[meta-freescale] [PATCH v2 05/14] imx-gpu-viv: Upgrade to 6.2.2.p0
Otavio Salvador
otavio.salvador at ossystems.com.br
Mon Nov 20 03:26:05 PST 2017
Hello Prabhu,
On Fri, Nov 17, 2017 at 8:37 PM, Prabhu Sundararaj
<prabhu.sundararaj at nxp.com> wrote:
> Hi Otavio, Gary,
> We tested only on SABRE boards and it got only 2GB MAX DDR.
Ok, this is a valid point but this does not mean it covers all
possible use-cases.
>> Using 4.1.15 of_reserved_mem.c works too but that is just a
>> workaround, the viv driver should manage that case since Etnaviv does.
> Wondering whether this restriction came in 4.9?
>
> This is not regression from in my point of view. May be we consider this a requirement and fix it in separate patch.
Sorry but it is a regression. The previous GA release works with 4GiB
of RAM. The new one does not.
> Fixing this and validation would require some time.
Gary did figure out the root cause. He found two important information pieces:
- A similar issue was seen and fixed on Etnaviv[1]. I guess a
similar patch is required for this driver as well. Forcing the cma
area to be under 0x90000000 (for imx6qdl) makes the driver work.
[1] https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-stable.git/commit/?id=471070ab
- A new patch[2], that is included in 4.9, changes the behavior of it
[2] https://github.com/boundarydevices/linux-imx6/commit/e53b50c0cbe392c946807abf7d07615a3c588642
Another important aspect is that the 6.2.3.p0[3] available for i.MX8
adds a new gc_hal_kernel_allocator_reserved_mem.c to address this very
issue.
[3] http://git.freescale.com/git/cgit.cgi/imx/linux-imx.git/commit/drivers/mxc/gpu-viv?id=fe649d5ca8e6088d54eee89825705257f404576d
So I think it all provides a good information about the issue. It
seems that backporting the gc_hal_kernel_allocator_reserved_mem.c is
the best option as it allows for no diverting of Linux kernel official
behavior.
--
Otavio Salvador O.S. Systems
http://www.ossystems.com.br http://code.ossystems.com.br
Mobile: +55 (53) 9981-7854 Mobile: +1 (347) 903-9750
More information about the meta-freescale
mailing list