Re: [U-Boot] [PATCH] M28: Added guarding for reserved bits in GPIO driver
Hi Marek,
Hey, this looks reasonable. Did you send similar patch to Linux too ?
No, I'm not at that yet. I'm just starting to explore the New U-Boot: It's been since version 1.1.2 that I have actually contributed something.
My current project requires a mechanism to manipulate GPIO pins, to aid the bring-up of a new board in a couple of weeks. Until last weeks I've been using the ancient FreeScale supplied U-Boot. When I took that in, your work didn't appear in the repository yet, so finding that was a big and pleasant surprise. Especially because the FreeScale version had a couple of bugs in it. Not to start about the bugs in the supplied bootlets...
So taking in U-Boot with your work is killing at least 3 birds with one stone.
I will do some work on the Kernel later. For now we're using the FreeScale supplied Kernel. I have tried 3.1rc4 before and it ran our application perfectly, so that's looking good already. But changing Kernel 2 weeks before board bring-up is a bit too exciting for me. For that reason I will probably hold on to the FreeScale bootlets and U-Boot version too.
But now I've got your attention anyway: Is there a reason why U-Boot expects the bootstream at sector 1024 of the SD-Card?
To make things simple for users, there's standard layout. It's actually at sector 2048 btw.
The internal bootrom doesn't know this limitation. It is clever enough to interpret the partition table, and start loading SRAM from the first sector of the first partition of type 53.
That's what happens internally, it's just that some pieces (like sector offset of the partition!) need to be filled into the bootstream header by the mxsboot utility -- see mxsboot -h for how to change that.
The 2048 sector offset was chosen because that's standard in Linux now for first partiiton.
Now I figure that the internal bootrom can only load the first part of U-Boot, because of SRAM limits, and that this first part has to power-up all power rails, initialize SD-RAM and load the remainder of U-Boot, but I don't see why it couldn't be as clever as the internal bootrom.
What do you mean?
The reason I'm asking is because our project intends to boot from MMC all the time. We don't think unmanaged NAND is reliable enough for our appliction. We would like to have the oppertunity to boot from an SD-Card, that also usable on Windows. And since that poor OS doesn't know the difference between a disk, a partition and a file system it gets confused if the first partition on an SD-Card isn't FAT formatted.
You can adjust that, see above.
Cheers,
Robert.
To make things simple for users, there's standard layout. It's actually at sector 2048 btw.
Got that.
That's what happens internally, it's just that some pieces (like sector offset of the partition!) need to be filled into the bootstream header by the mxsboot utility -- see mxsboot -h for how to change that.
You're completely right! This sector needs to be know build-time because it is specified in the Boot Stream Block of the Boot Stream header. I forgot about that, because I hid that in a script some time ago. That script was basically a rewritten version of the broken one that was supplied with the SDK.
Come to think of it, it's a bit weird that the sector the BSB is located in, is specified in the BSB: When reading it, it already knows this sector, because that's the sector it is reading from at that moment...
The 2048 sector offset was chosen because that's standard in Linux now for first partiiton.
Agreed.
What do you mean?
What I meant to ask was how you solved the paradoxic problem that U-Boot needs to initialize SD-RAM before it can be loaded (and obviously U-Boot needs to be loaded before it can initialize SD-RAM).
Freescale solved this problem by putting multiple bootlets in their bootstream. The first one is small enough to fit SRAM and it enables the PMIC's LDOs to power up all power domains. The second one too is small enough to fit SRAM. Loading it overwrites the first one and running it initializes SD-RAM. The third one is either U-Boot or a Kernel and is directly loaded into the now initialized SD-RAM.
I figured that your solution links U-Boot in a clever way that the lower part fitting SRAM will initialize the PMIC and SD-RAM, after which U-Boot will load the remainder of itself from MMC into SD-RAM.
You can adjust that, see above.
Understood.
Cheers,
Robert.
To make things simple for users, there's standard layout. It's actually at sector 2048 btw.
Got that.
That's what happens internally, it's just that some pieces (like sector offset of the partition!) need to be filled into the bootstream header by the mxsboot utility -- see mxsboot -h for how to change that.
You're completely right! This sector needs to be know build-time because it is specified in the Boot Stream Block of the Boot Stream header. I forgot about that, because I hid that in a script some time ago. That script was basically a rewritten version of the broken one that was supplied with the SDK.
Come to think of it, it's a bit weird that the sector the BSB is located in, is specified in the BSB: When reading it, it already knows this sector, because that's the sector it is reading from at that moment...
The 2048 sector offset was chosen because that's standard in Linux now for first partiiton.
Agreed.
What do you mean?
What I meant to ask was how you solved the paradoxic problem that U-Boot needs to initialize SD-RAM before it can be loaded (and obviously U-Boot needs to be loaded before it can initialize SD-RAM).
Well we have U-Boot SPL, which is loaded into SRAM, inits DRAM, pinmux etc., then we pass execution back to BootROM to load U-Boot into DRAM and execute it.
Freescale solved this problem by putting multiple bootlets in their bootstream. The first one is small enough to fit SRAM and it enables the PMIC's LDOs to power up all power domains. The second one too is small enough to fit SRAM. Loading it overwrites the first one and running it initializes SD-RAM. The third one is either U-Boot or a Kernel and is directly loaded into the now initialized SD-RAM.
Yea, you can load kernel directly via U-Boot SPL too.
I figured that your solution links U-Boot in a clever way that the lower part fitting SRAM will initialize the PMIC and SD-RAM, after which U-Boot will load the remainder of itself from MMC into SD-RAM.
See above.
You can adjust that, see above.
Understood.
Cheers,
Robert.
participants (2)
-
Marek Vasut -
Robert Deliën