HIGH MEM suport for up to 64GB

Abhishek sinha aby_sinha at yahoo.com
Sun Apr 21 18:08:10 PDT 2002

hi all

This is what i did to allocate more than 3 Gb out of 4 Gb of RAM to the 
userspace .
In the file /usr/src/linux2.4.*/include/asm-i386/page_offset.h

Under the line #ifdef CONFIG_1GB
Changed #define PAGE_OFFSET_RAW 0xC00000000 to PAGE_OFFSET_RAW 0xE00000000
then in processor.h changed #define TASK_UNMAPPED_BASE (TASK_SIZE/3) to 

and then recomplied the kernel with high mem support.
this was on 2.4.7-10 , in 2.4.18 there are minor changes

I am not a big kernel hacker and still learning so please use it with 
comments invited


William Harman wrote:

>Sounds right.  __PAGE_OFFSET is set to 0xC0000000 by default, so
>you get 3GB of address space for user processes.  And about 2GB
>max for the data segment.
>You can change the kernel to give you more than 3GB of address
>space and greater than 2GB of data segment for heap allocations.
>It's three lines to change. Two lines in header files and one
>line in the one of the assembly init files.
>Leandro Tavares Carneiro (leandro at ep.petrobras.com.br) wrote*:
>>Hi everyone,
>>	I am writing to ask to you all if anyone have tesed or used an machine
>>with more than 4GB of RAM or paging in virtual memory on intel machines.
>>	He have an linux beowulf cluster and one of ours developers have asked
>>us for how much memory an process can allocate to use. In the tests we
>>have made, we cannot allocate much more than 3GB, using an dual PIII
>>with 1GB of ram and 12Gb of swap area for testing.
>>	We can use 2 process alocating more or less 3Gb, but one process alone
>>canot pass this test.
>>	We have using redhat linux 7.2, kernel 2.4.9-21, recompiled with High
>>Mem suport.
>>	I have tested the same test aplication on an Itanium machine, with 1GB
>>of ram and 16Gb of swap area, and they passed. The aplication can
>>alocate more than 5GB of memory, using swap. In this machine, we are
>>using turbolinux 7, with kernel version 2.4.4-010508-18smp.
>>Thanks in advance for the help,
>>Best regards,
>>Beowulf mailing list, Beowulf at beowulf.org
>>To change your subscription (digest mode or unsubscribe) visit
>> Part 1.1
>> Content-Type:
>> text/plain
>> Content-Encoding:
>> 8bit

More information about the Beowulf mailing list