Re: simple-framebuffer enquire

From: Hans de Goede
Date: Tue Jun 26 2018 - 14:45:08 EST

Hi,

On 26-06-18 18:35, Michael Nazzareno Trimarchi wrote:
Hi Hans

On Tue, Jun 26, 2018 at 4:47 PM, Hans de Goede <hdegoede@xxxxxxxxxx> wrote:
Hi,


On 26-06-18 16:42, Michael Nazzareno Trimarchi wrote:

Hi Hans

On Tue, Jun 26, 2018 at 3:38 PM, Michael Nazzareno Trimarchi
<michael@xxxxxxxxxxxxxxxxxxxx> wrote:

Hi

On Tue, Jun 26, 2018 at 3:36 PM, Hans de Goede <hdegoede@xxxxxxxxxx>
wrote:

Hi,


On 26-06-18 15:29, Michael Nazzareno Trimarchi wrote:


Hi

to be more specific

On Tue, Jun 26, 2018 at 3:06 PM, Michael Nazzareno Trimarchi
<michael@xxxxxxxxxxxxxxxxxxxx> wrote:


Hi

On Tue., 26 Jun. 2018, 12:01 pm Hans de Goede, <hdegoede@xxxxxxxxxx>
wrote:



Hi,

On 25-06-18 15:29, Michael Nazzareno Trimarchi wrote:


Hi Hans

In order to let it even registered the simplefb I have added this
change. According on what I understand
from the code seems that this is the way to acquire memory with the
correct attribute

diff --git a/drivers/video/fbdev/simplefb.c
b/drivers/video/fbdev/simplefb.c
index a3c44ec..7e61ce3 100644
--- a/drivers/video/fbdev/simplefb.c
+++ b/drivers/video/fbdev/simplefb.c
@@ -466,8 +466,8 @@ static int simplefb_probe(struct platform_device
*pdev)

info->fbops = &simplefb_ops;
info->flags = FBINFO_DEFAULT | FBINFO_MISC_FIRMWARE;
- info->screen_base = ioremap_wc(info->fix.smem_start,
- info->fix.smem_len);
+ info->screen_base = arch_memremap_wb(info->fix.smem_start,
+ info->fix.smem_len);



I'm not sure why you need this? wb certainly is not optimal
for a framebuffer, the existing wc mapping is really what you
want.


Well in this way raise a WARN and get a nice NULL on memory remap on
imx6ull
SoC


[ 0.397484] WARNING: CPU: 0 PID: 1 at arch/arm/mm/ioremap.c:303
__arm_ioremap_pfn_caller+0x80/0x1cc




This is causes by a mismatch in memory attributes, which means the
memory is already mapped by the kernel as regular RAM and may
already be used for other purposes by the kernel!

Memory used by a simplefb framebuffer must be reserved by the
bootloader, so that it does not get used by the kernel as regular
RAM. See e.g.:


http://git.denx.de/?p=u-boot.git;a=blob;f=drivers/video/sunxi/sunxi_display.c

Near the end of the file where the framebuffer RAM gets excluded from
the memory-range reported to the kernel as usable RAM. Note this relies
on the u-boot sunxi video code putting the framebuffer at the end of the
RAM.


+ aliases {
+ display0 = &lcdif;
+ };
+
+ reserved-memory {
+ #address-cells = <1>;
+ #size-cells = <1>;
+ ranges;
+
+ display_reserved: framebuffer@86fd6080 {
+ reg = <0x86fd6080 (480 * 272 *4)>;
+ };
+

This should do the trick but I have still the same problem on memory
type. Any idea?


For starters your start address and size are not page-size
(multiple of 4k aligned), you need to fix that.


cat memblock/reserved
0: 0x80004000..0x80007fff
1: 0x80100000..0x81e030b3
2: 0x83000000..0x83007fff
3: 0x84000000..0x85ffffff
4: 0x86fa2000..0x87021fff

+ reserved-memory {
+ #address-cells = <1>;
+ #size-cells = <1>;
+ ranges;
+
+ display_reserved: framebuffer@86fa2000 {
+ reg = <0x86fa2000 0x80000>;
+ };
+
+ };
+
+ chosen {
+ #address-cells = <1>;
+ #size-cells = <1>;
+ ranges;
+
+ stdout-path = &uart1;
+ framebuffer0: framebuffer@86fa2000 {
+ compatible = "simple-framebuffer";
+ reg = <0x86fa2000 (480 * 272 * 4)>;
+ width = <480>;
+ height = <272>;
+ stride = <(480 * 4)>;
+ format = "a8r8g8b8";
+ clocks = <&clks IMX6UL_CLK_LCDIF_PIX>,
+ <&clks IMX6UL_CLK_LCDIF_APB>,
+ <&clks IMX6UL_CLK_DUMMY>,
+ <&clks IMX6UL_CLK_GPIO3>,
+ <&clks IMX6UL_CLK_GPIO4>;
+ nshut-supply = <&reg_lcd_nshut>;
+ nreset-supply = <&reg_lcd_nreset>;
+ display = <&lcdif>;

Still have the same on ioremap.

Hmm, I guess the kernel does map the entire region its get
passed and simply makes sure to not touch the reserved mem,
where as with the changes to the passed in mem-region the
sunxi u-boot code does the memory does not get mapped by
the kernel at all ?

I think at this point this may have more become more of a
question for the ARM folks then for the fbdev list.

Regards,

Hans






Michael


After that double check in the memory map reported by the
kernel during boot that your reservation actually works.

Regards,

Hans






+ linux,cma {
+ compatible = "shared-dma-pool";
+ reusable;
+ size = <0x1000000>;
+ linux,cma-default;
+ };
+ };
+
+ chosen {
+ #address-cells = <1>;
+ #size-cells = <1>;
+ ranges;
+
+ stdout-path = &uart1;
+ framebuffer0: framebuffer@86fd6080 {
+ compatible = "simple-framebuffer";
+ reg = <0x86fd6080 (480 * 272 * 4)>;

Here I try to use the same. I will create in uboot a dynamic way to
track down. I think that
we can even add to the simple buffer a way to get hand of reserved
region automatically

Michael

+ width = <480>;
+ height = <272>;
+ stride = <(480 * 4)>;
+ format = "a8r8g8b8";
+ clocks = <&clks IMX6UL_CLK_LCDIF_PIX>,
+ <&clks IMX6UL_CLK_LCDIF_APB>,
+ <&clks IMX6UL_CLK_DUMMY>,
+ <&clks IMX6UL_CLK_GPIO3>,
+ <&clks IMX6UL_CLK_GPIO4>;
+ nshut-supply = <&reg_lcd_nshut>;
+ nreset-supply = <&reg_lcd_nreset>;
+ display = <&lcdif>;
+ status = "okay";
+ };





Thank you very much for this lesson ;). I will try to document better
after my tour ;)

Michael

Regards,

Hans




--
| Michael Nazzareno Trimarchi Amarula Solutions BV |
| COO - Founder Cruquiuskade 47 |
| +31(0)851119172 Amsterdam 1018 AM NL |
| [`as] http://www.amarulasolutions.com |