed, 60 insertions(+)drivers/soc/qcom/qcom-geni-se.c +0
-0drivers/soc/qcom/qcom-geni-se.c struct geni_wrapper { · line 96
Quoted 7 lines
drivers/soc/qcom/qcom-geni-se.c +2
-0@@ -90,8 +90,11 @@ struct geni_wrapper {struct device *dev;void __iomem *base;struct clk_bulk_data ahb_clks[NUM_AHB_CLKS ]; +struct geni_icc_path to_core; }; +struct geni_wrapper *earlycon_wrapper;should be static
Quoted 22 lines
drivers/soc/qcom/qcom-geni-se.c +15
-0+#define QUP_HW_VER_REG 0x4/* Common SE registers */ @@ -818,6 +821,26 @@ int geni_icc_vote_off(struct geni_se *se) }EXPORT_SYMBOL (geni_icc_vote_off); +void geni_remove_earlycon_icc_vote (void ) +{ +struct geni_wrapper *wrapper = earlycon_wrapper; +struct device_node *parent =of_get_next_parent (wrapper->dev->of_node); +struct device_node *child; + +for_each_child_of_node (parent, child) { +if (of_device_is_compatible (child,"qcom,geni-se-qup" )) { + wrapper =platform_get_drvdata (of_find_device_by_node ( + child)); +icc_put (wrapper->to_core.path); + wrapper->to_core.path =NULL ; + } + }
Quoted 8 lines
drivers/soc/qcom/qcom-geni-se.c int geni_icc_vote_off(struct geni_se *se) · line 842
Quoted 5 lines
drivers/soc/qcom/qcom-geni-se.c +5
-0+of_node_put (parent); + + earlycon_wrapper =NULL ; +} +EXPORT_SYMBOL (geni_remove_earlycon_icc_vote);I didn't know that consoles have an exit handler, this is way nicer than the miscellaneous triggers we discussed earlier :)
No wonder I 'missed' this when looking at the console code for possible triggers, it is brand new and as of now only exists in -next:
commit ed31685c96e18f773ca11dd1a637974d62130673
Author: Andy Shevchenko <andriy.shevchenko-VuQAYsv1563Yd54FQh9/[email protected]> Date: Mon Feb 3 15:31:30 2020 +0200
console: Introduce ->exit() callback
sharp timing!
Re: [PATCH V3 2/8] soc: qcom: geni: Support for ICC voting
In reply to Matthias KaehlckeHi Matthias,
Quoted 103 lines · Matthias Kaehlcke
Hi Akash,
Quoted 73 lines · Akash Asthana
Add necessary macros and structure variables to support ICC BW voting from individual SE drivers.
Signed-off-by Akash Asthana <[email protected]>---
Changes in V2:
- As per Bjorn's comment dropped enums for ICC paths, given the three paths individual membersChanges in V3:
- Addgeni_icc_get,geni_icc_vote_onandgeni_icc_vote_offas helper API.
- Addgeni_icc_pathstructure in common headerdrivers/soc/qcom/qcom-geni-se.c | 98 +++++++++++++++++++++++++++++++++++++++++ include/linux/qcom-geni-se.h | 36 +++++++++++++++ 2 files changed, 134 insertions(+)drivers/soc/qcom/qcom-geni-se.c +48
-0@@ -720,6 +720,104 @@ void geni_se_rx_dma_unprep(struct geni_se *se, dma_addr_t iova, size_t len) }EXPORT_SYMBOL (geni_se_rx_dma_unprep); +int geni_icc_get (struct geni_se *se,const char *icc_core,const char *icc_cpu, +const char *icc_ddr) +{ +if (icc_core) { + se->to_core.path =devm_of_icc_get (se->dev,"qup-core" ); +if (IS_ERR (se->to_core.path)) +return PTR_ERR (se->to_core.path); + } + +if (icc_cpu) { + se->from_cpu.path =devm_of_icc_get (se->dev,"qup-config" ); +if (IS_ERR (se->from_cpu.path)) +return PTR_ERR (se->from_cpu.path); + } + +if (icc_ddr) { + se->to_ddr.path =devm_of_icc_get (se->dev,"qup-memory" ); +if (IS_ERR (se->to_ddr.path)) +return PTR_ERR (se->to_ddr.path); + } + +return 0 ; +} +EXPORT_SYMBOL (geni_icc_get); + +int geni_icc_vote_on (struct geni_se *se) +{ +int ret; + +if (se->to_core.path) { + ret =icc_set_bw (se->to_core.path, se->to_core.avg_bw, + se->to_core.peak_bw); +if (ret) { +dev_err_ratelimited (se->dev,"%s: ICC BW voting failed for core\n" , + __func__); +return ret; + } + } + +if (se->from_cpu.path) { + ret =icc_set_bw (se->from_cpu.path, se->from_cpu.avg_bw, + se->from_cpu.peak_bw); +if (ret) { +dev_err_ratelimited (se->dev,"%s: ICC BW voting failed for cpu\n" , + __func__); +return ret; + } + }drivers/soc/qcom/qcom-geni-se.c void geni_se_rx_dma_unprep(struct geni_se *se, dma_addr_t iova, size_t len) · line 780
Quoted 10 lines
drivers/soc/qcom/qcom-geni-se.c +10
-0+ +if (se->to_ddr.path) { + ret =icc_set_bw (se->to_ddr.path, se->to_ddr.avg_bw, + se->to_ddr.peak_bw); +if (ret) { +dev_err_ratelimited (se->dev,"%s: ICC BW voting failed for ddr\n" , + __func__); +return ret; + } + }With an array of 'struct
geni_icc_path' pointers the above could be reduced to:for (i = 0; i < ARRAY_SIZE(se->icc_paths); i++) { if (!se->icc_paths[i]) continue; ret = icc_set_bw(se->icc_paths[i]->path, se->icc_paths[i]->avg_bw, se->icc_paths[i]->peak_bw); if (ret) { dev_err_ratelimited(se->dev, "%s: ICC BW voting failed\n", __func__); return ret; } }
Quoted 3 lines
similar for
geni_icc_vote_off()It's just a suggestion, looks also good to me as is.
I thought giving individual path name will increase readability. But that doesn't seems to be adding much value on cost of repeated code.
So, I will make the suggested change in next version.
Thanks,
Akash
Quoted 2 lines
Reviewed-by Matthias Kaehlcke <[email protected]>
Signature · 1 line
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,\na Linux Foundation Collaborative Project
Re: [PATCH V3 2/8] soc: qcom: geni: Support for ICC voting
In reply to Bjorn AnderssonHi Bjorn,
Quoted 27 lines · Bjorn Andersson
Quoted 25 lines · Akash Asthana
Add necessary macros and structure variables to support ICC BW voting from individual SE drivers.
Signed-off-by Akash Asthana <[email protected]>---
Changes in V2:
- As per Bjorn's comment dropped enums for ICC paths, given the three paths individual membersChanges in V3:
- Addgeni_icc_get,geni_icc_vote_onandgeni_icc_vote_offas helper API.
- Addgeni_icc_pathstructure in common headerdrivers/soc/qcom/qcom-geni-se.c | 98 +++++++++++++++++++++++++++++++++++++++++ include/linux/qcom-geni-se.h | 36 +++++++++++++++ 2 files changed, 134 insertions(+)drivers/soc/qcom/qcom-geni-se.c +0
-0@@ -720,6 +720,104 @@ void geni_se_rx_dma_unprep(struct geni_se *se, dma_addr_t iova, size_t len) }EXPORT_SYMBOL (geni_se_rx_dma_unprep);
Quoted 6 lines
drivers/soc/qcom/qcom-geni-se.c void geni_se_rx_dma_unprep(struct geni_se *se, dma_addr_t iova, size_t len) · line 726
Quoted 4 lines
drivers/soc/qcom/qcom-geni-se.c +4
-0+int geni_icc_get (struct geni_se *se,const char *icc_core,const char *icc_cpu, +const char *icc_ddr) +{ +if (icc_core) {Afaict it's only this that might be passed as NULL, so please drop these conditionals (keep the last one).
IIUC you're suggesting to drop if (icc_core/cpu) but keep if (icc_ddr) ?
Quoted 5 lines
drivers/soc/qcom/qcom-geni-se.c void geni_se_rx_dma_unprep(struct geni_se *se, dma_addr_t iova, size_t len) · line 728
Quoted 3 lines
drivers/soc/qcom/qcom-geni-se.c +2
-0+ se->to_core.path =devm_of_icc_get (se->dev,"qup-core" ); +if (IS_ERR (se->to_core.path))It would be useful to print an error message here (if
PTR_ERR(path)!= -EPROBE_DEFER).
okay
Quoted 19 lines
Quoted 19 lines
drivers/soc/qcom/qcom-geni-se.c +18
-0+return PTR_ERR (se->to_core.path); + } + +if (icc_cpu) { + se->from_cpu.path =devm_of_icc_get (se->dev,"qup-config" ); +if (IS_ERR (se->from_cpu.path)) +return PTR_ERR (se->from_cpu.path); + } + +if (icc_ddr) { + se->to_ddr.path =devm_of_icc_get (se->dev,"qup-memory" ); +if (IS_ERR (se->to_ddr.path)) +return PTR_ERR (se->to_ddr.path); + } + +return 0 ; +} +EXPORT_SYMBOL (geni_icc_get);
Quoted 8 lines
drivers/soc/qcom/qcom-geni-se.c void geni_se_rx_dma_unprep(struct geni_se *se, dma_addr_t iova, size_t len) · line 752
Quoted 6 lines
drivers/soc/qcom/qcom-geni-se.c +6
-0+ +int geni_icc_vote_on (struct geni_se *se) +{ +int ret; + +if (se->to_core.path) {
icc_set_bw(NULL, ...)is valid and will return 0, so these checks doesn't add any value.
Yeah, ok
Quoted 7 lines
drivers/soc/qcom/qcom-geni-se.c void geni_se_rx_dma_unprep(struct geni_se *se, dma_addr_t iova, size_t len) · line 757
Quoted 6 lines
drivers/soc/qcom/qcom-geni-se.c +5
-0+ ret =icc_set_bw (se->to_core.path, se->to_core.avg_bw, + se->to_core.peak_bw); +if (ret) { +dev_err_ratelimited (se->dev,"%s: ICC BW voting failed for core\n" , + __func__);Please drop the
__func__, the message is specific enough.
ok
Regards,
Akash
Quoted 161 lines
Quoted 53 lines
drivers/soc/qcom/qcom-geni-se.c +52
-0+return ret; + } + } + +if (se->from_cpu.path) { + ret =icc_set_bw (se->from_cpu.path, se->from_cpu.avg_bw, + se->from_cpu.peak_bw); +if (ret) { +dev_err_ratelimited (se->dev,"%s: ICC BW voting failed for cpu\n" , + __func__); +return ret; + } + } + +if (se->to_ddr.path) { + ret =icc_set_bw (se->to_ddr.path, se->to_ddr.avg_bw, + se->to_ddr.peak_bw); +if (ret) { +dev_err_ratelimited (se->dev,"%s: ICC BW voting failed for ddr\n" , + __func__); +return ret; + } + } + +return 0 ; +} +EXPORT_SYMBOL (geni_icc_vote_on); + +int geni_icc_vote_off (struct geni_se *se) +{ +int ret; + +if (se->to_core.path) { + ret =icc_set_bw (se->to_core.path,0 ,0 ); +if (ret) { +dev_err_ratelimited (se->dev,"%s: ICC BW remove failed for core\n" , + __func__); +return ret; + } + } + +if (se->from_cpu.path) { + ret =icc_set_bw (se->from_cpu.path,0 ,0 ); +if (ret) { +dev_err_ratelimited (se->dev,"%s: ICC BW remove failed for cpu\n" , + __func__); +return ret; + } + } + +if (se->to_ddr.path) { + ret =icc_set_bw (se->to_ddr.path,0 ,0 );drivers/soc/qcom/qcom-geni-se.c void geni_se_rx_dma_unprep(struct geni_se *se, dma_addr_t iova, size_t len) · line 819
Quoted 10 lines
drivers/soc/qcom/qcom-geni-se.c +10
-0+if (ret) { +dev_err_ratelimited (se->dev,"%s: ICC BW remove failed for ddr\n" , + __func__); +return ret; + } + } + +return 0 ; +} +EXPORT_SYMBOL (geni_icc_vote_off);Given that these two functions only switch the bandwidth request between some value and 0, I really think we should carry a "bool enabled" on the path and replace these two functions with icc_bulk_enable()/icc_bulk_disable().
The added benefit of this would be that you call
icc_set_bw()instead of changing the geni_icc_path->{avg_bw,peak_bw} and don't need to keep track of them here.Regards,
BjornQuoted 87 lines
drivers/soc/qcom/qcom-geni-se.c +1
-0+static int geni_se_probe (struct platform_device *pdev) {struct device *dev = &pdev->dev;include/linux/qcom-geni-se.h +36
-0@@ -6,6 +6,8 @@#ifndef _LINUX_QCOM_GENI_SE#define _LINUX_QCOM_GENI_SE +#include <linux/interconnect.h> +/* Transfer mode supported by GENI Serial Engines */ enum geni_se_xfer_mode { GENI_SE_INVALID,@@ -25,6 +27,12 @@ enum geni_se_protocol_type {struct geni_wrapper ;struct clk ; +struct geni_icc_path { +struct icc_path *path; +unsigned int avg_bw; +unsigned int peak_bw; +}; +/** * struct geni_se - GENI Serial Engine * @base: Base Address of the Serial Engine's register block @@ -33,6 +41,9 @@ struct clk;* @clk: Handle to the core serial engine clock * @num_clk_levels: Number of valid clock levels in clk_perf_tbl * @clk_perf_tbl: Table of clock frequency input to serial engine clock +* @to_core: ICC path structure for geni to core +* @from_cpu: ICC path structure for cpu to geni +* @to_ddr: ICC path structure for geni to ddr */ struct geni_se {void __iomem *base;@@ -41,6 +52,9 @@ struct geni_se {struct clk *clk;unsigned int num_clk_levels;unsigned long *clk_perf_tbl; +struct geni_icc_path to_core; +struct geni_icc_path from_cpu; +struct geni_icc_path to_ddr; };/* Common SE registers */ @@ -229,6 +243,21 @@ struct geni_se {#define GENI_SE_VERSION_MINOR (ver) ((ver & HW_VER_MINOR_MASK) >> HW_VER_MINOR_SHFT)#define GENI_SE_VERSION_STEP (ver) (ver & HW_VER_STEP_MASK) +/* +* Define bandwidth thresholds that cause the underlying Core 2X interconnect +* clock to run at the named frequency. These baseline values are recommended +* by the hardware team, and are not dynamically scaled with GENI bandwidth +* beyond basic on/off. +*/ +#define CORE_2X_19_2_MHZ 960 +#define CORE_2X_50_MHZ 2500 +#define CORE_2X_100_MHZ 5000 +#define CORE_2X_150_MHZ 7500 +#define CORE_2X_200_MHZ 10000 +#define CORE_2X_236_MHZ 16383 + +#define GENI_DEFAULT_BW Bps_to_icc(1000) +#if IS_ENABLED (CONFIG_QCOM_GENI_SE )u32 geni_se_get_qup_hw_version (struct geni_se *se);@@ -416,5 +445,12 @@ int geni_se_rx_dma_prep(struct geni_se *se, void *buf, size_t len,void geni_se_tx_dma_unprep (struct geni_se *se,dma_addr_t iova,size_t len);void geni_se_rx_dma_unprep (struct geni_se *se,dma_addr_t iova,size_t len); + +int geni_icc_get (struct geni_se *se,const char *icc_core,const char *icc_cpu, +const char *icc_ddr); + +int geni_icc_vote_on (struct geni_se *se); + +int geni_icc_vote_off (struct geni_se *se);#endif #endif Signature · 1 line
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,\na Linux Foundation Collaborative Project
Signature · 1 line
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,\na Linux Foundation Collaborative Project
Re: [PATCH V3 2/8] soc: qcom: geni: Support for ICC voting
In reply to Evan GreenHi Bjorn, Evan,
Quoted 4 lines
Quoted 4 lines
Given that these two functions only switch the bandwidth request between some value and 0, I really think we should carry a "bool enabled" on the path and replace these two functions with icc_bulk_enable()/icc_bulk_disable().
So, if above is implementation "bool enabled" on path can be used directly in aggregation of ICC votes on particular node without using icc_set_bw call, if yes then I am not aware how? or we'll be using icc_set_bw API indirectly inside icc_bulk APIs?
Quoted 3 lines
Quoted 3 lines
The added benefit of this would be that you call
icc_set_bw()instead of changing the geni_icc_path->{avg_bw,peak_bw} and don't need to keep track of them here.
Ok IIUC, we need to call icc_set_bw() from GENI driver only if we change (avg_bw | peak_bw)?
Regards,
Akash
Quoted 6 lines
Yes yes! I had the same thought here [1].
Georgi, what do you think?
-Evan
Signature · 1 line
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,\na Linux Foundation Collaborative Project
Re: [PATCH V3 3/8] soc: qcom-geni-se: Add interconnect support to fix earlycon crash
In reply to Matthias KaehlckeHi Matthias
Quoted 39 lines · Matthias Kaehlcke
Quoted 38 lines · Matthias Kaehlcke
Hi Akash,
Quoted 35 lines · Akash Asthana
QUP core clock is shared among all the SE drivers present on particular QUP wrapper, the system will reset(unclocked access) if earlycon used after QUP core clock is put to 0 from other SE drivers before real console comes up.
As earlycon can't vote for it's QUP core need, to fix this add ICC support to common/QUP wrapper driver and put vote for QUP core from probe on behalf of earlycon and remove vote during earlycon exit call.
Signed-off-by Akash Asthana <[email protected]>Reported-by Matthias Kaehlcke <[email protected]>---
Change is V3:
- Addgeni_remove_earlycon_icc_voteAPI that will be used by earlycon exit function to remove ICC vote for earlyconsole.
- Remove suspend/resume hook for geni-se driver as we are no longer removing earlyconsole ICC vote from system suspend, we are removing from earlycon exit.drivers/soc/qcom/qcom-geni-se.c | 51 +++++++++++++++++++++++++++++++++++ drivers/tty/serial/qcom_geni_serial.c | 7 +++++ include/linux/qcom-geni-se.h | 2 ++ 3 files changed, 60 insertions(+)drivers/soc/qcom/qcom-geni-se.c +1
-0@@ -90,8 +90,11 @@ struct geni_wrapper {struct device *dev;void __iomem *base;struct clk_bulk_data ahb_clks[NUM_AHB_CLKS ]; +struct geni_icc_path to_core; };
Quoted 2 lines
Quoted 2 lines
drivers/soc/qcom/qcom-geni-se.c struct geni_wrapper { · line 96
Quoted 1 line
drivers/soc/qcom/qcom-geni-se.c +1
-0+struct geni_wrapper *earlycon_wrapper;should be static
Yeah ok, I missed it.
Quoted 33 lines
Quoted 22 lines
Quoted 22 lines
drivers/soc/qcom/qcom-geni-se.c +14
-0+#define QUP_HW_VER_REG 0x4/* Common SE registers */ @@ -818,6 +821,26 @@ int geni_icc_vote_off(struct geni_se *se) }EXPORT_SYMBOL (geni_icc_vote_off); +void geni_remove_earlycon_icc_vote (void ) +{ +struct geni_wrapper *wrapper = earlycon_wrapper; +struct device_node *parent =of_get_next_parent (wrapper->dev->of_node); +struct device_node *child; + +for_each_child_of_node (parent, child) { +if (of_device_is_compatible (child,"qcom,geni-se-qup" )) { + wrapper =platform_get_drvdata (of_find_device_by_node ( + child)); +icc_put (wrapper->to_core.path); + wrapper->to_core.path =NULL ; + }Quoted 8 lines
drivers/soc/qcom/qcom-geni-se.c int geni_icc_vote_off(struct geni_se *se) · line 842
Quoted 6 lines
drivers/soc/qcom/qcom-geni-se.c +6
-0+ } +of_node_put (parent); + + earlycon_wrapper =NULL ; +} +EXPORT_SYMBOL (geni_remove_earlycon_icc_vote);I didn't know that consoles have an exit handler, this is way nicer than the miscellaneous triggers we discussed earlier :)
No wonder I 'missed' this when looking at the console code for possible triggers, it is brand new and as of now only exists in -next:
Quoted 8 lines
commit
ed31685c96e18f773ca11dd1a637974d62130673Author: Andy Shevchenko <[email protected]> Date: Mon Feb 3 15:31:30 2020 +0200console: Introduce ->exit() callbacksharp timing!
Yeah this is added recently, even I was not aware of it, Bjorn suggested me to use this. Indeed sharp timing! :)
Regards,
Akash
Signature · 1 line
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,\na Linux Foundation Collaborative Project
Re: [PATCH V3 4/8] i2c: i2c-qcom-geni: Add interconnect support
In reply to Matthias KaehlckeHi Matthias,
Quoted 46 lines · Matthias Kaehlcke
Hi Akash,
Quoted 43 lines · Akash Asthana
Get the interconnect paths for I2C based Serial Engine device and vote according to the bus speed of the driver.
Signed-off-by Akash Asthana <[email protected]>---
Changes in V2:
- As per Bjorn's comment, removed se == NULL check fromgeni_i2c_icc_get
- As per Bjorn's comment, removed code to set se->icc_path* to NULL in failure
- As per Bjorn's comment, introduced and usingdevm_of_icc_getAPI for getting path handle
- As per Matthias comment, added error handling foricc_set_bwcallChanges in V3:
- As per Matthias comment, use common library APIs defined in geni-se driver for ICC functionality.drivers/i2c/busses/i2c-qcom-geni.c | 30 +++++++++++++++++++++++++++++- 1 file changed, 29 insertions(+), 1 deletion(-)drivers/i2c/busses/i2c-qcom-geni.c +16
-0@@ -557,6 +557,26 @@ static int geni_i2c_probe(struct platform_device *pdev) gi2c->adap.dev.of_node = dev->of_node;strlcpy (gi2c->adap.name,"Geni-I2C" ,sizeof (gi2c->adap.name)); + ret =geni_icc_get (&gi2c->se,"qup-core" ,"qup-config" ,"qup-memory" ); +if (ret) +return ret; +/* +* Set the bus quota for core and cpu to a reasonable value for +* register access. +* Set quota for DDR based on bus speed, assume peak requirement +* as twice of avg bw. +*/ + gi2c->se.to_core.avg_bw =GENI_DEFAULT_BW ; + gi2c->se.to_core.peak_bw =Bps_to_icc (CORE_2X_100_MHZ ); + gi2c->se.from_cpu.avg_bw =GENI_DEFAULT_BW ; + gi2c->se.from_cpu.peak_bw =GENI_DEFAULT_BW ; + gi2c->se.to_ddr.avg_bw =Bps_to_icc (gi2c->clk_freq_out); + gi2c->se.to_ddr.peak_bw =Bps_to_icc (2 * gi2c->clk_freq_out); +
Quoted 8 lines
drivers/i2c/busses/i2c-qcom-geni.c static int geni_i2c_probe(struct platform_device *pdev) · line 582
Quoted 7 lines
drivers/i2c/busses/i2c-qcom-geni.c +4
-0+ ret =geni_icc_vote_on (&gi2c->se); +if (ret) +return ret; + ret =geni_se_resources_on (&gi2c->se);if (ret) {dev_err (dev,"Error turning on resources %d\n" , ret);I think you need to call
geni_icc_vote_off()here and in other error paths.
No need to add geni_icc_vote_off() on error paths because we are using devm_of_icc_get to get ICC path handle.
If we are going to fail probe then "devm_icc_release---->icc_put" will set votes to 0 on each path. https://patchwork.kernel.org/patch/11467507/
Regards,
Akash
Signature · 1 line
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,\na Linux Foundation Collaborative Project
Re: [PATCH V3 5/8] spi: spi-geni-qcom: Add interconnect support
In reply to Matthias KaehlckeHi Matthias,
Quoted 7 lines
drivers/spi/spi-geni-qcom.c static int spi_geni_probe(struct platform_device *pdev) · line 593
Quoted 4 lines
drivers/spi/spi-geni-qcom.c +3
-0+ ret =geni_icc_get (&mas->se,"qup-core" ,"qup-config" ,NULL ); +if (ret) + goto spi_geni_probe_runtime_disable;This fails without providing any hints why, besides the error code. It might be worth to add error logging to
geni_icc_get().
As per Bjorn's comment, I will add error logs inside geni_icc_get API incase it returns something other than -EPROBE_DEFER . https://patchwork.kernel.org/patch/11467511/
regards,
Akash
Signature · 1 line
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,\na Linux Foundation Collaborative Project
Re: [PATCH V3 6/8] tty: serial: qcom_geni_serial: Add interconnect support
In reply to Matthias KaehlckeHi Matthias,
Quoted 88 lines · Matthias Kaehlcke
Hi Akash,
Quoted 58 lines · Akash Asthana
Get the interconnect paths for Uart based Serial Engine device and vote according to the baud rate requirement of the driver.
Signed-off-by Akash Asthana <[email protected]>---
Changes in V2:
- As per Bjorn's comment, removed se == NULL check fromgeni_serial_icc_get
- As per Bjorn's comment, removed code to set se->icc_path* to NULL in failure
- As per Bjorn's comment, introduced and usingdevm_of_icc_getAPI for getting path handle
- As per Matthias comment, added error handling foricc_set_bwcallChanges in V3:
- As per Matthias comment, use common library APIs defined in geni-se driver for ICC functionality.drivers/tty/serial/qcom_geni_serial.c | 28 +++++++++++++++++++++++++--- 1 file changed, 25 insertions(+), 3 deletions(-)drivers/tty/serial/qcom_geni_serial.c +14
-3@@ -965,6 +965,14 @@ static void qcom_geni_serial_set_termios(struct uart_port *uport, ser_clk_cfg =SER_CLK_EN ; ser_clk_cfg |= clk_div <<CLK_DIV_SHFT ; +/* +* Bump up BW vote on CPU path as driver supports FIFO mode only. +* Assume peak_bw as twice of avg_bw. +*/ + port->se.from_cpu.avg_bw =Bps_to_icc (baud); + port->se.from_cpu.peak_bw =Bps_to_icc (2 * baud); +geni_icc_vote_on (&port->se); +/* parity */ tx_trans_cfg =readl (uport->membase +SE_UART_TX_TRANS_CFG ); tx_parity_cfg =readl (uport->membase +SE_UART_TX_PARITY_CFG );@@ -1202,11 +1210,14 @@ static void qcom_geni_serial_pm(struct uart_port *uport,if (old_state ==UART_PM_STATE_UNDEFINED ) old_state =UART_PM_STATE_OFF ; -if (new_state ==UART_PM_STATE_ON && old_state ==UART_PM_STATE_OFF ) +if (new_state ==UART_PM_STATE_ON && old_state ==UART_PM_STATE_OFF ) { +geni_icc_vote_on (&port->se);geni_se_resources_on (&port->se); -else if (new_state ==UART_PM_STATE_OFF && - old_state ==UART_PM_STATE_ON ) + }else if (new_state ==UART_PM_STATE_OFF && + old_state ==UART_PM_STATE_ON ) {geni_se_resources_off (&port->se); +geni_icc_vote_off (&port->se); + } }static const struct uart_ops qcom_geni_console_pops = {@@ -1304,6 +1315,17 @@ static int qcom_geni_serial_probe(struct platform_device *pdev)return -ENOMEM ;drivers/tty/serial/qcom_geni_serial.c static int qcom_geni_serial_probe(struct platform_device *pdev) · line 1325
Quoted 10 lines
drivers/tty/serial/qcom_geni_serial.c +8
-0} + ret =geni_icc_get (&port->se,"qup-core" ,"qup-config" ,NULL ); +if (ret) +return ret; +/* Set the bus quota to a reasonable value */ + port->se.to_core.avg_bw = console ?GENI_DEFAULT_BW : +Bps_to_icc (CORE_2X_50_MHZ ); + port->se.to_core.peak_bw = console ?GENI_DEFAULT_BW : +Bps_to_icc (CORE_2X_100_MHZ );I'm still unconvinced about the setting of the core bandwidth based on whether the port is used as console or not. It could possibly break consoles working at speeds > 115kbs and reserve more bandwidth than necessary for ports with 'slow' devices.
Why not scale the core bandwidth dynamically? You said earlier that there is no clear/linear translation of port speed to bandwidth, but you could use the same logic that is implicitly used here:
if (baudrate <= 115200) { avg_bw = GENI_DEFAULT_BW; peak_bw = GENI_DEFAULT_BW; } else { avg_bw = Bps_to_icc(CORE_2X_50_MHZ); peak_bw = Bps_to_icc(CORE_2X_100_MHZ); }
Quoted 3 lines
This would be more robust, power efficient and future readers of the code don't have to wonder "why is the console special?" when our discussions on this will be long forgotten.
Okay, I will add this piece of code in set_termios call of the driver because I don't have baudrate information during probe. It covers the console case mentioned in probe function.
Regards,
Akash
Signature · 1 line
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,\na Linux Foundation Collaborative Project
Re: [PATCH V3 6/8] tty: serial: qcom_geni_serial: Add interconnect support
Hi Matthias,
Quoted 18 lines
Quoted 18 lines
drivers/tty/serial/qcom_geni_serial.c static int qcom_geni_serial_probe(struct · line 1325
Quoted 13 lines
static const struct
uart_opsqcom_geni_console_pops= {drivers/tty/serial/qcom_geni_serial.c +7
-0@@ -1304,6 +1315,17 @@ static int qcom_geni_serial_probe(struct platform_device *pdev) return -ENOMEM; } + ret = geni_icc_get(&port->se, "qup-core", "qup-config", NULL); + if (ret) + return ret; + /* Set the bus quota to a reasonable value */ + port->se.to_core.avg_bw = console ? GENI_DEFAULT_BW : + Bps_to_icc(CORE_2X_50_MHZ); + port->se.to_core.peak_bw = console ? GENI_DEFAULT_BW : + Bps_to_icc(CORE_2X_100_MHZ);I'm still unconvinced about the setting of the core bandwidth based on whether the port is used as console or not. It could possibly break consoles working at speeds > 115kbs and reserve more bandwidth than necessary for ports with 'slow' devices.
Quoted 7 lines
Quoted 7 lines
Why not scale the core bandwidth dynamically? You said earlier that there is no clear/linear translation of port speed to bandwidth, but you could use the same logic that is implicitly used here:
if (baudrate <= 115200) {
avg_bw=GENI_DEFAULT_BW;
peak_bw=GENI_DEFAULT_BW;
I will make peak_bw = 2 * DEFAULT to generalize this logic and will factor it out in common driver.
Anyway with peak_bw = GENI_DEFAULT_BW or 2 * GENI_DEFAULT_BW core clock is going to tick at 50 MHz.
9600(19.2 MHz) < GENI_DEFAULT_BW, 2 * GENI_DEFAULT_BW < 2500(50 MHz).
Regards,
Akash
Quoted 15 lines
Quoted 8 lines
} else {
avg_bw=Bps_to_icc(CORE_2X_50_MHZ);
peak_bw=Bps_to_icc(CORE_2X_100_MHZ);
}This would be more robust, power efficient and future readers of the code don't have to wonder "why is the console special?" when our discussions on this will be long forgotten.
Okay, I will add this piece of code in
set_termioscall of the driver because I don't have baudrate information during probe. It covers the console case mentioned in probe function.Regards,
Akash
Signature · 1 line
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,\na Linux Foundation Collaborative Project
Re: [PATCH V3 7/8] spi: spi-qcom-qspi: Add interconnect support
In reply to Mark BrownHi Mark,
Quoted 10 lines · Mark Brown
drivers/spi/spi-qcom-qspi.c static int qcom_qspi_transfer_one(struct spi_master *master, · line 253
Quoted 6 lines · Akash Asthana
drivers/spi/spi-qcom-qspi.c +6
-0+/* +* Set BW quota for CPU as driver supports FIFO mode only. +* Assume peak bw as twice of avg bw. +*/ + ctrl->avg_bw_cpu =Bps_to_icc (speed_hz); + ctrl->peak_bw_cpu =Bps_to_icc (2 * speed_hz);I thought you were going to factor this best guess handling of peak bandwidth out into the core?
I can centralize this for SPI, I2C and UART in Common driver(QUP wrapper) but still for QSPI I have to keep this piece of code as is because It is not child of QUP wrapper(it doesn't use common code).
I am not sure whether I can move this " Assume peak_bw as twice of avg_bw if nothing is mentioned explicitly" to ICC core because the factor of 2 is chosen randomly by me.
Regards,
Akash
Signature · 1 line
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,\na Linux Foundation Collaborative Project
Re: [PATCH V3 2/8] soc: qcom: geni: Support for ICC voting
In reply to Akash AsthanaHi,
Quoted 10 lines · Akash Asthana
Hi Bjorn, Evan,
Quoted 4 lines
Quoted 4 lines
Given that these two functions only switch the bandwidth request between some value and 0, I really think we should carry a "bool enabled" on the path and replace these two functions with icc_bulk_enable()/icc_bulk_disable().
So, if above is implementation "bool enabled" on path can be used directly in aggregation of ICC votes on particular node without using
icc_set_bwcall, if yes then I am not aware how? or we'll be usingicc_set_bwAPI indirectly insideicc_bulkAPIs?
If there is a repeated pattern to switch between some bandwidth value and zero, it really makes sense to introduce such functions in the framework core. I think that this might be very useful especially for suspend and resume cases. Something like icc_{enable,disable}(struct icc_path *path) functions and also the bulk versions, that will flag the path as disabled, re-aggregate and do icc_set_bw().
Quoted 6 lines
Quoted 3 lines
Quoted 3 lines
The added benefit of this would be that you call
icc_set_bw()instead of changing the geni_icc_path->{avg_bw,peak_bw} and don't need to keep track of them here.Ok IIUC, we need to call
icc_set_bw()from GENI driver only if we change (avg_bw|peak_bw)?
Yes, exactly.
Thanks,
Georgi
Quoted 14 lines
Regards,
Akash
Quoted 8 lines
Yes yes! I had the same thought here [1].
Georgi, what do you think?
-Evan
Mark Brown
maintainerRe: [PATCH V3 7/8] spi: spi-qcom-qspi: Add interconnect support
In reply to Akash AsthanaQuoted 11 lines · Akash Asthana
Quoted 5 lines · Mark Brown
drivers/spi/spi-qcom-qspi.c static int qcom_qspi_transfer_one(struct spi_master *master, · line 253
Quoted 2 lines
drivers/spi/spi-qcom-qspi.c +2
-0+ ctrl->avg_bw_cpu =Bps_to_icc (speed_hz); + ctrl->peak_bw_cpu =Bps_to_icc (2 * speed_hz);I thought you were going to factor this best guess handling of peak bandwidth out into the core?
I can centralize this for SPI, I2C and UART in Common driver(QUP wrapper) but still for QSPI I have to keep this piece of code as is because It is not child of QUP wrapper(it doesn't use common code).
Why not?
Quoted 3 lines
I am not sure whether I can move this " Assume
peak_bwas twice ofavg_bwif nothing is mentioned explicitly" to ICC core because the factor of 2 is chosen randomly by me.
That's the whole point - if this is just a random number then we may as well at least be consistently random.
Re: [PATCH V3 3/8] soc: qcom-geni-se: Add interconnect support to fix earlycon crash
In reply to Matthias KaehlckeHi Matthias,
Quoted 32 lines
Quoted 12 lines
static int geni_se_probe(struct platform_device *pdev) { struct device *dev = &pdev->dev;drivers/soc/qcom/qcom-geni-se.c +5
-0@@ -845,6 +868,34 @@ static int geni_se_probe(struct platform_device *pdev) } } +#ifdef CONFIG_SERIAL_EARLYCON + wrapper->to_core.path =devm_of_icc_get (dev,"qup-core" ); +if (IS_ERR (wrapper->to_core.path)) +return PTR_ERR (wrapper->to_core.path); +/* drivers/soc/qcom/qcom-geni-se.c static int geni_se_probe(struct platform_device *pdev) · line 885
Quoted 10 lines
drivers/soc/qcom/qcom-geni-se.c +10
-0+* Put minmal BW request on core clocks on behalf of early console. +* The vote will be removed earlycon exit function. +* +* Note: We are putting vote on each QUP wrapper instead only to which +* earlycon is connected because QUP core clock of different wrapper +* share same voltage domain. If core1 is put to 0, then core2 will +* also run at 0, if not voted. Default ICC vote will be removed ASA +* we touch any of the core clock. +* core1 = core2 = max(core1, core2) +*/ I don't really understand this part. According to the comment if we vote (let's say) for core2 but not for core1 then:
core1: 0
core2:GENI_DEFAULT_BWcore1 = core2 =
max(core1, core2)
or
core1 = core2 =max(0, GENI_DEFAULT_BW)
Quoted 5 lines
hence
core1 = core2 =
GENI_DEFAULT_BWWhat am I missing, why is it necessary to vote for both/all?
say core1 is for earlycon usecase
There is common switch to control both the QUP core clock. I guess most appropriate description would be switch = max(vote_on_core1, vote_on_core2) + default_vote.
During early bootup, vote_on_core1 = 0, vote_on_core2 = 0;
As earlycon was working even without voting it's core need because there was some default vote present on the core switch by ICC during bootup.
So if any child(say SPI) of other QUP wrapper resumed and suspended before earlycon wrapper comes up. This will make core clock to run at zero and will cause NOC issue because vote_on_core1 = 0, vote_on_core2 = 0; and it seems default votes from core switch is removed ASA it's voted on any core.
Regards,
Akash
Signature · 1 line
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,\na Linux Foundation Collaborative Project
Re: [PATCH V3 3/8] soc: qcom-geni-se: Add interconnect support to fix earlycon crash
Hi Akash,
Quoted 47 lines · Akash Asthana
Hi Matthias,
Quoted 31 lines
Quoted 11 lines
static int geni_se_probe(struct platform_device *pdev) { struct device *dev = &pdev->dev;drivers/soc/qcom/qcom-geni-se.c +5
-0@@ -845,6 +868,34 @@ static int geni_se_probe(struct platform_device *pdev) } } +#ifdef CONFIG_SERIAL_EARLYCON + wrapper->to_core.path =devm_of_icc_get (dev,"qup-core" ); +if (IS_ERR (wrapper->to_core.path)) +return PTR_ERR (wrapper->to_core.path); +/* drivers/soc/qcom/qcom-geni-se.c static int geni_se_probe(struct platform_device *pdev) · line 884
Quoted 10 lines
drivers/soc/qcom/qcom-geni-se.c +10
-0+* Put minmal BW request on core clocks on behalf of early console. +* The vote will be removed earlycon exit function. +* +* Note: We are putting vote on each QUP wrapper instead only to which +* earlycon is connected because QUP core clock of different wrapper +* share same voltage domain. If core1 is put to 0, then core2 will +* also run at 0, if not voted. Default ICC vote will be removed ASA +* we touch any of the core clock. +* core1 = core2 = max(core1, core2) +*/ I don't really understand this part. According to the comment if we vote (let's say) for core2 but not for core1 then:
core1: 0
core2:GENI_DEFAULT_BWcore1 = core2 =
max(core1, core2)
or
core1 = core2 =max(0, GENI_DEFAULT_BW)Quoted 5 lines
hence
core1 = core2 =
GENI_DEFAULT_BWWhat am I missing, why is it necessary to vote for both/all?
say core1 is for earlycon usecase
There is common switch to control both the QUP core clock. I guess most appropriate description would be switch =
max(vote_on_core1, vote_on_core2)+default_vote.During early bootup,
vote_on_core1= 0,vote_on_core2= 0;
Quoted 7 lines
As earlycon was working even without voting it's core need because there was some default vote present on the core switch by ICC during bootup.
So if any child(say SPI) of other QUP wrapper resumed and suspended before earlycon wrapper comes up. This will make core clock to run at zero and will cause NOC issue because
vote_on_core1= 0,vote_on_core2= 0; and it seems default votes from core switch is removed ASA it's voted on any core.
Thanks for the explication!
You are probably totally right, but for some reason my brain still resists to get it ...
With the above my current interpretation is (assuming earlycon only votes on core1):
core1 core2 default switch
early boot 0 0 1 1 SPI resume (core2) 0 1 0 1 SPI suspend (core2) 0 0 0 0
earlycon init 1 0 0 1
What is wrong in the above table?
Thanks for bearing with me :)
Matthias
Re: [PATCH V3 2/8] soc: qcom: geni: Support for ICC voting
In reply to Akash AsthanaQuoted 35 lines · Akash Asthana
Hi Bjorn,
Quoted 19 lines · Bjorn Andersson
Quoted 17 lines · Akash Asthana
Add necessary macros and structure variables to support ICC BW voting from individual SE drivers.
Signed-off-by Akash Asthana <[email protected]>---
Changes in V2:
- As per Bjorn's comment dropped enums for ICC paths, given the three paths individual membersChanges in V3:
- Addgeni_icc_get,geni_icc_vote_onandgeni_icc_vote_offas helper API.
- Addgeni_icc_pathstructure in common headerdrivers/soc/qcom/qcom-geni-se.c | 98 +++++++++++++++++++++++++++++++++++++++++ include/linux/qcom-geni-se.h | 36 +++++++++++++++ 2 files changed, 134 insertions(+)Quoted 13 lines
drivers/soc/qcom/qcom-geni-se.c void geni_se_rx_dma_unprep(struct geni_se *se, dma_addr_t iova, size_t len) · line 725
Quoted 11 lines
drivers/soc/qcom/qcom-geni-se.c +4
-0@@ -720,6 +720,104 @@ void geni_se_rx_dma_unprep(struct geni_se *se, dma_addr_t iova, size_t len) }EXPORT_SYMBOL (geni_se_rx_dma_unprep); +int geni_icc_get (struct geni_se *se,const char *icc_core,const char *icc_cpu, +const char *icc_ddr) +{ +if (icc_core) {Afaict it's only this that might be passed as NULL, so please drop these conditionals (keep the last one).
Quoted 1 line
IIUC you're suggesting to drop if (icc_core/cpu) but keep if (
icc_ddr) ?
Correct
Thanks,
Bjorn
Re: [PATCH V3 2/8] soc: qcom: geni: Support for ICC voting
In reply to Georgi DjakovHi Georgi, Bjorn, Evan,
Quoted 13 lines · Georgi Djakov
Hi,
Quoted 10 lines · Akash Asthana
Hi Bjorn, Evan,
Quoted 4 lines
Quoted 4 lines
Given that these two functions only switch the bandwidth request between some value and 0, I really think we should carry a "bool enabled" on the path and replace these two functions with icc_bulk_enable()/icc_bulk_disable().
So, if above is implementation "bool enabled" on path can be used directly in aggregation of ICC votes on particular node without using
icc_set_bwcall, if yes then I am not aware how? or we'll be usingicc_set_bwAPI indirectly insideicc_bulkAPIs?
Quoted 6 lines
If there is a repeated pattern to switch between some bandwidth value and zero, it really makes sense to introduce such functions in the framework core. I think that this might be very useful especially for suspend and resume cases. Something like icc_{enable,disable}(struct
icc_path*path) functions and also the bulk versions, that will flag the path as disabled, re-aggregate and doicc_set_bw().
This appears to be a non-trivial change to ICC core, as my understanding of ICC core is limited as of now hence, I am not very clear of the implementation of icc_bulk APIs.
Will it be okay if I keep geni_icc_vote_on/off API as is@https://patchwork.kernel.org/patch/11467511/ for now and later will switch to icc_bulk once it's introduced in ICC core.
Regards,
Akash
Quoted 22 lines
Quoted 5 lines
Quoted 3 lines
Quoted 3 lines
The added benefit of this would be that you call
icc_set_bw()instead of changing the geni_icc_path->{avg_bw,peak_bw} and don't need to keep track of them here.Ok IIUC, we need to call
icc_set_bw()from GENI driver only if we change (avg_bw|peak_bw)?Yes, exactly.
Thanks,
GeorgiQuoted 12 lines
Regards,
Akash
Quoted 8 lines
Yes yes! I had the same thought here [1].
Georgi, what do you think?
-Evan
Signature · 1 line
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,\na Linux Foundation Collaborative Project
Re: [PATCH V3 3/8] soc: qcom-geni-se: Add interconnect support to fix earlycon crash
In reply to Matthias KaehlckeHi Matthias,
Quoted 69 lines · Matthias Kaehlcke
Hi Akash,
Quoted 47 lines · Akash Asthana
Hi Matthias,
Quoted 31 lines
Quoted 11 lines
static int geni_se_probe(struct platform_device *pdev) { struct device *dev = &pdev->dev;drivers/soc/qcom/qcom-geni-se.c +5
-0@@ -845,6 +868,34 @@ static int geni_se_probe(struct platform_device *pdev) } } +#ifdef CONFIG_SERIAL_EARLYCON + wrapper->to_core.path =devm_of_icc_get (dev,"qup-core" ); +if (IS_ERR (wrapper->to_core.path)) +return PTR_ERR (wrapper->to_core.path); +/* drivers/soc/qcom/qcom-geni-se.c static int geni_se_probe(struct platform_device *pdev) · line 884
Quoted 10 lines
drivers/soc/qcom/qcom-geni-se.c +10
-0+* Put minmal BW request on core clocks on behalf of early console. +* The vote will be removed earlycon exit function. +* +* Note: We are putting vote on each QUP wrapper instead only to which +* earlycon is connected because QUP core clock of different wrapper +* share same voltage domain. If core1 is put to 0, then core2 will +* also run at 0, if not voted. Default ICC vote will be removed ASA +* we touch any of the core clock. +* core1 = core2 = max(core1, core2) +*/ I don't really understand this part. According to the comment if we vote (let's say) for core2 but not for core1 then:
core1: 0
core2:GENI_DEFAULT_BWcore1 = core2 =
max(core1, core2)
or
core1 = core2 =max(0, GENI_DEFAULT_BW)Quoted 5 lines
hence
core1 = core2 =
GENI_DEFAULT_BWWhat am I missing, why is it necessary to vote for both/all?
say core1 is for earlycon usecase
There is common switch to control both the QUP core clock. I guess most appropriate description would be switch =
max(vote_on_core1, vote_on_core2)+default_vote.During early bootup,
vote_on_core1= 0,vote_on_core2= 0;Quoted 7 lines
As earlycon was working even without voting it's core need because there was some default vote present on the core switch by ICC during bootup.
So if any child(say SPI) of other QUP wrapper resumed and suspended before earlycon wrapper comes up. This will make core clock to run at zero and will cause NOC issue because
vote_on_core1= 0,vote_on_core2= 0; and it seems default votes from core switch is removed ASA it's voted on any core.Thanks for the explication!
You are probably totally right, but for some reason my brain still resists to get it ...
With the above my current interpretation is (assuming earlycon only votes on core1):
core1 core2 default switch
early boot 0 0 1 1 SPI resume (core2) 0 1 0 1 SPI suspend (core2) 0 0 0 0
Quoted 6 lines
earlycon init 1 0 0 1
What is wrong in the above table?
Thanks for bearing with me :)
NP :)
I guess you meant QUP WRAPPER 1 probe by "earlycon init".
core1 core2 default switch Time
early boot 0 0 1 1 0s SPI resume (core2) 0 1 0 1 3.2s SPI suspend (core2) 0 0 0 0 3.3s QUP WRAPPER 1(probe) 1 0 0 1 5s (say)
So switch is at 0 in time interval [3.3, 5] that will make core clock to run at 0. If we use earlycon during this time interval it will reset the board.
Did above answered the query?
Regards,
Akash
Quoted 2 lines
Matthias
Signature · 1 line
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,\na Linux Foundation Collaborative Project
Re: [PATCH V3 7/8] spi: spi-qcom-qspi: Add interconnect support
In reply to Mark BrownHi Mark, Evan, Georgi,
Quoted 9 lines · Mark Brown
Quoted 8 lines · Akash Asthana
Quoted 4 lines · Mark Brown
drivers/spi/spi-qcom-qspi.c static int qcom_qspi_transfer_one(struct spi_master *master, · line 253
Quoted 2 lines
drivers/spi/spi-qcom-qspi.c +2
-0+ ctrl->avg_bw_cpu =Bps_to_icc (speed_hz); + ctrl->peak_bw_cpu =Bps_to_icc (2 * speed_hz);I thought you were going to factor this best guess handling of peak bandwidth out into the core?
I can centralize this for SPI, I2C and UART in Common driver(QUP wrapper) but still for QSPI I have to keep this piece of code as is because It is not child of QUP wrapper(it doesn't use common code).
Quoted 7 lines
Why not?
Quoted 3 lines
I am not sure whether I can move this " Assume
peak_bwas twice ofavg_bwif nothing is mentioned explicitly" to ICC core because the factor of 2 is chosen randomly by me.That's the whole point - if this is just a random number then we may as well at least be consistently random.
Can we centralize below logic of peak_bw selection for all the clients to ICC core?
"Assume peak_bw requirement as twice of avg_bw, if it is not mentioned explicitly"
===========================================================================
int icc_set_bw(struct icc_path *path, u32 avg_bw, u32 peak_bw) {
struct icc_node *node;
u32 old_avg, old_peak;
size_t i;
int ret;
if (!path)
return 0;
if (WARN_ON(IS_ERR(path) || !path->num_nodes)) return -EINVAL;
+ /*
+ * Assume peak_bw requirement as twice of avg_bw, if it is not + * mentioned explicitly
+ */
+ peak_bw = peak_bw ? peak_bw : 2 * avg_bw;
===========================================================================
In case if some client really don't want to put peak requirement they can pass avg_bw = peak_bw. As peak_bw <= avg_bw is kind of no-ops.
Regards,
Akash
Signature · 1 line
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,\na Linux Foundation Collaborative Project
Re: [PATCH V3 3/8] soc: qcom-geni-se: Add interconnect support to fix earlycon crash
In reply to Akash AsthanaQuoted 88 lines · Akash Asthana
Hi Matthias,
Quoted 69 lines · Matthias Kaehlcke
Hi Akash,
Quoted 47 lines · Akash Asthana
Hi Matthias,
Quoted 31 lines
Quoted 11 lines
static int geni_se_probe(struct platform_device *pdev) { struct device *dev = &pdev->dev;drivers/soc/qcom/qcom-geni-se.c +5
-0@@ -845,6 +868,34 @@ static int geni_se_probe(struct platform_device *pdev) } } +#ifdef CONFIG_SERIAL_EARLYCON + wrapper->to_core.path =devm_of_icc_get (dev,"qup-core" ); +if (IS_ERR (wrapper->to_core.path)) +return PTR_ERR (wrapper->to_core.path); +/* drivers/soc/qcom/qcom-geni-se.c static int geni_se_probe(struct platform_device *pdev) · line 884
Quoted 10 lines
drivers/soc/qcom/qcom-geni-se.c +10
-0+* Put minmal BW request on core clocks on behalf of early console. +* The vote will be removed earlycon exit function. +* +* Note: We are putting vote on each QUP wrapper instead only to which +* earlycon is connected because QUP core clock of different wrapper +* share same voltage domain. If core1 is put to 0, then core2 will +* also run at 0, if not voted. Default ICC vote will be removed ASA +* we touch any of the core clock. +* core1 = core2 = max(core1, core2) +*/ I don't really understand this part. According to the comment if we vote (let's say) for core2 but not for core1 then:
core1: 0
core2:GENI_DEFAULT_BWcore1 = core2 =
max(core1, core2)
or
core1 = core2 =max(0, GENI_DEFAULT_BW)Quoted 5 lines
hence
core1 = core2 =
GENI_DEFAULT_BWWhat am I missing, why is it necessary to vote for both/all?
say core1 is for earlycon usecase
There is common switch to control both the QUP core clock. I guess most appropriate description would be switch =
max(vote_on_core1, vote_on_core2)+default_vote.During early bootup,
vote_on_core1= 0,vote_on_core2= 0;Quoted 7 lines
As earlycon was working even without voting it's core need because there was some default vote present on the core switch by ICC during bootup.
So if any child(say SPI) of other QUP wrapper resumed and suspended before earlycon wrapper comes up. This will make core clock to run at zero and will cause NOC issue because
vote_on_core1= 0,vote_on_core2= 0; and it seems default votes from core switch is removed ASA it's voted on any core.Thanks for the explication!
You are probably totally right, but for some reason my brain still resists to get it ...
With the above my current interpretation is (assuming earlycon only votes on core1):
core1 core2 default switch
early boot 0 0 1 1 SPI resume (core2) 0 1 0 1 SPI suspend (core2) 0 0 0 0Quoted 6 lines
earlycon init 1 0 0 1
What is wrong in the above table?
Thanks for bearing with me :)
NP :)
I guess you meant QUP WRAPPER 1 probe by "earlycon init".
core1 core2 default switch Time
early boot 0 0 1 1 0s SPI resume (core2) 0 1 0 1 3.2s SPI suspend (core2) 0 0 0 0 3.3s QUP WRAPPER 1(probe) 1 0 0 1 5s (say)
Quoted 4 lines
So switch is at 0 in time interval [3.3, 5] that will make core clock to run at 0. If we use earlycon during this time interval it will reset the board.
Did above answered the query?
now it finally made 'click', thanks :)
Re: [PATCH V3 7/8] spi: spi-qcom-qspi: Add interconnect support
In reply to Akash AsthanaHi Akash,
Quoted 14 lines · Akash Asthana
Hi Mark, Evan, Georgi,
Quoted 11 lines · Mark Brown
Quoted 8 lines · Akash Asthana
Quoted 4 lines · Mark Brown
Quoted 2 lines
+
ctrl->avg_bw_cpu=Bps_to_icc(speed_hz); +ctrl->peak_bw_cpu=Bps_to_icc(2 * speed_hz);I thought you were going to factor this best guess handling of peak bandwidth out into the core?
I can centralize this for SPI, I2C and UART in Common driver(QUP wrapper) but still for QSPI I have to keep this piece of code as is because It is not child of QUP wrapper(it doesn't use common code).
Why not?
Quoted 7 lines
Quoted 5 lines
Quoted 3 lines
I am not sure whether I can move this " Assume
peak_bwas twice ofavg_bwif nothing is mentioned explicitly" to ICC core because the factor of 2 is chosen randomly by me.That's the whole point - if this is just a random number then we may as well at least be consistently random.
Can we centralize below logic of
peak_bwselection for all the clients to ICC core?
I don't think this is a good idea for now, because this is very hardware specific. A scaling factor that works for one client might not work for another.
My questions here is how did you decide on this "multiply by two"? I can imagine that the traffic can be bursty on some interfaces, but is the factor here really a "random number" or is this based on some data patterns or performance analysis?
Thanks,
Georgi
Mark Brown
maintainerRe: [PATCH V3 7/8] spi: spi-qcom-qspi: Add interconnect support
Quoted 6 lines · Georgi Djakov
Quoted 1 line · Akash Asthana
Can we centralize below logic of
peak_bwselection for all the clients to ICC core?I don't think this is a good idea for now, because this is very hardware specific. A scaling factor that works for one client might not work for another.
AIUI a driver can always override the setting if it's got a better idea.
Quoted 4 lines
My questions here is how did you decide on this "multiply by two"? I can imagine that the traffic can be bursty on some interfaces, but is the factor here really a "random number" or is this based on some data patterns or performance analysis?
The reason I'm pushing for this to go into the core is that the numbers seem to be just made up and not device specific at all (or at least there's a lot of devices with the same values).
Re: [PATCH V3 7/8] spi: spi-qcom-qspi: Add interconnect support
Hi,
Quoted 9 lines · Mark Brown
Quoted 6 lines · Georgi Djakov
Quoted 1 line · Akash Asthana
Can we centralize below logic of
peak_bwselection for all the clients to ICC core?I don't think this is a good idea for now, because this is very hardware specific. A scaling factor that works for one client might not work for another.
AIUI a driver can always override the setting if it's got a better idea.
True.
Quoted 8 lines
Quoted 4 lines
My questions here is how did you decide on this "multiply by two"? I can imagine that the traffic can be bursty on some interfaces, but is the factor here really a "random number" or is this based on some data patterns or performance analysis?
The reason I'm pushing for this to go into the core is that the numbers seem to be just made up and not device specific at all (or at least there's a lot of devices with the same values).
I might be a bit too cautious here, but the main question is what this common number should be then. Maybe the numbers are not device specific, but communication interface/bus specific? I would rather make the peak 30% higher than the average, but for low speed interfaces i guess the factor of two should also be fine?
Thanks,
Georgi
Re: [PATCH V3 7/8] spi: spi-qcom-qspi: Add interconnect support
In reply to a message not in the archiveHi Akash,
Quoted 34 lines · Akash Asthana
Hi Georgi,
Quoted 22 lines · Georgi Djakov
Hi Akash,
Quoted 19 lines · Akash Asthana
Hi Mark, Evan, Georgi,
Quoted 16 lines · Mark Brown
Quoted 8 lines · Akash Asthana
Quoted 4 lines · Mark Brown
Quoted 2 lines
+
ctrl->avg_bw_cpu=Bps_to_icc(speed_hz); +ctrl->peak_bw_cpu=Bps_to_icc(2 * speed_hz);I thought you were going to factor this best guess handling of peak bandwidth out into the core?
I can centralize this for SPI, I2C and UART in Common driver(QUP wrapper) but still for QSPI I have to keep this piece of code as is because It is not child of QUP wrapper(it doesn't use common code).
Why not?
Quoted 3 lines
I am not sure whether I can move this " Assume
peak_bwas twice ofavg_bwif nothing is mentioned explicitly" to ICC core because the factor of 2 is chosen randomly by me.That's the whole point - if this is just a random number then we may as well at least be consistently random.
Quoted 8 lines
Quoted 1 line
Can we centralize below logic of
peak_bwselection for all the clients to ICC core?I don't think this is a good idea for now, because this is very hardware specific. A scaling factor that works for one client might not work for another.
My questions here is how did you decide on this "multiply by two"? I can imagine that the traffic can be bursty on some interfaces, but is the factor here really a "random number" or is this based on some data patterns or performance analysis?
Quoted 7 lines
Factor of 2 is random number.
We are taking care of actual throughput requirement in
avg_bwvote and the intention of putting peak as twice of avg is to ensure that if high speed peripherals(ex:USB) removes their votes, we shouldn't see any latency issue because of other ICC client who don't vote for their BW requirement or *actual* BW requirement.
Thanks for clarifying, but is this latency a confirmed issue on real hardware? I guess voting for twice as average will work, but still wondering wouldn't it be more appropriate to handle it in the interconnect platform driver instead? Also why is the other client not voting, can we fix it?
Thanks,
Georgi