Re: [PATCH v5 08/14] serial: 8250_mxpcie: speed up TX using memory-mapped FIFO window
Andy Shevchenko <[email protected]> Mon, 3 Aug 2026 10:26:58 +0300
| Newsgroups | gmane.linux.serial,gmane.linux.kernel |
|---|---|
| Message-ID | <CAHp75VdSqLZo+XYr_V6QO-yMUJCvd5+8hKSkHDsNQWFF=KXERQ@mail.gmail.com> |
On Mon, Aug 3, 2026 at 7:46=E2=80=AFAM Jiri Slaby <[email protected]> wr= ote: > > On 02. 08. 26, 11:10, Andy Shevchenko wrote: > > On Fri, Jul 31, 2026 at 10:49=E2=80=AFAM Crescent Hsieh > > <[email protected]> wrote: > >> > >> The MUEx50 UART provides a memory-mapped TX FIFO data window along wit= h > >> a TX FIFO level counter. > >> > >> Fill the TX FIFO in bulk via the MMIO FIFO window based on available > >> FIFO space, using uart_port_tx_limited() for the common serial-core TX > >> handling. > > > > ... > > > >> +static void mxpcie8250_tx_chars(struct uart_8250_port *up) > >> +{ > >> + struct uart_port *port =3D &up->port; > >> + unsigned int offset =3D 0; > >> + unsigned char c; > >> + > >> + uart_port_tx_limited(port, c, port->fifosize - serial_in(up, M= OXA_PUART_TX_FIFO_CNT), > >> + true, > >> + serial_out(up, MOXA_PUART_TX_FIFO_MEM + o= ffset++, c), > >> + ({})); > >> +} > > > > At least three randomly chosen drivers use the empty tx_done, perhaps > > it's time to have a helper that embeds this ugly piece in it and hides > > from the driver? > > I haven't tried, can we make the helper a variadic macro? Since it's last argument it can be done with the variadic, but you can also reconsider the others from the same group (the wrappers on top of __uart_port_tx() call) to regroup argument list to make more than a single optional argument. This can mimic the case of https://git.kernel.org/pub/scm/linux/kernel/git/next/linux-next.git/tree/li= b/kstrtox.h#n21 where two optional arguments are possible. --=20 With Best Regards, Andy Shevchenko