Re: [PATCH linux dev-4.10] drivers/misc: (aspeed-lpc-snoop): Add ast2400 to compat

From: Patrick Venture
Date: Wed Jul 05 2017 - 13:49:21 EST

On Wed, Jul 5, 2017 at 10:33 AM, Rob Lippert <roblip@xxxxxxxxx> wrote:
> I checked the datasheets when I wrote this and ast2400 does not have
> the (undocumented) HICRB register bits 14,15 that enables the BMC to
> actually respond to the snoop'ed address.

You're right, it is marked as "reserved" in the datasheet for the ast2400.

> Without setting that bit in the ast2500 the transactions to that I/O
> port would timeout on the host side (although the BMC snoop logic
> would still see it and log it).
> Probably not an issue for x86 systems that don't have any LPC I/O
> error handling anyways but LPC timeouts causes issues with POWER
> systems since it sets a hardware FIR bit which can cause boot
> failures.

Interesting. I've been running experiments on x86 and I haven't seen
any errors, so that adds more credence to your point. If a device
doesn't respond within X time, three times in a row, you get a triple
fault. But, on x86, I don't think I've seen any mechanism with an
expectation that a port IO write will have a guaranteed response.

For the use-case I'm chasing, my goal being to snoop PoST codes from
the host, there is in the datasheet a post-code control register set,
but I haven't explored configuring them or whether someone has written
the fifo driver for them.

> -Rob
> On Tue, Jul 4, 2017 at 8:45 AM, Patrick Venture <venture@xxxxxxxxxx> wrote:
>> This driver can be used on the aspeed ast2400.
>> Tested: ast2400 on quanta-q71l
>> Signed-off-by: Patrick Venture <venture@xxxxxxxxxx>
>> ---
>> drivers/misc/aspeed-lpc-snoop.c | 1 +
>> 1 file changed, 1 insertion(+)
>> diff --git a/drivers/misc/aspeed-lpc-snoop.c b/drivers/misc/aspeed-lpc-snoop.c
>> index 593905565b74..0647cff6280a 100644
>> --- a/drivers/misc/aspeed-lpc-snoop.c
>> +++ b/drivers/misc/aspeed-lpc-snoop.c
>> @@ -241,6 +241,7 @@ static int aspeed_lpc_snoop_remove(struct platform_device *pdev)
>> static const struct of_device_id aspeed_lpc_snoop_match[] = {
>> { .compatible = "aspeed,ast2500-lpc-snoop" },
>> + { .compatible = "aspeed,ast2400-lpc-snoop" },
>> { },

An approach would be to ditch this change and instead refer to the
ast2500-lpc-snoop in my device-tree to avoid anyone non-x86 from
running this configuration and hitting issues.

>> };
>> --