[U-Boot] [PATCH 1/4] sunxi: nand: Add basic sunxi NAND driver with DMA support

[Scott Wood]

On Fri, 2015-07-17 at 00:06 +0200, Marek Vasut wrote:
> On Thursday, July 16, 2015 at 11:36:08 PM, Scott Wood wrote:
> > On Thu, 2015-07-16 at 23:26 +0200, Marek Vasut wrote:
> > > On Thursday, July 16, 2015 at 11:15:58 PM, Scott Wood wrote:
> > > 
> > > [...]
> > > 
> > > > > +/* temporary buffer in internal ram */
> > > > > +#ifdef CONFIG_SPL_BUILD
> > > > > +/* in SPL temporary buffer cannot be @ 0x0 */
> > > > > +unsigned char temp_buf[SPL_WRITE_SIZE] __aligned(0x10)
> > > > > __section(".text#"); +#else
> > > > > +/* put temporary buffer @ 0x0 */
> > > > > +unsigned char *temp_buf = (unsigned char *)0x0;
> > > > > +#endif
> > > > 
> > > > If 0x0 is the address of an SRAM, its address should be symbolically
> > > > defined. Also consider mapping it to a different virtual address, to
> > > > avoid potential compiler mischief.
> > > 
> > > The DMA I believe accesses it via PA anyway, so mapping it elsewhere
> > > would only confuse everyone who's hacking on the driver. Just my 5 cents
> > > ;-)
> > 
> > Hey, if it wakes people up to the fact that they ought to be using
> > virt_to_phys() (or better yet, something that distinguishes DMA addresses
> > from physical), great! :-)
> 
> Not really useful here, but whatever.
> 
> > And yes, people go on to do the same "everything is u32" crap in
> > non-platform- specific files.  Just look at drivers/block/ahci.c.
> 
> Errr, this file is platform specific.

(I'm assuming that by "this file" you mean the sunxi file, not the ahci file.)

The point is that establishing good habits everywhere reduces the likelihood 
of the bad habits migrating to places where things break.

And no, I don't seriously expect my suggestion of remapping the virtual 
address to be implemented, but GCC has been known to do all sorts of weird 
stuff when it thinks it knows a NULL pointer is being dereferenced.

-Scott