Re: [PATCH 5/9] usb: xhci: move ring initialization

[Michal Pecio <michal.pecio@gmail.com>]

On Mon, 30 Mar 2026 11:53:31 +0300, Neronin, Niklas wrote:
> xhci_clear_command_ring() clears the command ring specifically.
> 
> The idea is to have (naming not set in stone):
> ring_alloc() <- allocate necessary ring memory
> ring_free()  <- free all ring memory
> ring_init()  <- initialize necessary ring memory
> ring_reset() <- resets the ring

Not sure what reset() means, simply clear()? Would it include init()?

What I'm saying is that alloc() and init() almost always need to happen
together and the verbosity of this patch demonstrates widespread demand
for one function which can do both things.

Is this patch even correct? To know, I would need to find all callers
of ring_alloc() and verify that ring_init() has been added. OTOH, a
rename of ring_alloc() to ring_create() wouldn't compile if the patch
frogot to update some callers.

Similarly, separation of setting up links (ring_init) from clearing the
rest adds complexity and bug opportunities. Usually (always?) callers
will need to do both of those things for correct outcome.

I think it would be safer not to have rings in an invalid state, unless
it's some exceptional case like a very short-lived ring allocated for
expansion. And  even then I'm not sure why expansion allocates a new
ring and splices it into the old one, instead of just adding segments.

Regards,
Michal