i2c-mlxbf.c: bug fixes and new feature support
Asmaa Mnebhi
maintainerv2 cover i2c-mlxbf.c: bug fixes and new feature support
This is a series of patches fixing several bugs and implementing new features.
Bug fixes:
1) Fix the frequency calculation
2) remove unnecessary IRQF_ONESHOT flag
3) Fix incorrect base address passed during io write
4) prevent stack overflow in mlxbf_i2c_smbus_start_transaction()
5) Support lock mechanism
New features:
6) Update SPDX license
7) Support multi slave functionality
8) Support BlueField-3 SoC
9) Update binding devicetree
Asmaa Mnebhi (9):
i2c-mlxbf.c: Fix frequency calculation i2c-mlxbf.c: remove IRQF_ONESHOT
i2c-mlxbf.c: incorrect base address passed during io write
i2c-mlxbf: prevent stack overflow in
mlxbf_i2c_smbus_start_transaction()
i2c-mlxbf.c: support lock mechanism
i2c-mlxbf.c: Update SPDX license
i2c-mlxbf: add multi slave functionality i2c-mlxbf.c: support BlueField-3 SoC i2c-mlxbf.c: Update binding devicetree
.../bindings/i2c/mellanox,i2c-mlxbf.yaml | 49 +- MAINTAINERS | 1 + drivers/i2c/busses/i2c-mlxbf.c | 872 ++++++++++-------- 3 files changed, 525 insertions(+), 397 deletions(-)
Signature · 1 line
2.30.1
Asmaa Mnebhi
maintainerPATCH v2 1/9 i2c-mlxbf.c: Fix frequency calculation
The i2c-mlxbf.c driver is currently broken because there is a bug in the calculation of the frequency. core_f, core_r and core_od are components read from hardware registers and are used to compute the frequency used to compute different timing parameters. The shifting mechanism used to get core_f, core_r and core_od is wrong. Use FIELD_GET to mask and shift the bitfields properly.
--- drivers/i2c/busses/i2c-mlxbf.c | 59 ++++++++++++---------------------- 1 file changed, 21 insertions(+), 38 deletions(-)
drivers/i2c/busses/i2c-mlxbf.c +21 -38
@@ -6,6 +6,7 @@*/ #include <linux/acpi.h> +#include <linux/bitfield.h> #include <linux/delay.h> #include <linux/err.h> #include <linux/interrupt.h>
@@ -63,13 +64,14 @@*/ #define MLXBF_I2C_TYU_PLL_OUT_FREQ (400 * 1000 * 1000)/* Reference clock for Bluefield - 156 MHz. */ -#define MLXBF_I2C_PLL_IN_FREQ (156 * 1000 * 1000) +#define MLXBF_I2C_PLL_IN_FREQ 156250000ULL/* Constant used to determine the PLL frequency. */ -#define MLNXBF_I2C_COREPLL_CONST 16384 +#define MLNXBF_I2C_COREPLL_CONST 16384ULL + +#define MLXBF_I2C_FREQUENCY_1GHZ 1000000000ULL/* PLL registers. */ -#define MLXBF_I2C_CORE_PLL_REG0 0x0#define MLXBF_I2C_CORE_PLL_REG1 0x4#define MLXBF_I2C_CORE_PLL_REG2 0x8
@@ -181,22 +183,15 @@#define MLXBF_I2C_COREPLL_FREQ MLXBF_I2C_TYU_PLL_OUT_FREQ/* Core PLL TYU configuration. */ -#define MLXBF_I2C_COREPLL_CORE_F_TYU_MASK GENMASK(12, 0) -#define MLXBF_I2C_COREPLL_CORE_OD_TYU_MASK GENMASK(3, 0) -#define MLXBF_I2C_COREPLL_CORE_R_TYU_MASK GENMASK(5, 0) - -#define MLXBF_I2C_COREPLL_CORE_F_TYU_SHIFT 3 -#define MLXBF_I2C_COREPLL_CORE_OD_TYU_SHIFT 16 -#define MLXBF_I2C_COREPLL_CORE_R_TYU_SHIFT 20 +#define MLXBF_I2C_COREPLL_CORE_F_TYU_MASK GENMASK(15, 3) +#define MLXBF_I2C_COREPLL_CORE_OD_TYU_MASK GENMASK(19, 16) +#define MLXBF_I2C_COREPLL_CORE_R_TYU_MASK GENMASK(25, 20)/* Core PLL YU configuration. */ #define MLXBF_I2C_COREPLL_CORE_F_YU_MASK GENMASK(25, 0)#define MLXBF_I2C_COREPLL_CORE_OD_YU_MASK GENMASK(3, 0) -#define MLXBF_I2C_COREPLL_CORE_R_YU_MASK GENMASK(5, 0) +#define MLXBF_I2C_COREPLL_CORE_R_YU_MASK GENMASK(31, 26) -#define MLXBF_I2C_COREPLL_CORE_F_YU_SHIFT 0 -#define MLXBF_I2C_COREPLL_CORE_OD_YU_SHIFT 1 -#define MLXBF_I2C_COREPLL_CORE_R_YU_SHIFT 26/* Core PLL frequency. */ static u64 mlxbf_i2c_corepll_frequency;
@@ -479,8 +474,6 @@ static struct mutex mlxbf_i2c_bus_lock;#define MLXBF_I2C_MASK_8 GENMASK(7, 0)#define MLXBF_I2C_MASK_16 GENMASK(15, 0) -#define MLXBF_I2C_FREQUENCY_1GHZ 1000000000 -/* * Function to poll a set of bits at a specific address; it checks whether * the bits are equal to zero when eq_zero is set to 'true', and not equal
@@ -1407,24 +1400,19 @@ static int mlxbf_i2c_init_master(struct platform_device *pdev,return 0 ; } -static u64 mlxbf_calculate_freq_from_tyu (struct mlxbf_i2c_resource *corepll_res) +static u64 mlxbf_i2c_calculate_freq_from_tyu (struct mlxbf_i2c_resource *corepll_res) { -u64 core_frequency, pad_frequency; +u64 core_frequency;u8 core_od, core_r;u32 corepll_val;u16 core_f; - pad_frequency =MLXBF_I2C_PLL_IN_FREQ ; - corepll_val =readl (corepll_res->io +MLXBF_I2C_CORE_PLL_REG1 );/* Get Core PLL configuration bits. */ - core_f =rol32 (corepll_val,MLXBF_I2C_COREPLL_CORE_F_TYU_SHIFT ) & -MLXBF_I2C_COREPLL_CORE_F_TYU_MASK ; - core_od =rol32 (corepll_val,MLXBF_I2C_COREPLL_CORE_OD_TYU_SHIFT ) & -MLXBF_I2C_COREPLL_CORE_OD_TYU_MASK ; - core_r =rol32 (corepll_val,MLXBF_I2C_COREPLL_CORE_R_TYU_SHIFT ) & -MLXBF_I2C_COREPLL_CORE_R_TYU_MASK ; + core_f =FIELD_GET (MLXBF_I2C_COREPLL_CORE_F_TYU_MASK , corepll_val); + core_od =FIELD_GET (MLXBF_I2C_COREPLL_CORE_OD_TYU_MASK , corepll_val); + core_r =FIELD_GET (MLXBF_I2C_COREPLL_CORE_R_TYU_MASK , corepll_val);/* * Compute PLL output frequency as follow:
@@ -1436,31 +1424,26 @@ static u64 mlxbf_calculate_freq_from_tyu(struct mlxbf_i2c_resource *corepll_res)* Where PLL_OUT_FREQ and PLL_IN_FREQ refer to CoreFrequency * and PadFrequency, respectively. */ - core_frequency = pad_frequency * (++core_f); + core_frequency =MLXBF_I2C_PLL_IN_FREQ * (++core_f); core_frequency /= (++core_r) * (++core_od);return core_frequency; } -static u64 mlxbf_calculate_freq_from_yu (struct mlxbf_i2c_resource *corepll_res) +static u64 mlxbf_i2c_calculate_freq_from_yu (struct mlxbf_i2c_resource *corepll_res) {u32 corepll_reg1_val, corepll_reg2_val; -u64 corepll_frequency, pad_frequency; +u64 corepll_frequency;u8 core_od, core_r;u32 core_f; - pad_frequency =MLXBF_I2C_PLL_IN_FREQ ; - corepll_reg1_val =readl (corepll_res->io +MLXBF_I2C_CORE_PLL_REG1 ); corepll_reg2_val =readl (corepll_res->io +MLXBF_I2C_CORE_PLL_REG2 );/* Get Core PLL configuration bits */ - core_f =rol32 (corepll_reg1_val,MLXBF_I2C_COREPLL_CORE_F_YU_SHIFT ) & -MLXBF_I2C_COREPLL_CORE_F_YU_MASK ; - core_r =rol32 (corepll_reg1_val,MLXBF_I2C_COREPLL_CORE_R_YU_SHIFT ) & -MLXBF_I2C_COREPLL_CORE_R_YU_MASK ; - core_od =rol32 (corepll_reg2_val,MLXBF_I2C_COREPLL_CORE_OD_YU_SHIFT ) & -MLXBF_I2C_COREPLL_CORE_OD_YU_MASK ; + core_f =FIELD_GET (MLXBF_I2C_COREPLL_CORE_F_YU_MASK , corepll_reg1_val); + core_r =FIELD_GET (MLXBF_I2C_COREPLL_CORE_R_YU_MASK , corepll_reg1_val); + core_od =FIELD_GET (MLXBF_I2C_COREPLL_CORE_OD_YU_MASK , corepll_reg2_val);/* * Compute PLL output frequency as follow:
@@ -1472,7 +1455,7 @@ static u64 mlxbf_calculate_freq_from_yu(struct mlxbf_i2c_resource *corepll_res)* Where PLL_OUT_FREQ and PLL_IN_FREQ refer to CoreFrequency * and PadFrequency, respectively. */ - corepll_frequency = (pad_frequency * core_f) /MLNXBF_I2C_COREPLL_CONST ; + corepll_frequency = (MLXBF_I2C_PLL_IN_FREQ * core_f) /MLNXBF_I2C_COREPLL_CONST ; corepll_frequency /= (++core_r) * (++core_od);return corepll_frequency;
Signature · 1 line
2.30.1
Asmaa Mnebhi
maintainerPATCH v2 2/9 i2c-mlxbf.c: remove IRQF_ONESHOT
In reply to Asmaa MnebhiIRQF_ONESHOT is not needed so remove it.
--- drivers/i2c/busses/i2c-mlxbf.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/i2c/busses/i2c-mlxbf.c +1 -1
@@ -2356,7 +2356,7 @@ static int mlxbf_i2c_probe(struct platform_device *pdev)if (irq <0 )return irq; ret =devm_request_irq (dev, irq, mlxbf_smbus_irq, -IRQF_ONESHOT |IRQF_SHARED |IRQF_PROBE_SHARED , +IRQF_SHARED |IRQF_PROBE_SHARED ,dev_name (dev), priv);if (ret <0 ) {dev_err (dev,"Cannot get irq %d\n" , irq);
Signature · 1 line
2.30.1
Asmaa Mnebhi
maintainerPATCH v2 3/9 i2c-mlxbf.c: incorrect base address passed during io write
In reply to Asmaa MnebhiCorrect the base address used during io write.
This bug had no impact over the overall functionality of the read and write transactions. MLXBF_I2C_CAUSE_OR_CLEAR=0x18 so writing to (smbus->io + 0x18) instead of (mst_cause->ioi + 0x18) actually writes to the sc_low_timeout register which just sets the timeout value before a read/write aborts.
--- drivers/i2c/busses/i2c-mlxbf.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/i2c/busses/i2c-mlxbf.c +1 -1
@@ -662,7 +662,7 @@ static int mlxbf_i2c_smbus_enable(struct mlxbf_i2c_priv *priv, u8 slave,/* Clear status bits. */ writel (0x0 , priv->smbus->io +MLXBF_I2C_SMBUS_MASTER_STATUS );/* Set the cause data. */ -writel (~0x0 , priv->smbus->io +MLXBF_I2C_CAUSE_OR_CLEAR ); +writel (~0x0 , priv->mst_cause->io +MLXBF_I2C_CAUSE_OR_CLEAR );/* Zero PEC byte. */ writel (0x0 , priv->smbus->io +MLXBF_I2C_SMBUS_MASTER_PEC );/* Zero byte count. */
Signature · 1 line
2.30.1
Asmaa Mnebhi
maintainerPATCH v2 4/9 i2c-mlxbf: prevent stack overflow in mlxbf_i2c_smbus_start_transaction()
In reply to Asmaa Mnebhimemcpy() is called in a loop while 'operation->length' upper bound is not checked and 'data_idx' also increments.
--- drivers/i2c/busses/i2c-mlxbf.c | 3 +++ 1 file changed, 3 insertions(+)
drivers/i2c/busses/i2c-mlxbf.c +3 -0
@@ -731,6 +731,9 @@ mlxbf_i2c_smbus_start_transaction(struct mlxbf_i2c_priv *priv,if (flags &MLXBF_I2C_F_WRITE ) { write_en =1 ; write_len += operation->length; +if (data_idx + operation->length > +MLXBF_I2C_MASTER_DATA_DESC_SIZE ) +return -ENOBUFS ;memcpy (data_desc + data_idx, operation->buffer, operation->length); data_idx += operation->length;
Signature · 1 line
2.30.1
Asmaa Mnebhi
maintainerPATCH v2 5/9 i2c-mlxbf.c: support lock mechanism
In reply to Asmaa MnebhiLinux is not the only entity using the BlueField I2C busses so support a lock mechanism provided by hardware to avoid issues when multiple entities are trying to access the same bus.
The lock is acquired whenever written explicitely or the lock register is read. So make sure it is always released at the end of a successful or failed transaction.
--- drivers/i2c/busses/i2c-mlxbf.c | 42 ++++++++++++++++++++++++++++++---- 1 file changed, 38 insertions(+), 4 deletions(-)
drivers/i2c/busses/i2c-mlxbf.c +38 -4
@@ -306,6 +306,7 @@ static u64 mlxbf_i2c_corepll_frequency;* exact. */ #define MLXBF_I2C_SMBUS_TIMEOUT (300 * 1000)/* 300ms */ +#define MLXBF_I2C_SMBUS_LOCK_POLL_TIMEOUT (300 * 1000)/* 300ms */ /* Encapsulates timing parameters. */ struct mlxbf_i2c_timings {
@@ -514,6 +515,25 @@ static bool mlxbf_smbus_master_wait_for_idle(struct mlxbf_i2c_priv *priv)return false; } +/* +* wait for the lock to be released before acquiring it. +*/ +static bool mlxbf_i2c_smbus_master_lock (struct mlxbf_i2c_priv *priv) +{ +if (mlxbf_smbus_poll (priv->smbus->io,MLXBF_I2C_SMBUS_MASTER_GW , +MLXBF_I2C_MASTER_LOCK_BIT , true, +MLXBF_I2C_SMBUS_LOCK_POLL_TIMEOUT )) +return true; + +return false; +} + +static void mlxbf_i2c_smbus_master_unlock (struct mlxbf_i2c_priv *priv) +{ +/* Clear the gw to clear the lock */ +writel (0 , priv->smbus->io +MLXBF_I2C_SMBUS_MASTER_GW ); +} +static bool mlxbf_i2c_smbus_transaction_success (u32 master_status,u32 cause_status) {
@@ -705,10 +725,18 @@ mlxbf_i2c_smbus_start_transaction(struct mlxbf_i2c_priv *priv, slave = request->slave &GENMASK (6 ,0 ); addr = slave <<1 ; -/* First of all, check whether the HW is idle. */ -if (WARN_ON (!mlxbf_smbus_master_wait_for_idle (priv))) +/* Try to acquire the smbus gw lock before any reads of the GW register since +* a read sets the lock. +*/ +if (WARN_ON (!mlxbf_i2c_smbus_master_lock (priv)))return -EBUSY ; +/* Check whether the HW is idle */ +if (WARN_ON (!mlxbf_smbus_master_wait_for_idle (priv))) { +mlxbf_i2c_smbus_master_unlock (priv); +return -EBUSY ; + } +/* Set first byte. */ data_desc[data_idx++] = addr;
@@ -732,8 +760,10 @@ mlxbf_i2c_smbus_start_transaction(struct mlxbf_i2c_priv *priv, write_en =1 ; write_len += operation->length;if (data_idx + operation->length > -MLXBF_I2C_MASTER_DATA_DESC_SIZE ) +MLXBF_I2C_MASTER_DATA_DESC_SIZE ) { +mlxbf_i2c_smbus_master_unlock (priv);return -ENOBUFS ; + }memcpy (data_desc + data_idx, operation->buffer, operation->length); data_idx += operation->length;
@@ -764,8 +794,10 @@ mlxbf_i2c_smbus_start_transaction(struct mlxbf_i2c_priv *priv,if (write_en) { ret =mlxbf_i2c_smbus_enable (priv, slave, write_len, block_en, pec_en,0 ); -if (ret) +if (ret) { +mlxbf_i2c_smbus_master_unlock (priv);return ret; + } }if (read_en) {
@@ -792,6 +824,8 @@ mlxbf_i2c_smbus_start_transaction(struct mlxbf_i2c_priv *priv, priv->smbus->io + MLXBF_I2C_SMBUS_MASTER_FSM); } +mlxbf_i2c_smbus_master_unlock (priv); +return ret; }
Signature · 1 line
2.30.1
Asmaa Mnebhi
maintainerPATCH v2 6/9 i2c-mlxbf.c: Update SPDX license
In reply to Asmaa MnebhiUpdate the SPDX license to GPL-2.0 or BSD-3-Clause and update the copyright to NVIDIA CORPORATION.
--- drivers/i2c/busses/i2c-mlxbf.c | 10 +++------- 1 file changed, 3 insertions(+), 7 deletions(-)
drivers/i2c/busses/i2c-mlxbf.c +3 -7
@@ -1,9 +1,5 @@ -// SPDX-License-Identifier: GPL-2.0 -/* -* Mellanox BlueField I2C bus driver -* -* Copyright (C) 2020 Mellanox Technologies, Ltd. -*/ +// SPDX-License-Identifier: GPL-2.0 or BSD-3-Clause +/* Copyright (c) 2020 NVIDIA CORPORATION. All rights reserved. */ #include <linux/acpi.h> #include <linux/bitfield.h>
@@ -2487,4 +2483,4 @@ module_exit(mlxbf_i2c_exit);MODULE_DESCRIPTION ("Mellanox BlueField I2C bus driver" );MODULE_AUTHOR ("Khalil Blaiech <[email protected]>" ); -MODULE_LICENSE ("GPL v2" ); +MODULE_LICENSE ("Dual BSD/GPL" );
Signature · 1 line
2.30.1
Asmaa Mnebhi
maintainerPATCH v2 7/9 i2c-mlxbf: add multi slave functionality
In reply to Asmaa MnebhiSupport the multi slave functionality which enables the BlueField to be registered at up to 16 i2c slave addresses.
--- drivers/i2c/busses/i2c-mlxbf.c | 320 +++++++++++++++------------------ 1 file changed, 149 insertions(+), 171 deletions(-)
drivers/i2c/busses/i2c-mlxbf.c +149 -171
@@ -294,9 +294,6 @@ static u64 mlxbf_i2c_corepll_frequency;#define MLXBF_I2C_SMBUS_SLAVE_ADDR_EN_BIT 7#define MLXBF_I2C_SMBUS_SLAVE_ADDR_MASK GENMASK(6, 0) -#define MLXBF_I2C_SLAVE_ADDR_ENABLED (addr) \ - ((addr) & (1 << MLXBF_I2C_SMBUS_SLAVE_ADDR_EN_BIT)) -/* * Timeout is given in microsends. Note also that timeout handling is not * exact.
@@ -422,7 +419,7 @@ struct mlxbf_i2c_priv {u64 frequency;/* Core frequency in Hz. */ int bus;/* Physical bus identifier. */ int irq; -struct i2c_client *slave; +struct i2c_client *slave[MLXBF_I2C_SMBUS_SLAVE_ADDR_CNT ]; };static struct mlxbf_i2c_resource mlxbf_i2c_coalesce_res[] = {
@@ -1539,25 +1536,23 @@ static int mlxbf_i2c_calculate_corepll_freq(struct platform_device *pdev,return 0 ; } -static int mlxbf_slave_enable (struct mlxbf_i2c_priv *priv,u8 addr) +static int mlxbf_i2c_slave_enable (struct mlxbf_i2c_priv *priv, +struct i2c_client *slave) { -u32 slave_reg, slave_reg_tmp, slave_reg_avail, slave_addr_mask; -u8 reg, reg_cnt, byte, addr_tmp, reg_avail, byte_avail; -bool avail, disabled; - - disabled = false; - avail = false; +u8 reg, reg_cnt, byte, addr_tmp; +u32 slave_reg, slave_reg_tmp;if (!priv)return -EPERM ; reg_cnt =MLXBF_I2C_SMBUS_SLAVE_ADDR_CNT >>2 ; - slave_addr_mask =MLXBF_I2C_SMBUS_SLAVE_ADDR_MASK ;/* * Read the slave registers. There are 4 * 32-bit slave registers. -* Each slave register can hold up to 4 * 8-bit slave configuration -* (7-bit address, 1 status bit (1 if enabled, 0 if not)). +* Each slave register can hold up to 4 * 8-bit slave configuration: +* 1) A 7-bit address +* 2) And a status bit (1 if enabled, 0 if not). +* Look for the next available slave register slot. */ for (reg =0 ; reg < reg_cnt; reg++) { slave_reg = readl(priv->smbus->io +
@@ -1572,121 +1567,87 @@ static int mlxbf_slave_enable(struct mlxbf_i2c_priv *priv, u8 addr) addr_tmp = slave_reg_tmp &GENMASK (7 ,0 );/* -* Mark the first available slave address slot, i.e. its -* enabled bit should be unset. This slot might be used -* later on to register our slave. -*/ -if (!avail && !MLXBF_I2C_SLAVE_ADDR_ENABLED (addr_tmp)) { - avail = true; - reg_avail = reg; - byte_avail = byte; - slave_reg_avail = slave_reg; - } - -/* -* Parse slave address bytes and check whether the -* slave address already exists and it's enabled, -* i.e. most significant bit is set. +* If an enable bit is not set in the +* MLXBF_I2C_SMBUS_SLAVE_ADDR_CFG register, then the +* slave address slot associated with that bit is +* free. So set the enable bit and write the +* slave address bits. */ -if ((addr_tmp & slave_addr_mask) == addr) { -if (MLXBF_I2C_SLAVE_ADDR_ENABLED (addr_tmp)) -return 0 ; - disabled = true; -break ; +if (!(addr_tmp &MLXBF_I2C_SMBUS_SLAVE_ADDR_EN_BIT )) { + slave_reg &= ~(MLXBF_I2C_SMBUS_SLAVE_ADDR_MASK << (byte *8 )); + slave_reg |= (slave->addr << (byte *8 )); + slave_reg |=MLXBF_I2C_SMBUS_SLAVE_ADDR_EN_BIT << (byte *8 ); +writel (slave_reg, priv->smbus->io + +MLXBF_I2C_SMBUS_SLAVE_ADDR_CFG + + (reg *0x4 )); + +/* +* Set the slave at the corresponding index. +*/ + priv->slave[(reg *4 ) + byte] = slave; + +return 0 ; }/* Parse next byte. */ slave_reg_tmp >>=8 ; } - -/* Exit the loop if the slave address is found. */ -if (disabled) -break ; } -if (!avail && !disabled) -return -EINVAL ;/* No room for a new slave address. */ - -if (avail && !disabled) { - reg = reg_avail; - byte = byte_avail; -/* Set the slave address. */ - slave_reg_avail &= ~(slave_addr_mask << (byte *8 )); - slave_reg_avail |= addr << (byte *8 ); - slave_reg = slave_reg_avail; - } - -/* Enable the slave address and update the register. */ - slave_reg |= (1 <<MLXBF_I2C_SMBUS_SLAVE_ADDR_EN_BIT ) << (byte *8 ); -writel (slave_reg, priv->smbus->io +MLXBF_I2C_SMBUS_SLAVE_ADDR_CFG + - reg *0x4 ); - -return 0 ; +return -EBUSY ; } -static int mlxbf_slave_disable (struct mlxbf_i2c_priv *priv) +static int mlxbf_i2c_slave_disable (struct mlxbf_i2c_priv *priv,u8 addr) { -u32 slave_reg, slave_reg_tmp, slave_addr_mask; -u8 addr, addr_tmp, reg, reg_cnt, slave_byte; -struct i2c_client *client = priv->slave; -bool exist; +u8 addr_tmp, reg, reg_cnt, byte; +u32 slave_reg, slave_reg_tmp; - exist = false; - - addr = client->addr; reg_cnt =MLXBF_I2C_SMBUS_SLAVE_ADDR_CNT >>2 ; - slave_addr_mask =MLXBF_I2C_SMBUS_SLAVE_ADDR_MASK ;/* * Read the slave registers. There are 4 * 32-bit slave registers. -* Each slave register can hold up to 4 * 8-bit slave configuration -* (7-bit address, 1 status bit (1 if enabled, 0 if not)). +* Each slave register can hold up to 4 * 8-bit slave configuration: +* 1) A 7-bit address +* 2) And a status bit (1 if enabled, 0 if not). +* Check if addr is present in the registers. */ for (reg =0 ; reg < reg_cnt; reg++) { slave_reg =readl (priv->smbus->io +MLXBF_I2C_SMBUS_SLAVE_ADDR_CFG + reg *0x4 );/* Check whether the address slots are empty. */ -if (slave_reg ==0 ) +if (!slave_reg)continue ;/* -* Each register holds 4 slave addresses. So, we have to keep -* the byte order consistent with the value read in order to -* update the register correctly, if needed. +* Check if addr matches any of the 4 slave addresses +* in the register. */ slave_reg_tmp = slave_reg; - slave_byte =0 ; -while (slave_reg_tmp !=0 ) { - addr_tmp = slave_reg_tmp & slave_addr_mask; +for (byte =0 ; byte <4 ; byte++) { + addr_tmp = slave_reg_tmp &MLXBF_I2C_SMBUS_SLAVE_ADDR_MASK ;/* * Parse slave address bytes and check whether the * slave address already exists. */ if (addr_tmp == addr) { - exist = true; -break ; +/* Clear the slave address slot. */ + slave_reg &= ~(GENMASK (7 ,0 ) << (byte *8 )); +writel (slave_reg, priv->smbus->io + +MLXBF_I2C_SMBUS_SLAVE_ADDR_CFG + + (reg *0x4 )); +/* Free slave at the corresponding index */ + priv->slave[(reg *4 ) + byte] =NULL ; + +return 0 ; }/* Parse next byte. */ slave_reg_tmp >>=8 ; - slave_byte +=1 ; } - -/* Exit the loop if the slave address is found. */ -if (exist) -break ; } -if (!exist) -return 0 ;/* Slave is not registered, nothing to do. */ - -/* Cleanup the slave address slot. */ - slave_reg &= ~(GENMASK (7 ,0 ) << (slave_byte *8 )); -writel (slave_reg, priv->smbus->io +MLXBF_I2C_SMBUS_SLAVE_ADDR_CFG + - reg *0x4 ); - -return 0 ; +return -ENXIO ; } static int mlxbf_i2c_init_coalesce(struct platform_device *pdev,
@@ -1848,72 +1809,81 @@ static bool mlxbf_smbus_slave_wait_for_idle(struct mlxbf_i2c_priv *priv,return false; } -/* Send byte to 'external' smbus master. */ -static int mlxbf_smbus_irq_send (struct mlxbf_i2c_priv *priv,u8 recv_bytes) +static struct i2c_client *mlxbf_i2c_get_slave_from_addr ( +struct mlxbf_i2c_priv *priv,u8 addr) { -u8 data_desc[MLXBF_I2C_SLAVE_DATA_DESC_SIZE ] = {0 }; -u8 write_size, pec_en, addr, byte, value, byte_cnt, desc_size; -struct i2c_client *slave = priv->slave; -u32 control32, data32; -int ret; +int i; -if (!slave) -return -EINVAL ; +for (i =0 ; i <MLXBF_I2C_SMBUS_SLAVE_ADDR_CNT ; i++) { +if (!priv->slave[i]) +continue ; - addr =0 ; - byte =0 ; - desc_size =MLXBF_I2C_SLAVE_DATA_DESC_SIZE ; +if (priv->slave[i]->addr == addr) +return priv->slave[i]; + } + +return NULL ; +} + +/* +* Send byte to 'external' smbus master. This function is executed when +* an external smbus master wants to read data from the BlueField. +*/ +static int mlxbf_i2c_irq_send (struct mlxbf_i2c_priv *priv,u8 recv_bytes) +{ +u8 data_desc[MLXBF_I2C_SLAVE_DATA_DESC_SIZE ] = {0 }; +u8 write_size, pec_en, addr, value, byte_cnt; +struct i2c_client *slave; +u32 control32, data32; +int ret =0 ;/* -* Read bytes received from the external master. These bytes should -* be located in the first data descriptor register of the slave GW. -* These bytes are the slave address byte and the internal register -* address, if supplied. +* Read the first byte received from the external master to +* determine the slave address. This byte is located in the +* first data descriptor register of the slave GW. */ -if (recv_bytes >0 ) { - data32 =ioread32be (priv->smbus->io + -MLXBF_I2C_SLAVE_DATA_DESC_ADDR ); - -/* Parse the received bytes. */ -switch (recv_bytes) { -case 2 : - byte = (data32 >>8 ) &GENMASK (7 ,0 ); - fallthrough; -case 1 : - addr = (data32 &GENMASK (7 ,0 )) >>1 ; - } + data32 =ioread32be (priv->smbus->io + +MLXBF_I2C_SLAVE_DATA_DESC_ADDR ); + addr = (data32 &GENMASK (7 ,0 )) >>1 ; -/* Check whether it's our slave address. */ -if (slave->addr != addr) -return -EINVAL ; +/* +* Check if the slave address received in the data descriptor register +* matches any of the slave addresses registered. If there is a match, +* set the slave. +*/ + slave =mlxbf_i2c_get_slave_from_addr (priv, addr); +if (!slave) { + ret = -ENXIO ; + goto clear_csr; }/* -* I2C read transactions may start by a WRITE followed by a READ. -* Indeed, most slave devices would expect the internal address -* following the slave address byte. So, write that byte first, -* and then, send the requested data bytes to the master. +* An I2C read can consist of a WRITE bit transaction followed by +* a READ bit transaction. Indeed, slave devices often expect +* the slave address to be followed by the internal address. +* So, write the internal address byte first, and then, send the +* requested data to the master. */ if (recv_bytes >1 ) {i2c_slave_event (slave,I2C_SLAVE_WRITE_REQUESTED , &value); - value = byte; + value = (data32 >>8 ) &GENMASK (7 ,0 ); ret =i2c_slave_event (slave,I2C_SLAVE_WRITE_RECEIVED , &value);i2c_slave_event (slave,I2C_SLAVE_STOP , &value);if (ret <0 ) -return ret; + goto clear_csr; }/* -* Now, send data to the master; currently, the driver supports -* READ_BYTE, READ_WORD and BLOCK READ protocols. Note that the -* hardware can send up to 128 bytes per transfer. That is the -* size of its data registers. +* Send data to the master. Currently, the driver supports +* READ_BYTE, READ_WORD and BLOCK READ protocols. The +* hardware can send up to 128 bytes per transfer which is +* the total size of the data registers. */ i2c_slave_event (slave,I2C_SLAVE_READ_REQUESTED , &value); -for (byte_cnt =0 ; byte_cnt < desc_size; byte_cnt++) { +for (byte_cnt =0 ; byte_cnt <MLXBF_I2C_SLAVE_DATA_DESC_SIZE ; byte_cnt++) { data_desc[byte_cnt] = value;i2c_slave_event (slave,I2C_SLAVE_READ_PROCESSED , &value); }
@@ -1921,8 +1891,6 @@ static int mlxbf_smbus_irq_send(struct mlxbf_i2c_priv *priv, u8 recv_bytes)/* Send a stop condition to the backend. */ i2c_slave_event (slave,I2C_SLAVE_STOP , &value); -/* Handle the actual transfer. */ -/* Set the number of bytes to write to master. */ write_size = (byte_cnt -1 ) &0x7f ;
@@ -1945,38 +1913,44 @@ static int mlxbf_smbus_irq_send(struct mlxbf_i2c_priv *priv, u8 recv_bytes)*/ mlxbf_smbus_slave_wait_for_idle (priv,MLXBF_I2C_SMBUS_TIMEOUT ); +clear_csr:/* Release the Slave GW. */ writel (0x0 , priv->smbus->io +MLXBF_I2C_SMBUS_SLAVE_RS_MASTER_BYTES );writel (0x0 , priv->smbus->io +MLXBF_I2C_SMBUS_SLAVE_PEC );writel (0x1 , priv->smbus->io +MLXBF_I2C_SMBUS_SLAVE_READY ); -return 0 ; +return ret; } -/* Receive bytes from 'external' smbus master. */ -static int mlxbf_smbus_irq_recv (struct mlxbf_i2c_priv *priv,u8 recv_bytes) +/* +* Receive bytes from 'external' smbus master. This function is executed when +* an external smbus master wants to write data to the BlueField. +*/ +static int mlxbf_i2c_irq_recv (struct mlxbf_i2c_priv *priv,u8 recv_bytes) {u8 data_desc[MLXBF_I2C_SLAVE_DATA_DESC_SIZE ] = {0 }; -struct i2c_client *slave = priv->slave; +struct i2c_client *slave;u8 value, byte, addr;int ret =0 ; -if (!slave) -return -EINVAL ; -/* Read data from Slave GW data descriptor. */ mlxbf_i2c_smbus_read_data (priv, data_desc, recv_bytes,MLXBF_I2C_SLAVE_DATA_DESC_ADDR ); - -/* Check whether its our slave address. */ addr = data_desc[0 ] >>1 ; -if (slave->addr != addr) -return -EINVAL ;/* -* Notify the slave backend; another I2C master wants to write data -* to us. This event is sent once the slave address and the write bit -* is detected. +* Check if the slave address received in the data descriptor register +* matches any of the slave addresses registered. +*/ + slave =mlxbf_i2c_get_slave_from_addr (priv, addr); +if (!slave) { + ret = -EINVAL ; + goto clear_csr; + } + +/* +* Notify the slave backend that an smbus master wants to write data +* to the BlueField. */ i2c_slave_event (slave,I2C_SLAVE_WRITE_REQUESTED , &value);
@@ -1989,9 +1963,13 @@ static int mlxbf_smbus_irq_recv(struct mlxbf_i2c_priv *priv, u8 recv_bytes)break ; } -/* Send a stop condition to the backend. */ +/* +* Send a stop event to the slave backend, to signal +* the end of the write transactions. +*/ i2c_slave_event (slave,I2C_SLAVE_STOP , &value); +clear_csr:/* Release the Slave GW. */ writel (0x0 , priv->smbus->io +MLXBF_I2C_SMBUS_SLAVE_RS_MASTER_BYTES );writel (0x0 , priv->smbus->io +MLXBF_I2C_SMBUS_SLAVE_PEC );
@@ -2000,7 +1978,7 @@ static int mlxbf_smbus_irq_recv(struct mlxbf_i2c_priv *priv, u8 recv_bytes)return ret; } -static irqreturn_t mlxbf_smbus_irq (int irq,void *ptr) +static irqreturn_t mlxbf_i2c_irq (int irq,void *ptr) {struct mlxbf_i2c_priv *priv = ptr;bool read, write, irq_is_set;
@@ -2048,9 +2026,9 @@ static irqreturn_t mlxbf_smbus_irq(int irq, void *ptr) MLXBF_I2C_SLAVE_DATA_DESC_SIZE : recv_bytes;if (read) -mlxbf_smbus_irq_send (priv, recv_bytes); +mlxbf_i2c_irq_send (priv, recv_bytes);else -mlxbf_smbus_irq_recv (priv, recv_bytes); +mlxbf_i2c_irq_recv (priv, recv_bytes);return IRQ_HANDLED ; }
@@ -2145,23 +2123,21 @@ static s32 mlxbf_i2c_smbus_xfer(struct i2c_adapter *adap, u16 addr,static int mlxbf_i2c_reg_slave (struct i2c_client *slave) {struct mlxbf_i2c_priv *priv =i2c_get_adapdata (slave->adapter); +struct device *dev = &slave->dev;int ret; -if (priv->slave) -return -EBUSY ; -/* * Do not support ten bit chip address and do not use Packet Error * Checking (PEC). */ -if (slave->flags & (I2C_CLIENT_TEN |I2C_CLIENT_PEC )) +if (slave->flags & (I2C_CLIENT_TEN |I2C_CLIENT_PEC )) { +dev_err (dev,"SMBus PEC and 10 bit address not supported\n" );return -EAFNOSUPPORT ; + } - ret =mlxbf_slave_enable (priv, slave->addr); -if (ret <0 ) -return ret; - - priv->slave = slave; + ret =mlxbf_i2c_slave_enable (priv, slave); +if (ret) +dev_err (dev,"Surpassed max number of registered slaves allowed\n" );return 0 ; }
@@ -2169,18 +2145,19 @@ static int mlxbf_i2c_reg_slave(struct i2c_client *slave)static int mlxbf_i2c_unreg_slave (struct i2c_client *slave) {struct mlxbf_i2c_priv *priv =i2c_get_adapdata (slave->adapter); +struct device *dev = &slave->dev;int ret; -WARN_ON (!priv->slave); - -/* Unregister slave, i.e. disable the slave address in hardware. */ - ret =mlxbf_slave_disable (priv); -if (ret <0 ) -return ret; - - priv->slave =NULL ; +/* +* Unregister slave by: +* 1) Disabling the slave address in hardware +* 2) Freeing priv->slave at the corresponding index +*/ + ret =mlxbf_i2c_slave_disable (priv, slave->addr); +if (ret) +dev_err (dev,"Unable to find slave 0x%x\n" , slave->addr); -return 0 ; +return ret; }static u32 mlxbf_i2c_functionality (struct i2c_adapter *adap)
@@ -2388,7 +2365,7 @@ static int mlxbf_i2c_probe(struct platform_device *pdev) irq =platform_get_irq (pdev,0 );if (irq <0 )return irq; - ret =devm_request_irq (dev, irq, mlxbf_smbus_irq, + ret =devm_request_irq (dev, irq, mlxbf_i2c_irq,IRQF_SHARED |IRQF_PROBE_SHARED ,dev_name (dev), priv);if (ret <0 ) {
@@ -2483,4 +2460,5 @@ module_exit(mlxbf_i2c_exit);MODULE_DESCRIPTION ("Mellanox BlueField I2C bus driver" );MODULE_AUTHOR ("Khalil Blaiech <[email protected]>" ); +MODULE_AUTHOR ("Asmaa Mnebhi <[email protected]>" );MODULE_LICENSE ("Dual BSD/GPL" );
Signature · 1 line
2.30.1
Asmaa Mnebhi
maintainerPATCH v2 8/9 i2c-mlxbf.c: support BlueField-3 SoC
In reply to Asmaa MnebhiBlueField-3 SoC has the same I2C IP logic as previous BlueField-1 and 2 SoCs but it has different registers' addresses. This is an effort to keep this driver generic accross all BlueField generations.
This patch breaks down the "smbus" resource into 3 separate resources to enable us to use common registers' offsets for all BlueField SoCs:
struct mlxbf_i2c_resource *timer;
struct mlxbf_i2c_resource *mst;
struct mlxbf_i2c_resource *slv;
Of course, all offsets had to be adjusted accordingly, and we took this chance to reorganize the macros depending on the register block they target.
There are only 2 registers' offsets that do not fit within this schema so their offsets are passed as SoC-specific parameters:smbus_master_rs_bytes_offsmbus_master_fsm_off
--- MAINTAINERS | 1 + drivers/i2c/busses/i2c-mlxbf.c | 450 ++++++++++++++++++++------------- 2 files changed, 280 insertions(+), 171 deletions(-)
MAINTAINERS +1 -0
@@ -12978,6 +12978,7 @@ F: drivers/input/touchscreen/melfas_mip4.c MELLANOX BLUEFIELD I2C DRIVER M: Khalil Blaiech <[email protected]> +M: Asmaa Mnebhi <[email protected]> L: [email protected] S: Supported F: Documentation/devicetree/bindings/i2c/mellanox,i2c-mlxbf.yaml
drivers/i2c/busses/i2c-mlxbf.c +279 -171
848 lines not on this page · raw message
Signature · 1 line
2.30.1
Asmaa Mnebhi
maintainerPATCH v2 9/9 i2c-mlxbf.c: Update binding devicetree
In reply to Asmaa MnebhiIn the latest version of the i2c-mlxbf.c driver, the "Smbus block" resource was broken down to 3 separate resources "Smbus timer", "Smbus master" and "Smbus slave" to accommodate for BlueField-3 SoC registers' changes.
--- .../bindings/i2c/mellanox,i2c-mlxbf.yaml | 49 ++++++++++++++----- 1 file changed, 37 insertions(+), 12 deletions(-)
Documentation/devicetree/bindings/i2c/mellanox,i2c-mlxbf.yaml +37 -12
@@ -8,6 +8,7 @@ title: Mellanox I2C SMBus on BlueField SoCs maintainers: - Khalil Blaiech <[email protected]> + - Asmaa Mnebhi <[email protected]> allOf: - $ref: /schemas/i2c/i2c-controller.yaml#
@@ -17,14 +18,19 @@ properties: enum: - mellanox,i2c-mlxbf1 - mellanox,i2c-mlxbf2 + - mellanox,i2c-mlxbf3 reg: minItems: 3 + maxItems: 6 items: - description: Smbus block registers - description: Cause master registers - description: Cause slave registers - description: Cause coalesce registers + - description: Smbus timer registers + - description: Smbus master registers + - description: Smbus slave registers interrupts: maxItems: 1
@@ -35,6 +41,13 @@ properties: bus frequency used to configure timing registers; The frequency is expressed in Hz. Default is 100000. + resource_version: + enum: [ 0, 1 ] + description: + Version of the device tree. resource_version = 0 when the driver uses + Smbus block resource. resource_version = 1 when the driver uses Smbus + timer, Smbus master and Smbus slave resources. + required: - compatible - reg
@@ -42,18 +55,6 @@ required: unevaluatedProperties: false -if: - properties: - compatible: - contains: - enum: - - mellanox,i2c-mlxbf1 - -then: - properties: - reg: - maxItems: 3 - examples: - | i2c@2804000 {
@@ -61,8 +62,13 @@ examples: reg = <0x02804000 0x800>, <0x02801200 0x020>, <0x02801260 0x020>; + <0x00000001 0x1>; + <0x02804000 0x40>, + <0x02804200 0x200>, + <0x02804400 0x200>, interrupts = <57>; clock-frequency = <100000>; + resource_version = <1>; }; - |
@@ -72,6 +78,25 @@ examples: <0x02808e00 0x020>, <0x02808e20 0x020>, <0x02808e40 0x010>; + <0x02808800 0x040>; + <0x02808a00 0x200>, + <0x02808c00 0x200>, interrupts = <57>; clock-frequency = <400000>; + resource_version = <1>; + }; + + - | + i2c@2808800 { + compatible = "mellanox,i2c-mlxbf3"; + reg = <0x00000001 0x1>, + <0x13404400 0x020>, + <0x13404420 0x020>, + <0x13404440 0x010>; + <0x13404480 0x40>, + <0x13404200 0x200>, + <0x13404000 0x200>, + interrupts = <35>; + clock-frequency = <400000>; + resource_version = <1>; };
Signature · 1 line
2.30.1
Quoted 2 lines · Asmaa Mnebhi
This is a series of patches fixing several bugs and implementing new features.
What did change since v1 and where did Khalil's Rev-by tags go?
Quoted 3 lines
Bug fixes:
1) Fix the frequency calculation
2) remove unnecessaryIRQF_ONESHOTflag
Is this really a bugfix?
Quoted 3 lines
3) Fix incorrect base address passed during io write
4) prevent stack overflow inmlxbf_i2c_smbus_start_transaction()
5) Support lock mechanism
Here, I am also not sure if this is a bugfix.
Thanks for the update,
Wolfram
Asmaa Mnebhi
maintainerQuoted 2 lines
This is a series of patches fixing several bugs and implementing new features.
What did change since v1 and where did Khalil's Rev-by tags go?
Khalil told me to remove the rev-by : ) since it is supposed to be reserved for the reviewers. I will add it back and describe what has changed in my next set of patches. As a note , this is what changed from v1->v2:
1) moved all the bug fixes to the top commits and left the features for last
2) split the BlueField-3 SoC patch into 2 to address Rob's comment: one for the driver code and another patch for the device tree binding yaml documentation file
3) addressed Rob's comment regarding keeping the device tree/acpi tables backward compatible. So add the new resources at the end of the enum.
4) update the license in a separate patch
Quoted 3 lines
Bug fixes:
1) Fix the frequency calculation
2) remove unnecessaryIRQF_ONESHOTflag
Is this really a bugfix?
This is not a bug fix. We removed it because it is no longer needed.
Quoted 3 lines
3) Fix incorrect base address passed during io write
4) prevent stack overflow inmlxbf_i2c_smbus_start_transaction()
5) Support lock mechanism
Here, I am also not sure if this is a bugfix.
I think this is a bug fix because we have an i2c driver also in UEFI so we need this lock mechanism to avoid race conditions over acquiring the i2c bus. Not using this lock resulted in unexpected behavior.
Thanks for the update,
Wolfram
Quoted 4 lines
What did change since v1 and where did Khalil's Rev-by tags go?
Khalil told me to remove the rev-by : ) since it is supposed to be reserved for the reviewers.
But he reviewed the patches, or? Which makes him a reviewer? I mean he is the maintainer of the driver, so I basically a Rev-by from him.
Quoted 1 line
I will add it back and describe what has changed in my next set of patches.
So, you are planning a v3 already?
Quoted 1 line
As a note , this is what changed from
v1->v2:
Cool, thank you for doing that.
Quoted 3 lines
I think this is a bug fix because we have an i2c driver also in UEFI so we need this lock mechanism to avoid race conditions over acquiring the i2c bus. Not using this lock resulted in unexpected behavior.
OK, thank you!
Asmaa Mnebhi
maintainerQuoted 3 lines
What did change since v1 and where did Khalil's Rev-by tags go?
Khalil told me to remove the rev-by : ) since it is supposed to be reserved for the reviewers.
But he reviewed the patches, or? Which makes him a reviewer? I mean he is the maintainer of the driver, so I basically a Rev-by from him.
Yes he did review the patch internally so I will add the Rev-by in v3
Quoted 1 line
I will add it back and describe what has changed in my next set of patches.
So, you are planning a v3 already?
I was going to send a v3 patch after I get some feedback on the other patches unless you'd prefer I send it now?
Quoted 1 line
As a note , this is what changed from
v1->v2:
Cool, thank you for doing that.
Quoted 3 lines
I think this is a bug fix because we have an i2c driver also in UEFI so we need this lock mechanism to avoid race conditions over acquiring the i2c bus. Not using this lock resulted in unexpected behavior.
OK, thank you!
Quoted 3 lines
I was going to send a v3 patch after I get some feedback on the other patches unless you'd prefer I send it now?
Your choice. You can wait for some more comments if you want. But frankly, I don't really expect more comments.
Asmaa Mnebhi
maintainerQuoted 2 lines
I was going to send a v3 patch after I get some feedback on the other patches unless you'd prefer I send it now?
Your choice. You can wait for some more comments if you want. But frankly, I don't really expect more comments.
Ok! Sent v3!