serial: 8250: Fix ...console_fifo_write on BCM283x
PATCH 1/1 serial: 8250: Fix ...console_fifo_write on BCM283x
The mini-UART on BCM283x is doubly crippled - it has 8-byte FIFOs and the THRE bit indicates that the TX FIFO is not-full rather than empty.
The optimisation to enable the use of the FIFO assumes that it is safe to write fifosize bytes whenever THRE is set, but the BCM283x quirk (indicated by the presence of UART_CAP_MINI) makes it necessary to check the FIFO state after each byte.
See: https://github.com/raspberrypi/linux/issues/4849
--- drivers/tty/serial/8250/8250_port.c | 7 +++++++ 1 file changed, 7 insertions(+)
drivers/tty/serial/8250/8250_port.c +7 -0
@@ -3357,6 +3357,13 @@ static void serial8250_console_fifo_write(struct uart_8250_port *up,serial_out (up,UART_TX , *s++); cr_sent = false; } +/* +* The BCM2835 MINI UART THRE bit is really a not-full +* bit, so be prepared to bail out early. +*/ +if ((up->capabilities &UART_CAP_MINI ) && + !(serial_in (up,UART_LSR ) &UART_LSR_THRE )) +break ; } } }
Signature · 1 line
2.25.1
Greg Kroah-Hartman
maintainerQuoted 11 lines · Phil Elwell
The mini-UART on BCM283x is doubly crippled - it has 8-byte FIFOs and the THRE bit indicates that the TX FIFO is not-full rather than empty.
The optimisation to enable the use of the FIFO assumes that it is safe to write fifosize bytes whenever THRE is set, but the BCM283x quirk (indicated by the presence of
UART_CAP_MINI) makes it necessary to check the FIFO state after each byte.See: https://github.com/raspberrypi/linux/issues/4849
Fixes 5021d709b31b ("tty: serial: Use fifo in 8250 console driver")
I have just reverted this commit, it's broken for many platforms now, not just yours.
So there shouldn't be a need for this change now, right?
thanks,
greg k-h
Hi Greg,
Quoted 9 lines · Greg Kroah-Hartman
Quoted 8 lines · Phil Elwell
The mini-UART on BCM283x is doubly crippled - it has 8-byte FIFOs and the THRE bit indicates that the TX FIFO is not-full rather than empty.
The optimisation to enable the use of the FIFO assumes that it is safe to write fifosize bytes whenever THRE is set, but the BCM283x quirk (indicated by the presence of
UART_CAP_MINI) makes it necessary to check the FIFO state after each byte.
Quoted 8 lines
Quoted 3 lines
See: https://github.com/raspberrypi/linux/issues/4849
Fixes 5021d709b31b ("tty: serial: Use fifo in 8250 console driver")I have just reverted this commit, it's broken for many platforms now, not just yours.
So there shouldn't be a need for this change now, right?
This patch is sufficient on the BCM283x family of devices, but a Fix for a non-existent patch is clearly superfluous. The author of the original patch can consider this as a hint about the kind of issues that need to be considered before resubmitting.
Thanks,
Phil