于 2012年05月15日 15:55, Artem Bityutskiy 写道:
I am CCing few other guys who take care of several drivers which use
similar way of busy-waiting - probably you could change it?
Bastian: drivers/mtd/nand/sh_flctl.c
Lars-Peter: drivers/mtd/nand/jz4740_nand.c
Huang: drivers/mtd/nand/gpmi-nand/gpmi-lib.c
Lei Wen: drivers/mtd/nand/pxa3xx_nand.c
On Sat, 2012-05-12 at 15:29 +0200, Roland Stigge wrote:
+ /*
+ * The DMA is finished, but the NAND controller may still have
+ * buffered data. Wait until all the data is sent.
When all the data is sent, is there an interrupt for this?
Best Regards
Huang Shijie
+ */
+ timeout = LPC32XX_DMA_SIMPLE_TIMEOUT;
+ while ((readl(SLC_STAT(host->io_base))& SLCSTAT_DMA_FIFO)
+&& (timeout> 0))
+ timeout--;
+ if (!timeout) {
+ dev_err(mtd->dev.parent, "FIFO held data too long\n");
+ status = -EIO;
+ }
I know the MTD tree is full of this, but this is bad, I think. The
timeout should be time-backed, not CPU-cycles-backed.
I do not know the best way to do this, hopefully someone in the arm list
could suggest, but the following pattern is at least better:
/* Chip reaction time timeout in milliseconds */
#define LPC32XX_DMA_TIMEOUT 100
timeout = loops_per_jiffy * msecs_to_jiffies(LPC32XX_DMA_TIMEOUT);
while ((readl(...))&& timeout--> 0)
cpu_relax();
if (!timeout)
error;
So basically I turned your hard-coded iterations count into a time-based
timeout. I also used cpu_relax() which is commonly used in tight-loops
like this. Here is a piece of documentation about cpu_relax():
"
The right way to perform a busy wait is:
while (my_variable != what_i_want)
cpu_relax();
The cpu_relax() call can lower CPU power consumption or yield to a
hyperthreaded twin processor; it also happens to serve as a compiler
barrier, so, once again, volatile is unnecessary. Of course, busy-
waiting is generally an anti-social act to begin with.
"
_______________________________________________
devicetree-discuss mailing list
devicetree-discuss@lists.ozlabs.org
https://lists.ozlabs.org/listinfo/devicetree-discuss
