Re: [PATCH] x86/head: Refactor 32-bit pgtable setup

[Ingo Molnar <mingo@kernel.org>]

* Boris Ostrovsky <boris.ostrovsky@oracle.com> wrote:

> The new Xen PVH entry point requires page tables to be setup by the
> kernel since it is entered with paging disabled.
> 
> Pull the common code out of head_32.S so that mk_early_pgtbl_32 can be
> invoked from both the new Xen entry point and the existing startup_32
> code.
> 
> Convert resulting common code to C.
> 
> Signed-off-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
> ---
> This is replacement for https://lkml.org/lkml/2016/10/14/434, with
> assembly code re-written in C as requested by Ingo.
> 
> 
>  arch/x86/include/asm/pgtable_32.h |  32 ++++++++++
>  arch/x86/kernel/head32.c          |  62 +++++++++++++++++++
>  arch/x86/kernel/head_32.S         | 122 +++-----------------------------------
>  3 files changed, 101 insertions(+), 115 deletions(-)

Yeah, so (belatedly) I tried to merge this to latest upstream, via the commit 
below (note the slight edits to the changelog) - but 32-bit defconfig fails to 
build:

  arch/x86/kernel/head_32.S:615: Error: can't resolve `init_thread_union' {*UND* section} - `SIZEOF_PTREGS' {*UND* section}

Thanks,

	Ingo

=======================>
>From 071878c634fda7d9be3f015f05a89f5468e7e2e4 Mon Sep 17 00:00:00 2001
From: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Date: Thu, 8 Dec 2016 11:44:31 -0500
Subject: [PATCH] x86/boot/32: Convert the 32-bit pgtable setup code from assembly to C

The new Xen PVH entry point requires page tables to be set up by the
kernel since it is entered with paging disabled.

Pull the common code out of head_32.S so that mk_early_pgtbl_32() can be
invoked from both the new Xen entry point and the existing startup_32()
code.

Convert resulting common code to C.

Signed-off-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Brian Gerst <brgerst@gmail.com>
Cc: Denys Vlasenko <dvlasenk@redhat.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Josh Poimboeuf <jpoimboe@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: matt@codeblueprint.co.uk
Cc: xen-devel@lists.xenproject.org
Link: http://lkml.kernel.org/r/1481215471-9639-1-git-send-email-boris.ostrovsky@oracle.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
---
 arch/x86/include/asm/pgtable_32.h |  32 ++++++++++
 arch/x86/kernel/head32.c          |  62 +++++++++++++++++++
 arch/x86/kernel/head_32.S         | 122 +++-----------------------------------
 3 files changed, 101 insertions(+), 115 deletions(-)

diff --git a/arch/x86/include/asm/pgtable_32.h b/arch/x86/include/asm/pgtable_32.h
index b6c0b404898a..fbc73360aea0 100644
--- a/arch/x86/include/asm/pgtable_32.h
+++ b/arch/x86/include/asm/pgtable_32.h
@@ -27,6 +27,7 @@ struct vm_area_struct;
 
 extern pgd_t swapper_pg_dir[1024];
 extern pgd_t initial_page_table[1024];
+extern pmd_t initial_pg_pmd[];
 
 static inline void pgtable_cache_init(void) { }
 static inline void check_pgt_cache(void) { }
@@ -75,4 +76,35 @@ do {						\
 #define kern_addr_valid(kaddr)	(0)
 #endif
 
+/*
+ * This is how much memory in addition to the memory covered up to
+ * and including _end we need mapped initially.
+ * We need:
+ *     (KERNEL_IMAGE_SIZE/4096) / 1024 pages (worst case, non PAE)
+ *     (KERNEL_IMAGE_SIZE/4096) / 512 + 4 pages (worst case for PAE)
+ *
+ * Modulo rounding, each megabyte assigned here requires a kilobyte of
+ * memory, which is currently unreclaimed.
+ *
+ * This should be a multiple of a page.
+ *
+ * KERNEL_IMAGE_SIZE should be greater than pa(_end)
+ * and small than max_low_pfn, otherwise will waste some page table entries
+ */
+#if PTRS_PER_PMD > 1
+#define PAGE_TABLE_SIZE(pages) (((pages) / PTRS_PER_PMD) + PTRS_PER_PGD)
+#else
+#define PAGE_TABLE_SIZE(pages) ((pages) / PTRS_PER_PGD)
+#endif
+
+/*
+ * Number of possible pages in the lowmem region.
+ *
+ * We shift 2 by 31 instead of 1 by 32 to the left in order to avoid a
+ * gas warning about overflowing shift count when gas has been compiled
+ * with only a host target support using a 32-bit type for internal
+ * representation.
+ */
+#define LOWMEM_PAGES ((((2<<31) - __PAGE_OFFSET) >> PAGE_SHIFT))
+
 #endif /* _ASM_X86_PGTABLE_32_H */
diff --git a/arch/x86/kernel/head32.c b/arch/x86/kernel/head32.c
index f16c55bfc090..e5fb436a6548 100644
--- a/arch/x86/kernel/head32.c
+++ b/arch/x86/kernel/head32.c
@@ -49,3 +49,65 @@ asmlinkage __visible void __init i386_start_kernel(void)
 
 	start_kernel();
 }
+
+/*
+ * Initialize page tables.  This creates a PDE and a set of page
+ * tables, which are located immediately beyond __brk_base.  The variable
+ * _brk_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.
+ *
+ * In PAE mode initial_page_table is statically defined to contain
+ * enough entries to cover the VMSPLIT option (that is the top 1, 2 or 3
+ * entries). The identity mapping is handled by pointing two PGD entries
+ * to the first kernel PMD. Note the upper half of each PMD or PTE are
+ * always zero at this stage.
+ */
+void __init mk_early_pgtbl_32(void)
+{
+#ifdef __pa
+#undef __pa
+#endif
+#define __pa(x)  ((unsigned long)(x) - PAGE_OFFSET)
+	pte_t pte, *ptep;
+	int i;
+	unsigned long *ptr;
+	/* Enough space to fit pagetables for the low memory linear map */
+	const unsigned long limit = __pa(_end) +
+		(PAGE_TABLE_SIZE(LOWMEM_PAGES) << PAGE_SHIFT);
+#ifdef CONFIG_X86_PAE
+	pmd_t pl2, *pl2p = (pmd_t *)__pa(initial_pg_pmd);
+#define SET_PL2(pl2, val)    { (pl2).pmd = (val); }
+#else
+	pgd_t pl2, *pl2p = (pgd_t *)__pa(initial_page_table);
+#define SET_PL2(pl2, val)   { (pl2).pgd = (val); }
+#endif
+
+	ptep = (pte_t *)__pa(__brk_base);
+	pte.pte = PTE_IDENT_ATTR;
+
+	while ((pte.pte & PTE_PFN_MASK) < limit) {
+
+		SET_PL2(pl2, (unsigned long)ptep | PDE_IDENT_ATTR);
+		*pl2p = pl2;
+#ifndef CONFIG_X86_PAE
+		/* Kernel PDE entry */
+		*(pl2p +  ((PAGE_OFFSET >> PGDIR_SHIFT))) = pl2;
+#endif
+		for (i = 0; i < PTRS_PER_PTE; i++) {
+			*ptep = pte;
+			pte.pte += PAGE_SIZE;
+			ptep++;
+		}
+
+		pl2p++;
+	}
+
+	ptr = (unsigned long *)__pa(&max_pfn_mapped);
+	/* Can't use pte_pfn() since it's a call with CONFIG_PARAVIRT */
+	*ptr = (pte.pte & PTE_PFN_MASK) >> PAGE_SHIFT;
+
+	ptr = (unsigned long *)__pa(&_brk_end);
+	*ptr = (unsigned long)ptep + PAGE_OFFSET;
+}
+
diff --git a/arch/x86/kernel/head_32.S b/arch/x86/kernel/head_32.S
index 4e8577d03372..eca1c93c38e8 100644
--- a/arch/x86/kernel/head_32.S
+++ b/arch/x86/kernel/head_32.S
@@ -24,6 +24,7 @@
 #include <asm/nops.h>
 #include <asm/bootparam.h>
 #include <asm/export.h>
+#include <asm/pgtable_32.h>
 
 /* Physical address */
 #define pa(X) ((X) - __PAGE_OFFSET)
@@ -42,43 +43,8 @@
 #define X86_VENDOR_ID	new_cpu_data+CPUINFO_x86_vendor_id
 
 /*
- * This is how much memory in addition to the memory covered up to
- * and including _end we need mapped initially.
- * We need:
- *     (KERNEL_IMAGE_SIZE/4096) / 1024 pages (worst case, non PAE)
- *     (KERNEL_IMAGE_SIZE/4096) / 512 + 4 pages (worst case for PAE)
- *
- * Modulo rounding, each megabyte assigned here requires a kilobyte of
- * memory, which is currently unreclaimed.
- *
- * This should be a multiple of a page.
- *
- * KERNEL_IMAGE_SIZE should be greater than pa(_end)
- * and small than max_low_pfn, otherwise will waste some page table entries
- */
-
-#if PTRS_PER_PMD > 1
-#define PAGE_TABLE_SIZE(pages) (((pages) / PTRS_PER_PMD) + PTRS_PER_PGD)
-#else
-#define PAGE_TABLE_SIZE(pages) ((pages) / PTRS_PER_PGD)
-#endif
-
 #define SIZEOF_PTREGS 17*4
 
-/*
- * Number of possible pages in the lowmem region.
- *
- * We shift 2 by 31 instead of 1 by 32 to the left in order to avoid a
- * gas warning about overflowing shift count when gas has been compiled
- * with only a host target support using a 32-bit type for internal
- * representation.
- */
-LOWMEM_PAGES = (((2<<31) - __PAGE_OFFSET) >> PAGE_SHIFT)
-
-/* Enough space to fit pagetables for the low memory linear map */
-MAPPING_BEYOND_END = PAGE_TABLE_SIZE(LOWMEM_PAGES) << PAGE_SHIFT
-
-/*
  * Worst-case size of the kernel mapping we need to make:
  * a relocatable kernel can live anywhere in lowmem, so we need to be able
  * to map all of lowmem.
@@ -160,90 +126,15 @@ ENTRY(startup_32)
 	call load_ucode_bsp
 #endif
 
-/*
- * Initialize page tables.  This creates a PDE and a set of page
- * tables, which are located immediately beyond __brk_base.  The variable
- * _brk_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.
- */
-#ifdef CONFIG_X86_PAE
-
-	/*
-	 * In PAE mode initial_page_table is statically defined to contain
-	 * enough entries to cover the VMSPLIT option (that is the top 1, 2 or 3
-	 * entries). The identity mapping is handled by pointing two PGD entries
-	 * to the first kernel PMD.
-	 *
-	 * Note the upper half of each PMD or PTE are always zero at this stage.
-	 */
-
-#define KPMDS (((-__PAGE_OFFSET) >> 30) & 3) /* Number of kernel PMDs */
-
-	xorl %ebx,%ebx				/* %ebx is kept at zero */
-
-	movl $pa(__brk_base), %edi
-	movl $pa(initial_pg_pmd), %edx
-	movl $PTE_IDENT_ATTR, %eax
-10:
-	leal PDE_IDENT_ATTR(%edi),%ecx		/* Create PMD entry */
-	movl %ecx,(%edx)			/* Store PMD entry */
-						/* Upper half already zero */
-	addl $8,%edx
-	movl $512,%ecx
-11:
-	stosl
-	xchgl %eax,%ebx
-	stosl
-	xchgl %eax,%ebx
-	addl $0x1000,%eax
-	loop 11b
-
-	/*
-	 * End condition: we must map up to the end + MAPPING_BEYOND_END.
-	 */
-	movl $pa(_end) + MAPPING_BEYOND_END + PTE_IDENT_ATTR, %ebp
-	cmpl %ebp,%eax
-	jb 10b
-1:
-	addl $__PAGE_OFFSET, %edi
-	movl %edi, pa(_brk_end)
-	shrl $12, %eax
-	movl %eax, pa(max_pfn_mapped)
+	/* Create early pagetables. */
+	call  mk_early_pgtbl_32
 
 	/* Do early initialization of the fixmap area */
 	movl $pa(initial_pg_fixmap)+PDE_IDENT_ATTR,%eax
+#ifdef  CONFIG_X86_PAE
+#define KPMDS (((-__PAGE_OFFSET) >> 30) & 3) /* Number of kernel PMDs */
 	movl %eax,pa(initial_pg_pmd+0x1000*KPMDS-8)
-#else	/* Not PAE */
-
-page_pde_offset = (__PAGE_OFFSET >> 20);
-
-	movl $pa(__brk_base), %edi
-	movl $pa(initial_page_table), %edx
-	movl $PTE_IDENT_ATTR, %eax
-10:
-	leal PDE_IDENT_ATTR(%edi),%ecx		/* Create PDE entry */
-	movl %ecx,(%edx)			/* Store identity PDE entry */
-	movl %ecx,page_pde_offset(%edx)		/* Store kernel PDE entry */
-	addl $4,%edx
-	movl $1024, %ecx
-11:
-	stosl
-	addl $0x1000,%eax
-	loop 11b
-	/*
-	 * End condition: we must map up to the end + MAPPING_BEYOND_END.
-	 */
-	movl $pa(_end) + MAPPING_BEYOND_END + PTE_IDENT_ATTR, %ebp
-	cmpl %ebp,%eax
-	jb 10b
-	addl $__PAGE_OFFSET, %edi
-	movl %edi, pa(_brk_end)
-	shrl $12, %eax
-	movl %eax, pa(max_pfn_mapped)
-
-	/* Do early initialization of the fixmap area */
-	movl $pa(initial_pg_fixmap)+PDE_IDENT_ATTR,%eax
+#else
 	movl %eax,pa(initial_page_table+0xffc)
 #endif
 
@@ -666,6 +557,7 @@ ENTRY(setup_once_ref)
 __PAGE_ALIGNED_BSS
 	.align PAGE_SIZE
 #ifdef CONFIG_X86_PAE
+.globl initial_pg_pmd
 initial_pg_pmd:
 	.fill 1024*KPMDS,4,0
 #else