ARM: shmobile: sh73a0 and kzm9g Multiplatform support
cover ARM: shmobile: sh73a0 and kzm9g Multiplatform support
ARM: shmobile: sh73a0 and kzm9g Multiplatform support
[PATCH 01/03] ARM: shmobile: sh73a0: Introduce generic setup callback [PATCH 02/03] ARM: shmobile: sh73a0: Add Multiplatform support [PATCH 03/03] ARM: shmobile: kzm9g: Build DTS for Multiplatform
This patch series enables sh73a0 Multiplatform support. At this point the KZM9G board code is left as-is, but future updates will make sure to consolidate the DTS and get rid of C board code.
---
Built on top of renesas-devel-20140904-v3.17-rc3
arch/arm/boot/dts/Makefile | 3 ++- arch/arm/mach-shmobile/Kconfig | 5 +++++ arch/arm/mach-shmobile/Makefile | 4 ++-- arch/arm/mach-shmobile/setup-sh73a0.c | 21 +++++++++++++++++---- 4 files changed, 26 insertions(+), 7 deletions(-)
Hi Magnus,
Quoted 9 lines · Magnus Damm
ARM: shmobile: sh73a0 and kzm9g Multiplatform support
[PATCH 01/03] ARM: shmobile: sh73a0: Introduce generic setup callback [PATCH 02/03] ARM: shmobile: sh73a0: Add Multiplatform support [PATCH 03/03] ARM: shmobile: kzm9g: Build DTS for Multiplatform
This patch series enables sh73a0 Multiplatform support. At this point the KZM9G board code is left as-is, but future updates will make sure to consolidate the DTS and get rid of C board code.
Did you manage to boot this successfully?
For me, it hangs after:
On node 0 totalpages: 124928
free_area_init_node: node 0, pgdat c04f9a40, node_mem_map df429000
Normal zone: 976 pages used for memmap
Normal zone: 0 pages reserved
Normal zone: 124928 pages, LIFO batch:31Both kzm9g-legacy and kzm9g-reference do work.
So it must be some funky difference between reference and multi-platform.
More debugging shows it hangs in memmap_init_zone(), which loops over all pfns, calling pfn_to_page(pfn):
pfn 266240 -> page 0xdf448000
...
pfn 388095 -> page 0xdf7fffe0
pfn 388096 -> page 0xdf800000As 0xdf800000 is just beyond the end of RAM, the kernel locks up.
With -reference, the output is:
free_area_init_node: node 0, pgdat c0508740, node_mem_map de429000
Normal zone: 976 pages used for memmap
Normal zone: 0 pages reserved
Normal zone: 124928 pages, LIFO batch:31
Note that node_mem_map is 0x1000000 less!
And the loop goes like:
pfn 266240 -> page 0xde428000
...
pfn 391167 -> page 0xde7f7fe0And all pages fit in RAM.
On armadillo, node_mem_map is the same for -legacy and -multiplatform.
The kzm9g memory node has "reg = <0x41000000 0x1e800000>;". The armadillo memory node has "reg = <0x40000000 0x20000000>;". But, if I limit memory on armadillo to match kzm9g, it also locks up.
Quoted 4 lines
Signed-off-by Magnus Damm <[email protected]>---
Built on top of renesas-devel-20140904-v3.17-rc3
To check if this is a recent regression, I also tried on top of renesas-devel-20140904v2-v3.17-rc3, but I get the same lock up.
Gr{oetje,eeting}s,
Geert
--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- [email protected]
In personal conversations with technical people, I call myself a hacker. But when I'm talking to journalists I just say "programmer" or something like that.
-- Linus Torvalds
Hi Magnus,
Quoted 15 lines · Geert Uytterhoeven
Quoted 9 lines · Magnus Damm
ARM: shmobile: sh73a0 and kzm9g Multiplatform support
[PATCH 01/03] ARM: shmobile: sh73a0: Introduce generic setup callback [PATCH 02/03] ARM: shmobile: sh73a0: Add Multiplatform support [PATCH 03/03] ARM: shmobile: kzm9g: Build DTS for Multiplatform
This patch series enables sh73a0 Multiplatform support. At this point the KZM9G board code is left as-is, but future updates will make sure to consolidate the DTS and get rid of C board code.
Did you manage to boot this successfully?
For me, it hangs after:
Quoted 7 lines
On node 0 totalpages: 124928 free_area_init_node: node 0, pgdat c04f9a40, node_mem_map df429000 Normal zone: 976 pages used for memmap Normal zone: 0 pages reserved Normal zone: 124928 pages, LIFO batch:31The kzm9g memory node has "reg = <
0x410000000x1e800000>;".
... which is incompatible with CONFIG_AUTO_ZRELADDR=y, which requires that the start address of physical memory is a multiple of 128 MiB, and which is enabled for multi-platform kernels...
The rationale for stealing the first 16 MiB of RAM is explained in the u-boot history:
commit 9415cf93bc5ded82c6b915d5ab3b8c8f61a75c46
Author: Tetsuyuki Kobayashi <[email protected]> Date: Thu Jul 5 01:43:44 2012 +0000
arm: rmobile: kzm9g: Modify sdram area
Reserve first 16MB for RT-CPU (as same as kernel config).
Do you remember what's the reason for stealing the last 8 MiB of RAM?
It works fine with the full 512 MiB of RAM, if enabled.
Gr{oetje,eeting}s,
Geert
--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- [email protected]
In personal conversations with technical people, I call myself a hacker. But when I'm talking to journalists I just say "programmer" or something like that.
-- Linus Torvalds