Skip to contentSkip to conversation

lguest: PAE support

PATCH Applied Matias Zabaljauregui 2 participants 5 messages Jun 5, 2009 – Jun 11, 2009 3 messages not in the archive

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 lguest: PAE support

Hi, this version requires that host and guest have the same PAE status. NX cap is not offered to the guest, yet.

Thanks,
Matias

Lguest PAE support

---
 Documentation/lguest/lguest.txt     |    1 -
 arch/x86/include/asm/lguest.h       |    7 +-
 arch/x86/include/asm/lguest_hcall.h |    3 +-
 arch/x86/lguest/Kconfig             |    1 -
 arch/x86/lguest/boot.c              |   79 +++++++-
 drivers/lguest/Kconfig              |    2 +-
 drivers/lguest/hypercalls.c         |   10 +
 drivers/lguest/lg.h                 |    5 +
 drivers/lguest/page_tables.c        |  356 +++++++++++++++++++++++++++++++----
 9 files changed, 410 insertions(+), 54 deletions(-)
Documentation/lguest/lguest.txt +0 -1
index 28c7473..efb3a6a 100644
@@ -37,7 +37,6 @@ Running Lguest:
      "Paravirtualized guest support" = Y
         "Lguest guest support" = Y
      "High Memory Support" = off/4GB
-     "PAE (Physical Address Extension) Support" = N
      "Alignment value to which kernel should be aligned" = 0x100000
         (CONFIG_PARAVIRT=y, CONFIG_LGUEST_GUEST=y, CONFIG_HIGHMEM64G=n and
          CONFIG_PHYSICAL_ALIGN=0x100000)
arch/x86/include/asm/lguest.h +6 -1
index 1caf576..313389c 100644
@@ -17,8 +17,13 @@
 /* Pages for switcher itself, then two pages per cpu */
 #define TOTAL_SWITCHER_PAGES (SHARED_SWITCHER_PAGES + 2 * nr_cpu_ids)
 
-/* We map at -4M for ease of mapping into the guest (one PTE page). */
+/* We map at -4M (-2M when PAE is activated) for ease of mapping
+ * into the guest (one PTE page). */
+#ifdef CONFIG_X86_PAE
+#define SWITCHER_ADDR 0xFFE00000
+#else
 #define SWITCHER_ADDR 0xFFC00000
+#endif
 
 /* Found in switcher.S */
 extern unsigned long default_idt_entries[];
arch/x86/include/asm/lguest_hcall.h +2 -1
index 553810f..7ba1d81 100644
@@ -12,6 +12,7 @@
 #define LHCALL_TS		8
 #define LHCALL_SET_CLOCKEVENT	9
 #define LHCALL_HALT		10
+#define LHCALL_SET_PMD		13
 #define LHCALL_SET_PTE		14
 #define LHCALL_SET_PGD		15
 #define LHCALL_LOAD_TLS		16
@@ -32,7 +33,7 @@
  * operations?  There are two ways: the direct way is to make a "hypercall",
  * to make requests of the Host Itself.
  *
- * We use the KVM hypercall mechanism. Eighteen hypercalls are
+ * We use the KVM hypercall mechanism. Sixteen hypercalls are
  * available: the hypercall number is put in the %eax register, and the
  * arguments (when required) are placed in %ebx, %ecx, %edx and %esi.
  * If a return value makes sense, it's returned in %eax.
arch/x86/lguest/Kconfig +0 -1
index 8dab8f7..3871804 100644
@@ -2,7 +2,6 @@ config LGUEST_GUEST
 	bool "Lguest guest support"
 	select PARAVIRT
 	depends on X86_32
-	depends on !X86_PAE
 	select VIRTIO
 	select VIRTIO_RING
 	select VIRTIO_CONSOLE
arch/x86/lguest/boot.c +70 -9

144 lines not on this page · raw message

drivers/lguest/Kconfig +1 -1
index a3d3cba..8f63845 100644
@@ -1,6 +1,6 @@
 config LGUEST
 	tristate "Linux hypervisor example code"
-	depends on X86_32 && EXPERIMENTAL && !X86_PAE && FUTEX
+	depends on X86_32 && EXPERIMENTAL && FUTEX
 	select HVC_DRIVER
 	---help---
 	  This is a very simple module which allows you to run
drivers/lguest/hypercalls.c +10 -0
index 1cc3377..5415da1 100644
@@ -73,11 +73,21 @@ static void do_hcall(struct lg_cpu *cpu, struct hcall_args *args)
 		guest_set_stack(cpu, args->arg1, args->arg2, args->arg3);
 		break;
 	case LHCALL_SET_PTE:
+#ifdef CONFIG_X86_PAE
+		guest_set_pte(cpu, args->arg1, args->arg2,
+				__pte(args->arg3 | (u64)args->arg4 << 32));
+#else
 		guest_set_pte(cpu, args->arg1, args->arg2, __pte(args->arg3));
+#endif
 		break;
 	case LHCALL_SET_PGD:
 		guest_set_pgd(cpu->lg, args->arg1, args->arg2);
 		break;
+#ifdef CONFIG_X86_PAE
+	case LHCALL_SET_PMD:
+		guest_set_pmd(cpu->lg, args->arg1, args->arg2);
+		break;
+#endif
 	case LHCALL_SET_CLOCKEVENT:
 		guest_set_clockevent(cpu, args->arg1);
 		break;
drivers/lguest/lg.h +5 -0
index 5e7a27d..ce56ef8 100644
@@ -137,6 +137,8 @@ int run_guest(struct lg_cpu *cpu, unsigned long __user *user);
  * in the kernel. */
 #define pgd_flags(x)	(pgd_val(x) & ~PAGE_MASK)
 #define pgd_pfn(x)	(pgd_val(x) >> PAGE_SHIFT)
+#define pmd_flags(x)    (pmd_val(x) & ~PAGE_MASK)
+#define pmd_pfn(x)	(pmd_val(x) >> PAGE_SHIFT)
 
 /* interrupts_and_traps.c: */
 void maybe_do_interrupt(struct lg_cpu *cpu);
@@ -169,6 +171,9 @@ int init_guest_pagetable(struct lguest *lg);
 void free_guest_pagetable(struct lguest *lg);
 void guest_new_pagetable(struct lg_cpu *cpu, unsigned long pgtable);
 void guest_set_pgd(struct lguest *lg, unsigned long gpgdir, u32 i);
+#ifdef CONFIG_X86_PAE
+void guest_set_pmd(struct lguest *lg, unsigned long gpgdir, u32 i);
+#endif
 void guest_pagetable_clear_all(struct lg_cpu *cpu);
 void guest_pagetable_flush_user(struct lg_cpu *cpu);
 void guest_set_pte(struct lg_cpu *cpu, unsigned long gpgdir,
drivers/lguest/page_tables.c +316 -40

575 lines not on this page · raw message

Signature · 1 line
1.6.0.4
#1 lore raw
Quoted 2 lines · Matias Zabaljauregui

Hi, this version requires that host and guest have the same PAE status. NX cap is not offered to the guest, yet.

Thanks, applied!

I extracted the following parts for moving into your previous "native ops" patch:

arch/x86/lguest/boot.c +4 -4
@@ -519,7 +519,7 @@ static void lguest_pte_update(struct mm_
 static void lguest_set_pte_at(struct mm_struct *mm, unsigned long addr,
 			      pte_t *ptep, pte_t pteval)
 {
-	*ptep = pteval;
+	native_set_pte(ptep, pteval);
 	lguest_pte_update(mm, addr, ptep);
 }
 
@@ -528,9 +528,9 @@ static void lguest_set_pte_at(struct mm_
  * changed. */
 static void lguest_set_pmd(pmd_t *pmdp, pmd_t pmdval)
 {
-	*pmdp = pmdval;
+	native_set_pmd(pmdp, pmdval);
 	lazy_hcall2(LHCALL_SET_PMD, __pa(pmdp) & PAGE_MASK,
-		   (__pa(pmdp) & (PAGE_SIZE - 1)) / 4);
+		   (__pa(pmdp) & (PAGE_SIZE - 1)) / sizeof(pmd_t));
 }
 
 /* There are a couple of legacy places where the kernel sets a PTE, but we
@@ -544,7 +544,7 @@ static void lguest_set_pmd(pmd_t *pmdp, 
  * which brings boot back to 0.25 seconds. */
 static void lguest_set_pte(pte_t *ptep, pte_t pteval)
 {
-	*ptep = pteval;
+	native_set_pte(ptep, pteval);
 	if (cr3_changed)
 		lazy_hcall1(LHCALL_FLUSH_TLB, 1);
 }
drivers/lguest/page_tables.c +3 -2
@@ -726,8 +726,9 @@ void map_switcher_in_guest(struct lg_cpu
 	 * page is already mapped there, we don't have to copy them out
 	 * again. */
 	pfn = __pa(cpu->regs_page) >> PAGE_SHIFT;
-	regs_pte = pfn_pte(pfn, __pgprot(__PAGE_KERNEL));
-	switcher_pte_page[(unsigned long)pages/PAGE_SIZE%PTRS_PER_PTE] = regs_pte;
+	native_set_pte(&regs_pte, pfn_pte(pfn, PAGE_KERNEL));
+	native_set_pte(&switcher_pte_page[pte_index((unsigned long)pages)],
+			regs_pte);
 }
 /*:*/

A message that is not in the archive (sent off-list or not yet received): [email protected]

#3 lore raw
Quoted 8 lines · Rusty Russell
Quoted 2 lines · Matias Zabaljauregui

Hi, this version requires that host and guest have the same PAE status. NX cap is not offered to the guest, yet.

Thanks, applied!

I extracted the following parts for moving into your previous "native ops" patch:

great, thank you (forgive my laziness)

Quoted 49 lines
arch/x86/lguest/boot.c +4 -4
@@ -519,7 +519,7 @@ static void lguest_pte_update(struct mm_
 static void lguest_set_pte_at(struct mm_struct *mm, unsigned long addr,
 			      pte_t *ptep, pte_t pteval)
 {
-	*ptep = pteval;
+	native_set_pte(ptep, pteval);
 	lguest_pte_update(mm, addr, ptep);
 }
 
@@ -528,9 +528,9 @@ static void lguest_set_pte_at(struct mm_
  * changed. */
 static void lguest_set_pmd(pmd_t *pmdp, pmd_t pmdval)
 {
-	*pmdp = pmdval;
+	native_set_pmd(pmdp, pmdval);
 	lazy_hcall2(LHCALL_SET_PMD, __pa(pmdp) & PAGE_MASK,
-		   (__pa(pmdp) & (PAGE_SIZE - 1)) / 4);
+		   (__pa(pmdp) & (PAGE_SIZE - 1)) / sizeof(pmd_t));
 }
 
 /* There are a couple of legacy places where the kernel sets a PTE, but we
@@ -544,7 +544,7 @@ static void lguest_set_pmd(pmd_t *pmdp, 
  * which brings boot back to 0.25 seconds. */
 static void lguest_set_pte(pte_t *ptep, pte_t pteval)
 {
-	*ptep = pteval;
+	native_set_pte(ptep, pteval);
 	if (cr3_changed)
 		lazy_hcall1(LHCALL_FLUSH_TLB, 1);
 }
drivers/lguest/page_tables.c +3 -2
@@ -726,8 +726,9 @@ void map_switcher_in_guest(struct lg_cpu
 	 * page is already mapped there, we don't have to copy them out
 	 * again. */
 	pfn = __pa(cpu->regs_page) >> PAGE_SHIFT;
-	regs_pte = pfn_pte(pfn, __pgprot(__PAGE_KERNEL));
-	switcher_pte_page[(unsigned long)pages/PAGE_SIZE%PTRS_PER_PTE] = regs_pte;
+	native_set_pte(&regs_pte, pfn_pte(pfn, PAGE_KERNEL));
+	native_set_pte(&switcher_pte_page[pte_index((unsigned long)pages)],
+			regs_pte);
 }
 /*:*/
 

A message that is not in the archive (sent off-list or not yet received): 1244397067.21275.33.camel@zetabook

#5 lore raw

Debugged why guest was crashing immediately with unhandled page fault, also fixed a potential future problem. Please review.

lguest: PAE fixes

1) j wasn't initialized in setup_pagetables, so they weren't set up for me causing immediate guest crashes.

2) gpte_addr should not re-read the pmd from the Guest. Especially not BUG_ON() based on the value. If we ever supported SMP guests, they could trigger that. And the Launcher could also trigger it (tho currently root-only).

drivers/lguest/page_tables.c +21 -17
@@ -154,26 +154,25 @@ static unsigned long gpmd_addr(pgd_t gpg
 	BUG_ON(!(pgd_flags(gpgd) & _PAGE_PRESENT));
 	return gpage + pmd_index(vaddr) * sizeof(pmd_t);
 }
-#endif
 
 static unsigned long gpte_addr(struct lg_cpu *cpu,
+			       pmd_t gpmd, unsigned long vaddr)
+{
+	unsigned long gpage = pmd_pfn(gpmd) << PAGE_SHIFT;
+
+	BUG_ON(!(pmd_flags(gpmd) & _PAGE_PRESENT));
+	return gpage + pte_index(vaddr) * sizeof(pte_t);
+}
+#else
+static unsigned long gpte_addr(struct lg_cpu *cpu,
 				pgd_t gpgd, unsigned long vaddr)
 {
-#ifdef CONFIG_X86_PAE
-	pmd_t gpmd;
-#endif
-	unsigned long gpage;
+	unsigned long gpage = pgd_pfn(gpgd) << PAGE_SHIFT;
 
 	BUG_ON(!(pgd_flags(gpgd) & _PAGE_PRESENT));
-#ifdef CONFIG_X86_PAE
-	gpmd = lgread(cpu, gpmd_addr(gpgd, vaddr), pmd_t);
-	gpage = pmd_pfn(gpmd) << PAGE_SHIFT;
-	BUG_ON(!(pmd_flags(gpmd) & _PAGE_PRESENT));
-#else
-	gpage = pgd_pfn(gpgd) << PAGE_SHIFT;
-#endif
 	return gpage + pte_index(vaddr) * sizeof(pte_t);
 }
+#endif
 /*:*/
 
 /*M:014 get_pfn is slow: we could probably try to grab batches of pages here as
@@ -339,10 +338,15 @@ bool demand_page(struct lg_cpu *cpu, uns
 		 * number in the shadow PMD is the page we just allocated. */
 		native_set_pmd(spmd, __pmd(__pa(ptepage) | pmd_flags(gpmd)));
 	}
-#endif
+
+	/* OK, now we look at the lower level in the Guest page table: keep its
+	 * address, because we might update it later. */
+	gpte_ptr = gpte_addr(cpu, gpmd, vaddr);
+#else
 	/* OK, now we look at the lower level in the Guest page table: keep its
 	 * address, because we might update it later. */
 	gpte_ptr = gpte_addr(cpu, gpgd, vaddr);
+#endif
 	gpte = lgread(cpu, gpte_ptr, pte_t);
 
 	/* If this page isn't in the Guest page tables, we can't page it in. */
@@ -522,7 +526,6 @@ unsigned long guest_pa(struct lg_cpu *cp
 {
 	pgd_t gpgd;
 	pte_t gpte;
-
 #ifdef CONFIG_X86_PAE
 	pmd_t gpmd;
 #endif
@@ -534,13 +537,14 @@ unsigned long guest_pa(struct lg_cpu *cp
 		return -1UL;
 	}
 
-	gpte = lgread(cpu, gpte_addr(cpu, gpgd, vaddr), pte_t);
 #ifdef CONFIG_X86_PAE
 	gpmd = lgread(cpu, gpmd_addr(gpgd, vaddr), pmd_t);
 	if (!(pmd_flags(gpmd) & _PAGE_PRESENT))
 		kill_guest(cpu, "Bad address %#lx", vaddr);
+	gpte = lgread(cpu, gpte_addr(cpu, gpmd, vaddr), pte_t);
+#else
+	gpte = lgread(cpu, gpte_addr(cpu, gpgd, vaddr), pte_t);
 #endif
-	gpte = lgread(cpu, gpte_addr(cpu, gpgd, vaddr), pte_t);
 	if (!(pte_flags(gpte) & _PAGE_PRESENT))
 		kill_guest(cpu, "Bad address %#lx", vaddr);
 
@@ -847,7 +851,7 @@ static unsigned long setup_pagetables(st
 	/* The top level points to the linear page table pages above.
 	 * We setup the identity and linear mappings here. */
 #ifdef CONFIG_X86_PAE
-	for (i = 0, j; i < mapped_pages && j < PTRS_PER_PMD;
+	for (i = j = 0; i < mapped_pages && j < PTRS_PER_PMD;
 	     i += PTRS_PER_PTE, j++) {
 		native_set_pmd(&pmd, __pmd(((unsigned long)(linear + i)
 		- mem_base) | _PAGE_PRESENT | _PAGE_RW | _PAGE_USER));

A message that is not in the archive (sent off-list or not yet received): [email protected]

#7 lore raw
Quoted 7 lines · Rusty Russell

Debugged why guest was crashing immediately with unhandled page fault, also fixed a potential future problem. Please review.

lguest: PAE fixes

1) j wasn't initialized in setup_pagetables, so they weren't set up for me causing immediate guest crashes.

Oops, don't understand how it worked on my test boxes all these days.

Quoted 6 lines

2) gpte_addr should not re-read the pmd from the Guest. Especially not BUG_ON() based on the value. If we ever supported SMP guests, they could trigger that. And the Launcher could also trigger it (tho currently root-only).

thanks for fixing this as well.

Matias

Quoted 95 lines
drivers/lguest/page_tables.c +21 -17
@@ -154,26 +154,25 @@ static unsigned long gpmd_addr(pgd_t gpg
 	BUG_ON(!(pgd_flags(gpgd) & _PAGE_PRESENT));
 	return gpage + pmd_index(vaddr) * sizeof(pmd_t);
 }
-#endif
 
 static unsigned long gpte_addr(struct lg_cpu *cpu,
+			       pmd_t gpmd, unsigned long vaddr)
+{
+	unsigned long gpage = pmd_pfn(gpmd) << PAGE_SHIFT;
+
+	BUG_ON(!(pmd_flags(gpmd) & _PAGE_PRESENT));
+	return gpage + pte_index(vaddr) * sizeof(pte_t);
+}
+#else
+static unsigned long gpte_addr(struct lg_cpu *cpu,
 				pgd_t gpgd, unsigned long vaddr)
 {
-#ifdef CONFIG_X86_PAE
-	pmd_t gpmd;
-#endif
-	unsigned long gpage;
+	unsigned long gpage = pgd_pfn(gpgd) << PAGE_SHIFT;
 
 	BUG_ON(!(pgd_flags(gpgd) & _PAGE_PRESENT));
-#ifdef CONFIG_X86_PAE
-	gpmd = lgread(cpu, gpmd_addr(gpgd, vaddr), pmd_t);
-	gpage = pmd_pfn(gpmd) << PAGE_SHIFT;
-	BUG_ON(!(pmd_flags(gpmd) & _PAGE_PRESENT));
-#else
-	gpage = pgd_pfn(gpgd) << PAGE_SHIFT;
-#endif
 	return gpage + pte_index(vaddr) * sizeof(pte_t);
 }
+#endif
 /*:*/
 
 /*M:014 get_pfn is slow: we could probably try to grab batches of pages here as
@@ -339,10 +338,15 @@ bool demand_page(struct lg_cpu *cpu, uns
 		 * number in the shadow PMD is the page we just allocated. */
 		native_set_pmd(spmd, __pmd(__pa(ptepage) | pmd_flags(gpmd)));
 	}
-#endif
+
+	/* OK, now we look at the lower level in the Guest page table: keep its
+	 * address, because we might update it later. */
+	gpte_ptr = gpte_addr(cpu, gpmd, vaddr);
+#else
 	/* OK, now we look at the lower level in the Guest page table: keep its
 	 * address, because we might update it later. */
 	gpte_ptr = gpte_addr(cpu, gpgd, vaddr);
+#endif
 	gpte = lgread(cpu, gpte_ptr, pte_t);
 
 	/* If this page isn't in the Guest page tables, we can't page it in. */
@@ -522,7 +526,6 @@ unsigned long guest_pa(struct lg_cpu *cp
 {
 	pgd_t gpgd;
 	pte_t gpte;
-
 #ifdef CONFIG_X86_PAE
 	pmd_t gpmd;
 #endif
@@ -534,13 +537,14 @@ unsigned long guest_pa(struct lg_cpu *cp
 		return -1UL;
 	}
 
-	gpte = lgread(cpu, gpte_addr(cpu, gpgd, vaddr), pte_t);
 #ifdef CONFIG_X86_PAE
 	gpmd = lgread(cpu, gpmd_addr(gpgd, vaddr), pmd_t);
 	if (!(pmd_flags(gpmd) & _PAGE_PRESENT))
 		kill_guest(cpu, "Bad address %#lx", vaddr);
+	gpte = lgread(cpu, gpte_addr(cpu, gpmd, vaddr), pte_t);
+#else
+	gpte = lgread(cpu, gpte_addr(cpu, gpgd, vaddr), pte_t);
 #endif
-	gpte = lgread(cpu, gpte_addr(cpu, gpgd, vaddr), pte_t);
 	if (!(pte_flags(gpte) & _PAGE_PRESENT))
 		kill_guest(cpu, "Bad address %#lx", vaddr);
 
@@ -847,7 +851,7 @@ static unsigned long setup_pagetables(st
 	/* The top level points to the linear page table pages above.
 	 * We setup the identity and linear mappings here. */
 #ifdef CONFIG_X86_PAE
-	for (i = 0, j; i < mapped_pages && j < PTRS_PER_PMD;
+	for (i = j = 0; i < mapped_pages && j < PTRS_PER_PMD;
 	     i += PTRS_PER_PTE, j++) {
 		native_set_pmd(&pmd, __pmd(((unsigned long)(linear + i)
 		- mem_base) | _PAGE_PRESENT | _PAGE_RW | _PAGE_USER));