i2c: designware: Generic polling mode code
Subsystems (3)
cover i2c: designware: Generic polling mode code
Hi Jiawen, Sanket and Basavaraj
Could you test on your Wangxun 10Gb NIC and AMD NAVI GPU harware this patchset since it's touching both.
For the AMD NAVI GPU and changes are less but for the Wangxun 10Gb NIC patchset replaces the txgbe_i2c_dw_xfer_quirk() with generic polling mode code.
I've been testing this on our HW which all have interrupt connected and tried to cover also FIFO depth defines for Wangxun 10Gb NIC. Obviously I would like to know how this set works on your real HW.
Jarkko Nikula (6):
i2c: designware: Uniform initialization flow for polling mode
i2c: designware: Do not enable interrupts shortly in polling mode
i2c: designware: Use accessors to DW_IC_INTR_MASK register
i2c: designware: Move interrupt handling functions before
i2c_dw_xfer()
i2c: designware: Fix RX FIFO depth define on Wangxun 10Gb NIC
i2c: designware: Implement generic polling mode code for Wangxun 10Gb
NIC
drivers/i2c/busses/i2c-designware-common.c | 2 +- drivers/i2c/busses/i2c-designware-core.h | 23 +- drivers/i2c/busses/i2c-designware-master.c | 417 +++++++++----------- drivers/i2c/busses/i2c-designware-pcidrv.c | 2 +- drivers/i2c/busses/i2c-designware-platdrv.c | 2 +- 5 files changed, 208 insertions(+), 238 deletions(-)
Signature · 1 line
2.43.0
PATCH 1/6 i2c: designware: Uniform initialization flow for polling mode
Currently initialization flow in i2c_dw_probe_master() skips a few steps and has code duplication for polling mode implementation.
Simplify this by adding a new ACCESS_POLLING flag that is set for those two platforms that currently use polling mode and use it to skip interrupt handler setup.
--- drivers/i2c/busses/i2c-designware-core.h | 1 + drivers/i2c/busses/i2c-designware-master.c | 42 ++++----------------- drivers/i2c/busses/i2c-designware-pcidrv.c | 2 +- drivers/i2c/busses/i2c-designware-platdrv.c | 2 +- 4 files changed, 11 insertions(+), 36 deletions(-)
drivers/i2c/busses/i2c-designware-core.h +1 -0
@@ -303,6 +303,7 @@ struct dw_i2c_dev {#define ACCESS_INTR_MASK BIT(0)#define ACCESS_NO_IRQ_SUSPEND BIT(1)#define ARBITRATION_SEMAPHORE BIT(2) +#define ACCESS_POLLING BIT(3)#define MODEL_MSCC_OCELOT BIT(8)#define MODEL_BAIKAL_BT1 BIT(9)
drivers/i2c/busses/i2c-designware-master.c +8 -34
60 lines not on this page · raw message
drivers/i2c/busses/i2c-designware-pcidrv.c +1 -1
@@ -154,7 +154,7 @@ static int navi_amd_setup(struct pci_dev *pdev, struct dw_pci_controller *c) {struct dw_i2c_dev *dev =dev_get_drvdata (&pdev->dev); - dev->flags |=MODEL_AMD_NAVI_GPU ; + dev->flags |=MODEL_AMD_NAVI_GPU |ACCESS_POLLING ; dev->timings.bus_freq_hz =I2C_MAX_STANDARD_MODE_FREQ ;return 0 ; }
drivers/i2c/busses/i2c-designware-platdrv.c +1 -1
@@ -290,7 +290,7 @@ static int dw_i2c_plat_probe(struct platform_device *pdev) dev->flags = (uintptr_t )device_get_match_data (&pdev->dev);if (device_property_present (&pdev->dev,"wx,i2c-snps-model" )) - dev->flags =MODEL_WANGXUN_SP ; + dev->flags =MODEL_WANGXUN_SP |ACCESS_POLLING ; dev->dev = &pdev->dev; dev->irq = irq;
Signature · 1 line
2.43.0
Andy Shevchenko
reviewerRe: [PATCH 1/6] i2c: designware: Uniform initialization flow for polling mode
Quoted 6 lines · Jarkko Nikula
Currently initialization flow in
i2c_dw_probe_master()skips a few steps and has code duplication for polling mode implementation.Simplify this by adding a new
ACCESS_POLLINGflag that is set for those two platforms that currently use polling mode and use it to skip interrupt handler setup.
...
Quoted 4 lines
#define
ACCESS_INTR_MASKBIT(0)
#defineACCESS_NO_IRQ_SUSPENDBIT(1)
#defineARBITRATION_SEMAPHOREBIT(2)
+#defineACCESS_POLLINGBIT(3)
For the sake of consistency, you may incorporate internal patch from somebody else to move to the same prefix (namespace) for ARBITRATION_SEMAPHORE.
...
drivers/i2c/busses/i2c-designware-master.c int i2c_dw_probe_master(struct dw_i2c_dev *dev) · line 1024
Quoted 1 line
drivers/i2c/busses/i2c-designware-master.c +1
-0+if (!(dev->flags &ACCESS_POLLING )) {
Actually seems to me that is better to have in a separate helper and you then can do here something like
ret = i2c_dw_request_irq(...); if (ret) return ret;
Quoted 7 lines
drivers/i2c/busses/i2c-designware-master.c +7
-0+ ret =devm_request_irq (dev->dev, dev->irq, i2c_dw_isr, + irq_flags,dev_name (dev->dev), dev); +if (ret) { +dev_err (dev->dev,"failure requesting irq %i: %d\n" , + dev->irq, ret); +return ret; + }
Signature · 2 lines
With Best Regards, Andy Shevchenko
Andi Shyti
maintainerRe: [PATCH 1/6] i2c: designware: Uniform initialization flow for polling mode
Hi Andy,
Quoted 8 lines
drivers/i2c/busses/i2c-designware-master.c int i2c_dw_probe_master(struct dw_i2c_dev *dev) · line 1024
Quoted 1 line
drivers/i2c/busses/i2c-designware-master.c +1
-0+if (!(dev->flags &ACCESS_POLLING )) {Actually seems to me that is better to have in a separate helper and you then can do here something like
ret = i2c_dw_request_irq(...); if (ret) return ret;
I find it a bit misleading because the first question that comes to mind upon reading such a line is, 'What if access is polling?' without looking inside the function. In this case, a more appropriate name would then be i2c_sw_request_irq_if_needed(). :)
Andi
PATCH 2/6 i2c: designware: Do not enable interrupts shortly in polling mode
In reply to Jarkko NikulaI was testing the polling mode txgbe_i2c_dw_xfer_quirk() on a HW where the i2c-designware has interrupt connected and shared with other device. I noticed there is a bogus interrupt for each transfer.
Reason for this that both polling mode functions call the i2c_dw_xfer_init() which enable interrupts then followed by immediate disable by the same polling mode functions. This is enough to trigger TX_EMPTY interrupt.
Fix this by introducing a __i2c_dw_write_intr_mask() helper that unmasks interrupts conditionally and use it in i2c_dw_xfer_init().
--- drivers/i2c/busses/i2c-designware-core.h | 8 ++++++++ drivers/i2c/busses/i2c-designware-master.c | 4 +--- 2 files changed, 9 insertions(+), 3 deletions(-)
drivers/i2c/busses/i2c-designware-core.h +8 -0
@@ -352,6 +352,14 @@ static inline void __i2c_dw_disable_nowait(struct dw_i2c_dev *dev) dev->status &= ~STATUS_ACTIVE ; } +static inline void __i2c_dw_write_intr_mask (struct dw_i2c_dev *dev, +unsigned int intr_mask) +{ +unsigned int val = dev->flags &ACCESS_POLLING ?0 : intr_mask; + +regmap_write (dev->map,DW_IC_INTR_MASK , val); +} +void __i2c_dw_disable (struct dw_i2c_dev *dev);extern void i2c_dw_configure_master (struct dw_i2c_dev *dev);
drivers/i2c/busses/i2c-designware-master.c +1 -3
@@ -250,7 +250,7 @@ static void i2c_dw_xfer_init(struct dw_i2c_dev *dev)/* Clear and enable interrupts */ regmap_read (dev->map,DW_IC_CLR_INTR , &dummy); -regmap_write (dev->map,DW_IC_INTR_MASK ,DW_IC_INTR_MASTER_MASK ); +__i2c_dw_write_intr_mask (dev,DW_IC_INTR_MASTER_MASK ); }static int i2c_dw_check_stopbit (struct dw_i2c_dev *dev)
@@ -300,7 +300,6 @@ static int amd_i2c_dw_xfer_quirk(struct i2c_adapter *adap, struct i2c_msg *msgs, dev->msgs = msgs; dev->msgs_num = num_msgs;i2c_dw_xfer_init (dev); -regmap_write (dev->map,DW_IC_INTR_MASK ,0 );/* Initiate messages read/write transaction */ for (msg_wrt_idx =0 ; msg_wrt_idx < num_msgs; msg_wrt_idx++) {
@@ -384,7 +383,6 @@ static int txgbe_i2c_dw_xfer_quirk(struct i2c_adapter *adap, struct i2c_msg *msg dev->msgs = msgs; dev->msgs_num = num_msgs;i2c_dw_xfer_init (dev); -regmap_write (dev->map,DW_IC_INTR_MASK ,0 );for (msg_idx =0 ; msg_idx < num_msgs; msg_idx++) { buf = msgs[msg_idx].buf;
Signature · 1 line
2.43.0
PATCH 3/6 i2c: designware: Use accessors to DW_IC_INTR_MASK register
In reply to Jarkko NikulaConvert access to DW_IC_INTR_MASK register using the existing __i2c_dw_write_intr_mask() and a __i2c_dw_read_intr_mask() introduced here. Motivation to this is to prepare for generic polling mode code where polling mode will use a SW mask instead of DW_IC_INTR_MASK.
--- drivers/i2c/busses/i2c-designware-common.c | 2 +- drivers/i2c/busses/i2c-designware-core.h | 7 +++++++ drivers/i2c/busses/i2c-designware-master.c | 22 ++++++++++++---------- 3 files changed, 20 insertions(+), 11 deletions(-)
drivers/i2c/busses/i2c-designware-common.c +1 -1
@@ -648,7 +648,7 @@ void i2c_dw_disable(struct dw_i2c_dev *dev)__i2c_dw_disable (dev);/* Disable all interrupts */ -regmap_write (dev->map,DW_IC_INTR_MASK ,0 ); +__i2c_dw_write_intr_mask (dev,0 );regmap_read (dev->map,DW_IC_CLR_INTR , &dummy);i2c_dw_release_lock (dev);
drivers/i2c/busses/i2c-designware-core.h +7 -0
@@ -360,6 +360,13 @@ static inline void __i2c_dw_write_intr_mask(struct dw_i2c_dev *dev,regmap_write (dev->map,DW_IC_INTR_MASK , val); } +static inline void __i2c_dw_read_intr_mask (struct dw_i2c_dev *dev, +unsigned int *intr_mask) +{ +if (!(dev->flags &ACCESS_POLLING )) +regmap_read (dev->map,DW_IC_INTR_MASK , intr_mask); +} +void __i2c_dw_disable (struct dw_i2c_dev *dev);extern void i2c_dw_configure_master (struct dw_i2c_dev *dev);
drivers/i2c/busses/i2c-designware-master.c +12 -10
@@ -240,7 +240,7 @@ static void i2c_dw_xfer_init(struct dw_i2c_dev *dev) msgs[dev->msg_write_idx].addr | ic_tar);/* Enforce disabled interrupts (due to HW issues) */ -regmap_write (dev->map,DW_IC_INTR_MASK ,0 ); +__i2c_dw_write_intr_mask (dev,0 );/* Enable the adapter */ __i2c_dw_enable (dev);
@@ -544,7 +544,7 @@ i2c_dw_xfer_msg(struct dw_i2c_dev *dev)if (dev->msg_err) intr_mask =0 ; -regmap_write (dev->map,DW_IC_INTR_MASK , intr_mask); +__i2c_dw_write_intr_mask (dev, intr_mask); } static u8
@@ -552,6 +552,7 @@ i2c_dw_recv_len(struct dw_i2c_dev *dev, u8 len) {struct i2c_msg *msgs = dev->msgs;u32 flags = msgs[dev->msg_read_idx].flags; +unsigned int intr_mask;/* * Adjust the buffer length and mask the flag
@@ -566,8 +567,9 @@ i2c_dw_recv_len(struct dw_i2c_dev *dev, u8 len)* Received buffer length, re-enable TX_EMPTY interrupt * to resume the SMBUS transaction. */ -regmap_update_bits (dev->map,DW_IC_INTR_MASK ,DW_IC_INTR_TX_EMPTY , -DW_IC_INTR_TX_EMPTY ); +__i2c_dw_read_intr_mask (dev, &intr_mask); + intr_mask |=DW_IC_INTR_TX_EMPTY ; +__i2c_dw_write_intr_mask (dev, intr_mask);return len; }
@@ -827,7 +829,7 @@ static irqreturn_t i2c_dw_isr(int this_irq, void *dev_id)* interrupt really came from this HW (E.g. firmware has left * the HW active). */ -regmap_write (dev->map,DW_IC_INTR_MASK ,0 ); +__i2c_dw_write_intr_mask (dev,0 );return IRQ_HANDLED ; }
@@ -840,7 +842,7 @@ static irqreturn_t i2c_dw_isr(int this_irq, void *dev_id)* Anytime TX_ABRT is set, the contents of the tx/rx * buffers are flushed. Make sure to skip them. */ -regmap_write (dev->map,DW_IC_INTR_MASK ,0 ); +__i2c_dw_write_intr_mask (dev,0 ); goto tx_aborted; }
@@ -862,9 +864,9 @@ static irqreturn_t i2c_dw_isr(int this_irq, void *dev_id)complete (&dev->cmd_complete); else if (unlikely(dev->flags & ACCESS_INTR_MASK)) {/* Workaround to trigger pending interrupt */ -regmap_read (dev->map,DW_IC_INTR_MASK , &stat); -regmap_write (dev->map,DW_IC_INTR_MASK ,0 ); -regmap_write (dev->map,DW_IC_INTR_MASK , stat); +__i2c_dw_read_intr_mask (dev, &stat); +__i2c_dw_write_intr_mask (dev,0 ); +__i2c_dw_write_intr_mask (dev, stat); }return IRQ_HANDLED ;
@@ -1016,7 +1018,7 @@ int i2c_dw_probe_master(struct dw_i2c_dev *dev)if (ret)return ret; -regmap_write (dev->map,DW_IC_INTR_MASK ,0 ); +__i2c_dw_write_intr_mask (dev,0 );i2c_dw_release_lock (dev);if (!(dev->flags &ACCESS_POLLING )) {
Signature · 1 line
2.43.0
PATCH 4/6 i2c: designware: Move interrupt handling functions before i2c_dw_xfer()
In reply to Jarkko NikulaCode is more logically arranged when i2c_dw_read_clear_intrbits() and i2c_dw_isr() are located before i2c_dw_xfer().
Real reason for this is to prepare for more shared code between interrupt and polling mode code.
--- drivers/i2c/busses/i2c-designware-master.c | 226 ++++++++++----------- 1 file changed, 113 insertions(+), 113 deletions(-)
drivers/i2c/busses/i2c-designware-master.c +113 -113
@@ -633,119 +633,6 @@ i2c_dw_read(struct dw_i2c_dev *dev) } } -/* -* Prepare controller for a transaction and call i2c_dw_xfer_msg. -*/ -static int -i2c_dw_xfer (struct i2c_adapter *adap,struct i2c_msg msgs[],int num) -{ -struct dw_i2c_dev *dev =i2c_get_adapdata (adap); -int ret; - -dev_dbg (dev->dev,"%s: msgs: %d\n" , __func__, num); - -pm_runtime_get_sync (dev->dev); - -/* -* Initiate I2C message transfer when polling mode is enabled, -* As it is polling based transfer mechanism, which does not support -* interrupt based functionalities of existing DesignWare driver. -*/ -switch (dev->flags &MODEL_MASK ) { -case MODEL_AMD_NAVI_GPU : - ret =amd_i2c_dw_xfer_quirk (adap, msgs, num); - goto done_nolock; -case MODEL_WANGXUN_SP : - ret =txgbe_i2c_dw_xfer_quirk (adap, msgs, num); - goto done_nolock; -default : -break ; - } - -reinit_completion (&dev->cmd_complete); - dev->msgs = msgs; - dev->msgs_num = num; - dev->cmd_err =0 ; - dev->msg_write_idx =0 ; - dev->msg_read_idx =0 ; - dev->msg_err =0 ; - dev->status =0 ; - dev->abort_source =0 ; - dev->rx_outstanding =0 ; - - ret =i2c_dw_acquire_lock (dev); -if (ret) - goto done_nolock; - - ret =i2c_dw_wait_bus_not_busy (dev); -if (ret <0 ) - goto done; - -/* Start the transfers */ -i2c_dw_xfer_init (dev); - -/* Wait for tx to complete */ -if (!wait_for_completion_timeout (&dev->cmd_complete, adap->timeout)) { -dev_err (dev->dev,"controller timed out\n" ); -/* i2c_dw_init implicitly disables the adapter */ -i2c_recover_bus (&dev->adapter); -i2c_dw_init_master (dev); - ret = -ETIMEDOUT ; - goto done; - } - -/* -* We must disable the adapter before returning and signaling the end -* of the current transfer. Otherwise the hardware might continue -* generating interrupts which in turn causes a race condition with -* the following transfer. Needs some more investigation if the -* additional interrupts are a hardware bug or this driver doesn't -* handle them correctly yet. -*/ -__i2c_dw_disable_nowait (dev); - -if (dev->msg_err) { - ret = dev->msg_err; - goto done; - } - -/* No error */ -if (likely (!dev->cmd_err && !dev->status)) { - ret = num; - goto done; - } - -/* We have an error */ -if (dev->cmd_err ==DW_IC_ERR_TX_ABRT ) { - ret =i2c_dw_handle_tx_abort (dev); - goto done; - } - -if (dev->status) -dev_err (dev->dev, -"transfer terminated early - interrupt latency too high?\n" ); - - ret = -EIO ; - -done: -i2c_dw_release_lock (dev); - -done_nolock: -pm_runtime_mark_last_busy (dev->dev); -pm_runtime_put_autosuspend (dev->dev); - -return ret; -} - -static const struct i2c_algorithm i2c_dw_algo = { - .master_xfer = i2c_dw_xfer, - .functionality = i2c_dw_func, -}; - -static const struct i2c_adapter_quirks i2c_dw_quirks = { - .flags =I2C_AQ_NO_ZERO_LEN , -}; -static u32 i2c_dw_read_clear_intrbits (struct dw_i2c_dev *dev) {unsigned int stat, dummy;
@@ -872,6 +759,119 @@ static irqreturn_t i2c_dw_isr(int this_irq, void *dev_id)return IRQ_HANDLED ; } +/* +* Prepare controller for a transaction and call i2c_dw_xfer_msg. +*/ +static int +i2c_dw_xfer (struct i2c_adapter *adap,struct i2c_msg msgs[],int num) +{ +struct dw_i2c_dev *dev =i2c_get_adapdata (adap); +int ret; + +dev_dbg (dev->dev,"%s: msgs: %d\n" , __func__, num); + +pm_runtime_get_sync (dev->dev); + +/* +* Initiate I2C message transfer when polling mode is enabled, +* As it is polling based transfer mechanism, which does not support +* interrupt based functionalities of existing DesignWare driver. +*/ +switch (dev->flags &MODEL_MASK ) { +case MODEL_AMD_NAVI_GPU : + ret =amd_i2c_dw_xfer_quirk (adap, msgs, num); + goto done_nolock; +case MODEL_WANGXUN_SP : + ret =txgbe_i2c_dw_xfer_quirk (adap, msgs, num); + goto done_nolock; +default : +break ; + } + +reinit_completion (&dev->cmd_complete); + dev->msgs = msgs; + dev->msgs_num = num; + dev->cmd_err =0 ; + dev->msg_write_idx =0 ; + dev->msg_read_idx =0 ; + dev->msg_err =0 ; + dev->status =0 ; + dev->abort_source =0 ; + dev->rx_outstanding =0 ; + + ret =i2c_dw_acquire_lock (dev); +if (ret) + goto done_nolock; + + ret =i2c_dw_wait_bus_not_busy (dev); +if (ret <0 ) + goto done; + +/* Start the transfers */ +i2c_dw_xfer_init (dev); + +/* Wait for tx to complete */ +if (!wait_for_completion_timeout (&dev->cmd_complete, adap->timeout)) { +dev_err (dev->dev,"controller timed out\n" ); +/* i2c_dw_init implicitly disables the adapter */ +i2c_recover_bus (&dev->adapter); +i2c_dw_init_master (dev); + ret = -ETIMEDOUT ; + goto done; + } + +/* +* We must disable the adapter before returning and signaling the end +* of the current transfer. Otherwise the hardware might continue +* generating interrupts which in turn causes a race condition with +* the following transfer. Needs some more investigation if the +* additional interrupts are a hardware bug or this driver doesn't +* handle them correctly yet. +*/ +__i2c_dw_disable_nowait (dev); + +if (dev->msg_err) { + ret = dev->msg_err; + goto done; + } + +/* No error */ +if (likely (!dev->cmd_err && !dev->status)) { + ret = num; + goto done; + } + +/* We have an error */ +if (dev->cmd_err ==DW_IC_ERR_TX_ABRT ) { + ret =i2c_dw_handle_tx_abort (dev); + goto done; + } + +if (dev->status) +dev_err (dev->dev, +"transfer terminated early - interrupt latency too high?\n" ); + + ret = -EIO ; + +done: +i2c_dw_release_lock (dev); + +done_nolock: +pm_runtime_mark_last_busy (dev->dev); +pm_runtime_put_autosuspend (dev->dev); + +return ret; +} + +static const struct i2c_algorithm i2c_dw_algo = { + .master_xfer = i2c_dw_xfer, + .functionality = i2c_dw_func, +}; + +static const struct i2c_adapter_quirks i2c_dw_quirks = { + .flags =I2C_AQ_NO_ZERO_LEN , +}; +void i2c_dw_configure_master (struct dw_i2c_dev *dev) {struct i2c_timings *t = &dev->timings;
Signature · 1 line
2.43.0
PATCH 5/6 i2c: designware: Fix RX FIFO depth define on Wangxun 10Gb NIC
In reply to Jarkko NikulaI believe RX FIFO depth define 0 is incorrect on Wangxun 10Gb NIC. It must be at least 1 since code is able to read received data from the DW_IC_DATA_CMD register.
For now this define is irrelevant since the txgbe_i2c_dw_xfer_quirk() doesn't use the rx_fifo_depth member variable of struct dw_i2c_dev but is needed when converting code into generic polling mode implementation.
--- drivers/i2c/busses/i2c-designware-core.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/i2c/busses/i2c-designware-core.h +1 -1
@@ -319,7 +319,7 @@ struct dw_i2c_dev {#define AMD_UCSI_INTR_EN 0xd#define TXGBE_TX_FIFO_DEPTH 4 -#define TXGBE_RX_FIFO_DEPTH 0 +#define TXGBE_RX_FIFO_DEPTH 1struct i2c_dw_semaphore_callbacks {int (*probe)(struct dw_i2c_dev *dev);
Signature · 1 line
2.43.0
Andy Shevchenko
reviewerRe: [PATCH 5/6] i2c: designware: Fix RX FIFO depth define on Wangxun 10Gb NIC
Quoted 7 lines · Jarkko Nikula
I believe RX FIFO depth define 0 is incorrect on Wangxun 10Gb NIC. It must be at least 1 since code is able to read received data from the
DW_IC_DATA_CMDregister.For now this define is irrelevant since the
txgbe_i2c_dw_xfer_quirk()doesn't use therx_fifo_depthmember variable of structdw_i2c_devbut is needed when converting code into generic polling mode implementation.
Shouldn't this formally have a Fixes tag?
Signature · 2 lines
With Best Regards, Andy Shevchenko
Re: [PATCH 5/6] i2c: designware: Fix RX FIFO depth define on Wangxun 10Gb NIC
Quoted 11 lines · Andy Shevchenko
Quoted 7 lines · Jarkko Nikula
I believe RX FIFO depth define 0 is incorrect on Wangxun 10Gb NIC. It must be at least 1 since code is able to read received data from the
DW_IC_DATA_CMDregister.For now this define is irrelevant since the
txgbe_i2c_dw_xfer_quirk()doesn't use therx_fifo_depthmember variable of structdw_i2c_devbut is needed when converting code into generic polling mode implementation.Shouldn't this formally have a Fixes tag?
I don't think so since this define is not in use before my last patch. May needlessly bother stable bot & folks.
PATCH 5/5 i2c: designware: Implement generic polling mode code for Wangxun 10Gb NIC
In reply to Jarkko NikulaI got an idea the i2c-designware should not need duplicated state machines for the interrupt and polling modes. The IP is practically the same and state transitions happens in response to the events that can be observed from the DW_IC_RAW_INTR_STAT register. Either by interrupts or by polling.
Another reasons are the interrupt mode is the most tested, has handling for many exceptions as well as transmit abort handling and those are missing from two polling mode quirks.
Patch implements generic polling mode code which shares the same code with interrupt mode code. This is done by moving event handling from the i2c_dw_isr() into a new i2c_dw_process_transfer() that will be called both from the i2c_dw_isr() and polling loop.
Polling loop is implemented in a new i2c_dw_wait_transfer() that is shared between both modes. In interrupt mode it waits for the completion object as before. In polling mode both completion object and DW_IC_RAW_INTR_STAT are polled.
I decided to convert the txgbe_i2c_dw_xfer_quirk() straight to generic polling mode code in this patch. It doesn't have HW dependent quirks like the amd_i2c_dw_xfer_quirk() does have and without users this patch is needless.
--- drivers/i2c/busses/i2c-designware-core.h | 5 + drivers/i2c/busses/i2c-designware-master.c | 169 +++++++++------------ 2 files changed, 77 insertions(+), 97 deletions(-)
drivers/i2c/busses/i2c-designware-core.h +5 -0
@@ -212,6 +212,7 @@ struct reset_control;* @msg_err: error status of the current transfer * @status: i2c master status, one of STATUS_* * @abort_source: copy of the TX_ABRT_SOURCE register +* @sw_mask: SW mask of DW_IC_INTR_MASK used in polling mode * @irq: interrupt number for the i2c master * @flags: platform specific flags like type of IO accessors or model * @adapter: i2c subsystem adapter node
@@ -270,6 +271,7 @@ struct dw_i2c_dev {int msg_err;unsigned int status;unsigned int abort_source; +unsigned int sw_mask;int irq;u32 flags;struct i2c_adapter adapter;
@@ -358,6 +360,7 @@ static inline void __i2c_dw_write_intr_mask(struct dw_i2c_dev *dev,unsigned int val = dev->flags &ACCESS_POLLING ?0 : intr_mask;regmap_write (dev->map,DW_IC_INTR_MASK , val); + dev->sw_mask = intr_mask; } static inline void __i2c_dw_read_intr_mask(struct dw_i2c_dev *dev,
@@ -365,6 +368,8 @@ static inline void __i2c_dw_read_intr_mask(struct dw_i2c_dev *dev, {if (!(dev->flags &ACCESS_POLLING ))regmap_read (dev->map,DW_IC_INTR_MASK , intr_mask); +else + *intr_mask = dev->sw_mask; }void __i2c_dw_disable (struct dw_i2c_dev *dev);
drivers/i2c/busses/i2c-designware-master.c +72 -97
214 lines not on this page · raw message
Signature · 1 line
2.43.0
Andy Shevchenko
reviewerRe: [PATCH 5/5] i2c: designware: Implement generic polling mode code for Wangxun 10Gb NIC
Quoted 24 lines · Jarkko Nikula
I got an idea the i2c-designware should not need duplicated state machines for the interrupt and polling modes. The IP is practically the same and state transitions happens in response to the events that can be observed from the
DW_IC_RAW_INTR_STATregister. Either by interrupts or by polling.Another reasons are the interrupt mode is the most tested, has handling for many exceptions as well as transmit abort handling and those are missing from two polling mode quirks.
Patch implements generic polling mode code which shares the same code with interrupt mode code. This is done by moving event handling from the
i2c_dw_isr()into a newi2c_dw_process_transfer()that will be called both from thei2c_dw_isr()and polling loop.Polling loop is implemented in a new
i2c_dw_wait_transfer()that is shared between both modes. In interrupt mode it waits for the completion object as before. In polling mode both completion object andDW_IC_RAW_INTR_STATare polled.I decided to convert the
txgbe_i2c_dw_xfer_quirk()straight to generic polling mode code in this patch. It doesn't have HW dependent quirks like theamd_i2c_dw_xfer_quirk()does have and without users this patch is needless.
...
drivers/i2c/busses/i2c-designware-master.c static irqreturn_t i2c_dw_isr(int this_irq, void *dev_id) · line 689
Quoted 6 lines
drivers/i2c/busses/i2c-designware-master.c +6
-0+regmap_read (dev->map,DW_IC_ENABLE , &enabled); +regmap_read (dev->map,DW_IC_RAW_INTR_STAT , &stat); +if (!enabled || !(stat & ~DW_IC_INTR_ACTIVITY )) +return IRQ_NONE ; +if (pm_runtime_suspended (dev->dev) || stat ==GENMASK (31 ,0 )) +return IRQ_NONE ;
Hmm... This was in the original code, but to me checking for stat = ~0 is logically to have done before checking for (set / cleared) bits in it.
...
drivers/i2c/busses/i2c-designware-master.c static irqreturn_t i2c_dw_isr(int this_irq, void *dev_id) · line 720
Quoted 13 lines
drivers/i2c/busses/i2c-designware-master.c +8
-0+{ +unsigned long timeout = dev->adapter.timeout; +unsigned int stat; +int ret =0 ; +if (!(dev->flags &ACCESS_POLLING )) { + ret =wait_for_completion_timeout (&dev->cmd_complete, + timeout) ?0 : -ETIMEDOUT ; + }else {
...this is redundant 'else' as redundant label below.
You can return earlier. But it might be done this way to have single point of the (same) error code.
if (...) {
ret = ...;
if (ret)
return 0;
} else {
...
if (try_wait_for_completion(&dev->cmd_complete))
return 0;
...
}
return ETIMEDOUT;Quoted 18 lines
drivers/i2c/busses/i2c-designware-master.c +18
-0+ timeout += jiffies; +do { +if (try_wait_for_completion (&dev->cmd_complete)) + goto out; + + stat =i2c_dw_read_clear_intrbits (dev); +if (stat) +i2c_dw_process_transfer (dev, stat); +else +/* Try save some power */ +usleep_range (3 ,25 ); + }while (time_before (jiffies, timeout)); + ret =ETIMEDOUT ; + } + +out: +return ret; +}
Signature · 2 lines
With Best Regards, Andy Shevchenko
Re: [PATCH 5/5] i2c: designware: Implement generic polling mode code for Wangxun 10Gb NIC
In reply to Jarkko NikulaQuoted 25 lines · Jarkko Nikula
I got an idea the i2c-designware should not need duplicated state machines for the interrupt and polling modes. The IP is practically the same and state transitions happens in response to the events that can be observed from the
DW_IC_RAW_INTR_STATregister. Either by interrupts or by polling.Another reasons are the interrupt mode is the most tested, has handling for many exceptions as well as transmit abort handling and those are missing from two polling mode quirks.
Patch implements generic polling mode code which shares the same code with interrupt mode code. This is done by moving event handling from the
i2c_dw_isr()into a newi2c_dw_process_transfer()that will be called both from thei2c_dw_isr()and polling loop.Polling loop is implemented in a new
i2c_dw_wait_transfer()that is shared between both modes. In interrupt mode it waits for the completion object as before. In polling mode both completion object andDW_IC_RAW_INTR_STATare polled.I decided to convert the
txgbe_i2c_dw_xfer_quirk()straight to generic polling mode code in this patch. It doesn't have HW dependent quirks like theamd_i2c_dw_xfer_quirk()does have and without users this patch is needless.
Quoted 6 lines
Signed-off-by Jarkko Nikula <[email protected]>--- drivers/i2c/busses/i2c-designware-core.h | 5 + drivers/i2c/busses/i2c-designware-master.c | 169 +++++++++------------ 2 files changed, 77 insertions(+), 97 deletions(-)
Please discard this 5/5 which was accidental left over file in the same directory with the patches 0-6/6.
PATCH 6/6 i2c: designware: Implement generic polling mode code for Wangxun 10Gb NIC
In reply to Jarkko NikulaI got an idea the i2c-designware should not need duplicated state machines for the interrupt and polling modes. The IP is practically the same and state transitions happens in response to the events that can be observed from the DW_IC_RAW_INTR_STAT register. Either by interrupts or by polling.
Another reasons are the interrupt mode is the most tested, has handling for special cases as well as transmit abort handling and those are missing from two polling mode quirks.
Patch implements a generic polling mode by using existing code for interrupt mode. This is done by moving event handling from the i2c_dw_isr() into a new i2c_dw_process_transfer() that will be called both from the i2c_dw_isr() and a polling loop.
Polling loop is implemented in a new i2c_dw_wait_transfer() that is shared between both modes. In interrupt mode it waits for the completion object as before. In polling mode both completion object and DW_IC_RAW_INTR_STAT are polled to determine completed transfer and state transitions.
Loop tries to save power by sleeping "stetson guessed" range between 3 and 25 µS which falls between 10 cycles of High-speed mode 3.4 Mb/s and Fast mode 400 kHz. With it the CPU usage was reduced under heavy Fast mode I2C transfer without much increase in total transfer time but otherwise no more effort has been put to optimize this.
I decided to convert the txgbe_i2c_dw_xfer_quirk() straight to generic polling mode code in this patch. It doesn't have HW dependent quirks like the amd_i2c_dw_xfer_quirk() does have and without users this patch is needless.
--- drivers/i2c/busses/i2c-designware-core.h | 5 + drivers/i2c/busses/i2c-designware-master.c | 169 +++++++++------------ 2 files changed, 77 insertions(+), 97 deletions(-)
drivers/i2c/busses/i2c-designware-core.h +5 -0
@@ -212,6 +212,7 @@ struct reset_control;* @msg_err: error status of the current transfer * @status: i2c master status, one of STATUS_* * @abort_source: copy of the TX_ABRT_SOURCE register +* @sw_mask: SW mask of DW_IC_INTR_MASK used in polling mode * @irq: interrupt number for the i2c master * @flags: platform specific flags like type of IO accessors or model * @adapter: i2c subsystem adapter node
@@ -270,6 +271,7 @@ struct dw_i2c_dev {int msg_err;unsigned int status;unsigned int abort_source; +unsigned int sw_mask;int irq;u32 flags;struct i2c_adapter adapter;
@@ -358,6 +360,7 @@ static inline void __i2c_dw_write_intr_mask(struct dw_i2c_dev *dev,unsigned int val = dev->flags &ACCESS_POLLING ?0 : intr_mask;regmap_write (dev->map,DW_IC_INTR_MASK , val); + dev->sw_mask = intr_mask; } static inline void __i2c_dw_read_intr_mask(struct dw_i2c_dev *dev,
@@ -365,6 +368,8 @@ static inline void __i2c_dw_read_intr_mask(struct dw_i2c_dev *dev, {if (!(dev->flags &ACCESS_POLLING ))regmap_read (dev->map,DW_IC_INTR_MASK , intr_mask); +else + *intr_mask = dev->sw_mask; }void __i2c_dw_disable (struct dw_i2c_dev *dev);
drivers/i2c/busses/i2c-designware-master.c +72 -97
214 lines not on this page · raw message
Signature · 1 line
2.43.0
Andy Shevchenko
reviewerQuoted 12 lines · Jarkko Nikula
Hi Jiawen, Sanket and Basavaraj
Could you test on your Wangxun 10Gb NIC and AMD NAVI GPU harware this patchset since it's touching both.
For the AMD NAVI GPU and changes are less but for the Wangxun 10Gb NIC patchset replaces the
txgbe_i2c_dw_xfer_quirk()with generic polling mode code.I've been testing this on our HW which all have interrupt connected and tried to cover also FIFO depth defines for Wangxun 10Gb NIC. Obviously I would like to know how this set works on your real HW.
...
Quoted 9 lines
Jarkko Nikula (6):
i2c: designware: Uniform initialization flow for polling mode
i2c: designware: Do not enable interrupts shortly in polling mode
i2c: designware: Use accessors toDW_IC_INTR_MASKregister
i2c: designware: Move interrupt handling functions beforei2c_dw_xfer()i2c: designware: Fix RX FIFO depth define on Wangxun 10Gb NIC
i2c: designware: Implement generic polling mode code for Wangxun 10Gb
NIC
Last two patches named the same, Something is screwed up.
Signature · 2 lines
With Best Regards, Andy Shevchenko
Quoted 26 lines · Andy Shevchenko
Quoted 12 lines · Jarkko Nikula
Hi Jiawen, Sanket and Basavaraj
Could you test on your Wangxun 10Gb NIC and AMD NAVI GPU harware this patchset since it's touching both.
For the AMD NAVI GPU and changes are less but for the Wangxun 10Gb NIC patchset replaces the
txgbe_i2c_dw_xfer_quirk()with generic polling mode code.I've been testing this on our HW which all have interrupt connected and tried to cover also FIFO depth defines for Wangxun 10Gb NIC. Obviously I would like to know how this set works on your real HW.
...
Quoted 9 lines
Jarkko Nikula (6):
i2c: designware: Uniform initialization flow for polling mode
i2c: designware: Do not enable interrupts shortly in polling mode
i2c: designware: Use accessors toDW_IC_INTR_MASKregister
i2c: designware: Move interrupt handling functions beforei2c_dw_xfer()i2c: designware: Fix RX FIFO depth define on Wangxun 10Gb NIC
i2c: designware: Implement generic polling mode code for Wangxun 10Gb
NIC
Quoted 2 lines
Last two patches named the same, Something is screwed up.
Indeed, thanks Andy. 5/5 was accidental left over file.
Quoted 5 lines · Jarkko Nikula
Hi Jiawen, Sanket and Basavaraj
Could you test on your Wangxun 10Gb NIC and AMD NAVI GPU harware this patchset since it's touching both.
Quoted 7 lines
For the AMD NAVI GPU and changes are less but for the Wangxun 10Gb NIC patchset replaces the
txgbe_i2c_dw_xfer_quirk()with generic polling mode code.I've been testing this on our HW which all have interrupt connected and tried to cover also FIFO depth defines for Wangxun 10Gb NIC. Obviously I would like to know how this set works on your real HW.
Hi Jarkko,
Sorry for the late reply.
I've been testing the patch series on Wangxun 10Gb NIC, it works well. Thanks!
Hi
Quoted 11 lines · Jiawen Wu
Quoted 3 lines · Jarkko Nikula
I've been testing this on our HW which all have interrupt connected and tried to cover also FIFO depth defines for Wangxun 10Gb NIC. Obviously I would like to know how this set works on your real HW.
Hi Jarkko,
Sorry for the late reply.
I've been testing the patch series on Wangxun 10Gb NIC, it works well. Thanks!
Thank you very much! The first priority for this set it must not cause a regression on your HW and I'm glad my idea works :-)