Re: [PATCH 0/7] soc: aspeed: Add AST2600 eSPI controller support

["Shulzhenko, Oleksandr" <oleksandr.shulzhenko@linux.intel.com>]

On 5/12/2026 9:08 AM, YH Chung wrote:
> Hi Shulzhenko,
>
> Thanks for the follow-up.
>
>> Integrating this driver into the SPI subsystem may allow reusing some existing
>> definitions, e.g.|spi_controller|,|spi_message|, and perhaps parts related to
>> single/dual/quad I/O handling. At the same time, parts such as the Flash channel
>> (included in the current series), and OOB / Virtual Wire support (I would expect
>> to come later), appear to be specific to the Intel eSPI protocol. Modeling all of
>> that as just another SPI IP driver may introduce some awkward layering and
>> overhead.
> Agreed. eSPI introduces two additional pins, RESET# and ALERT#, beyond the
> standard SPI signals. More importantly, eSPI functionality is described
> primarily in terms of four logical channels, rather than generic low-level
> bus signaling or pure data transfers.
>
>> Also, the current series already seems to separate common eSPI logic from
>> AST2600-specific pieces, assuming that 2700 driver is also coming at some point.
>>
>> This makes me wonder whether a dedicated eSPI layer/subsystem could be a
>> better fit — either under the SPI or as something separate (but not SoC driver).
>>
>> Given my limited experience with SPI/eSPI, could you help clarify a few points for
>> me (and probably others as well)?
>>
>> * How much of the SPI subsystem can be reused for this implementation,
>> both for the current patchset and for likely future extensions?
> I believe only a limited portion of the SPI subsystem can be reused. Some
> generic framework elements, such as controller registration and basic
> scaffolding, may be useful initially. But this reuse appears to be mostly
> mechanical rather than semantic. Once eSPI-specific features like Flash
> channels, OOB messaging, and Virtual Wire semantics are involved, the SPI
> transaction model does not seem to map very naturally.
>
>> * Are there any pitfalls or abstraction mismatches in trying to reuse
>> the SPI core here?
> Our main concern is an abstraction mismatch. SPI is designed as a generic
> peripheral bus, while eSPI is more of a system-management interface with
> explicit host-BMC-specific semantics. Reusing the SPI core would likely
> require treating eSPI packets as generic bus-level transfers in the kernel.
>
> However, some eSPI transactions and protocol handling, such as LPC bridge
> accesses, are performed autonomously by the hardware rather than being fully
> driven as low-level bus operations by the driver. This makes the eSPI driver
> somewhat different from a conventional serial bus controller driver
> maintained under the SPI core.
>
Hi YH,

My main concern is trying to understand whether it is completely 
impossible (or introduces too much effort that we'd better not to take) 
integrating this to SPI subsystem.

 From your reply I understand there are two potential blockers:

a) Treating eSPI transfers as bus-level transfers (meaning that it will 
be necessary probably making separate driver for OOB/VW/Flash channels 
as they essentially use eSPI as a transport);

b) Some logic being done by the hardware (i.e. LPC bridge).

Please confirm my understanding:

(a) is feasible, but requires many effort to re-define architecture

(b) If something is done by the hardware - what is the driver impact? I 
recall eDAF use case when the driver wasn't involved at all - and flash 
access was fully done by the hardware (unless the controller is 
configured to handle it in SW mode).


P.S. I guess we can talk about host-BMC communication only when talking 
about hardware-dependent stuff (i.e. ast2600-espi files). eSPI core 
should be (it seems to be already is) at least BMC agnostic and this is 
the reason not having it under SOC/aspeed (ast2600-espi.* may stay here 
though).