spi: omap2-mcspi: Improve performance waiting for CHSTAT

This reverts commit 13d515c796 (spi: omap2-mcspi: Switch to
readl_poll_timeout()).

The amount of time spent polling for the MCSPI_CHSTAT bits to be set on
AM335x-icev2 platform is less than 1us (about 0.6us) in most cases, with
or without using DMA. So, in most cases the function need not sleep.
Also, setting the sleep_usecs to zero would not be optimal here because
ktime_add_us() used in readl_poll_timeout() is slower compared to the
direct addition used after the revert. So, it is sub-optimal to use
readl_poll_timeout in this case.

When DMA is not enabled, this revert results in an increase of about 27%
in throughput and decrease of about 20% in CPU usage. However, the CPU
usage and throughput are almost the same when used with DMA.

Therefore, fix this by reverting the commit which switched to using
readl_poll_timeout().

Fixes: 13d515c796 ("spi: omap2-mcspi: Switch to readl_poll_timeout()")
Signed-off-by: Aswath Govindraju <a-govindraju@ti.com>
Link: https://lore.kernel.org/r/20200910122624.8769-1-a-govindraju@ti.com
Signed-off-by: Mark Brown <broonie@kernel.org>
This commit is contained in:
Aswath Govindraju 2020-09-10 17:56:24 +05:30 committed by Mark Brown
parent 4c3a14fbc0
commit 7b1d968133
No known key found for this signature in database
GPG Key ID: 24D68B725D5487D0
1 changed files with 12 additions and 3 deletions

View File

@ -24,7 +24,6 @@
#include <linux/of.h>
#include <linux/of_device.h>
#include <linux/gcd.h>
#include <linux/iopoll.h>
#include <linux/spi/spi.h>
@ -348,9 +347,19 @@ disable_fifo:
static int mcspi_wait_for_reg_bit(void __iomem *reg, unsigned long bit)
{
u32 val;
unsigned long timeout;
return readl_poll_timeout(reg, val, val & bit, 1, MSEC_PER_SEC);
timeout = jiffies + msecs_to_jiffies(1000);
while (!(readl_relaxed(reg) & bit)) {
if (time_after(jiffies, timeout)) {
if (!(readl_relaxed(reg) & bit))
return -ETIMEDOUT;
else
return 0;
}
cpu_relax();
}
return 0;
}
static int mcspi_wait_for_completion(struct omap2_mcspi *mcspi,