On 02/19/2013 05:14 PM, Wolfgang Denk wrote:
Dear Rob Herring,
In message 5123E311.1070607@gmail.com you wrote:
[snip]
What I would really prefer to do is like powerpc where platforms override this if they need the fixup and the default behavior is no fixup. But that would require knowing which platforms do and don't need the fixup. It only going to get harder to change that.
Normally we use get_ramsize() to check / verify the amount of available RAM. And on PPC, we nearly always set the ram size in the DT. I don't know why you think the default was no fixup; this is not true. It's the other way round.
No it is not.
The fixup is in bootm.c for all of arm, yet appears in all the board/soc files for powerpc. It may be nearly always fixed up because every platform duplicates the fixup code, but it is not the default. It is a per platform decision on powerpc and it is not on arm.
$ git grep fixup_memory arch/arm/lib/bootm.c:static int fixup_memory_node(void *blob) arch/arm/lib/bootm.c: return fdt_fixup_memory_banks(blob, start, size, CONFIG_NR_DRAM_BANKS); arch/arm/lib/bootm.c: fixup_memory_node(*of_flat_tree); arch/powerpc/cpu/74xx_7xx/cpu.c: fdt_fixup_memory(blob, (u64)bd->bi_memstart, (u64)bd->bi_memsize); arch/powerpc/cpu/mpc512x/cpu.c: fdt_fixup_memory(blob, (u64)bd->bi_memstart, (u64)bd->bi_memsize); arch/powerpc/cpu/mpc5xxx/cpu.c: fdt_fixup_memory(blob, (u64)bd->bi_memstart, (u64)bd->bi_memsize); arch/powerpc/cpu/mpc8260/cpu.c: fdt_fixup_memory(blob, (u64)bd->bi_memstart, (u64)bd->bi_memsize); arch/powerpc/cpu/mpc83xx/fdt.c: fdt_fixup_memory(blob, (u64)bd->bi_memstart, (u64)bd->bi_memsize); arch/powerpc/cpu/mpc85xx/fdt.c: fdt_fixup_memory(blob, (u64)bd->bi_memstart, (u64)bd->bi_memsize); arch/powerpc/cpu/mpc86xx/fdt.c: fdt_fixup_memory(blob, (u64)bd->bi_memstart, (u64)bd->bi_memsize); arch/powerpc/cpu/mpc8xx/fdt.c: fdt_fixup_memory(blob, (u64)bd->bi_memstart, (u64)bd->bi_memsize); arch/powerpc/cpu/ppc4xx/fdt.c: fdt_fixup_memory(blob, (u64)bd->bi_memstart, (u64)bd->bi_memsize); board/exmeritus/hww1u1a/hww1u1a.c: fdt_fixup_memory(blob, (u64)base, (u64)size); board/freescale/bsc9131rdb/bsc9131rdb.c: fdt_fixup_memory(blob, (u64)base, (u64)size); board/freescale/corenet_ds/corenet_ds.c: fdt_fixup_memory(blob, (u64)base, (u64)size); board/freescale/mpc7448hpc2/mpc7448hpc2.c: fdt_fixup_memory(blob, (u64)bd->bi_memstart, (u64)bd->bi_memsize); board/freescale/mpc8572ds/mpc8572ds.c: fdt_fixup_memory(blob, (u64)base, (u64)size); board/freescale/p1010rdb/p1010rdb.c: fdt_fixup_memory(blob, (u64)base, (u64)size); board/freescale/p1022ds/p1022ds.c: fdt_fixup_memory(blob, (u64)base, (u64)size); board/freescale/p1023rds/p1023rds.c: fdt_fixup_memory(blob, (u64)base, (u64)size); board/freescale/p1_p2_rdb/p1_p2_rdb.c: fdt_fixup_memory(blob, (u64)base, (u64)size); board/freescale/p1_p2_rdb_pc/p1_p2_rdb_pc.c: fdt_fixup_memory(blob, (u64)base, (u64)size); board/freescale/p2020come/p2020come.c: fdt_fixup_memory(blob, (u64)base, (u64)size); board/freescale/p2020ds/p2020ds.c: fdt_fixup_memory(blob, (u64)base, (u64)size); board/freescale/p2041rdb/p2041rdb.c: fdt_fixup_memory(blob, (u64)base, (u64)size); board/freescale/t4qds/t4qds.c: fdt_fixup_memory(blob, (u64)base, (u64)size); board/ipek01/ipek01.c: fdt_fixup_memory (blob, (u64) bd->bi_memstart, (u64) bd->bi_memsize); board/pdm360ng/pdm360ng.c: fdt_fixup_memory(blob, (u64)bd->bi_memstart, (u64)bd->bi_memsize); common/cmd_fdt.c: err = fdt_fixup_memory(working_fdt, addr, size); common/fdt_support.c:int fdt_fixup_memory_banks(void *blob, u64 start[], u64 size[], int banks) common/fdt_support.c:int fdt_fixup_memory(void *blob, u64 start, u64 size) common/fdt_support.c: return fdt_fixup_memory_banks(blob, &start, &size, 1); include/fdt_support.h:int fdt_fixup_memory(void *blob, u64 start, u64 size); include/fdt_support.h:int fdt_fixup_memory_banks(void *blob, u64 start[], u64 size[], int banks);
Actually you are reaching here a point where it seems necessary that U-Boot itself is able to read the DT to configure itself correctly, i. e. the memory size information it holds should then also be extracted from the DT.
I thought about doing this. That's really an orthogonal issue. The problem is you cannot adjust the amount of memory u-boot uses to be different than the amount of RAM in the /memory node. If you adjust the size in u-boot, the adjusted size is passed to the kernel no matter what.
There's issues with LPAE systems having >4GB of RAM as all the size and address values are 32-bit. This could be fixed, but I'm not sure there is much value in u-boot knowing about memory above 4GB as it can't really access all of it and it would be purely informational.
Well, let's face it - no matter what you do, if the RAM sizes as used in U-Boot and in Linux differ, at least one of these is wrong. This is a bug that should be fixed. What you attempt to do is implementing a method to paper over such bugs. This is not a good thing to do.
We'll probably just have to disagree that u-boot needs to know about memory it can't access on 32-bit PAE system unless you want to implement paging.
Rob