SPARSEMEM support for Atmel AT91SAM9G45 and M10

Fri May 28 03:42:10 EDT 2010

Marek Vasut schrieb:
> Dne Čt 27. května 2010 10:15:59 Yegor Yefremov napsal(a):
>> Marek Vasut schrieb:
>>> Dne St 26. května 2010 15:46:21 Yegor Yefremov napsal(a):
>>>>>     I'm resurrecting an old thread
>>>>>
>>>>> (http://lists.arm.linux.org.uk/lurker/message/20090914.072902.97115112.
>>>>> en .html) about using the two memory banks available
>>>>> on Atmel AT91SAM9G45/M10. I managed to make it work on 2.6.30 by
>>>>> following Russell King advice on how to configure SPARSEMEM.
>>>>>
>>>>>     I submitted a working patch for 2.6.34 here:
>>>>> http://www.arm.linux.org.uk/developer/patches/viewpatch.php?id=6143/1
>>>>>
>>>>> The problem is that I'm unsure how to get rid of the #define that
>>>>> forces high_memory value in mm/init.c, should I add a fixup function
>>>>> to board-sam9m10g45ek.c to provide meminfo information ?
>>>> I've already tested the patch on 2.6.33 and it is working properly.
>>>> Concerning high_memory value problem I think the cause is how
>>>> find_node_limits() in arch/arm/m/init.c is implemented. We have two
>>>> memory banks one starting at 0x70000000 and another at 0x200000000. The
>>>> virtual mapping looks like this:
>>>>
>>>> 0x70000000 -> 0xc0000000
>>>> 0x20000000 -> 0xc8000000
>>>>
>>>> However find_node_limits() routine is working with physical addresses so
>>>> the upper limit is 0x70000000 and not 0x20000000 as it should be. As a
>>>> result max_low is pointing to 0x70000000 and that's why only first bank
>>>> will be mapped correctly.
>>>>
>>>> How can find_node_limits() be changed to work with virtual addresses?
>>>>
>>>> Regards,
>>>> Yegor
>>> Hey, try searching for "[PATCH 1/5] pxa/vpac270: Enable SparseMEM for 256
>>> MB of RAM" ... The device has also one memory bank mapped under the
>>> primary one. Maybe that can help. Cheers
>> Marek, I've looked at patch and apart from defining HIGH_MEMORY_VIRT it
>> does the same. Nevertheless if I don't define HIGH_MEMORY_VIRT I get the
>> following crash (I've added debug string showing calculated high_memory.
>> It should be 0xd0000000 and not 0xc8000000):
>>
>> Linux version 2.6.33.2 (OpenRISC at VScom) (gcc version 4.3.2
>> (crosstool-NG-1.3.2) ) #30 Thu May 27 10:06:42 CEST 2010
>> CPU: ARM926EJ-S [41069265] revision 5 (ARMv5TEJ), cr=00053177
>> CPU: VIVT data cache, VIVT instruction cache
>> Machine: Atmel AT91SAM9G45-EKES
>> Memory policy: ECC disabled, Data cache writeback
>> DBG: high_memory 0xc8000000
>> Clocks: CPU 400 MHz, master 133 MHz, main 12.000 MHz
>> Built 1 zonelists in Zone order, mobility grouping on.  Total pages: 62720
>> Kernel command line: console=ttyS0,115200 root=/dev/mtdblock2 rootwait
>> PID hash table entries: 1024 (order: 0, 4096 bytes)
>> Dentry cache hash table entries: 32768 (order: 5, 131072 bytes)
>> Inode-cache hash table entries: 16384 (order: 4, 65536 bytes)
>> Memory: 128MB 128MB = 256MB total
>> Memory: 255036KB available (3944K code, 303K data, 140K init, 0K highmem)
>> Hierarchical RCU implementation.
>> NR_IRQS:192
>> AT91: 160 gpio irqs in 5 banks
>> Console: colour dummy device 80x30
>> console [ttyS0] enabled
>> Calibrating delay loop... 199.47 BogoMIPS (lpj=997376)
>> Mount-cache hash table entries: 512
>> CPU: Testing write buffer coherency:
>> ------------[ cut here ]------------
>> WARNING: at mm/vmalloc.c:107 vmap_page_range_noflush+0x128/0x1d8()
>> Modules linked in:
>> [<c0031894>] (unwind_backtrace+0x0/0xdc) from [<c0043c34>]
>> (warn_slowpath_common+0x4c/0x80) [<c0043c34>]
>> (warn_slowpath_common+0x4c/0x80) from [<c008df9c>]
>> (vmap_page_range_noflush+0x128/0x1d8) [<c008df9c>]
>> (vmap_page_range_noflush+0x128/0x1d8) from [<c008e074>]
>> (map_vm_area+0x28/0x50) [<c008e074>] (map_vm_area+0x28/0x50) from
>> [<c008e4f8>] (vmap+0x48/0x6c) [<c008e4f8>] (vmap+0x48/0x6c) from
>> [<c000cf60>] (check_writebuffer_bugs+0x5c/0x134) [<c000cf60>]
>> (check_writebuffer_bugs+0x5c/0x134) from [<c0008b9c>]
>> (start_kernel+0x31c/0x388) [<c0008b9c>] (start_kernel+0x31c/0x388) from
>> [<70008034>] (0x70008034) ---[ end trace 1b75b31a2719ed1c ]---
>> ------------[ cut here ]------------
>> WARNING: at mm/vmalloc.c:107 vmap_page_range_noflush+0x128/0x1d8()
>> Modules linked in:
>> [<c0031894>] (unwind_backtrace+0x0/0xdc) from [<c0043c34>]
>> (warn_slowpath_common+0x4c/0x80) [<c0043c34>]
>> (warn_slowpath_common+0x4c/0x80) from [<c008df9c>]
>> (vmap_page_range_noflush+0x128/0x1d8) [<c008df9c>]
>> (vmap_page_range_noflush+0x128/0x1d8) from [<c008e074>]
>> (map_vm_area+0x28/0x50) [<c008e074>] (map_vm_area+0x28/0x50) from
>> [<c008e4f8>] (vmap+0x48/0x6c) [<c008e4f8>] (vmap+0x48/0x6c) from
>> [<c000cf78>] (check_writebuffer_bugs+0x74/0x134) [<c000cf78>]
>> (check_writebuffer_bugs+0x74/0x134) from [<c0008b9c>]
>> (start_kernel+0x31c/0x388) [<c0008b9c>] (start_kernel+0x31c/0x388) from
>> [<70008034>] (0x70008034) ---[ end trace 1b75b31a2719ed1d ]---
>> failed, unable to map memory
>>
>> Any ideas?
>>
>> Yegor
> 
> Do you have CONFIG_HIGHMEM enabled ? If so, why ?

No, I don't. Could you post your bootlog and your .config?

#
# Kernel Features
#
# CONFIG_NO_HZ is not set
# CONFIG_HIGH_RES_TIMERS is not set
CONFIG_GENERIC_CLOCKEVENTS_BUILD=y
CONFIG_VMSPLIT_3G=y
# CONFIG_VMSPLIT_2G is not set
# CONFIG_VMSPLIT_1G is not set
CONFIG_PAGE_OFFSET=0xC0000000
CONFIG_PREEMPT_NONE=y
# CONFIG_PREEMPT_VOLUNTARY is not set
# CONFIG_PREEMPT is not set
CONFIG_HZ=100
CONFIG_AEABI=y
CONFIG_OABI_COMPAT=y
CONFIG_ARCH_SPARSEMEM_ENABLE=y
CONFIG_ARCH_SPARSEMEM_DEFAULT=y
CONFIG_ARCH_SELECT_MEMORY_MODEL=y
# CONFIG_HIGHMEM is not set
CONFIG_SELECT_MEMORY_MODEL=y
# CONFIG_FLATMEM_MANUAL is not set
# CONFIG_DISCONTIGMEM_MANUAL is not set
CONFIG_SPARSEMEM_MANUAL=y
CONFIG_SPARSEMEM=y
CONFIG_HAVE_MEMORY_PRESENT=y
CONFIG_SPARSEMEM_EXTREME=y
CONFIG_SPLIT_PTLOCK_CPUS=999999
# CONFIG_PHYS_ADDR_T_64BIT is not set
CONFIG_ZONE_DMA_FLAG=0
CONFIG_VIRT_TO_BUS=y
# CONFIG_KSM is not set
CONFIG_DEFAULT_MMAP_MIN_ADDR=4096
# CONFIG_LEDS is not set
CONFIG_ALIGNMENT_TRAP=y
# CONFIG_UACCESS_WITH_MEMCPY is not set