Linus, Alan,
This patch to i386 setup.S resulted from my attempts to load big
bzImages from DOS. Loadlin fails at aroung 1020000 bytes, I didn't
want to mess with that much asm in loadlin src and wrote new
loader from scratch mostly in c.
I had a problem booting from DOS on pc with 196 mb or ram. It turned
out int15/e801 and int15/88 reported zero ext mem -> "Less than 4MB"
type hang. Digging in loadlin sources turned out some horrors
(loadlin hooks itself on int 15/88 and reads CMOS)
I think it's cleaner to do this in setup.S rather than jump thru the
hoops in DOS loader. Attached patch does that *only if* int15 methods
fail. It does not break existing practice, only adds just another
fallback in case of brain damage.
I couldn't resist and fixed some typos, cleaned up some label names
and replaced some instructions with shorter code :-)
Tested on three boxes. Works for me. :-)
--
Best regards, vda
mailto:[email protected]
diff -ubB --expand-tabs setup.S.orig setup.S
--- setup.S.orig Mon Sep 24 00:58:58 2001
+++ setup.S Thu Oct 18 11:36:24 2001
@@ -39,9 +39,13 @@
* from Ralf Brown interrupt list seem to indicate AX/BX should be used
* anyway. So to avoid breaking many machines (presumably there was a reason
* to orginally use CX/DX instead of AX/BX), we do a kludge to see
- * if CX/DX have been changed in the e801 call and if so use AX/BX .
+ * if CX/DX have been changed in the e801 call and if so use AX/BX.
* Michael Miller, April 2001 <[email protected]>
*
+ * Even more fixes for memory detection when started from DOS
+ * TODO: maybe we can just resort to memory scan on our own
+ * to stop using int15/e820,int15/e801,int15/88,cmos once and for all?
+ * <[email protected]> october 2001
*/
#include <linux/config.h>
@@ -297,13 +301,13 @@
#define SMAP 0x534d4150
-meme820:
+mem_e820:
xorl %ebx, %ebx # continuation counter
movw $E820MAP, %di # point into the whitelist
# so we can have the bios
# directly write into it.
-jmpe820:
+jmp820:
movl $0x0000e820, %eax # e820, upper word zeroed
movl $SMAP, %edx # ascii 'SMAP'
movl $20, %ecx # size of the e820rec
@@ -327,12 +331,10 @@
jnl bail820
incb (E820NR)
- movw %di, %ax
- addw $20, %ax
- movw %ax, %di
+ addw $20, %di
again820:
- cmpl $0, %ebx # check to see if
- jne jmpe820 # %ebx is set to EOF
+ testl %ebx, %ebx # check to see if
+ jnz jmp820 # %ebx is set to EOF
bail820:
@@ -344,10 +346,10 @@
# alternative new memory detection scheme, and it's sensible
# to write everything into the same place.)
-meme801:
+mem_e801:
stc # fix to work around buggy
- xorw %cx,%cx # BIOSes which dont clear/set
- xorw %dx,%dx # carry on pass/error of
+ xorw %cx, %cx # BIOSes which dont clear/set
+ xorw %dx, %dx # carry on pass/error of
# e801h memory size call
# or merely pass cx,dx though
# without changing them.
@@ -355,28 +357,62 @@
int $0x15
jc mem88
- cmpw $0x0, %cx # Kludge to handle BIOSes
- jne e801usecxdx # which report their extended
- cmpw $0x0, %dx # memory in AX/BX rather than
- jne e801usecxdx # CX/DX. The spec I have read
+ testw %cx, %cx # Kludge to handle BIOSes
+ jnz e801use_cxdx # which report their extended
+ testw %dx, %dx # memory in AX/BX rather than
+ jnz e801use_cxdx # CX/DX. The spec I have read
movw %ax, %cx # seems to indicate AX/BX
movw %bx, %dx # are more reasonable anyway...
-e801usecxdx:
- andl $0xffff, %edx # clear sign extend
+e801use_cxdx:
+ movzwl %dx, %edx # clear sign extend
shll $6, %edx # and go from 64k to 1k chunks
- movl %edx, (0x1e0) # store extended memory size
- andl $0xffff, %ecx # clear sign extend
- addl %ecx, (0x1e0) # and add lower memory into
- # total size.
+ movzwl %cx, %ecx # clear sign extend
+ addl %ecx, %edx # add lower memory to
+ movl %edx, (0x1e0) # extended, store
# Ye Olde Traditional Methode. Returns the memory size (up to 16mb or
# 64mb, depending on the bios) in ax.
mem88:
#endif
+ #stc # guard against brain damage -
+ # int 15 must clear cf to indicate success
+ clc # unbelievable: some BIOSes/DOSes can leave
+ # cf as is so I had to abandon stc trick
movb $0x88, %ah
int $0x15
+ jc mem_cmos
+ testw %ax, %ax
+ jnz mem_store
+
+# Fallback: if int15/88 fails, get same data from CMOS
+# this works around extremely stupid case I had with PC not booting
+# from DOS when I put 196 megs of RAM in it
+# (it reported 0 via int15/e801 and int15/88)
+# Also this makes unnecessary for loadlin to jump thru the hoops
+# just in order to let us know ext mem size
+# (it hooks itself on int 15/88 and does CMOS reads for us)
+mem_cmos:
+ pushf
+ cli
+ movb $0x18, %al
+ outb %al, $0x70
+ #iodelay?
+ inb $0x71, %al
+ movb %al, %ah
+ #iodelay?
+ movb $0x17, %al
+ outb %al, $0x70
+ #iodelay?
+ inb $0x71, %al
+ popf
+
+mem_store:
+ cmpw $0xffff-0x400, %ax # We want to be sure it won't roll over
+ jbe mem_store2 # 16 bit value when low 1024k gets added
+ movw $0xffff-0x400, %ax # thus max memtop is 0xffff*1k = 64m-1k
+mem_store2:
movw %ax, (2)
# Set the keyboard repeat rate to the max