[patch 02/13] jffs2 summary allocation: don't use vmalloc()

[David Brownell]

On Wednesday 30 July 2008, Andrew Morton wrote:
> On Wed, 30 Jul 2008 15:39:24 -0700
> David Brownell <david-b at pacbell.net> wrote:
> 
> > I'm still asking whether MTD folk have any plans to make that stack DMA-safe...
> > more than just the SPI flash drivers (mtd_dataflash, m25p80) could benefit
> > from DMA support, so I'd hope it's at least being considered.
> > 
> > If the answer is "no" then (a) the MTD interface specs need to finally say
> > they pass DMA-unsafe addresses, and (b) those SPI flash drivers are going
> > to need updates.
> 
> Well yes.  It's been four months since this bug (it goes oops!) was
> reported and afaik there's been no discussion or consideration or
> anything else.

Well, the appended is at least starting to address (a) ... under the
assumption the MTD folk aren't making the stack more DMA-friendly.
Presumably the same thing should be true of panic_write(), the OOB
operations (for NAND), and the OTP operations.

I actually think it makes the most technical sense to make the MTD
stack talk to its low level drivers the way most code seems to:
only pass DMA-safe buffers.


> I don't know how much of a problem this bug is in the real world, but
> it's taking an awful long time to get fixed?

I'm not keen on spending time on a "fix" that both worsens performance
and is also wrong ... which is why I've asked the broader question about
the MTD stack and DMA.  It's taking a long time to get an answer to what
should be a simple question...

That, and the fact that the board I have two types of SPI flash hooked up
to currently has a misbehaving SPI driver, which I think someone (else)
is going to be fixing soon.  So testing such changes is impractical, even
if I assume the MTD stack is unlkely to change...

Not all the SPI controllers use DMA, so only some boards suffer from
this problem in any case.

- Dave


---
 include/linux/mtd/mtd.h |   10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)

--- a/include/linux/mtd/mtd.h	2008-07-30 20:46:02.000000000 -0700
+++ b/include/linux/mtd/mtd.h	2008-07-30 21:20:37.000000000 -0700
@@ -149,8 +149,14 @@ struct mtd_info {
 	void (*unpoint) (struct mtd_info *mtd, loff_t from, size_t len);
 
 
-	int (*read) (struct mtd_info *mtd, loff_t from, size_t len, size_t *retlen, u_char *buf);
-	int (*write) (struct mtd_info *mtd, loff_t to, size_t len, size_t *retlen, const u_char *buf);
+	/*
+	 * Read/write operations are given *VIRTUAL* addresses, which
+	 * must not be passed as-is to dma mapping operations.
+	 */
+	int (*read) (struct mtd_info *mtd, loff_t from,
+			size_t len, size_t *retlen, u_char *virt);
+	int (*write) (struct mtd_info *mtd, loff_t to,
+			size_t len, size_t *retlen, const u_char *virt);
 
 	/* In blackbox flight recorder like scenarios we want to make successful
 	   writes in interrupt context. panic_write() is only intended to be