[U-Boot] Using spi_alloc_slave() from SPL

[Marek Vasut]

On Thursday, August 06, 2015 at 07:03:24 PM, Stefano Babic wrote:
> Hi Marek,
> 
> On 06/08/2015 16:28, Marek Vasut wrote:
> > On Thursday, August 06, 2015 at 04:14:34 PM, Fabio Estevam wrote:
> >> Hi Stefano and Marek,
> >> 
> >> Thanks for the suggestions.
> >> 
> >> On Thu, Aug 6, 2015 at 10:38 AM, Stefano Babic <sbabic at denx.de> wrote:
> >>> There is the possibility to set a malloc area inside SPL:
> >>> 
> >>> CONFIG_SYS_SPL_MALLOC_START
> >>> CONFIG_SYS_SPL_MALLOC_SIZE
> >>> you do not need a lot of space, and you can try to put it inside the
> >>> IRAM.
> >>> 
> >>> This should guarantee that spi_alloc_slave() works.
> >> 
> >> So I tried moving them to the internal RAM:
> >> 
> >> --- a/include/configs/imx6_spl.h
> >> +++ b/include/configs/imx6_spl.h
> >> @@ -70,8 +70,8 @@
> >> 
> >>  #else
> >>  #define CONFIG_SPL_BSS_START_ADDR      0x18200000
> >>  #define CONFIG_SPL_BSS_MAX_SIZE                0x100000        /* 1 MB
> >>  */
> >> 
> >> -#define CONFIG_SYS_SPL_MALLOC_START    0x18300000
> >> -#define CONFIG_SYS_SPL_MALLOC_SIZE     0x3200000       /* 50 MB */
> >> +#define CONFIG_SYS_SPL_MALLOC_START    0x900000
> >> +#define CONFIG_SYS_SPL_MALLOC_SIZE     0x8000
> >> 
> >>  #define CONFIG_SYS_TEXT_BASE           0x17800000
> >>  #endif
> >>  #endif
> >> 
> >> but still getting spi_alloc_slave() to fail.
> > 
> > You want to avoid this "CONFIG_SYS_SPL_MALLOC_*" stuff, as it increases
> > the SPL size by 3kiB compared to MALLOC_F . Also, MALLOC_F needs just
> > the base address of the malloc area to work (see my email).
> 
> It does not matter at the moment, because Fabio's issue is not yet
> solved. But is it not CONFIG_SYS_SPL_MALLOC the preferred way for SPL ?
> Setting the array with MALLOC_F looks like a hack. And
> CONFIG_SYS_SPL_MALLOC is *already* set for i.MX6, lool at
> include/configs/imx6_spl.h:
> 
> #define CONFIG_SYS_SPL_MALLOC_START     0x18300000
> #define CONFIG_SYS_SPL_MALLOC_SIZE      0x3200000       /* 50 MB */
> 
> This is in RAM, of course. Increasing the size by 3KiB is not IMHO for
> i.MX6 a problem, there is enough space in IRAM. But what is surprising
> is that Fabio gets a Null pointer by malloc().

The malloc is set up only at the beginning of board_init_r(), but Fabio
needs to use the SPI in board_init_f(), so of course he does not have the
malloc operational at that point.

The MALLOC_F if intended to be used before RAM is operational (ie. before
board_init_r() is invoked), so it looks like the correct tool to use here.
Setting up a full malloc() in your small OCRAM is on the other hand a really
bad idea.

Best regards,
Marek Vasut