mflash's polling driver operate in standard request_fn_proc's context,
sleep in this isn't permitted.
Signed-off-by: unsik Kim <[email protected]>
---
drivers/block/mg_disk.c | 2 --
1 files changed, 0 insertions(+), 2 deletions(-)
diff --git a/drivers/block/mg_disk.c b/drivers/block/mg_disk.c
index 64f90f5..26371d5 100644
--- a/drivers/block/mg_disk.c
+++ b/drivers/block/mg_disk.c
@@ -244,8 +244,6 @@ static unsigned int mg_wait(struct mg_host *host, u32 expect, u32 msec)
mg_dump_status("not ready", status, host);
return MG_ERR_INV_STAT;
}
- if (prv_data->use_polling)
- msleep(1);
status = inb((unsigned long)host->dev_base + MG_REG_STATUS);
} while (time_before(cur_jiffies, expire));
--
1.6.0.6
When using polling driver, little delay is required to access
status register. Without this, host might read invalid status.
Signed-off-by: unsik Kim <[email protected]>
---
drivers/block/mg_disk.c | 10 ++++++++++
1 files changed, 10 insertions(+), 0 deletions(-)
diff --git a/drivers/block/mg_disk.c b/drivers/block/mg_disk.c
index 26371d5..6d7fbaa 100644
--- a/drivers/block/mg_disk.c
+++ b/drivers/block/mg_disk.c
@@ -218,6 +218,16 @@ static unsigned int mg_wait(struct mg_host *host, u32 expect, u32 msec)
host->error = MG_ERR_NONE;
expire = jiffies + msecs_to_jiffies(msec);
+ /* These 2 times dummy status read prevents reading invalid
+ * status. A very little time (3 times of mflash operating clk)
+ * is required for busy bit is set. Use dummy read instead of
+ * busy wait, because mflash's PLL is machine dependent.
+ */
+ if (prv_data->use_polling) {
+ status = inb((unsigned long)host->dev_base + MG_REG_STATUS);
+ status = inb((unsigned long)host->dev_base + MG_REG_STATUS);
+ }
+
status = inb((unsigned long)host->dev_base + MG_REG_STATUS);
do {
--
1.6.0.6
On Thu, 23 Jul 2009 13:57:45 +0900
unsik Kim <[email protected]> wrote:
> mflash's polling driver operate in standard request_fn_proc's context,
> sleep in this isn't permitted.
> Signed-off-by: unsik Kim <[email protected]>
> ---
> drivers/block/mg_disk.c | 2 --
> 1 files changed, 0 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/block/mg_disk.c b/drivers/block/mg_disk.c
> index 64f90f5..26371d5 100644
> --- a/drivers/block/mg_disk.c
> +++ b/drivers/block/mg_disk.c
> @@ -244,8 +244,6 @@ static unsigned int mg_wait(struct mg_host *host, u32 expect, u32 msec)
> mg_dump_status("not ready", status, host);
> return MG_ERR_INV_STAT;
> }
> - if (prv_data->use_polling)
> - msleep(1);
>
> status = inb((unsigned long)host->dev_base + MG_REG_STATUS);
> } while (time_before(cur_jiffies, expire));
This change has the potential to cause the driver to use much much more
CPU time and power. Are you really sure that the driver is optimal in
this area?
On possibility is to pass a flag into mg_wait() telling it whether it
is permitted to sleep:
if (may_sleep && prv_data->use_polling)
msleep(1);
else
mdelay(1);
or
if (may_sleep && prv_data->use_polling)
msleep(1);
else
/* nothing */;
so that then we only consume extra power if the caller is the
request_fn handler.