2012-02-20 08:14:23

by Alexander Stein

[permalink] [raw]

Subject: [PATCH v2] i2c-eg20t: Rework pch_i2c_wait_for_bus_idle to reduce wait time

---

If you insert several i2c transfers, the driver might start the next one
while the STOP bit of the previous transfer is still on the bus, marking
the bus as busy.
pch_i2c_wait_for_bus_idle does an msleep(20) delaying the next transfer
by >=20ms. Reduce wait time by actively waiting 5 us once, then using
usleep_range.

Signed-off-by: Alexander Stein <[email protected]>
---
Changes in v2:
* Remove redundant code

drivers/i2c/busses/i2c-eg20t.c | 42 ++++++++++++++++++++++-----------------
1 files changed, 24 insertions(+), 18 deletions(-)

diff --git a/drivers/i2c/busses/i2c-eg20t.c b/drivers/i2c/busses/i2c-eg20t.c
index ca88776..1eb60ca 100644
--- a/drivers/i2c/busses/i2c-eg20t.c
+++ b/drivers/i2c/busses/i2c-eg20t.c
@@ -271,30 +271,36 @@ static inline bool ktime_lt(const ktime_t cmp1, const ktime_t cmp2)
/**
* pch_i2c_wait_for_bus_idle() - check the status of bus.
* @adap: Pointer to struct i2c_algo_pch_data.
- * @timeout: waiting time counter (us).
+ * @timeout: waiting time counter (ms).
*/
static s32 pch_i2c_wait_for_bus_idle(struct i2c_algo_pch_data *adap,
s32 timeout)
{
void __iomem *p = adap->pch_base_address;
- ktime_t ns_val;
+ int schedule = 0;
+ unsigned long end = jiffies + msecs_to_jiffies(timeout);
+
+ while (ioread32(p + PCH_I2CSR) & I2CMBB_BIT) {
+ if (time_after(jiffies, end)) {
+ pch_dbg(adap, "I2CSR = %x\n", ioread32(p + PCH_I2CSR));
+ pch_err(adap, "%s: Timeout Error.return%d\n",
+ __func__, -ETIME);
+ pch_i2c_init(adap);

- if ((ioread32(p + PCH_I2CSR) & I2CMBB_BIT) == 0)
- return 0;
+ return -ETIME;
+ }

- /* MAX timeout value is timeout*1000*1000nsec */
- ns_val = ktime_add_ns(ktime_get(), timeout*1000*1000);
- do {
- msleep(20);
- if ((ioread32(p + PCH_I2CSR) & I2CMBB_BIT) == 0)
- return 0;
- } while (ktime_lt(ktime_get(), ns_val));
+ if (!schedule)
+ /* Retry after some usecs */
+ udelay(5);
+ else
+ /* Wait a bit more without consuming CPU */
+ usleep_range(20, 1000);

- pch_dbg(adap, "I2CSR = %x\n", ioread32(p + PCH_I2CSR));
- pch_err(adap, "%s: Timeout Error.return%d\n", __func__, -ETIME);
- pch_i2c_init(adap);
+ schedule = 1;
+ }

- return -ETIME;
+ return 0;
}

/**
--
1.7.3.4

2012-02-24 12:18:07

by Wolfram Sang

[permalink] [raw]

Subject: Re: [PATCH v2] i2c-eg20t: Rework pch_i2c_wait_for_bus_idle to reduce wait time

---

On Mon, Feb 20, 2012 at 09:14:16AM +0100, Alexander Stein wrote:
> If you insert several i2c transfers, the driver might start the next one
> while the STOP bit of the previous transfer is still on the bus, marking
> the bus as busy.
> pch_i2c_wait_for_bus_idle does an msleep(20) delaying the next transfer
> by >=20ms. Reduce wait time by actively waiting 5 us once, then using
> usleep_range.
>
> Signed-off-by: Alexander Stein <[email protected]>

Tomoya, are you happy with it? (I hope he reads the list, since his
companies mailserver doesn't seem to accept my mails).

Regards,

Wolfram

--
Pengutronix e.K. | Wolfram Sang |
Industrial Linux Solutions | http://www.pengutronix.de/ |

---

Attachments:

(No filename) (773.00 B)
signature.asc (198.00 B)
Digital signature Download all attachments

2012-02-27 23:53:11

by Tomoya MORINAGA

[permalink] [raw]

Subject: Re: [PATCH v2] i2c-eg20t: Rework pch_i2c_wait_for_bus_idle to reduce wait time

---

Hi

Our team moved to ROHM Co., Ltd.
So, previous email address already disappeared.
Current my mail address is "[email protected]".

thanks.
ROHM Co., Ltd.
tomoya