[Qemu-devel] does a device model need to call qemu_chr_fe_accept_input() in its reset method?

[Qemu-devel] does a device model need to call qemu_chr_fe_accept_input() in its reset method?

[Peter Maydell]

Hi; I noticed that one of our UART models, hw/char/cadence_uart.c,
calls qemu_chr_fe_accept_input() in its reset method. This sort
of makes sense, since device reset is a state transition from
"I'm in a state where I've previously said I can't accept input"
to "I'm in a state where I can accept input". However, it looks
like none of the other UARTs do this.

Is this a bug in our other UARTs, or does the chardev framework
guarantee to forget about 'nope' responses from the fd_can_read
callback on system reset?

thanks
-- PMM

Re: [Qemu-devel] does a device model need to call qemu_chr_fe_accept_input() in its reset method?

[Paolo Bonzini]

On 03/05/2018 14:49, Peter Maydell wrote:
> Hi; I noticed that one of our UART models, hw/char/cadence_uart.c,
> calls qemu_chr_fe_accept_input() in its reset method. This sort
> of makes sense, since device reset is a state transition from
> "I'm in a state where I've previously said I can't accept input"
> to "I'm in a state where I can accept input". However, it looks
> like none of the other UARTs do this.
> 
> Is this a bug in our other UARTs, or does the chardev framework
> guarantee to forget about 'nope' responses from the fd_can_read
> callback on system reset?

No, system reset is not a particularly interesting event for most of the
backends.

So you have found a bug, and in fact it should be called on every rising
edge of can_read, so for example digic-uart.c is even more broken.  I
vaguely remember looking at it years ago...

Paolo