Re: [PATCH 11/11] ipmi: Add NVIDIA LSTP IPMI BMC driver

[Adrian Ambrozewicz <aambrozewicz@nvidia.com>]

Hi Corey,

On Fri, Oct 16, 2026 at 07:26:53AM -0500, Corey Minyard wrote:
> Going a little higher level, I'm wondering if there is any advantage to
> having this in the kernel.  Since it's USB you could do this all in
> userland.  There's no kernel subsystem that this would tie in to.

It could end up in userspace, but there are reasons against it.

Existing BMC userspace is written against a BMC-side IPMI character
device. With CONFIG_USB_LSTP_IPMI_MSG_NUM disabled, read() and write()
use the ssif_bmc layout; in the next version I plan to make that the
default.

The IPMI channel shares one USB interface with the GPIO, I2C and other
channels. Those have kernel drivers, like other bridges, so userspace
stays the same when a board is redesigned. An interface is
bound to a kernel driver or to usbfs, not both, so the IPMI channel
cannot move to userspace alone. Even if it could, serializing access
and protocol state between kernel and userspace drivers would be
inefficient and hard.

Thanks,
Adrian