Re: earlycon issues in -next with amba-pl011 updates
From: Peter Hurley <hidden>
Date: 2015-08-17 16:04:16
Also in:
linux-arm-kernel
On 08/11/2015 08:40 AM, Russell King - ARM Linux wrote:
On Mon, Aug 10, 2015 at 08:31:03PM -0400, Peter Hurley wrote:quoted
diff --git a/drivers/tty/serial/amba-pl011.c b/drivers/tty/serial/amba-pl011.c index 2af09ab..fd54991 100644 --- a/drivers/tty/serial/amba-pl011.c +++ b/drivers/tty/serial/amba-pl011.c@@ -2348,13 +2348,10 @@ static struct console amba_console = { static void pl011_putc(struct uart_port *port, int c) { - struct uart_amba_port *uap = - container_of(port, struct uart_amba_port, port); - - while (pl011_readw(uap, REG_FR) & UART01x_FR_TXFF) + while (readl(port->membase + UART01x_FR) & UART01x_FR_TXFF) ; - pl011_writeb(uap, c, REG_DR); - while (pl011_readw(uap, REG_FR) & uap->fr_busy) + writeb(c, port->membase + UART01x_DR); + while (readl(port->membase + UART01x_FR) & UART01x_FR_BUSY) ;These loops are wrong. Kernel coding style is that loops like this will use cpu_relax() in them, just like the driver used to. This was introduced by: commit 0d3c673e7881e691991b2a4745bd4f149603baa2 Author: Rob Herring [off-list ref] Date: Fri Apr 18 17:19:57 2014 -0500 tty/serial: pl011: add generic earlycon support Also, the above readl()s should be readl_relaxed(), we don't need barriered reads or to hit the L2 cache on those reads in the above code.
This patch just reverts pl011 earlycon back to the existing mainline state. The amba-pl011 driver has many places where cpu_relax() is not used, so that seems like a good patch for -next during the 4.3-rc cycle. Similarly for read_relaxed() use in earlycon. Regards, Peter Hurley