[syslinux] Limiting memory used by syslinux

[jochen.desmet at emc.com]

Chris Dion came up with the following patch which seems
to do the trick for now:

--- syslinux-3.82.orig//com32/lib/malloc.c      2009-06-09 13:19:25.000000000 -0400
+++ syslinux-3.82/com32/lib/malloc.c    2012-06-03 10:42:48.000000000 -0400
@@ -34,7 +34,7 @@
   return sp;
 }

-#define E820_MEM_MAX 0xfff00000        /* 4 GB - 1 MB */
+#define E820_MEM_MAX 0x40000000        /* 1GB */

 static int consider_memory_area(void *dummy, addr_t start,
                                addr_t len, bool valid)
--- syslinux-3.82.orig//core/highmem.inc        2009-06-09 13:19:25.000000000 -0400
+++ syslinux-3.82/core/highmem.inc      2012-06-03 11:14:04.000000000 -0400
@@ -38,7 +38,7 @@
 ; report main memory as multiple contiguous ranges...
 ;
 get_e820:
-               mov dword [E820Max],-(1 << 20)  ; Max amount of high memory
+               mov dword [E820Max],(1 << 30)   ; Max amount of high memory
                mov dword [E820Mem],(1 << 20)   ; End of detected high memory
 .start_over:
                mov di,E820Buf


but he's looking into making it configurable.

J.




-----Original Message-----
From: H. Peter Anvin [mailto:hpa at zytor.com] 
Sent: Saturday, June 02, 2012 13:06
To: For discussion of Syslinux and tftp-hpa
Cc: Gene Cumm; De Smet, Jochen
Subject: Re: [syslinux] Limiting memory used by syslinux

On 06/02/2012 09:28 AM, Gene Cumm wrote:
> On Jun 2, 2012 11:12 AM, "H. Peter Anvin" <hpa at zytor.com> wrote:
>>
>> On 06/01/2012 12:07 PM, jochen.desmet at emc.com wrote:
> 
>>> The problem we're having is that syslinux is overwriting some of that
>>> memory (specifically, at first glance it looks like the com32 stack). Is
>>> there a way to tell syslinux to not use memory above a certain address ?
> 
>> There isn't in the stock syslinux, but it wouldn't be hard to do; there
>> is actually a hard-coded limit of 0xfff00000 already (in two places, you
>> probably need to grep for it.)  The problem with making this
>> configurable is that it would have to happen very early on, before a lot
>> of the configuration information is available.
> 
> Wouldn't this be best done in the ADV or should it be configurable in the
> config file?
> 

Those are problematic, since they both are too late in the memory setup.

The best would of course to mark the persistent memory as unavailable in
the e820 setup.

	-hpa