| Message ID | 1383060519-14868-1-git-send-email-u.kleine-koenig@pengutronix.de |
|---|---|
| State | New |
| Headers | show |
diff --git a/arch/arm/kernel/debug.S b/arch/arm/kernel/debug.S index e18cb30..17156ee 100644 --- a/arch/arm/kernel/debug.S +++ b/arch/arm/kernel/debug.S @@ -83,7 +83,6 @@ ENTRY(printascii) b 2f 1: waituart r2, r3 senduart r1, r3 - busyuart r2, r3 teq r1, #'\n' moveq r1, #'\r' beq 1b @@ -91,6 +90,7 @@ ENTRY(printascii) ldrneb r1, [r0], #1 teqne r1, #0 bne 1b + busyuart r2, r3 mov pc, lr ENDPROC(printascii)
Instead of waiting after every single character that the fifo and shifter become empty only wait after the whole string. Without this optimization the fifo isn't used at all for more than a single character. The added benefit is that debug output speed is improved. Signed-off-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de> --- Hello, I didn't do any benchmarks with this patch but noticed this optimization when I copied the printch/printascii code for the bootloader on efm32. Obviously it depends on waituart to reliably signal if there is space available in the fifo while without this patch debug output works fine if only busyuart is functional. So this breaks at least arch/arm/include/debug/imx.S and arch/arm/include/debug/8250.S . Looking at the 8250 code I wonder if I got the semantics of waituart right. I thought it's about waiting until there is space available in the fifo but arch/arm/include/debug/8250.S suggests it's about hand shaking? (In that case my patch would be wrong.) Best regards Uwe --- arch/arm/kernel/debug.S | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)