[PATCH] more i386 head.S cleanups

[PATCH] more i386 head.S cleanups

[Brian Gerst]

[-- Attachment #1: Type: text/plain, Size: 327 bytes --]

- Move empty_zero_page and swapper_pg_dir to BSS.  This requires that 
BSS is cleared earlier, but reclaims over 3k that was lost due to page 
alignment.
- Move stack_start, ready, and int_msg, boot_gdt_descr, idt_descr, and 
cpu_gdt_descr to .data.  They were interfering with disassembly while in 
.text.

--
				Brian Gerst

[-- Attachment #2: head-sections-1 --]
[-- Type: text/plain, Size: 4047 bytes --]

diff -urN linux-bk/arch/i386/kernel/head.S linux/arch/i386/kernel/head.S
--- linux-bk/arch/i386/kernel/head.S	2004-04-01 10:17:01.000000000 -0500
+++ linux/arch/i386/kernel/head.S	2004-04-03 09:08:28.053518856 -0500
@@ -69,9 +69,22 @@
 	movl %eax,%gs
 
 /*
+ * Clear BSS first so that there are no surprises...
+ * No need to cld as DF is already clear from cld above...
+ */
+	xorl %eax,%eax
+	movl $__bss_start - __PAGE_OFFSET,%edi
+	movl $__bss_stop - __PAGE_OFFSET,%ecx
+	subl %edi,%ecx
+	shrl $2,%ecx
+	rep ; stosl
+
+/*
  * Initialize page tables.  This creates a PDE and a set of page
  * tables, which are located immediately beyond _end.  The variable
  * init_pg_tables_end is set up to point to the first "safe" location.
+ * Mappings are created both at virtual address 0 (identity mapping)
+ * and PAGE_OFFSET for up to _end+sizeof(page tables)+INIT_MAP_BEYOND_END.
  *
  * Warning: don't use %esi or the stack in this code.  However, %esp
  * can be used as a GPR if you really need it...
@@ -173,17 +186,6 @@
 #endif /* CONFIG_SMP */
 
 /*
- * Clear BSS first so that there are no surprises...
- * No need to cld as DF is already clear from cld above...
- */
-	xorl %eax,%eax
-	movl $__bss_start,%edi
-	movl $__bss_stop,%ecx
-	subl %edi,%ecx
-	shrl $2,%ecx
-	rep ; stosl
-
-/*
  * start system 32-bit setup. We need to re-do some of the things done
  * in 16-bit mode for the "real" operations.
  */
@@ -304,8 +306,6 @@
 	jmp L6			# main should never return here, but
 				# just in case, we know what happens.
 
-ready:	.byte 0
-
 /*
  * We depend on ET to be correct. This checks for 287/387.
  */
@@ -353,13 +353,7 @@
 	jne rp_sidt
 	ret
 
-ENTRY(stack_start)
-	.long init_thread_union+THREAD_SIZE
-	.long __BOOT_DS
-
 /* This is the default interrupt "handler" :-) */
-int_msg:
-	.asciz "Unknown interrupt or fault at EIP %p %p %p\n"
 	ALIGN
 ignore_int:
 	cld
@@ -386,6 +380,35 @@
 	iret
 
 /*
+ * Real beginning of normal "text" segment
+ */
+ENTRY(stext)
+ENTRY(_stext)
+
+/*
+ * BSS section
+ */
+.section ".bss.page_aligned","w"
+ENTRY(swapper_pg_dir)
+	.fill 1024,4,0
+ENTRY(empty_zero_page)
+	.fill 4096,1,0
+
+/*
+ * This starts the data section.
+ */
+.data
+
+ENTRY(stack_start)
+	.long init_thread_union+THREAD_SIZE
+	.long __BOOT_DS
+
+ready:	.byte 0
+
+int_msg:
+	.asciz "Unknown interrupt or fault at EIP %p %p %p\n"
+
+/*
  * The IDT and GDT 'descriptors' are a strange 48-bit object
  * only used by the lidt and lgdt instructions. They are not
  * like usual segment descriptors - they consist of a 16-bit
@@ -417,39 +440,6 @@
 	.fill NR_CPUS-1,8,0		# space for the other GDT descriptors
 
 /*
- * swapper_pg_dir is the main page directory, address 0x00101000
- *
- * This is initialized to create an identity-mapping at 0 (for bootup
- * purposes) and another mapping at virtual address PAGE_OFFSET.  The
- * values put here should be all invalid (zero); the valid
- * entries are created dynamically at boot time.
- *
- * The code creates enough page tables to map 0-_end, the page tables
- * themselves, plus INIT_MAP_BEYOND_END bytes; see comment at beginning.
- */
-.org 0x1000
-ENTRY(swapper_pg_dir)
-	.fill 1024,4,0
-
-.org 0x2000
-ENTRY(empty_zero_page)
-	.fill 4096,1,0
-
-.org 0x3000
-/*
- * Real beginning of normal "text" segment
- */
-ENTRY(stext)
-ENTRY(_stext)
-
-/*
- * This starts the data section. Note that the above is all
- * in the text section because it has alignment requirements
- * that we cannot fulfill any other way.
- */
-.data
-
-/*
  * The boot_gdt_table must mirror the equivalent in setup.S and is
  * used only for booting.
  */
diff -urN linux-bk/arch/i386/kernel/vmlinux.lds.S linux/arch/i386/kernel/vmlinux.lds.S
--- linux-bk/arch/i386/kernel/vmlinux.lds.S	2004-04-01 10:17:01.000000000 -0500
+++ linux/arch/i386/kernel/vmlinux.lds.S	2004-04-02 09:25:14.000000000 -0500
@@ -105,7 +105,10 @@
   /* freed after init ends here */
 	
   __bss_start = .;		/* BSS */
-  .bss : { *(.bss) }
+  .bss : {
+	*(.bss.page_aligned)
+	*(.bss)
+  }
   . = ALIGN(4);
   __bss_stop = .; 
 

Re: [PATCH] more i386 head.S cleanups

[Matt Mackall]

On Sat, Apr 03, 2004 at 09:32:07AM -0500, Brian Gerst wrote:
> - Move empty_zero_page and swapper_pg_dir to BSS.  This requires that 
> BSS is cleared earlier, but reclaims over 3k that was lost due to page 
> alignment.
> - Move stack_start, ready, and int_msg, boot_gdt_descr, idt_descr, and 
> cpu_gdt_descr to .data.  They were interfering with disassembly while in 
> .text.

Nice. Do you mean 3k here or 0x3000? 

On a related note, I've been sitting on this patch which reorders the
bootstrap code so we can free most of it once we're up:


drop i386 bootstrap text

From: Zwane Mwaikambo <zwane@arm.linux.org.uk>
Subject: [PATCH][2.6-tiny] free i386 bootstrap text

Shuffle the idt/gdt descriptors around so that we get a full page and then
free it.

Index: linux-2.6.1-rc1-tiny2/arch/i386/kernel/head.S
===================================================================
RCS file: /home/cvsroot/linux-2.6.1-rc1-tiny2/arch/i386/kernel/head.S,v
retrieving revision 1.1.1.1
diff -u -p -B -r1.1.1.1 head.S


 tiny-mpm/arch/i386/kernel/head.S |   40 +++++++++++++++++++--------------------
 tiny-mpm/arch/i386/mm/init.c     |    8 +++++++
 2 files changed, 28 insertions(+), 20 deletions(-)

diff -puN arch/i386/kernel/head.S~bootstrap-free arch/i386/kernel/head.S
--- tiny/arch/i386/kernel/head.S~bootstrap-free	2004-03-20 12:14:33.000000000 -0600
+++ tiny-mpm/arch/i386/kernel/head.S	2004-03-20 12:14:33.000000000 -0600
@@ -386,6 +386,26 @@ ignore_int:
 	iret
 
 /*
+ * swapper_pg_dir is the main page directory, address 0x00101000
+ *
+ * This is initialized to create an identity-mapping at 0 (for bootup
+ * purposes) and another mapping at virtual address PAGE_OFFSET.  The
+ * values put here should be all invalid (zero); the valid
+ * entries are created dynamically at boot time.
+ *
+ * The code creates enough page tables to map 0-_end, the page tables
+ * themselves, plus INIT_MAP_BEYOND_END bytes; see comment at beginning.
+ */
+.org 0x1000
+ENTRY(swapper_pg_dir)
+	.fill 1024,4,0
+
+.org 0x2000
+ENTRY(empty_zero_page)
+	.fill 4096,1,0
+
+.org 0x3000
+/*
  * The IDT and GDT 'descriptors' are a strange 48-bit object
  * only used by the lidt and lgdt instructions. They are not
  * like usual segment descriptors - they consist of a 16-bit
@@ -417,26 +437,6 @@ cpu_gdt_descr:
 	.fill NR_CPUS-1,8,0		# space for the other GDT descriptors
 
 /*
- * swapper_pg_dir is the main page directory, address 0x00101000
- *
- * This is initialized to create an identity-mapping at 0 (for bootup
- * purposes) and another mapping at virtual address PAGE_OFFSET.  The
- * values put here should be all invalid (zero); the valid
- * entries are created dynamically at boot time.
- *
- * The code creates enough page tables to map 0-_end, the page tables
- * themselves, plus INIT_MAP_BEYOND_END bytes; see comment at beginning.
- */
-.org 0x1000
-ENTRY(swapper_pg_dir)
-	.fill 1024,4,0
-
-.org 0x2000
-ENTRY(empty_zero_page)
-	.fill 4096,1,0
-
-.org 0x3000
-/*
  * Real beginning of normal "text" segment
  */
 ENTRY(stext)
diff -puN arch/i386/mm/init.c~bootstrap-free arch/i386/mm/init.c
--- tiny/arch/i386/mm/init.c~bootstrap-free	2004-03-20 12:14:33.000000000 -0600
+++ tiny-mpm/arch/i386/mm/init.c	2004-03-20 12:14:33.000000000 -0600
@@ -586,6 +586,14 @@ void free_initmem(void)
 		free_page(addr);
 		totalram_pages++;
 	}
+
+	/* free early bootstrap code in head.S */
+	addr = (unsigned long)&_text;
+	ClearPageReserved(virt_to_page(addr));
+	set_page_count(virt_to_page(addr), 1);
+	free_page(addr);
+	totalram_pages++;
+
 	printk (KERN_INFO "Freeing unused kernel memory: %dk freed\n", (__init_end - __init_begin) >> 10);
 }
 

_



-- 
Matt Mackall : http://www.selenic.com : Linux development and consulting

Re: [PATCH] more i386 head.S cleanups

[Brian Gerst]

Matt Mackall wrote:
> On Sat, Apr 03, 2004 at 09:32:07AM -0500, Brian Gerst wrote:
> 
>>- Move empty_zero_page and swapper_pg_dir to BSS.  This requires that 
>>BSS is cleared earlier, but reclaims over 3k that was lost due to page 
>>alignment.
>>- Move stack_start, ready, and int_msg, boot_gdt_descr, idt_descr, and 
>>cpu_gdt_descr to .data.  They were interfering with disassembly while in 
>>.text.
> 
> 
> Nice. Do you mean 3k here or 0x3000? 

3 kilobytes apporiximately.  Most of the first page was wasted because 
swapper_pg_dir aligns to the start of the next page.

> 
> On a related note, I've been sitting on this patch which reorders the
> bootstrap code so we can free most of it once we're up:
> 

The bootstrap page tables can easily be reclaimed if large pages are 
used, but the bootstrap code needs more care, especially with hotplug cpus.

--
				Brian Gerst

Re: [PATCH] more i386 head.S cleanups

[Zwane Mwaikambo]

On Sat, 3 Apr 2004, Brian Gerst wrote:

> > On a related note, I've been sitting on this patch which reorders the
> > bootstrap code so we can free most of it once we're up:
> >
>
> The bootstrap page tables can easily be reclaimed if large pages are
> used, but the bootstrap code needs more care, especially with hotplug cpus.

This would be fine for hotplug cpus as well, we never hit that path again.

Re: [PATCH] more i386 head.S cleanups

[Coywolf Qi Hunt]

Brian Gerst wrote:

> - Move empty_zero_page and swapper_pg_dir to BSS.  This requires that 
> BSS is cleared earlier, but reclaims over 3k that was lost due to page 
> alignment.
> - Move stack_start, ready, and int_msg, boot_gdt_descr, idt_descr, and 
> cpu_gdt_descr to .data.  They were interfering with disassembly while in 
> .text.
> 

Interesting, two years ago in 2002, you sent a patch basically of the
same idea.

http://www.uwsg.iu.edu/hypermail/linux/kernel/0208.0/0850.html


-- 
Coywolf Qi Hunt
Admin of http://GreatCN.org and http://LoveCN.org

Re: [PATCH] more i386 head.S cleanups

[Randy.Dunlap]

On Thu, 15 Apr 2004 22:14:19 +0800 Coywolf Qi Hunt wrote:

| Brian Gerst wrote:
| 
| > - Move empty_zero_page and swapper_pg_dir to BSS.  This requires that 
| > BSS is cleared earlier, but reclaims over 3k that was lost due to page 
| > alignment.
| > - Move stack_start, ready, and int_msg, boot_gdt_descr, idt_descr, and 
| > cpu_gdt_descr to .data.  They were interfering with disassembly while in 
| > .text.
| > 
| 
| Interesting, two years ago in 2002, you sent a patch basically of the
| same idea.
| 
| http://www.uwsg.iu.edu/hypermail/linux/kernel/0208.0/0850.html

and your point is??

--
~Randy

Re: [PATCH] more i386 head.S cleanups

[H. Peter Anvin]

Followup to:  <406ECAE7.1020407@quark.didntduck.org>
By author:    Brian Gerst <bgerst@didntduck.org>
In newsgroup: linux.dev.kernel
> 
> - Move empty_zero_page and swapper_pg_dir to BSS.  This requires that 
> BSS is cleared earlier, but reclaims over 3k that was lost due to page 
> alignment.
> - Move stack_start, ready, and int_msg, boot_gdt_descr, idt_descr, and 
> cpu_gdt_descr to .data.  They were interfering with disassembly while in 
> .text.
> 

Some of them should be in .rodata instead of .data.

.data should only be used for modifiable numbers.

All in all these patches are a good idea; I had the same thoughts when
I was doing the changes to the boot page tables, but didn't want to
change too much at once.

	-hpa