[Intel-gfx] [PATCH 1/2] Give up on edid retries when i2c tells us that bus is not there

[Michael Büsch]

On Thu, 20 Oct 2011 14:33:39 +0200
Jean Delvare <khali at linux-fr.org> wrote:

> retry mechanism: Chris Wilson and Michael Buesch (both Cc'd.) Chris,
> Michael, do you know of ways to reproduce the issue?

The error could easily reproduced by loading the machine heavily.
So my guess is that it is caused by electrical noise injected into
the i2c bus. Probably due to bad hardware design. But that doesn't matter.
I have to live with that. The error did not trigger again with the workaround
in place, though.

> Can you please
> also confirm that the error code you were receiving was not -ENXIO?

I really don't remember what error code it was.

> Note that the problem is more likely to happen with slow masters,
> because (1) every transaction takes longer and thus has a higher
> probability to be hit by interrupts and (2) long delays mean smaller
> margins to the spec limits, so interrupts are more likely to cause
> trouble. I see that both radeon_i2c and intel_i2c set udelay to 20 µs,
> which means a 25 kbps I2C bus. In general it is recommended to not
> drive the I2C bus below 10 kbps (that's even a hard limit for SMBus
> compliance.) nouveau_i2c is even worse with udelay = 40 µs which is
> equivalent to a 12.5 kbps I2C bus, very close to the low limit.

Hm. Not sure. I don't think the machine had heavy IRQ load.
Just high CPU and memory load (compile the kernel or something like that).

-- 
Greetings, Michael.