Re: RFC: fast transactions in EC [was: a problem about the two patches in bug 10724 & 11428]

[Alexey Starikovskiy <astarikovskiy@suse.de>]

Zhao Yakui wrote:
> On Thu, 2008-09-04 at 01:55 +0400, Alexey Starikovskiy wrote:
>> Hi,
>> Here is the patch, which moves almost all transaction functionality into interrupt handler, which is IMHO good.
>>
>> with the enabled DEBUG, the interrupt picture looks like this (single transaction):
> Thanks for your work and efforts. Maybe you have tested the patch on
> your laptop. But IMO this is not reasonable. In the following cases
> maybe the patch can't work well.
>    a. EC GPE storm.  According to ACPI spec the EC uses the pulse
This _is_ the machine with the EC GPE storm. Acer TM 2300. And the patch works...
And you may see it in the quote I gave.
> interrupt and interrupt is firmware generated using an EC GPIO output,
> which is connected with chipset GPIO input. If the pulse waveform is
> very wide, maybe several EC GPE interrupts will be triggered although EC
> firmware generates one pulse waveform. How can we read the corresponding
> data from EC in the GPE interrupt service handler? Maybe the read/write
> data is completely incorrect.
Well, it is not any different from the data I get with unpatched ec.c.
>    b. If there is no EC interrupt although OBF_1 bit is valid.(In theory
> when OBF_1 is valid, EC should trigger GPE interrupt). In such case the
> read/write flowchart will be wrong.
Oh, you mean it does not cover all the error paths yet?
>   At the same time it seems that EC transaction flowchart will become
> more complex in this patch.
How so? It even does not involve second thread any longer :)