On 2024/1/4 03:39, Andi Shyti wrote:
Hi Hans,
[...]
static int viai2c_wait_bus_ready(struct viai2c *i2c)
{
@@ -25,12 +26,35 @@ static int viai2c_wait_bus_ready(struct viai2c *i2c)
static int viai2c_wait_status(struct viai2c *i2c)
{
int ret = 0;
- unsigned long wait_result;
+ unsigned long time_left;
+ unsigned long delta_ms;
+
+ time_left = wait_for_completion_timeout(&i2c->complete,
+ VIAI2C_TIMEOUT);
+ if (!time_left) {
+ dev_err(i2c->dev, "bus transfer timeout\n");
+ return -EIO;
+ }
- wait_result = wait_for_completion_timeout(&i2c->complete,
- msecs_to_jiffies(500));
- if (!wait_result)
- return -ETIMEDOUT;
this change is unrelated. Why are you changing the timeout here
from 500 to 1000?
it does unrelated, will drop it.
+ /*
+ * During each byte access, the host performs clock stretching.
+ * In this case, the thread may be interrupted by preemption,
+ * resulting in a long stretching time.
+ * However, some touchpad can only tolerate host clock stretching
+ * of no more than 200 ms. We reduce the impact of this through
+ * a retransmission mechanism.
+ */
is the hardware sending the stretching on its own?
Yes, controller will handle it.
+ local_irq_disable();
+ i2c->to = ktime_get();
+ delta_ms = ktime_to_ms(ktime_sub(i2c->to, i2c->ti));
+ if (delta_ms > VIAI2C_STRETCHING_TIMEOUT) {
+ local_irq_enable();
+ dev_warn(i2c->dev, "thread blocked more than %ldms\n",
+ delta_ms);
+ return -EAGAIN;
+ }
+ i2c->ti = i2c->to;
+ local_irq_enable();
if (i2c->cmd_status & VIAI2C_ISR_NACK_ADDR)
ret = -EIO;
@@ -184,6 +208,7 @@ int viai2c_xfer(struct i2c_adapter *adap,
int ret = 0;
struct viai2c *i2c = i2c_get_adapdata(adap);
+ i2c->to = i2c->ti = ktime_get();
for (i = 0; ret >= 0 && i < num; i++) {
pmsg = &msgs[i];
if (!(pmsg->flags & I2C_M_NOSTART)) {
diff --git a/drivers/i2c/busses/i2c-viai2c-common.h b/drivers/i2c/busses/i2c-viai2c-common.h
index f171f81e4d0f..73a88398d763 100644
--- a/drivers/i2c/busses/i2c-viai2c-common.h
+++ b/drivers/i2c/busses/i2c-viai2c-common.h
@@ -58,6 +58,8 @@ struct viai2c {
u16 tcr;
int irq;
u16 cmd_status;
+ ktime_t ti;
+ ktime_t to;
don't these need some arbitration?
Andi
I don't think it needs to be arbitration.
the controllers are independent of each other,
each access is locked using __i2c_lock_bus_helper().
Am I missing something?
Hans
