Skip to contentSkip to conversation

linux/i386: make kernel tolerate variable start of hypervisor hole

PATCH New Jan Beulich 1 participant 1 message Nov 10, 2006

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 linux/i386: make kernel tolerate variable start of hypervisor hole

This is the kernel side of the needed change. The hypervisor side, as it's useful only with 32-on-64, will come with those patches. On an 8 Gb machine, I saw a lowmem increase of 160 Mb.
I didn't go the capabilities route, but rather introduced a separate note that specifies the lower boundary the kernel wants to tolerate for the hypervisor hole. That way one might one day support more than 16 Gb for x86-32/PAE (at the obvious price of further reducing lowmem in the guest).

arch/i386/kernel/head-xen.S +2 -1
Index: head-2006-11-06/arch/i386/kernel/head-xen.S
===================================================================
@@ -9,7 +9,7 @@
 #include <asm/page.h>
 #include <asm/thread_info.h>
 #include <asm/asm-offsets.h>
-#include <xen/interface/arch-x86_32.h>
+#include <xen/interface/xen.h>
 #include <xen/interface/elfnote.h>
 
 /*
@@ -192,6 +192,7 @@ ENTRY(cpu_gdt_table)
 #endif /* !CONFIG_XEN_COMPAT_030002 */
 	ELFNOTE(Xen, XEN_ELFNOTE_ENTRY,          .long,  startup_32)
 	ELFNOTE(Xen, XEN_ELFNOTE_HYPERCALL_PAGE, .long,  hypercall_page)
+	ELFNOTE(Xen, XEN_ELFNOTE_HV_START_LOW,   .long,  HYPERVISOR_VIRT_START)
 	ELFNOTE(Xen, XEN_ELFNOTE_FEATURES,       .asciz,

"writable_page_tables|writable_descriptor_tables|auto_translated_physmap|pae_pgdir_above_4gb|supervisor_mode_kernel")
#ifdef CONFIG_X86_PAE
ELFNOTE(Xen, XEN_ELFNOTE_PAE_MODE, .asciz, "yes")

include/asm-i386/mach-xen/setup_arch_post.h +4 -2
Index: head-2006-11-06/include/asm-i386/mach-xen/setup_arch_post.h
===================================================================
@@ -92,8 +92,10 @@ void __init machine_specific_arch_setup(
 #endif
 
 	if (HYPERVISOR_xen_version(XENVER_platform_parameters,
-				   &pp) == 0)
-		set_fixaddr_top(pp.virt_start - PAGE_SIZE);
+				   &pp) == 0) {
+		hypervisor_virt_start = pp.virt_start;
+		set_fixaddr_top();
+	}
 
 	machine_to_phys_mapping = (unsigned long *)MACH2PHYS_VIRT_START;
 	machine_to_phys_nr_ents = MACH2PHYS_NR_ENTRIES;
arch/i386/mm/fault-xen.c +1 -7
Index: head-2006-11-06/arch/i386/mm/fault-xen.c
===================================================================
@@ -282,12 +282,6 @@ static int spurious_fault(struct pt_regs
 	pmd_t *pmd;
 	pte_t *pte;
 
-#ifdef CONFIG_XEN
-	/* Faults in hypervisor area are never spurious. */
-	if (address >= HYPERVISOR_VIRT_START)
-		return 0;
-#endif
-
 	/* Reserved-bit violation or user access to kernel space? */
 	if (error_code & 0x0c)
 		return 0;
@@ -372,7 +366,7 @@ fastcall void __kprobes do_page_fault(st
 	if (unlikely(address >= TASK_SIZE)) { 
 #ifdef CONFIG_XEN
 		/* Faults in hypervisor area can never be patched up. */
-		if (address >= HYPERVISOR_VIRT_START)
+		if (address >= hypervisor_virt_start)
 			goto bad_area_nosemaphore;
 #endif
 		if (!(error_code & 5))
arch/i386/mm/init-xen.c +3 -3
Index: head-2006-11-06/arch/i386/mm/init-xen.c
===================================================================
@@ -130,7 +130,7 @@ static void __init page_table_range_init
 		pud = pud_offset(pgd, vaddr);
 		pmd = pmd_offset(pud, vaddr);
 		for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end); pmd++, pmd_idx++) {
-			if (vaddr < HYPERVISOR_VIRT_START && pmd_none(*pmd)) 
+			if (vaddr < hypervisor_virt_start && pmd_none(*pmd))
 				one_page_table_init(pmd);
 
 			vaddr += PMD_SIZE;
@@ -187,7 +187,7 @@ static void __init kernel_physical_mappi
 		pmd += pmd_idx;
 		for (; pmd_idx < PTRS_PER_PMD && pfn < max_low_pfn; pmd++, pmd_idx++) {
 			unsigned int address = pfn * PAGE_SIZE + PAGE_OFFSET;
-			if (address >= HYPERVISOR_VIRT_START)
+			if (address >= hypervisor_virt_start)
 				continue;
 
 			/* Map with big pages if possible, otherwise create normal page tables. */
@@ -410,7 +410,7 @@ static void __init pagetable_init (void)
 	 * created - mappings will be set by set_fixmap():
 	 */
 	vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
-	page_table_range_init(vaddr, 0, pgd_base);
+	page_table_range_init(vaddr, hypervisor_virt_start, pgd_base);
 
 	permanent_kmaps_init(pgd_base);
 }
arch/i386/mm/pgtable-xen.c +7 -6
Index: head-2006-11-06/arch/i386/mm/pgtable-xen.c
===================================================================
@@ -186,9 +186,16 @@ void set_pmd_pfn(unsigned long vaddr, un
 }
 
 static int nr_fixmaps = 0;
+unsigned long hypervisor_virt_start = HYPERVISOR_VIRT_START;
 unsigned long __FIXADDR_TOP = (HYPERVISOR_VIRT_START - 2 * PAGE_SIZE);
 EXPORT_SYMBOL(__FIXADDR_TOP);
 
+void __init set_fixaddr_top()
+{
+	BUG_ON(nr_fixmaps > 0);
+	__FIXADDR_TOP = hypervisor_virt_start - 2 * PAGE_SIZE;
+}
+
 void __set_fixmap (enum fixed_addresses idx, maddr_t phys, pgprot_t flags)
 {
 	unsigned long address = __fix_to_virt(idx);
@@ -211,12 +218,6 @@ void __set_fixmap (enum fixed_addresses 
 	nr_fixmaps++;
 }
 
-void set_fixaddr_top(unsigned long top)
-{
-	BUG_ON(nr_fixmaps > 0);
-	__FIXADDR_TOP = top - PAGE_SIZE;
-}
-
 pte_t *pte_alloc_one_kernel(struct mm_struct *mm, unsigned long address)
 {
 	pte_t *pte = (pte_t *)__get_free_page(GFP_KERNEL|__GFP_REPEAT|__GFP_ZERO);
include/asm-i386/mach-xen/asm/fixmap.h +1 -1
Index: head-2006-11-06/include/asm-i386/mach-xen/asm/fixmap.h
===================================================================
@@ -98,7 +98,7 @@ enum fixed_addresses {
 extern void __set_fixmap(enum fixed_addresses idx,
 					maddr_t phys, pgprot_t flags);
 
-extern void set_fixaddr_top(unsigned long top);
+extern void set_fixaddr_top(void);
 
 #define set_fixmap(idx, phys) \
 		__set_fixmap(idx, phys, PAGE_KERNEL)
include/asm-i386/mach-xen/asm/hypervisor.h +4 -0
Index: head-2006-11-06/include/asm-i386/mach-xen/asm/hypervisor.h
===================================================================
@@ -56,6 +56,10 @@
 
 extern shared_info_t *HYPERVISOR_shared_info;
 
+#ifdef CONFIG_X86_32
+extern unsigned long hypervisor_virt_start;
+#endif
+
 /* arch/xen/i386/kernel/setup.c */
 extern start_info_t *xen_start_info;
 #ifdef CONFIG_XEN_PRIVILEGED_GUEST
include/asm-i386/mach-xen/asm/pgtable-2level-defs.h +0 -1
Index: head-2006-11-06/include/asm-i386/mach-xen/asm/pgtable-2level-defs.h
===================================================================
@@ -9,7 +9,6 @@
 
 #define PGDIR_SHIFT	22
 #define PTRS_PER_PGD	1024
-#define PTRS_PER_PGD_NO_HV	(HYPERVISOR_VIRT_START >> PGDIR_SHIFT)
 
 /*
  * the i386 is two-level, so we don't really have any
include/asm-i386/mach-xen/asm/pgtable-3level-defs.h +0 -1
Index: head-2006-11-06/include/asm-i386/mach-xen/asm/pgtable-3level-defs.h
===================================================================
@@ -8,7 +8,6 @@
  */
 #define PGDIR_SHIFT	30
 #define PTRS_PER_PGD	4
-#define PTRS_PER_PGD_NO_HV 4
 
 /*
  * PMD_SHIFT determines the size of the area a middle-level
../xen/include/public/elfnote.h +9 -0
Index: ./../xen/include/public/elfnote.h
===================================================================
@@ -138,6 +138,15 @@
  */
 #define XEN_ELFNOTE_BSD_SYMTAB    11
 
+/*
+ * The lowest address the hypervisor hole can begin at (numeric).
+ *
+ * This must not be set higher than HYPERVISOR_VIRT_START. Its presence
+ * also indicates to the hypervisor that the kernel can deal with the
+ * hole starting at a higher address.
+ */
+#define XEN_ELFNOTE_HV_START_LOW  12
+
 #endif /* __XEN_PUBLIC_ELFNOTE_H__ */
 
 /*