ioremap on system memory

ioremap on system memory

[Marcus Folkesson]

Hi,

I'm working with a custom based platform based on OMAP-L138.
A driver we are using is ioremapping system memory. It does this to
share data between the GPP and DSP core.
This is not allowed in later kernel versions since ARMv6+ does not support this.

Is there any good(and easy) way to get around this or is it so
fundamentally wrong that the driver should be written from scratch?

Thanks in advance

Med vänliga hälsningar / Best regards
Marcus Folkesson

Re: ioremap on system memory

[Ryan Mallon]

On 10/11/11 19:06, Marcus Folkesson wrote:

> Hi,
> 
> I'm working with a custom based platform based on OMAP-L138.
> A driver we are using is ioremapping system memory. It does this to
> share data between the GPP and DSP core.
> This is not allowed in later kernel versions since ARMv6+ does not support this.
> 
> Is there any good(and easy) way to get around this or is it so
> fundamentally wrong that the driver should be written from scratch?


Cc'ed the ARM kernel list and Marek Szyprowski (CMA developer). The CMA
patches I think do what you want, but I am not sure if they are merged
yet. The other solution (again, I'm not an expert on this area) is to
not pass the region of memory you want to manage as part of Linux's
normal system memory. See how the OMAP framebuffer driver reserves a
region of SDRAM (arch/arm/plat-omap/fb.c).

Some one else on the ARM list can probably explain in more detail.

~Ryan

ioremap on system memory

[Ryan Mallon]

On 10/11/11 19:06, Marcus Folkesson wrote:

> Hi,
> 
> I'm working with a custom based platform based on OMAP-L138.
> A driver we are using is ioremapping system memory. It does this to
> share data between the GPP and DSP core.
> This is not allowed in later kernel versions since ARMv6+ does not support this.
> 
> Is there any good(and easy) way to get around this or is it so
> fundamentally wrong that the driver should be written from scratch?


Cc'ed the ARM kernel list and Marek Szyprowski (CMA developer). The CMA
patches I think do what you want, but I am not sure if they are merged
yet. The other solution (again, I'm not an expert on this area) is to
not pass the region of memory you want to manage as part of Linux's
normal system memory. See how the OMAP framebuffer driver reserves a
region of SDRAM (arch/arm/plat-omap/fb.c).

Some one else on the ARM list can probably explain in more detail.

~Ryan

Re: ioremap on system memory

[Marcus Folkesson]

2011/11/10 Ryan Mallon <rmallon@gmail.com>
>
> On 10/11/11 19:06, Marcus Folkesson wrote:
>
> > Hi,
> >
> > I'm working with a custom based platform based on OMAP-L138.
> > A driver we are using is ioremapping system memory. It does this to
> > share data between the GPP and DSP core.
> > This is not allowed in later kernel versions since ARMv6+ does not support this.
> >
> > Is there any good(and easy) way to get around this or is it so
> > fundamentally wrong that the driver should be written from scratch?
>
>
> Cc'ed the ARM kernel list and Marek Szyprowski (CMA developer). The CMA
> patches I think do what you want, but I am not sure if they are merged
> yet. The other solution (again, I'm not an expert on this area) is to
> not pass the region of memory you want to manage as part of Linux's
> normal system memory. See how the OMAP framebuffer driver reserves a
> region of SDRAM (arch/arm/plat-omap/fb.c).
>
> Some one else on the ARM list can probably explain in more detail.
>
> ~Ryan
>

Not passing the region of the concerned memory seems to solve the problem.
Thank you!

Med vänliga hälsningar / Best regards
Marcus Folkesson

ioremap on system memory

[Marcus Folkesson]

2011/11/10 Ryan Mallon <rmallon@gmail.com>
>
> On 10/11/11 19:06, Marcus Folkesson wrote:
>
> > Hi,
> >
> > I'm working with a custom based platform based on OMAP-L138.
> > A driver we are using is ioremapping system memory. It does this to
> > share data between the GPP and DSP core.
> > This is not allowed in later kernel versions since ARMv6+ does not support this.
> >
> > Is there any good(and easy) way to get around this or is it so
> > fundamentally wrong that the driver should be written from scratch?
>
>
> Cc'ed the ARM kernel list and Marek Szyprowski (CMA developer). The CMA
> patches I think do what you want, but I am not sure if they are merged
> yet. The other solution (again, I'm not an expert on this area) is to
> not pass the region of memory you want to manage as part of Linux's
> normal system memory. See how the OMAP framebuffer driver reserves a
> region of SDRAM (arch/arm/plat-omap/fb.c).
>
> Some one else on the ARM list can probably explain in more detail.
>
> ~Ryan
>

Not passing the region of the concerned memory seems to solve the problem.
Thank you!

Med v?nliga h?lsningar / Best regards
Marcus Folkesson