[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-Id: <be4faaed-be22-42e5-bcc2-6a992b4f9b58@app.fastmail.com>
Date: Fri, 04 Oct 2024 16:57:01 +0000
From: "Arnd Bergmann" <arnd@...nel.org>
To: "Maciej W. Rozycki" <macro@...am.me.uk>
Cc: "Niklas Schnelle" <schnelle@...ux.ibm.com>,
"Greg Kroah-Hartman" <gregkh@...uxfoundation.org>,
"Jiri Slaby" <jirislaby@...nel.org>,
Ilpo Järvinen <ilpo.jarvinen@...ux.intel.com>,
linux-serial@...r.kernel.org, "Heiko Carstens" <hca@...ux.ibm.com>,
linux-kernel@...r.kernel.org
Subject: Re: [PATCH 1/1] tty: serial: handle HAS_IOPORT dependencies
On Fri, Oct 4, 2024, at 16:24, Maciej W. Rozycki wrote:
> On Fri, 4 Oct 2024, Arnd Bergmann wrote:
>> There are also a lot of Arm systems that have no I/O space support at
>> all, such as the Apple M2 I'm using at the moment.
>
> Thanks for letting me know. Is it AArch64 only that has no port I/O
> support in the PCIe root complex nowadays, or is it 32-bit ARM as well?
I'm fairly sure we have some on 32-bit as well, it's certainly
up to the specific SoC rather than the architecture.
I think I've seen three different cases:
- Apple leaves out every feature from hardware that they don't
have to do for compliance, this also includes CPU stuff like
32-bit mode, big-endian or cacheable PCI mappings. I think IBMs
ppc64 chips are in a similar situation
- Some devices can probably do I/O space in hardware, but this is
left out from the driver or the DT because it has not been
validated to work
- Some devices have a limited number of mapping windows that
are shared between prefetchable-memory, non-prefetchable-memory,
config and io space. Leaving out I/O space completely is
easier than runtime remapping of the windows
Arnd
Powered by blists - more mailing lists