add busybox_ldscript.README.txt
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
diff --git a/busybox_ldscript.README.txt b/busybox_ldscript.README.txt
new file mode 100644
index 0000000..1625a97
--- /dev/null
+++ b/busybox_ldscript.README.txt
@@ -0,0 +1,47 @@
+/* Add SORT_BY_ALIGNMENT to linker script (found in busybox_unstripped.out):
+## .rodata : { *(.rodata SORT_BY_ALIGNMENT(.rodata.*) .gnu.linkonce.r.*) }
+## .data : { *(.data SORT_BY_ALIGNMENT(.data.*) .gnu.linkonce.d.*) }
+## .bss : { *(.bss SORT_BY_ALIGNMENT(.bss.*) .gnu.linkonce.b.*) }
+## This will eliminate most of the padding (~3kb).
+## Hmm, "ld --sort-section alignment" should do it too.
+##
+## There is a ld hack which is meant to decrease disk usage
+## at the cost of more RAM usage (??!!) in standard ld script:
+## . = ALIGN (0x1000) - ((0x1000 - .) & (0x1000 - 1)); . = DATA_SEGMENT_ALIGN (0x1000, 0x1000);
+## Replace it with:
+## . = ALIGN (0x1000); . = DATA_SEGMENT_ALIGN (0x1000, 0x1000);
+## to unconditionally align .data to the next page boundary,
+## instead of "next page, plus current offset in this page"
+*/
+
+/* To reduce the number of VMAs each bbox process has,
+## move *(.bss SORT_BY_ALIGNMENT(.bss.*) ...)
+## part from .bss : {...} block to .data : { ... } block.
+## (This usually increases .data section by only one page).
+## Result:
+##
+## text data bss dec hex filename
+## 1050792 560 7580 1058932 102874 busybox.bss
+## 1050792 8149 0 1058941 10287d busybox.nobss
+##
+## $ exec busybox.bss pmap $$
+## 0000000008048000 1028K r-xp /path/to/busybox.bss
+## 0000000008149000 8K rw-p /path/to/busybox.bss
+## 000000000814b000 4K rw-p [ anon ] <---- this VMA is eliminated
+## 00000000085f5000 4K ---p [heap]
+## 00000000085f6000 4K rw-p [heap]
+## 00000000f7778000 8K rw-p [ anon ]
+## 00000000f777a000 12K r--p [vvar]
+## 00000000f777d000 8K r-xp [vdso]
+## 00000000ff7e9000 132K rw-p [stack]
+##
+## $ exec busybox.nobss pmap $$
+## 0000000008048000 1028K r-xp /path/to/busybox.nobss
+## 0000000008149000 12K rw-p /path/to/busybox.nobss
+## 00000000086f0000 4K ---p [heap]
+## 00000000086f1000 4K rw-p [heap]
+## 00000000f7783000 8K rw-p [ anon ]
+## 00000000f7785000 12K r--p [vvar]
+## 00000000f7788000 8K r-xp [vdso]
+## 00000000ffac0000 132K rw-p [stack]
+*/