Re: [PATCH] i2c: imx: add slave support. v2 status
Hello, Please find some comments below. 31.10.2016 5:14, Frkuska, Joshua пишет: Hello Maxim, Thank you very much for the intermediate patch. I am in the process of reviewing it. Please let me clarify a few questions I have. 1. What alternative to "bus busy/bus free/IBB" polling do you have in mind? This seems like a reasonable approach to me. We didn't find any suitable alternatives. The only one we're considered was using timeout on receive (which is kind of polling of course) 2. What are the major points you consider in need of refactoring? As you can see we have implemented FSM in slave thread. - Due to lack of time all master functionality had not been included in State Machine. - wait_event_timeout() calls are used in every event handler (obviosly it is better to have only one wait function) - Need to review state switching code 3. You mention race conditions being fixed in this version relating to bus-locking by the slave and breaking slave transactions by the master. Does this mean mixed slave/master mode works to your satisfaction? If not, what work still needs to be done here? Yes mixed slave/master mode works ok. It had passed long-lasting stress tests (async message exchange of two imx6 boards connected together by i2c bus ) 4. You mention the need for a slave locking test and a work-around (checking IMX_I2C_I2DR and IBB) being in-place. Why is this work-around not sufficient? By the time we discovered I2DR workaround we went far from version 2 of driver and it wasn't tested. I'm sure that I2DR workaround will improve stability, but I do not know if it will fix all issues (i.e. passing of stress tests ) Best Regards, Maxim Syrchin Thanks again, Joshua On 10/28/2016 04:38 AM, Maxim Syrchin wrote: Hi, Sorry for huge delay in answering. Unfortunately we don't have enough resources now to prepare clean enough patch to be accepted by community. Please find the latest version attached. Driver has passed stress tests, but looks like it need seriuos refactoring (it is unnecessarily complicated). We still have polling in slave code. Since imx doesn't generate interrupt on "bus busy"/"bus free" events we have to test IBB bit in polling loop. Comparing to v2 version several race conditions were fixed (bus locking by slave, breaking slave transaction by starting master xfer). v2 is working pretty good in slave-only or master-only mode. It is reasonable to add slave locking test - sometimes imx6 hw doesn't release data line. As workaround we use dummy reading of IMX_I2C_I2DR if driver is in slave mode and IBB bit is set for a long time. Thanks, Maxim Syrchin 27.10.2016 10:31, Frkuska, Joshua пишет: Hi Maxim, Dmitriy, Wolfram, If there is no immediate plan for a third release of the below patch set, would it be possible to continue with merging v2 after addressing the remaining concerns? Thank you and regards, Joshua Hi Maxim, On 2016-03-04 11:06:10 in the thread "Re: [PATCH] i2c: imx: add slave support. v2" referenced here: https://patchwork.ozlabs.org/patch/573353/ you said: Hi Wolfram, I'm now working on creating new driver version. I think I'll be able to sent it soon. Do you still plan to release a driver update for an i2c imx driver slave support? Best regards, Jim Baxter
Re: [PATCH] i2c: imx: add slave support. v2 status
Hello, Please find some comments below. 31.10.2016 5:14, Frkuska, Joshua пишет: Hello Maxim, Thank you very much for the intermediate patch. I am in the process of reviewing it. Please let me clarify a few questions I have. 1. What alternative to "bus busy/bus free/IBB" polling do you have in mind? This seems like a reasonable approach to me. We didn't find any suitable alternatives. The only one we're considered was using timeout on receive (which is kind of polling of course) 2. What are the major points you consider in need of refactoring? As you can see we have implemented FSM in slave thread. - Due to lack of time all master functionality had not been included in State Machine. - wait_event_timeout() calls are used in every event handler (obviosly it is better to have only one wait function) - Need to review state switching code 3. You mention race conditions being fixed in this version relating to bus-locking by the slave and breaking slave transactions by the master. Does this mean mixed slave/master mode works to your satisfaction? If not, what work still needs to be done here? Yes mixed slave/master mode works ok. It had passed long-lasting stress tests (async message exchange of two imx6 boards connected together by i2c bus ) 4. You mention the need for a slave locking test and a work-around (checking IMX_I2C_I2DR and IBB) being in-place. Why is this work-around not sufficient? By the time we discovered I2DR workaround we went far from version 2 of driver and it wasn't tested. I'm sure that I2DR workaround will improve stability, but I do not know if it will fix all issues (i.e. passing of stress tests ) Best Regards, Maxim Syrchin Thanks again, Joshua On 10/28/2016 04:38 AM, Maxim Syrchin wrote: Hi, Sorry for huge delay in answering. Unfortunately we don't have enough resources now to prepare clean enough patch to be accepted by community. Please find the latest version attached. Driver has passed stress tests, but looks like it need seriuos refactoring (it is unnecessarily complicated). We still have polling in slave code. Since imx doesn't generate interrupt on "bus busy"/"bus free" events we have to test IBB bit in polling loop. Comparing to v2 version several race conditions were fixed (bus locking by slave, breaking slave transaction by starting master xfer). v2 is working pretty good in slave-only or master-only mode. It is reasonable to add slave locking test - sometimes imx6 hw doesn't release data line. As workaround we use dummy reading of IMX_I2C_I2DR if driver is in slave mode and IBB bit is set for a long time. Thanks, Maxim Syrchin 27.10.2016 10:31, Frkuska, Joshua пишет: Hi Maxim, Dmitriy, Wolfram, If there is no immediate plan for a third release of the below patch set, would it be possible to continue with merging v2 after addressing the remaining concerns? Thank you and regards, Joshua Hi Maxim, On 2016-03-04 11:06:10 in the thread "Re: [PATCH] i2c: imx: add slave support. v2" referenced here: https://patchwork.ozlabs.org/patch/573353/ you said: Hi Wolfram, I'm now working on creating new driver version. I think I'll be able to sent it soon. Do you still plan to release a driver update for an i2c imx driver slave support? Best regards, Jim Baxter
Re: [PATCH] i2c: imx: add slave support. v2 status
Hi,
Sorry for huge delay in answering. Unfortunately we don't have enough
resources now to prepare clean enough patch to be accepted by community.
Please find the latest version attached. Driver has passed stress
tests, but looks like it need seriuos refactoring (it is unnecessarily
complicated).
We still have polling in slave code. Since imx doesn't generate
interrupt on "bus busy"/"bus free" events we have to test IBB bit in
polling loop.
Comparing to v2 version several race conditions were fixed (bus locking
by slave, breaking slave transaction by starting master xfer). v2 is
working pretty good in slave-only or master-only mode. It is reasonable
to add slave locking test - sometimes imx6 hw doesn't release data
line. As workaround we use dummy reading of IMX_I2C_I2DR if driver is
in slave mode and IBB bit is set for a long time.
Thanks,
Maxim Syrchin
27.10.2016 10:31, Frkuska, Joshua пишет:
Hi Maxim, Dmitriy, Wolfram,
If there is no immediate plan for a third release of the below patch
set, would it be possible to continue with merging v2 after addressing
the remaining concerns?
Thank you and regards,
Joshua
Hi Maxim,
On 2016-03-04 11:06:10 in the thread "Re: [PATCH] i2c: imx: add slave
support. v2"
referenced here: https://patchwork.ozlabs.org/patch/573353/ you said:
Hi Wolfram,
I'm now working on creating new driver version. I think I'll be able to
sent it soon.
Do you still plan to release a driver update for an i2c imx driver
slave support?
Best regards,
Jim Baxter
From 61ae34268d78eb284bf8ee0cbe8f9f0c5e7df074 Mon Sep 17 00:00:00 2001
From: Maxim Syrchin <msyrc...@dev.rtsoft.ru>
Date: Thu, 27 Oct 2016 17:37:56 +0300
Subject: [PATCH] i2c: imx: add slave support. v3
Add I2C slave provider using the generic slave interface.
It also supports master transactions when the slave in the idle mode.
Signed-off-by: Maxim Syrchin <msyrc...@dev.rtsoft.ru>
---
drivers/i2c/busses/i2c-imx.c | 682 +--
1 file changed, 653 insertions(+), 29 deletions(-)
diff --git a/drivers/i2c/busses/i2c-imx.c b/drivers/i2c/busses/i2c-imx.c
index 592a8f2..11a2292 100644
--- a/drivers/i2c/busses/i2c-imx.c
+++ b/drivers/i2c/busses/i2c-imx.c
@@ -53,6 +53,7 @@
#include
#include
#include
+#include
/* This will be the driver name the kernel reports */
#define DRIVER_NAME "imx-i2c"
@@ -171,6 +172,82 @@ enum imx_i2c_type {
VF610_I2C,
};
+enum i2c_imx_state {
+ STATE_IDLE = 0,
+ STATE_SLAVE,
+ STATE_MASTER,
+ STATE_SP
+};
+
+#define MAX_EVENTS (1<<4)
+#define EUNDEFINED 4000
+
+enum i2c_imx_events {
+ EVT_AL = 0,
+ EVT_SI,
+ EVT_START,
+ EVT_STOP,
+ EVT_POLL,
+ EVT_INVALID,
+ EVT_ENTRY
+};
+
+typedef struct evt_queue {
+ atomic_t count;
+ atomic_t ir;
+ atomic_t iw;
+ unsigned int evt[MAX_EVENTS];
+} evt_queue;
+
+static inline int evt_find_next_idx(atomic_t *v)
+{
+ return atomic_inc_return(v) & (MAX_EVENTS - 1);
+}
+
+static unsigned int evt_put(evt_queue *stk, unsigned int evt)
+{
+ int count = atomic_inc_return(>count);
+ int idx;
+ if (count < MAX_EVENTS)
+ {
+ idx = evt_find_next_idx(>iw);
+ stk->evt[idx] = evt;
+
+ return 0;
+ } else {
+ atomic_dec(>count);
+ return EVT_INVALID;
+ }
+}
+
+static unsigned int evt_get(evt_queue *stk)
+{
+ int count = atomic_dec_return(>count);
+ int idx;
+
+ if (count >= 0)
+ {
+ idx = evt_find_next_idx(>ir);
+ return stk->evt[idx];
+ } else {
+ atomic_inc(>count);
+ return EVT_INVALID;
+ }
+}
+
+static unsigned int evt_is_empty(evt_queue *stk)
+{
+ int ret = atomic_read(>count);
+ return (ret <= 0);
+}
+
+static void evt_init(evt_queue *stk)
+{
+ atomic_set(>count,0);
+ atomic_set(>iw,0);
+ atomic_set(>ir,0);
+}
+
struct imx_i2c_hwdata {
enum imx_i2c_type devtype;
unsignedregshift;
@@ -193,6 +270,7 @@ struct imx_i2c_dma {
struct imx_i2c_struct {
struct i2c_adapter adapter;
+ struct i2c_client *slave;
struct clk *clk;
void __iomem*base;
wait_queue_head_t queue;
@@ -210,6 +288,18 @@ struct imx_i2c_struct {
struct pinctrl_state *pinctrl_pins_gpio;
struct imx_i2c_dma *dma;
+
+ unsigned intstate;
+ evt_queue events;
+ atomic_tlast_error;
+
+ int master_interrupt;
+ int start_retry_cnt;
+ int slave_poll_cnt;
+
+ struct task_struct *slave_task;
+ wait_queue_head_t sta
Re: [PATCH] i2c: imx: add slave support. v2 status
Hi,
Sorry for huge delay in answering. Unfortunately we don't have enough
resources now to prepare clean enough patch to be accepted by community.
Please find the latest version attached. Driver has passed stress
tests, but looks like it need seriuos refactoring (it is unnecessarily
complicated).
We still have polling in slave code. Since imx doesn't generate
interrupt on "bus busy"/"bus free" events we have to test IBB bit in
polling loop.
Comparing to v2 version several race conditions were fixed (bus locking
by slave, breaking slave transaction by starting master xfer). v2 is
working pretty good in slave-only or master-only mode. It is reasonable
to add slave locking test - sometimes imx6 hw doesn't release data
line. As workaround we use dummy reading of IMX_I2C_I2DR if driver is
in slave mode and IBB bit is set for a long time.
Thanks,
Maxim Syrchin
27.10.2016 10:31, Frkuska, Joshua пишет:
Hi Maxim, Dmitriy, Wolfram,
If there is no immediate plan for a third release of the below patch
set, would it be possible to continue with merging v2 after addressing
the remaining concerns?
Thank you and regards,
Joshua
Hi Maxim,
On 2016-03-04 11:06:10 in the thread "Re: [PATCH] i2c: imx: add slave
support. v2"
referenced here: https://patchwork.ozlabs.org/patch/573353/ you said:
Hi Wolfram,
I'm now working on creating new driver version. I think I'll be able to
sent it soon.
Do you still plan to release a driver update for an i2c imx driver
slave support?
Best regards,
Jim Baxter
From 61ae34268d78eb284bf8ee0cbe8f9f0c5e7df074 Mon Sep 17 00:00:00 2001
From: Maxim Syrchin
Date: Thu, 27 Oct 2016 17:37:56 +0300
Subject: [PATCH] i2c: imx: add slave support. v3
Add I2C slave provider using the generic slave interface.
It also supports master transactions when the slave in the idle mode.
Signed-off-by: Maxim Syrchin
---
drivers/i2c/busses/i2c-imx.c | 682 +--
1 file changed, 653 insertions(+), 29 deletions(-)
diff --git a/drivers/i2c/busses/i2c-imx.c b/drivers/i2c/busses/i2c-imx.c
index 592a8f2..11a2292 100644
--- a/drivers/i2c/busses/i2c-imx.c
+++ b/drivers/i2c/busses/i2c-imx.c
@@ -53,6 +53,7 @@
#include
#include
#include
+#include
/* This will be the driver name the kernel reports */
#define DRIVER_NAME "imx-i2c"
@@ -171,6 +172,82 @@ enum imx_i2c_type {
VF610_I2C,
};
+enum i2c_imx_state {
+ STATE_IDLE = 0,
+ STATE_SLAVE,
+ STATE_MASTER,
+ STATE_SP
+};
+
+#define MAX_EVENTS (1<<4)
+#define EUNDEFINED 4000
+
+enum i2c_imx_events {
+ EVT_AL = 0,
+ EVT_SI,
+ EVT_START,
+ EVT_STOP,
+ EVT_POLL,
+ EVT_INVALID,
+ EVT_ENTRY
+};
+
+typedef struct evt_queue {
+ atomic_t count;
+ atomic_t ir;
+ atomic_t iw;
+ unsigned int evt[MAX_EVENTS];
+} evt_queue;
+
+static inline int evt_find_next_idx(atomic_t *v)
+{
+ return atomic_inc_return(v) & (MAX_EVENTS - 1);
+}
+
+static unsigned int evt_put(evt_queue *stk, unsigned int evt)
+{
+ int count = atomic_inc_return(>count);
+ int idx;
+ if (count < MAX_EVENTS)
+ {
+ idx = evt_find_next_idx(>iw);
+ stk->evt[idx] = evt;
+
+ return 0;
+ } else {
+ atomic_dec(>count);
+ return EVT_INVALID;
+ }
+}
+
+static unsigned int evt_get(evt_queue *stk)
+{
+ int count = atomic_dec_return(>count);
+ int idx;
+
+ if (count >= 0)
+ {
+ idx = evt_find_next_idx(>ir);
+ return stk->evt[idx];
+ } else {
+ atomic_inc(>count);
+ return EVT_INVALID;
+ }
+}
+
+static unsigned int evt_is_empty(evt_queue *stk)
+{
+ int ret = atomic_read(>count);
+ return (ret <= 0);
+}
+
+static void evt_init(evt_queue *stk)
+{
+ atomic_set(>count,0);
+ atomic_set(>iw,0);
+ atomic_set(>ir,0);
+}
+
struct imx_i2c_hwdata {
enum imx_i2c_type devtype;
unsignedregshift;
@@ -193,6 +270,7 @@ struct imx_i2c_dma {
struct imx_i2c_struct {
struct i2c_adapter adapter;
+ struct i2c_client *slave;
struct clk *clk;
void __iomem*base;
wait_queue_head_t queue;
@@ -210,6 +288,18 @@ struct imx_i2c_struct {
struct pinctrl_state *pinctrl_pins_gpio;
struct imx_i2c_dma *dma;
+
+ unsigned intstate;
+ evt_queue events;
+ atomic_tlast_error;
+
+ int master_interrupt;
+ int start_retry_cnt;
+ int slave_poll_cnt;
+
+ struct task_struct *slave_task;
+ wait_queue_head_t state_queue;
+
};
static const struct imx_i2c_hwdata imx1_i2c_hwdata
Re: [PATCH] i2c: imx: add slave support. v2
Hi Wolfram,
I'm now working on creating new driver version. I think I'll be able to
sent it soon.
04.03.2016 0:35, Wolfram Sang пишет:
There are might be race conditions.
Can you name them
Most of races are fixed already. There were some issues with interrupt
latencies - sometimes slave interrupt appears in process of starting
master xfer.
+enum imx_i2c_slave_state {
+ I2C_IMX_SLAVE_IDLE,
+ I2C_IMX_SLAVE_IRQ,
+ I2C_IMX_SLAVE_POLLING
Highlevel question first: Why do you have polling? Why would anyone not
want to use interrupts here?
Since imx doesn't generate interrupt on "bus stop" condition we'd had to
implement polling scheme. Interrupts are used for starting polling and
for waking polling loop on new slave request. Without polling we can't
handle "end-of-packet" event correctly.
In current version states are:
I2C_IMX_SLAVE_IDLE // default state. slave process is waiting for
interrupt
I2C_IMX_SLAVE_POLLING // slave transfer is in process.
I2C_IMX_MASTER // master transfer is in process.
Re: [PATCH] i2c: imx: add slave support. v2
Hi Wolfram,
I'm now working on creating new driver version. I think I'll be able to
sent it soon.
04.03.2016 0:35, Wolfram Sang пишет:
There are might be race conditions.
Can you name them
Most of races are fixed already. There were some issues with interrupt
latencies - sometimes slave interrupt appears in process of starting
master xfer.
+enum imx_i2c_slave_state {
+ I2C_IMX_SLAVE_IDLE,
+ I2C_IMX_SLAVE_IRQ,
+ I2C_IMX_SLAVE_POLLING
Highlevel question first: Why do you have polling? Why would anyone not
want to use interrupts here?
Since imx doesn't generate interrupt on "bus stop" condition we'd had to
implement polling scheme. Interrupts are used for starting polling and
for waking polling loop on new slave request. Without polling we can't
handle "end-of-packet" event correctly.
In current version states are:
I2C_IMX_SLAVE_IDLE // default state. slave process is waiting for
interrupt
I2C_IMX_SLAVE_POLLING // slave transfer is in process.
I2C_IMX_MASTER // master transfer is in process.
