Mapped kernel changes: create elf sections as close as possible to

IRIX, to get the same kind of loader behavior. Read comments in
mapped_kernel.h for more details.

1 files changed, 15 insertions, 0 deletions

diff --git a/include/asm-mips64/sn/mapped_kernel.h b/include/asm-mips64/sn/mapped_kernel.h
index ad72d99db..07e799b96 100644
--- a/include/asm-mips64/sn/mapped_kernel.h
+++ b/include/asm-mips64/sn/mapped_kernel.h
@@ -5,6 +5,21 @@
 #ifndef __ASM_SN_MAPPED_KERNEL_H
 #define __ASM_SN_MAPPED_KERNEL_H
 
+/*
+ * Note on how mapped kernels work: the text and data section is 
+ * compiled at cksseg segment (LOADADDR = 0xc001c000), and the 
+ * init/setup/data section gets a 16M virtual address bump in the 
+ * ld.script file (so that tlblo0 and tlblo1 maps the sections). 
+ * The vmlinux.64 section addresses are put in the xkseg range 
+ * using the change-addresses makefile option. Use elfdump -of 
+ * on IRIX to see where the sections go. The Origin loader loads 
+ * the two sections contiguously in physical memory. The loader 
+ * sets the entry point into kernel_entry using a xkphys address, 
+ * but instead of using 0xa800000001160000, it uses the address 
+ * 0xa800000000160000, which is where it physically loaded that 
+ * code. So no jumps can be done before we have switched to using 
+ * cksseg addresses.
+ */
 #include <linux/config.h>
 #include <asm/addrspace.h>