[U-Boot-Users] AT91 with framebuffer in SDRAM
Hi,
I'm trying to cobble together a proto with the AT91SAM9261EK and a WVGA TFT screen using Ulf's 1.2.0-atmel branch. I have gotten the screen to work just fine. The problem is that I need to operate out of SDRAM, and I'm trashing the environment variables somehow. The environment variables are at PHYS_SDRAM + 0x100 and are 0x2000 long so I have tried PHYS_SDRAM + several larger numbers (0x10000, 0x20000, 0x100000) for fb_base but all result in corrupt environment variables when I try to save them. I believe that I am staying out of u-boot's operating RAM (0x23f00000+ correct?), but I could be mistaken.
The symptom is the constantly-recurring bad CRC warning and saveenv appears to have no effect. The settings seem to be working fine when I switch fb_base back to AT91C_IRAM, but obviously the display doesn't function properly. I've looked through for obvious mistakes in writing the logo and text below the logo to the fb but haven't been able to track anything down.
Also, as an aside, I think that the vl_pixclock variable is used incorrectly in the at91sam926X files. It is supposed to be (from the definition) in picoseconds but appears to only work in Hz.
Any help is appreciated.
Thanks, Zac
Hi:
I need to test Idle mode for Atmel at91rm9200 CPU.
Does anybody have "ready to go" code for that purpose which can be compiled as part of u-boot or as standalone application (running after u-boot or as a first code on the board)?
Any additional information on this subject is also appreciated.
Thanks,
Leonid.
Hi,
I'm trying to cobble together a proto with the AT91SAM9261EK and a WVGA TFT screen using Ulf's 1.2.0-atmel branch. I have gotten the screen to work just fine. The problem is that I need to operate out of SDRAM, and I'm trashing the environment variables somehow. The environment variables are at PHYS_SDRAM + 0x100 and are 0x2000 long so I have tried PHYS_SDRAM + several larger numbers (0x10000, 0x20000, 0x100000) for fb_base but all result in corrupt environment variables when I try to save them. I believe that I am staying out of u-boot's operating RAM (0x23f00000+ correct?), but I could be mistaken.
The environment variables should be stored in flash, not in SDRAM. - Even if you execute from SDRAM.
What type of flash are you using. My stuff is there to support dataflash, and anything else get little or no testing.
The symptom is the constantly-recurring bad CRC warning and saveenv appears to have no effect. The settings seem to be working fine when I switch fb_base back to AT91C_IRAM, but obviously the display doesn't function properly. I've looked through for obvious mistakes in writing the logo and text below the logo to the fb but haven't been able to track anything down.
Also, as an aside, I think that the vl_pixclock variable is used incorrectly in the at91sam926X files. It is supposed to be (from the definition) in picoseconds but appears to only work in Hz.
Any help is appreciated.
Thanks, Zac
Best Regards Ulf Samuelsson
On 10/22/07, Ulf Samuelsson ulf@atmel.com wrote:
The environment variables should be stored in flash, not in SDRAM.
- Even if you execute from SDRAM.
What type of flash are you using. My stuff is there to support dataflash, and anything else get little or no testing.
I am using dataflash for everything. Everything is pretty equivalent to your code base, just the LCD timing values and the fb_base have changed.
Maybe I should rephrase. Everything seems fine once it is in dataflash. But if I `setenv bootcmd something' and then `saveenv`, it says "saving to dataflash...". `printenv` prints back the correct value (bootcmd=something), which I understand it isn't actually reloading from dataflash. If I then reset the board that bootcmd=something is gone. The environment variables are not in protected dataflash.
In board/at91sam9261ek/at91sam9261ek.c If I change gd->fb_base = (unsigned long) PHYS_SDRAM + 0x10000; back to gd->fb_base = (unsigned long) AT91C_IRAM; the LCD goes wonky but saving to dataflash works again.
The problem is that my framebuffer needs to be 384k, which is too big for the internal ram. Looking at what I just wrote it seems that maybe it is the writes to dataflash that are breaking, and not actually parts of the SDRAM getting corrupted.
Zac
On 10/22/07, Ulf Samuelsson ulf@atmel.com wrote:
The environment variables should be stored in flash, not in SDRAM.
- Even if you execute from SDRAM.
What type of flash are you using. My stuff is there to support dataflash, and anything else get little or no testing.
I am using dataflash for everything. Everything is pretty equivalent to your code base, just the LCD timing values and the fb_base have changed.
Maybe I should rephrase. Everything seems fine once it is in dataflash. But if I `setenv bootcmd something' and then `saveenv`, it says "saving to dataflash...". `printenv` prints back the correct value (bootcmd=something), which I understand it isn't actually reloading from dataflash. If I then reset the board that bootcmd=something is gone. The environment variables are not in protected dataflash.
In board/at91sam9261ek/at91sam9261ek.c If I change gd->fb_base = (unsigned long) PHYS_SDRAM + 0x10000; back to gd->fb_base = (unsigned long) AT91C_IRAM; the LCD goes wonky but saving to dataflash works again.
Don't know if this is the problem, but you can at least try it:
The main difference I can see is that when you put the framebuffer in external SDRAM, you are affecting the bus bandwidth. When doing the LCD from internal SRAM, you have zero access.
Consider changing the speed of the SPI to much lower speed (1 Mbps). If this works, then increase the speed, until you see the problem. Reduce to something lower.
An alternative is to link the environment into the internal SRAM. Then the SPI transfer will not be affected by the LCD refresh.
Ensure that you run the SPI in mode 0, dont ask me why
The problem is that my framebuffer needs to be 384k, which is too big for the internal ram. Looking at what I just wrote it seems that maybe it is the writes to dataflash that are breaking, and not actually parts of the SDRAM getting corrupted.
Zac
Best Regards Ulf Samuelsson
On 10/23/07, Ulf Samuelsson ulf@atmel.com wrote:
On 10/22/07, Ulf Samuelsson ulf@atmel.com wrote:
The environment variables should be stored in flash, not in SDRAM.
- Even if you execute from SDRAM.
What type of flash are you using. My stuff is there to support dataflash, and anything else get little or no testing.
I am using dataflash for everything. Everything is pretty equivalent to your code base, just the LCD timing values and the fb_base have changed.
Maybe I should rephrase. Everything seems fine once it is in dataflash. But if I `setenv bootcmd something' and then `saveenv`, it says "saving to dataflash...". `printenv` prints back the correct value (bootcmd=something), which I understand it isn't actually reloading from dataflash. If I then reset the board that bootcmd=something is gone. The environment variables are not in protected dataflash.
In board/at91sam9261ek/at91sam9261ek.c If I change gd->fb_base = (unsigned long) PHYS_SDRAM + 0x10000; back to gd->fb_base = (unsigned long) AT91C_IRAM; the LCD goes wonky but saving to dataflash works again.
Don't know if this is the problem, but you can at least try it:
The main difference I can see is that when you put the framebuffer in external SDRAM, you are affecting the bus bandwidth. When doing the LCD from internal SRAM, you have zero access.
Consider changing the speed of the SPI to much lower speed (1 Mbps). If this works, then increase the speed, until you see the problem. Reduce to something lower.
This did the trick, thanks Ulf. I'm using 8 MHz as it seems stable.
An alternative is to link the environment into the internal SRAM. Then the SPI transfer will not be affected by the LCD refresh.
I couldn't take this approach because this only sort of skirts the problem. If I did something else (e.g. used u-boot to upgrade itself) it would corrupt the write and then I'm back to JTAG to reload it.
Ensure that you run the SPI in mode 0, dont ask me why
I won't :)
The problem is that my framebuffer needs to be 384k, which is too big for the internal ram. Looking at what I just wrote it seems that maybe it is the writes to dataflash that are breaking, and not actually parts of the SDRAM getting corrupted.
Zac
Best Regards Ulf Samuelsson
Thanks again, Zac
On 10/23/07, Ulf Samuelsson ulf@atmel.com wrote:
On 10/22/07, Ulf Samuelsson ulf@atmel.com wrote:
The environment variables should be stored in flash, not in SDRAM.
- Even if you execute from SDRAM.
What type of flash are you using. My stuff is there to support dataflash, and anything else get little or no testing.
I am using dataflash for everything. Everything is pretty equivalent to your code base, just the LCD timing values and the fb_base have changed.
Maybe I should rephrase. Everything seems fine once it is in dataflash. But if I `setenv bootcmd something' and then `saveenv`, it says "saving to dataflash...". `printenv` prints back the correct value (bootcmd=something), which I understand it isn't actually reloading from dataflash. If I then reset the board that bootcmd=something is gone. The environment variables are not in protected dataflash.
In board/at91sam9261ek/at91sam9261ek.c If I change gd->fb_base = (unsigned long) PHYS_SDRAM + 0x10000; back to gd->fb_base = (unsigned long) AT91C_IRAM; the LCD goes wonky but saving to dataflash works again.
Don't know if this is the problem, but you can at least try it:
The main difference I can see is that when you put the framebuffer in external SDRAM, you are affecting the bus bandwidth. When doing the LCD from internal SRAM, you have zero access.
Consider changing the speed of the SPI to much lower speed (1 Mbps). If this works, then increase the speed, until you see the problem. Reduce to something lower.
This did the trick, thanks Ulf. I'm using 8 MHz as it seems stable.
An alternative is to link the environment into the internal SRAM. Then the SPI transfer will not be affected by the LCD refresh.
I couldn't take this approach because this only sort of skirts the problem. If I did something else (e.g. used u-boot to upgrade itself) it would corrupt the write and then I'm back to JTAG to reload it.
I think that you should still do this. You may need to do a slight modification to U-Boot so that the dataflash write driver copies a chunk of data to the internal SRAM before it writes it to the dataflash.
The CPU is sitting idling while waiting for the PDC to complete the SPI transfer anyway.
Best Regards Ulf Samuelsson ulf@atmel.com Atmel Nordic AB Mail: Box 2033, 174 02 Sundbyberg, Sweden Visit: Kavallerivägen 24, 174 58 Sundbyberg, Sweden Phone +46 (8) 441 54 22 Fax +46 (8) 441 54 29 GSM +46 (706) 22 44 57
participants (3)
-
Leonid -
Ulf Samuelsson -
Zac Wheeler