aboutsummaryrefslogtreecommitdiff
path: root/include/rdma/ib_cache.h
diff options
context:
space:
mode:
authorJean Delvare <khali@linux-fr.org>2007-05-01 23:26:33 +0200
committerJean Delvare <khali@hyperion.delvare>2007-05-01 23:26:33 +0200
commit424ed67c7dae37e8115e1bebc3261e86a624dff2 (patch)
tree5873a171f3a54109ac680de55bf5dbf139b67a23 /include/rdma/ib_cache.h
parent7c175499822ba34ba60f32e4995fcc16c007d308 (diff)
i2c-algo-bit: Implement a 50/50 SCL duty cycle
The original i2c-algo-bit implementation uses a 33/66 SCL duty cycle when bits are being written on the bus. While the I2C specification doesn't forbid it, this prevents us from driving the I2C bus to its max speed, limiting us to 66 kbps max on standard I2C busses. Implementing a 50/50 duty cycle instead lets us max out the bandwidth up to the theoretical max of 100 kbps on standard I2C busses. This is particularly important when large amounts of data need to be transfered over the bus, as is the case with some TV adapters when the firmware is being uploaded. In fact this change even allows, at least in theory, fast-mode I2C support at 125, 166 and 250 kbps. There's no way to reach the theoretical max of 400 kbps with this implementation. But I don't think we want to put efforts in that direction anyway: software-driven I2C is very CPU-intensive and bad for latency. Other timing changes: * Don't set SDA high explicitly on error, we're going to issue a stop condition before we leave anyway. * If an error occurs when sending the slave address, yield the CPU before retrying, and remove the additional delay after the new start condition. Signed-off-by: Jean Delvare <khali@linux-fr.org>
Diffstat (limited to 'include/rdma/ib_cache.h')
0 files changed, 0 insertions, 0 deletions