On Tue, Jul 12, 2011 at 5:56 PM, Igor Grinberg [off-list ref] wrote:

On 07/12/11 15:02, Daniel Mack wrote:

quoted

On Tue, Jul 12, 2011 at 1:39 PM, Daniel Mack [off-list ref] wrote:

quoted

On Tue, Jul 12, 2011 at 12:28 PM, Lei Wen [off-list ref] wrote:

quoted

This patch add protection on the suspend&resume path to prevent
some unexpected behavior, like interrupt occur at the very second
of resume back and it don't follow normal command path, which lead
to bug.

Signed-off-by: Lei Wen <redacted>
---
?drivers/mtd/nand/pxa3xx_nand.c | ? 28 ++++++++++++++++++++++++++++
?1 files changed, 28 insertions(+), 0 deletions(-)

[...]

quoted

@@ -1267,6 +1283,18 @@ static int pxa3xx_nand_resume(struct platform_device *pdev)

? ? ? ?info->cs = 0xff;
? ? ? ?clk_enable(info->clk);

+ ? ? ? /*
+ ? ? ? ?* As the spec, the NDSR would be updated to 0x1800 when
+ ? ? ? ?* do the nand_clk disable/enable.
+ ? ? ? ?* To prevent it damage state machine of the driver, clear
+ ? ? ? ?* all status before resume
+ ? ? ? ?*/
+ ? ? ? nand_writel(nand, NDSR, NDSR_MASK);

This doesn't build:

?CC ? ? ?drivers/mtd/nand/pxa3xx_nand.o
drivers/mtd/nand/pxa3xx_nand.c: In function 'pxa3xx_nand_resume':
drivers/mtd/nand/pxa3xx_nand.c:1292: error: 'nand' undeclared (first
use in this function)
drivers/mtd/nand/pxa3xx_nand.c:1292: error: (Each undeclared
identifier is reported only once
drivers/mtd/nand/pxa3xx_nand.c:1292: error: for each function it appears in.)
drivers/mtd/nand/pxa3xx_nand.c:1294: error: 'mtd' undeclared (first
use in this function)
make[3]: *** [drivers/mtd/nand/pxa3xx_nand.o] Error 1

I guess this was not even compile tested? Anyway, I did a trivial
fix-up and will test.

Also, with this (fixed) patch applied, the system doesn't resume at
all. No messages, it simply doesn't come back.

I was skeptic about the clock being disabled in Lei's patch,
as I observed system hangs if that clock was disabled back then in 2.6.31,
but wanted to give it a try, because things has changed since then.

Now I see, that Lei already sent v7 without clock toggling...

Yes, we debugged this quickly via Jabber, and without the clock
disable, things work fine for me again.

Daniel