[PATCH v3 1/4] x86/mm: Adapt MODULES_END based on Fixmap section size

[PATCH v3 1/4] x86/mm: Adapt MODULES_END based on Fixmap section size

[Thomas Garnier]

This patch aligns MODULES_END to the beginning of the Fixmap section.
It optimizes the space available for both sections. The address is
pre-computed based on the number of pages required by the Fixmap
section.

It will allow GDT remapping in the Fixmap section. The current
MODULES_END static address does not provide enough space for the kernel
to support a large number of processors.

Signed-off-by: Thomas Garnier <thgarnie@google.com>
---
Based on next-20170213
---
 arch/x86/include/asm/fixmap.h           | 8 ++++++++
 arch/x86/include/asm/pgtable_64_types.h | 3 ---
 arch/x86/kernel/module.c                | 1 +
 arch/x86/mm/dump_pagetables.c           | 1 +
 arch/x86/mm/kasan_init_64.c             | 1 +
 5 files changed, 11 insertions(+), 3 deletions(-)

diff --git a/arch/x86/include/asm/fixmap.h b/arch/x86/include/asm/fixmap.h
index 8554f960e21b..20231189e0e3 100644
--- a/arch/x86/include/asm/fixmap.h
+++ b/arch/x86/include/asm/fixmap.h
@@ -132,6 +132,14 @@ enum fixed_addresses {
 
 extern void reserve_top_address(unsigned long reserve);
 
+/* On 64-bit, the module sections ends with the start of the fixmap */
+#ifdef CONFIG_X86_64
+#define MODULES_VADDR    (__START_KERNEL_map + KERNEL_IMAGE_SIZE)
+#define MODULES_END   __fix_to_virt(__end_of_fixed_addresses + 1)
+#define MODULES_LEN   (MODULES_END - MODULES_VADDR)
+#endif /* CONFIG_X86_64 */
+
+
 #define FIXADDR_SIZE	(__end_of_permanent_fixed_addresses << PAGE_SHIFT)
 #define FIXADDR_START		(FIXADDR_TOP - FIXADDR_SIZE)
 
diff --git a/arch/x86/include/asm/pgtable_64_types.h b/arch/x86/include/asm/pgtable_64_types.h
index 3a264200c62f..de8bace10200 100644
--- a/arch/x86/include/asm/pgtable_64_types.h
+++ b/arch/x86/include/asm/pgtable_64_types.h
@@ -66,9 +66,6 @@ typedef struct { pteval_t pte; } pte_t;
 #define VMEMMAP_START	__VMEMMAP_BASE
 #endif /* CONFIG_RANDOMIZE_MEMORY */
 #define VMALLOC_END	(VMALLOC_START + _AC((VMALLOC_SIZE_TB << 40) - 1, UL))
-#define MODULES_VADDR    (__START_KERNEL_map + KERNEL_IMAGE_SIZE)
-#define MODULES_END      _AC(0xffffffffff000000, UL)
-#define MODULES_LEN   (MODULES_END - MODULES_VADDR)
 #define ESPFIX_PGD_ENTRY _AC(-2, UL)
 #define ESPFIX_BASE_ADDR (ESPFIX_PGD_ENTRY << PGDIR_SHIFT)
 #define EFI_VA_START	 ( -4 * (_AC(1, UL) << 30))
diff --git a/arch/x86/kernel/module.c b/arch/x86/kernel/module.c
index 477ae806c2fa..fad61caac75e 100644
--- a/arch/x86/kernel/module.c
+++ b/arch/x86/kernel/module.c
@@ -35,6 +35,7 @@
 #include <asm/page.h>
 #include <asm/pgtable.h>
 #include <asm/setup.h>
+#include <asm/fixmap.h>
 
 #if 0
 #define DEBUGP(fmt, ...)				\
diff --git a/arch/x86/mm/dump_pagetables.c b/arch/x86/mm/dump_pagetables.c
index 8aa6bea1cd6c..90170415f08a 100644
--- a/arch/x86/mm/dump_pagetables.c
+++ b/arch/x86/mm/dump_pagetables.c
@@ -19,6 +19,7 @@
 #include <linux/seq_file.h>
 
 #include <asm/pgtable.h>
+#include <asm/fixmap.h>
 
 /*
  * The dumper groups pagetable entries of the same type into one, and for
diff --git a/arch/x86/mm/kasan_init_64.c b/arch/x86/mm/kasan_init_64.c
index 0493c17b8a51..34f167cf3316 100644
--- a/arch/x86/mm/kasan_init_64.c
+++ b/arch/x86/mm/kasan_init_64.c
@@ -8,6 +8,7 @@
 
 #include <asm/tlbflush.h>
 #include <asm/sections.h>
+#include <asm/fixmap.h>
 
 extern pgd_t early_level4_pgt[PTRS_PER_PGD];
 extern struct range pfn_mapped[E820_X_MAX];
-- 
2.11.0.483.g087da7b7c-goog

[PATCH v3 2/4] x86: Remap GDT tables in the Fixmap section

[Thomas Garnier]

Each processor holds a GDT in its per-cpu structure. The sgdt
instruction gives the base address of the current GDT. This address can
be used to bypass KASLR memory randomization. With another bug, an
attacker could target other per-cpu structures or deduce the base of
the main memory section (PAGE_OFFSET).

This patch relocates the GDT table for each processor inside the
Fixmap section. The space is reserved based on number of supported
processors.

For consistency, the remapping is done by default on 32 and 64-bit.

Each processor switches to its remapped GDT at the end of
initialization. For hibernation, the main processor returns with the
original GDT and switches back to the remapping at completion.

This patch was tested on both architectures. Hibernation and KVM were
both tested specially for their usage of the GDT.

Signed-off-by: Thomas Garnier <thgarnie@google.com>
---
Based on next-20170213
---
 arch/x86/entry/vdso/vma.c             |  2 +-
 arch/x86/include/asm/desc.h           | 33 +++++++++++++++++++++++++++++----
 arch/x86/include/asm/fixmap.h         |  4 ++++
 arch/x86/include/asm/processor.h      |  1 +
 arch/x86/include/asm/stackprotector.h |  2 +-
 arch/x86/kernel/acpi/sleep.c          |  2 +-
 arch/x86/kernel/apm_32.c              |  6 +++---
 arch/x86/kernel/cpu/common.c          | 26 ++++++++++++++++++++++++--
 arch/x86/kernel/setup_percpu.c        |  2 +-
 arch/x86/kernel/smpboot.c             |  2 +-
 arch/x86/platform/efi/efi_32.c        |  4 ++--
 arch/x86/power/cpu.c                  |  7 +++++--
 arch/x86/xen/enlighten.c              |  2 +-
 arch/x86/xen/smp.c                    |  2 +-
 drivers/lguest/x86/core.c             |  6 +++---
 drivers/pnp/pnpbios/bioscalls.c       | 10 +++++-----
 16 files changed, 83 insertions(+), 28 deletions(-)

diff --git a/arch/x86/entry/vdso/vma.c b/arch/x86/entry/vdso/vma.c
index 572cee3fccff..9c8bd4cfcc6e 100644
--- a/arch/x86/entry/vdso/vma.c
+++ b/arch/x86/entry/vdso/vma.c
@@ -353,7 +353,7 @@ static void vgetcpu_cpu_init(void *arg)
 	d.p = 1;		/* Present */
 	d.d = 1;		/* 32-bit */
 
-	write_gdt_entry(get_cpu_gdt_table(cpu), GDT_ENTRY_PER_CPU, &d, DESCTYPE_S);
+	write_gdt_entry(get_cpu_gdt_rw(cpu), GDT_ENTRY_PER_CPU, &d, DESCTYPE_S);
 }
 
 static int vgetcpu_online(unsigned int cpu)
diff --git a/arch/x86/include/asm/desc.h b/arch/x86/include/asm/desc.h
index 12080d87da3b..5d4ba1311737 100644
--- a/arch/x86/include/asm/desc.h
+++ b/arch/x86/include/asm/desc.h
@@ -4,6 +4,7 @@
 #include <asm/desc_defs.h>
 #include <asm/ldt.h>
 #include <asm/mmu.h>
+#include <asm/fixmap.h>
 
 #include <linux/smp.h>
 #include <linux/percpu.h>
@@ -45,11 +46,35 @@ struct gdt_page {
 
 DECLARE_PER_CPU_PAGE_ALIGNED(struct gdt_page, gdt_page);
 
-static inline struct desc_struct *get_cpu_gdt_table(unsigned int cpu)
+/* Provide the original GDT */
+static inline struct desc_struct *get_cpu_gdt_rw(unsigned int cpu)
 {
 	return per_cpu(gdt_page, cpu).gdt;
 }
 
+static inline unsigned long get_cpu_gdt_rw_vaddr(unsigned int cpu)
+{
+	return (unsigned long)get_cpu_gdt_rw(cpu);
+}
+
+/* Get the fixmap index for a specific processor */
+static inline unsigned int get_cpu_gdt_ro_index(int cpu)
+{
+	return FIX_GDT_REMAP_BEGIN + cpu;
+}
+
+/* Provide the fixmap address of the remapped GDT */
+static inline struct desc_struct *get_cpu_gdt_ro(int cpu)
+{
+	unsigned int idx = get_cpu_gdt_ro_index(cpu);
+	return (struct desc_struct *)__fix_to_virt(idx);
+}
+
+static inline unsigned long get_cpu_gdt_ro_vaddr(int cpu)
+{
+	return (unsigned long)get_cpu_gdt_ro(cpu);
+}
+
 #ifdef CONFIG_X86_64
 
 static inline void pack_gate(gate_desc *gate, unsigned type, unsigned long func,
@@ -174,7 +199,7 @@ static inline void set_tssldt_descriptor(void *d, unsigned long addr, unsigned t
 
 static inline void __set_tss_desc(unsigned cpu, unsigned int entry, void *addr)
 {
-	struct desc_struct *d = get_cpu_gdt_table(cpu);
+	struct desc_struct *d = get_cpu_gdt_rw(cpu);
 	tss_desc tss;
 
 	/*
@@ -202,7 +227,7 @@ static inline void native_set_ldt(const void *addr, unsigned int entries)
 
 		set_tssldt_descriptor(&ldt, (unsigned long)addr, DESC_LDT,
 				      entries * LDT_ENTRY_SIZE - 1);
-		write_gdt_entry(get_cpu_gdt_table(cpu), GDT_ENTRY_LDT,
+		write_gdt_entry(get_cpu_gdt_rw(cpu), GDT_ENTRY_LDT,
 				&ldt, DESC_LDT);
 		asm volatile("lldt %w0"::"q" (GDT_ENTRY_LDT*8));
 	}
@@ -244,7 +269,7 @@ static inline unsigned long native_store_tr(void)
 
 static inline void native_load_tls(struct thread_struct *t, unsigned int cpu)
 {
-	struct desc_struct *gdt = get_cpu_gdt_table(cpu);
+	struct desc_struct *gdt = get_cpu_gdt_rw(cpu);
 	unsigned int i;
 
 	for (i = 0; i < GDT_ENTRY_TLS_ENTRIES; i++)
diff --git a/arch/x86/include/asm/fixmap.h b/arch/x86/include/asm/fixmap.h
index 20231189e0e3..4c11425d856c 100644
--- a/arch/x86/include/asm/fixmap.h
+++ b/arch/x86/include/asm/fixmap.h
@@ -100,6 +100,10 @@ enum fixed_addresses {
 #ifdef	CONFIG_X86_INTEL_MID
 	FIX_LNW_VRTC,
 #endif
+	/* Fixmap entries to remap the GDTs, one per processor. */
+	FIX_GDT_REMAP_BEGIN,
+	FIX_GDT_REMAP_END = FIX_GDT_REMAP_BEGIN + NR_CPUS - 1,
+
 	__end_of_permanent_fixed_addresses,
 
 	/*
diff --git a/arch/x86/include/asm/processor.h b/arch/x86/include/asm/processor.h
index e6cfe7ba2d65..c441d1f7e275 100644
--- a/arch/x86/include/asm/processor.h
+++ b/arch/x86/include/asm/processor.h
@@ -706,6 +706,7 @@ extern struct desc_ptr		early_gdt_descr;
 
 extern void cpu_set_gdt(int);
 extern void switch_to_new_gdt(int);
+extern void load_fixmap_gdt(int);
 extern void load_percpu_segment(int);
 extern void cpu_init(void);
 
diff --git a/arch/x86/include/asm/stackprotector.h b/arch/x86/include/asm/stackprotector.h
index 58505f01962f..dcbd9bcce714 100644
--- a/arch/x86/include/asm/stackprotector.h
+++ b/arch/x86/include/asm/stackprotector.h
@@ -87,7 +87,7 @@ static inline void setup_stack_canary_segment(int cpu)
 {
 #ifdef CONFIG_X86_32
 	unsigned long canary = (unsigned long)&per_cpu(stack_canary, cpu);
-	struct desc_struct *gdt_table = get_cpu_gdt_table(cpu);
+	struct desc_struct *gdt_table = get_cpu_gdt_rw(cpu);
 	struct desc_struct desc;
 
 	desc = gdt_table[GDT_ENTRY_STACK_CANARY];
diff --git a/arch/x86/kernel/acpi/sleep.c b/arch/x86/kernel/acpi/sleep.c
index 48587335ede8..ed014814ea35 100644
--- a/arch/x86/kernel/acpi/sleep.c
+++ b/arch/x86/kernel/acpi/sleep.c
@@ -101,7 +101,7 @@ int x86_acpi_suspend_lowlevel(void)
 #ifdef CONFIG_SMP
 	initial_stack = (unsigned long)temp_stack + sizeof(temp_stack);
 	early_gdt_descr.address =
-			(unsigned long)get_cpu_gdt_table(smp_processor_id());
+			(unsigned long)get_cpu_gdt_rw(smp_processor_id());
 	initial_gs = per_cpu_offset(smp_processor_id());
 #endif
 	initial_code = (unsigned long)wakeup_long64;
diff --git a/arch/x86/kernel/apm_32.c b/arch/x86/kernel/apm_32.c
index 45d44c173cf9..dc4e89d93a10 100644
--- a/arch/x86/kernel/apm_32.c
+++ b/arch/x86/kernel/apm_32.c
@@ -608,7 +608,7 @@ static long __apm_bios_call(void *_call)
 
 	cpu = get_cpu();
 	BUG_ON(cpu != 0);
-	gdt = get_cpu_gdt_table(cpu);
+	gdt = get_cpu_gdt_rw(cpu);
 	save_desc_40 = gdt[0x40 / 8];
 	gdt[0x40 / 8] = bad_bios_desc;
 
@@ -684,7 +684,7 @@ static long __apm_bios_call_simple(void *_call)
 
 	cpu = get_cpu();
 	BUG_ON(cpu != 0);
-	gdt = get_cpu_gdt_table(cpu);
+	gdt = get_cpu_gdt_rw(cpu);
 	save_desc_40 = gdt[0x40 / 8];
 	gdt[0x40 / 8] = bad_bios_desc;
 
@@ -2351,7 +2351,7 @@ static int __init apm_init(void)
 	 * Note we only set APM segments on CPU zero, since we pin the APM
 	 * code to that CPU.
 	 */
-	gdt = get_cpu_gdt_table(0);
+	gdt = get_cpu_gdt_rw(0);
 	set_desc_base(&gdt[APM_CS >> 3],
 		 (unsigned long)__va((unsigned long)apm_info.bios.cseg << 4));
 	set_desc_base(&gdt[APM_CS_16 >> 3],
diff --git a/arch/x86/kernel/cpu/common.c b/arch/x86/kernel/cpu/common.c
index 1a4ee7082ad7..2853a42ded2d 100644
--- a/arch/x86/kernel/cpu/common.c
+++ b/arch/x86/kernel/cpu/common.c
@@ -444,6 +444,23 @@ void load_percpu_segment(int cpu)
 	load_stack_canary_segment();
 }
 
+/* Setup the fixmap mapping only once per-processor */
+static inline void setup_fixmap_gdt(int cpu)
+{
+	__set_fixmap(get_cpu_gdt_ro_index(cpu),
+		     __pa(get_cpu_gdt_rw(cpu)), PAGE_KERNEL);
+}
+
+/* Load a fixmap remapping of the per-cpu GDT */
+void load_fixmap_gdt(int cpu)
+{
+	struct desc_ptr gdt_descr;
+
+	gdt_descr.address = (long)get_cpu_gdt_ro(cpu);
+	gdt_descr.size = GDT_SIZE - 1;
+	load_gdt(&gdt_descr);
+}
+
 /*
  * Current gdt points %fs at the "master" per-cpu area: after this,
  * it's on the real one.
@@ -452,11 +469,10 @@ void switch_to_new_gdt(int cpu)
 {
 	struct desc_ptr gdt_descr;
 
-	gdt_descr.address = (long)get_cpu_gdt_table(cpu);
+	gdt_descr.address = (long)get_cpu_gdt_rw(cpu);
 	gdt_descr.size = GDT_SIZE - 1;
 	load_gdt(&gdt_descr);
 	/* Reload the per-cpu base */
-
 	load_percpu_segment(cpu);
 }
 
@@ -1524,6 +1540,9 @@ void cpu_init(void)
 
 	if (is_uv_system())
 		uv_cpu_init();
+
+	setup_fixmap_gdt(cpu);
+	load_fixmap_gdt(cpu);
 }
 
 #else
@@ -1579,6 +1598,9 @@ void cpu_init(void)
 	dbg_restore_debug_regs();
 
 	fpu__init_cpu();
+
+	setup_fixmap_gdt(cpu);
+	load_fixmap_gdt(cpu);
 }
 #endif
 
diff --git a/arch/x86/kernel/setup_percpu.c b/arch/x86/kernel/setup_percpu.c
index 9820d6d977c6..11338b0b3ad2 100644
--- a/arch/x86/kernel/setup_percpu.c
+++ b/arch/x86/kernel/setup_percpu.c
@@ -160,7 +160,7 @@ static inline void setup_percpu_segment(int cpu)
 	pack_descriptor(&gdt, per_cpu_offset(cpu), 0xFFFFF,
 			0x2 | DESCTYPE_S, 0x8);
 	gdt.s = 1;
-	write_gdt_entry(get_cpu_gdt_table(cpu),
+	write_gdt_entry(get_cpu_gdt_rw(cpu),
 			GDT_ENTRY_PERCPU, &gdt, DESCTYPE_S);
 #endif
 }
diff --git a/arch/x86/kernel/smpboot.c b/arch/x86/kernel/smpboot.c
index 99b920d0e516..2e61b4b6957b 100644
--- a/arch/x86/kernel/smpboot.c
+++ b/arch/x86/kernel/smpboot.c
@@ -980,7 +980,7 @@ static int do_boot_cpu(int apicid, int cpu, struct task_struct *idle)
 	unsigned long timeout;
 
 	idle->thread.sp = (unsigned long)task_pt_regs(idle);
-	early_gdt_descr.address = (unsigned long)get_cpu_gdt_table(cpu);
+	early_gdt_descr.address = (unsigned long)get_cpu_gdt_rw(cpu);
 	initial_code = (unsigned long)start_secondary;
 	initial_stack  = idle->thread.sp;
 
diff --git a/arch/x86/platform/efi/efi_32.c b/arch/x86/platform/efi/efi_32.c
index cef39b097649..950071171436 100644
--- a/arch/x86/platform/efi/efi_32.c
+++ b/arch/x86/platform/efi/efi_32.c
@@ -68,7 +68,7 @@ pgd_t * __init efi_call_phys_prolog(void)
 	load_cr3(initial_page_table);
 	__flush_tlb_all();
 
-	gdt_descr.address = __pa(get_cpu_gdt_table(0));
+	gdt_descr.address = __pa(get_cpu_gdt_rw(0));
 	gdt_descr.size = GDT_SIZE - 1;
 	load_gdt(&gdt_descr);
 
@@ -79,7 +79,7 @@ void __init efi_call_phys_epilog(pgd_t *save_pgd)
 {
 	struct desc_ptr gdt_descr;
 
-	gdt_descr.address = (unsigned long)get_cpu_gdt_table(0);
+	gdt_descr.address = (unsigned long)get_cpu_gdt_rw(0);
 	gdt_descr.size = GDT_SIZE - 1;
 	load_gdt(&gdt_descr);
 
diff --git a/arch/x86/power/cpu.c b/arch/x86/power/cpu.c
index 66ade16c7693..6b05a9219ea2 100644
--- a/arch/x86/power/cpu.c
+++ b/arch/x86/power/cpu.c
@@ -95,7 +95,7 @@ static void __save_processor_state(struct saved_context *ctxt)
 	 * 'pmode_gdt' in wakeup_start.
 	 */
 	ctxt->gdt_desc.size = GDT_SIZE - 1;
-	ctxt->gdt_desc.address = (unsigned long)get_cpu_gdt_table(smp_processor_id());
+	ctxt->gdt_desc.address = (unsigned long)get_cpu_gdt_rw(smp_processor_id());
 
 	store_tr(ctxt->tr);
 
@@ -162,7 +162,7 @@ static void fix_processor_context(void)
 	int cpu = smp_processor_id();
 	struct tss_struct *t = &per_cpu(cpu_tss, cpu);
 #ifdef CONFIG_X86_64
-	struct desc_struct *desc = get_cpu_gdt_table(cpu);
+	struct desc_struct *desc = get_cpu_gdt_rw(cpu);
 	tss_desc tss;
 #endif
 	set_tss_desc(cpu, t);	/*
@@ -183,6 +183,9 @@ static void fix_processor_context(void)
 	load_mm_ldt(current->active_mm);	/* This does lldt */
 
 	fpu__resume_cpu();
+
+	/* The processor is back on the direct GDT, load back the fixmap */
+	load_fixmap_gdt(cpu);
 }
 
 /**
diff --git a/arch/x86/xen/enlighten.c b/arch/x86/xen/enlighten.c
index ec1d5c46e58f..4951fcf95143 100644
--- a/arch/x86/xen/enlighten.c
+++ b/arch/x86/xen/enlighten.c
@@ -710,7 +710,7 @@ static void load_TLS_descriptor(struct thread_struct *t,
 
 	*shadow = t->tls_array[i];
 
-	gdt = get_cpu_gdt_table(cpu);
+	gdt = get_cpu_gdt_rw(cpu);
 	maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
 	mc = __xen_mc_entry(0);
 
diff --git a/arch/x86/xen/smp.c b/arch/x86/xen/smp.c
index 0dee6f59ea82..6399bab936cd 100644
--- a/arch/x86/xen/smp.c
+++ b/arch/x86/xen/smp.c
@@ -391,7 +391,7 @@ cpu_initialize_context(unsigned int cpu, struct task_struct *idle)
 	if (ctxt == NULL)
 		return -ENOMEM;
 
-	gdt = get_cpu_gdt_table(cpu);
+	gdt = get_cpu_gdt_rw(cpu);
 
 #ifdef CONFIG_X86_32
 	ctxt->user_regs.fs = __KERNEL_PERCPU;
diff --git a/drivers/lguest/x86/core.c b/drivers/lguest/x86/core.c
index d71f6323ac00..b4f79b923aea 100644
--- a/drivers/lguest/x86/core.c
+++ b/drivers/lguest/x86/core.c
@@ -504,7 +504,7 @@ void __init lguest_arch_host_init(void)
 		 * byte, not the size, hence the "-1").
 		 */
 		state->host_gdt_desc.size = GDT_SIZE-1;
-		state->host_gdt_desc.address = (long)get_cpu_gdt_table(i);
+		state->host_gdt_desc.address = (long)get_cpu_gdt_rw(i);
 
 		/*
 		 * All CPUs on the Host use the same Interrupt Descriptor
@@ -554,8 +554,8 @@ void __init lguest_arch_host_init(void)
 		 * The Host needs to be able to use the LGUEST segments on this
 		 * CPU, too, so put them in the Host GDT.
 		 */
-		get_cpu_gdt_table(i)[GDT_ENTRY_LGUEST_CS] = FULL_EXEC_SEGMENT;
-		get_cpu_gdt_table(i)[GDT_ENTRY_LGUEST_DS] = FULL_SEGMENT;
+		get_cpu_gdt_rw(i)[GDT_ENTRY_LGUEST_CS] = FULL_EXEC_SEGMENT;
+		get_cpu_gdt_rw(i)[GDT_ENTRY_LGUEST_DS] = FULL_SEGMENT;
 	}
 
 	/*
diff --git a/drivers/pnp/pnpbios/bioscalls.c b/drivers/pnp/pnpbios/bioscalls.c
index 438d4c72c7b3..ff563db025b3 100644
--- a/drivers/pnp/pnpbios/bioscalls.c
+++ b/drivers/pnp/pnpbios/bioscalls.c
@@ -54,7 +54,7 @@ __asm__(".text			\n"
 
 #define Q2_SET_SEL(cpu, selname, address, size) \
 do { \
-	struct desc_struct *gdt = get_cpu_gdt_table((cpu)); \
+	struct desc_struct *gdt = get_cpu_gdt_rw((cpu)); \
 	set_desc_base(&gdt[(selname) >> 3], (u32)(address)); \
 	set_desc_limit(&gdt[(selname) >> 3], (size) - 1); \
 } while(0)
@@ -95,8 +95,8 @@ static inline u16 call_pnp_bios(u16 func, u16 arg1, u16 arg2, u16 arg3,
 		return PNP_FUNCTION_NOT_SUPPORTED;
 
 	cpu = get_cpu();
-	save_desc_40 = get_cpu_gdt_table(cpu)[0x40 / 8];
-	get_cpu_gdt_table(cpu)[0x40 / 8] = bad_bios_desc;
+	save_desc_40 = get_cpu_gdt_rw(cpu)[0x40 / 8];
+	get_cpu_gdt_rw(cpu)[0x40 / 8] = bad_bios_desc;
 
 	/* On some boxes IRQ's during PnP BIOS calls are deadly.  */
 	spin_lock_irqsave(&pnp_bios_lock, flags);
@@ -134,7 +134,7 @@ static inline u16 call_pnp_bios(u16 func, u16 arg1, u16 arg2, u16 arg3,
 			     :"memory");
 	spin_unlock_irqrestore(&pnp_bios_lock, flags);
 
-	get_cpu_gdt_table(cpu)[0x40 / 8] = save_desc_40;
+	get_cpu_gdt_rw(cpu)[0x40 / 8] = save_desc_40;
 	put_cpu();
 
 	/* If we get here and this is set then the PnP BIOS faulted on us. */
@@ -477,7 +477,7 @@ void pnpbios_calls_init(union pnp_bios_install_struct *header)
 	pnp_bios_callpoint.segment = PNP_CS16;
 
 	for_each_possible_cpu(i) {
-		struct desc_struct *gdt = get_cpu_gdt_table(i);
+		struct desc_struct *gdt = get_cpu_gdt_rw(i);
 		if (!gdt)
 			continue;
 		set_desc_base(&gdt[GDT_ENTRY_PNPBIOS_CS32],
-- 
2.11.0.483.g087da7b7c-goog

Re: [PATCH v3 2/4] x86: Remap GDT tables in the Fixmap section

[Boris Ostrovsky]

On 02/14/2017 02:42 PM, Thomas Garnier wrote:
> diff --git a/arch/x86/xen/smp.c b/arch/x86/xen/smp.c
> index 0dee6f59ea82..6399bab936cd 100644
> --- a/arch/x86/xen/smp.c
> +++ b/arch/x86/xen/smp.c
> @@ -391,7 +391,7 @@ cpu_initialize_context(unsigned int cpu, struct task_struct *idle)
>  	if (ctxt == NULL)
>  		return -ENOMEM;
>  
> -	gdt = get_cpu_gdt_table(cpu);
> +	gdt = get_cpu_gdt_rw(cpu);
>  
>  #ifdef CONFIG_X86_32
>  	ctxt->user_regs.fs = __KERNEL_PERCPU;


Which tree are these patches against? The chunk above cannot be applied
to current Linus tree but it can be applied against Xen staging tree
that has not been pulled yet.

I can apply this manually but the series crashes PV guests so I want to
make sure I am using right bits.

-boris

Re: [PATCH v3 2/4] x86: Remap GDT tables in the Fixmap section

[Thomas Garnier]

On Wed, Feb 15, 2017 at 7:37 AM, Boris Ostrovsky
<boris.ostrovsky@oracle.com> wrote:
> On 02/14/2017 02:42 PM, Thomas Garnier wrote:
>> diff --git a/arch/x86/xen/smp.c b/arch/x86/xen/smp.c
>> index 0dee6f59ea82..6399bab936cd 100644
>> --- a/arch/x86/xen/smp.c
>> +++ b/arch/x86/xen/smp.c
>> @@ -391,7 +391,7 @@ cpu_initialize_context(unsigned int cpu, struct task_struct *idle)
>>       if (ctxt == NULL)
>>               return -ENOMEM;
>>
>> -     gdt = get_cpu_gdt_table(cpu);
>> +     gdt = get_cpu_gdt_rw(cpu);
>>
>>  #ifdef CONFIG_X86_32
>>       ctxt->user_regs.fs = __KERNEL_PERCPU;
>
>
> Which tree are these patches against? The chunk above cannot be applied
> to current Linus tree but it can be applied against Xen staging tree
> that has not been pulled yet.
>
> I can apply this manually but the series crashes PV guests so I want to
> make sure I am using right bits.
>

That was against linux-next (next-20170213 tag).

> -boris
>
>



-- 
Thomas

[PATCH v3 3/4] x86: Make the GDT remapping read-only on 64-bit

[Thomas Garnier]

This patch makes the GDT remapped pages read-only to prevent corruption.
This change is done only on 64-bit.

The native_load_tr_desc function was adapted to correctly handle a
read-only GDT. The LTR instruction always writes to the GDT TSS entry.
This generates a page fault if the GDT is read-only. This change checks
if the current GDT is a remap and swap GDTs as needed. This function was
tested by booting multiple machines and checking hibernation works
properly.

KVM SVM and VMX were adapted to use the writeable GDT. On VMX, the
per-cpu variable was removed for functions to fetch the original GDT.
Instead of reloading the previous GDT, VMX will reload the fixmap GDT as
expected. For testing, VMs were started and restored on multiple
configurations.

Signed-off-by: Thomas Garnier <thgarnie@google.com>
---
Based on next-20170213
---
 arch/x86/include/asm/desc.h      | 51 ++++++++++++++++++++++++++++++++++++----
 arch/x86/include/asm/processor.h |  1 +
 arch/x86/kernel/cpu/common.c     | 28 +++++++++++++++++-----
 arch/x86/kvm/svm.c               |  4 +---
 arch/x86/kvm/vmx.c               | 15 ++++--------
 5 files changed, 75 insertions(+), 24 deletions(-)

diff --git a/arch/x86/include/asm/desc.h b/arch/x86/include/asm/desc.h
index 5d4ba1311737..15b2a86c9267 100644
--- a/arch/x86/include/asm/desc.h
+++ b/arch/x86/include/asm/desc.h
@@ -57,6 +57,17 @@ static inline unsigned long get_cpu_gdt_rw_vaddr(unsigned int cpu)
 	return (unsigned long)get_cpu_gdt_rw(cpu);
 }
 
+/* Provide the current original GDT */
+static inline struct desc_struct *get_current_gdt_rw(void)
+{
+	return this_cpu_ptr(&gdt_page)->gdt;
+}
+
+static inline unsigned long get_current_gdt_rw_vaddr(void)
+{
+	return (unsigned long)get_current_gdt_rw();
+}
+
 /* Get the fixmap index for a specific processor */
 static inline unsigned int get_cpu_gdt_ro_index(int cpu)
 {
@@ -233,11 +244,6 @@ static inline void native_set_ldt(const void *addr, unsigned int entries)
 	}
 }
 
-static inline void native_load_tr_desc(void)
-{
-	asm volatile("ltr %w0"::"q" (GDT_ENTRY_TSS*8));
-}
-
 static inline void native_load_gdt(const struct desc_ptr *dtr)
 {
 	asm volatile("lgdt %0"::"m" (*dtr));
@@ -258,6 +264,41 @@ static inline void native_store_idt(struct desc_ptr *dtr)
 	asm volatile("sidt %0":"=m" (*dtr));
 }
 
+/*
+ * The LTR instruction marks the TSS GDT entry as busy. On 64-bit, the GDT is
+ * a read-only remapping. To prevent a page fault, the GDT is switched to the
+ * original writeable version when needed.
+ */
+#ifdef CONFIG_X86_64
+static inline void native_load_tr_desc(void)
+{
+	struct desc_ptr gdt;
+	int cpu = raw_smp_processor_id();
+	bool restore = 0;
+	struct desc_struct *fixmap_gdt;
+
+	native_store_gdt(&gdt);
+	fixmap_gdt = get_cpu_gdt_ro(cpu);
+
+	/*
+	 * If the current GDT is the read-only fixmap, swap to the original
+	 * writeable version. Swap back at the end.
+	 */
+	if (gdt.address == (unsigned long)fixmap_gdt) {
+		load_direct_gdt(cpu);
+		restore = 1;
+	}
+	asm volatile("ltr %w0"::"q" (GDT_ENTRY_TSS*8));
+	if (restore)
+		load_fixmap_gdt(cpu);
+}
+#else
+static inline void native_load_tr_desc(void)
+{
+	asm volatile("ltr %w0"::"q" (GDT_ENTRY_TSS*8));
+}
+#endif
+
 static inline unsigned long native_store_tr(void)
 {
 	unsigned long tr;
diff --git a/arch/x86/include/asm/processor.h b/arch/x86/include/asm/processor.h
index c441d1f7e275..6ea9e419a856 100644
--- a/arch/x86/include/asm/processor.h
+++ b/arch/x86/include/asm/processor.h
@@ -706,6 +706,7 @@ extern struct desc_ptr		early_gdt_descr;
 
 extern void cpu_set_gdt(int);
 extern void switch_to_new_gdt(int);
+extern void load_direct_gdt(int);
 extern void load_fixmap_gdt(int);
 extern void load_percpu_segment(int);
 extern void cpu_init(void);
diff --git a/arch/x86/kernel/cpu/common.c b/arch/x86/kernel/cpu/common.c
index 2853a42ded2d..bdf521383900 100644
--- a/arch/x86/kernel/cpu/common.c
+++ b/arch/x86/kernel/cpu/common.c
@@ -444,13 +444,31 @@ void load_percpu_segment(int cpu)
 	load_stack_canary_segment();
 }
 
+/* On 64-bit the GDT remapping is read-only */
+#ifdef CONFIG_X86_64
+#define PAGE_FIXMAP_GDT PAGE_KERNEL_RO
+#else
+#define PAGE_FIXMAP_GDT PAGE_KERNEL
+#endif
+
 /* Setup the fixmap mapping only once per-processor */
 static inline void setup_fixmap_gdt(int cpu)
 {
 	__set_fixmap(get_cpu_gdt_ro_index(cpu),
-		     __pa(get_cpu_gdt_rw(cpu)), PAGE_KERNEL);
+		     __pa(get_cpu_gdt_rw(cpu)), PAGE_FIXMAP_GDT);
 }
 
+/* Load the original GDT from the per-cpu structure */
+void load_direct_gdt(int cpu)
+{
+	struct desc_ptr gdt_descr;
+
+	gdt_descr.address = (long)get_cpu_gdt_rw(cpu);
+	gdt_descr.size = GDT_SIZE - 1;
+	load_gdt(&gdt_descr);
+}
+EXPORT_SYMBOL_GPL(load_direct_gdt);
+
 /* Load a fixmap remapping of the per-cpu GDT */
 void load_fixmap_gdt(int cpu)
 {
@@ -460,6 +478,7 @@ void load_fixmap_gdt(int cpu)
 	gdt_descr.size = GDT_SIZE - 1;
 	load_gdt(&gdt_descr);
 }
+EXPORT_SYMBOL_GPL(load_fixmap_gdt);
 
 /*
  * Current gdt points %fs at the "master" per-cpu area: after this,
@@ -467,11 +486,8 @@ void load_fixmap_gdt(int cpu)
  */
 void switch_to_new_gdt(int cpu)
 {
-	struct desc_ptr gdt_descr;
-
-	gdt_descr.address = (long)get_cpu_gdt_rw(cpu);
-	gdt_descr.size = GDT_SIZE - 1;
-	load_gdt(&gdt_descr);
+	/* Load the original GDT */
+	load_direct_gdt(cpu);
 	/* Reload the per-cpu base */
 	load_percpu_segment(cpu);
 }
diff --git a/arch/x86/kvm/svm.c b/arch/x86/kvm/svm.c
index d0414f054bdf..7b9a71e465b1 100644
--- a/arch/x86/kvm/svm.c
+++ b/arch/x86/kvm/svm.c
@@ -741,7 +741,6 @@ static int svm_hardware_enable(void)
 
 	struct svm_cpu_data *sd;
 	uint64_t efer;
-	struct desc_ptr gdt_descr;
 	struct desc_struct *gdt;
 	int me = raw_smp_processor_id();
 
@@ -763,8 +762,7 @@ static int svm_hardware_enable(void)
 	sd->max_asid = cpuid_ebx(SVM_CPUID_FUNC) - 1;
 	sd->next_asid = sd->max_asid + 1;
 
-	native_store_gdt(&gdt_descr);
-	gdt = (struct desc_struct *)gdt_descr.address;
+	gdt = get_current_gdt_rw();
 	sd->tss_desc = (struct kvm_ldttss_desc *)(gdt + GDT_ENTRY_TSS);
 
 	wrmsrl(MSR_EFER, efer | EFER_SVME);
diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
index 7c3e42623090..99167f20bc34 100644
--- a/arch/x86/kvm/vmx.c
+++ b/arch/x86/kvm/vmx.c
@@ -935,7 +935,6 @@ static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
  * when a CPU is brought down, and we need to VMCLEAR all VMCSs loaded on it.
  */
 static DEFINE_PER_CPU(struct list_head, loaded_vmcss_on_cpu);
-static DEFINE_PER_CPU(struct desc_ptr, host_gdt);
 
 /*
  * We maintian a per-CPU linked-list of vCPU, so in wakeup_handler() we
@@ -1997,10 +1996,9 @@ static void reload_tss(void)
 	/*
 	 * VT restores TR but not its size.  Useless.
 	 */
-	struct desc_ptr *gdt = this_cpu_ptr(&host_gdt);
 	struct desc_struct *descs;
 
-	descs = (void *)gdt->address;
+	descs = get_current_gdt_rw();
 	descs[GDT_ENTRY_TSS].type = 9; /* available TSS */
 	load_TR_desc();
 }
@@ -2061,7 +2059,6 @@ static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
 
 static unsigned long segment_base(u16 selector)
 {
-	struct desc_ptr *gdt = this_cpu_ptr(&host_gdt);
 	struct desc_struct *d;
 	unsigned long table_base;
 	unsigned long v;
@@ -2069,7 +2066,7 @@ static unsigned long segment_base(u16 selector)
 	if (!(selector & ~3))
 		return 0;
 
-	table_base = gdt->address;
+	table_base = get_current_gdt_rw_vaddr();
 
 	if (selector & 4) {           /* from ldt */
 		u16 ldt_selector = kvm_read_ldt();
@@ -2185,7 +2182,7 @@ static void __vmx_load_host_state(struct vcpu_vmx *vmx)
 #endif
 	if (vmx->host_state.msr_host_bndcfgs)
 		wrmsrl(MSR_IA32_BNDCFGS, vmx->host_state.msr_host_bndcfgs);
-	load_gdt(this_cpu_ptr(&host_gdt));
+	load_fixmap_gdt(raw_smp_processor_id());
 }
 
 static void vmx_load_host_state(struct vcpu_vmx *vmx)
@@ -2287,7 +2284,7 @@ static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 	}
 
 	if (!already_loaded) {
-		struct desc_ptr *gdt = this_cpu_ptr(&host_gdt);
+		unsigned long gdt = get_current_gdt_rw_vaddr();
 		unsigned long sysenter_esp;
 
 		kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
@@ -2297,7 +2294,7 @@ static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 		 * processors.
 		 */
 		vmcs_writel(HOST_TR_BASE, kvm_read_tr_base()); /* 22.2.4 */
-		vmcs_writel(HOST_GDTR_BASE, gdt->address);   /* 22.2.4 */
+		vmcs_writel(HOST_GDTR_BASE, gdt);   /* 22.2.4 */
 
 		rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
 		vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
@@ -3523,8 +3520,6 @@ static int hardware_enable(void)
 		ept_sync_global();
 	}
 
-	native_store_gdt(this_cpu_ptr(&host_gdt));
-
 	return 0;
 }
 
-- 
2.11.0.483.g087da7b7c-goog

[PATCH v3 4/4] KVM: VMX: Simplify segment_base

[Thomas Garnier]

The KVM segment_base function is confusing. This patch replaces integers
with appropriate flags, simplify constructs and add comments.

Signed-off-by: Thomas Garnier <thgarnie@google.com>
---
Based on next-20170213
---
 arch/x86/kvm/vmx.c | 26 ++++++++++++++++++--------
 1 file changed, 18 insertions(+), 8 deletions(-)

diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
index 99167f20bc34..edb8326108dd 100644
--- a/arch/x86/kvm/vmx.c
+++ b/arch/x86/kvm/vmx.c
@@ -2062,25 +2062,35 @@ static unsigned long segment_base(u16 selector)
 	struct desc_struct *d;
 	unsigned long table_base;
 	unsigned long v;
+	u32 high32;
 
-	if (!(selector & ~3))
+	if (!(selector & ~SEGMENT_RPL_MASK))
 		return 0;
 
-	table_base = get_current_gdt_rw_vaddr();
-
-	if (selector & 4) {           /* from ldt */
+	/* LDT selector */
+	if ((selector & SEGMENT_TI_MASK) == SEGMENT_LDT) {
 		u16 ldt_selector = kvm_read_ldt();
 
-		if (!(ldt_selector & ~3))
+		if (!(ldt_selector & ~SEGMENT_RPL_MASK))
 			return 0;
 
 		table_base = segment_base(ldt_selector);
+	} else {
+		table_base = get_current_gdt_rw_vaddr();
 	}
-	d = (struct desc_struct *)(table_base + (selector & ~7));
+
+	d = (struct desc_struct *)table_base + (selector >> 3);
 	v = get_desc_base(d);
 #ifdef CONFIG_X86_64
-       if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
-               v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
+	/*
+	 * Extend the virtual address if we have a system descriptor entry for
+	 * LDT or TSS (available or busy).
+	 */
+	if (d->s == 0 && (d->type == DESC_LDT || d->type == DESC_TSS ||
+			  d->type == 11/*Busy TSS */)) {
+		high32 = ((struct ldttss_desc64 *)d)->base3;
+		v |= (u64)high32 << 32;
+	}
 #endif
 	return v;
 }
-- 
2.11.0.483.g087da7b7c-goog

Re: [PATCH v3 4/4] KVM: VMX: Simplify segment_base

[Andy Lutomirski]

On Tue, Feb 14, 2017 at 11:42 AM, Thomas Garnier <thgarnie@google.com> wrote:
> The KVM segment_base function is confusing. This patch replaces integers
> with appropriate flags, simplify constructs and add comments.

It could pay to put this first in the series, but last is probably fine, too.

>
> Signed-off-by: Thomas Garnier <thgarnie@google.com>
> ---
> Based on next-20170213
> ---
>  arch/x86/kvm/vmx.c | 26 ++++++++++++++++++--------
>  1 file changed, 18 insertions(+), 8 deletions(-)
>
> diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
> index 99167f20bc34..edb8326108dd 100644
> --- a/arch/x86/kvm/vmx.c
> +++ b/arch/x86/kvm/vmx.c
> @@ -2062,25 +2062,35 @@ static unsigned long segment_base(u16 selector)
>         struct desc_struct *d;
>         unsigned long table_base;

This should go away IMO.  It should be struct desc_struct *table;

> +               table_base = get_current_gdt_rw_vaddr();

Then this can go away, too, and you can stop having the silly
get_current_gdt_rw_vaddr() function at all.

> +       d = (struct desc_struct *)table_base + (selector >> 3);

And this cast goes away too.

Re: [PATCH v3 4/4] KVM: VMX: Simplify segment_base

[Thomas Garnier]

On Tue, Feb 14, 2017 at 7:57 PM, Andy Lutomirski <luto@amacapital.net> wrote:
> On Tue, Feb 14, 2017 at 11:42 AM, Thomas Garnier <thgarnie@google.com> wrote:
>> The KVM segment_base function is confusing. This patch replaces integers
>> with appropriate flags, simplify constructs and add comments.
>
> It could pay to put this first in the series, but last is probably fine, too.
>

:)

>>
>> Signed-off-by: Thomas Garnier <thgarnie@google.com>
>> ---
>> Based on next-20170213
>> ---
>>  arch/x86/kvm/vmx.c | 26 ++++++++++++++++++--------
>>  1 file changed, 18 insertions(+), 8 deletions(-)
>>
>> diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
>> index 99167f20bc34..edb8326108dd 100644
>> --- a/arch/x86/kvm/vmx.c
>> +++ b/arch/x86/kvm/vmx.c
>> @@ -2062,25 +2062,35 @@ static unsigned long segment_base(u16 selector)
>>         struct desc_struct *d;
>>         unsigned long table_base;
>
> This should go away IMO.  It should be struct desc_struct *table;
>
>> +               table_base = get_current_gdt_rw_vaddr();
>
> Then this can go away, too, and you can stop having the silly
> get_current_gdt_rw_vaddr() function at all.
>
>> +       d = (struct desc_struct *)table_base + (selector >> 3);
>
> And this cast goes away too.

No problem.

-- 
Thomas

Re: [PATCH v3 4/4] KVM: VMX: Simplify segment_base

[Jim Mattson]

Can we use the read-only GDT here? When expanding the virtual address
for 64-bit system descriptors, isn't it sufficient to check (d->s == 0
&& d->type != 0)?

On Tue, Feb 14, 2017 at 11:42 AM, Thomas Garnier <thgarnie@google.com> wrote:
> The KVM segment_base function is confusing. This patch replaces integers
> with appropriate flags, simplify constructs and add comments.
>
> Signed-off-by: Thomas Garnier <thgarnie@google.com>
> ---
> Based on next-20170213
> ---
>  arch/x86/kvm/vmx.c | 26 ++++++++++++++++++--------
>  1 file changed, 18 insertions(+), 8 deletions(-)
>
> diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
> index 99167f20bc34..edb8326108dd 100644
> --- a/arch/x86/kvm/vmx.c
> +++ b/arch/x86/kvm/vmx.c
> @@ -2062,25 +2062,35 @@ static unsigned long segment_base(u16 selector)
>         struct desc_struct *d;
>         unsigned long table_base;
>         unsigned long v;
> +       u32 high32;
>
> -       if (!(selector & ~3))
> +       if (!(selector & ~SEGMENT_RPL_MASK))
>                 return 0;
>
> -       table_base = get_current_gdt_rw_vaddr();
> -
> -       if (selector & 4) {           /* from ldt */
> +       /* LDT selector */
> +       if ((selector & SEGMENT_TI_MASK) == SEGMENT_LDT) {
>                 u16 ldt_selector = kvm_read_ldt();
>
> -               if (!(ldt_selector & ~3))
> +               if (!(ldt_selector & ~SEGMENT_RPL_MASK))
>                         return 0;
>
>                 table_base = segment_base(ldt_selector);
> +       } else {
> +               table_base = get_current_gdt_rw_vaddr();
>         }
> -       d = (struct desc_struct *)(table_base + (selector & ~7));
> +
> +       d = (struct desc_struct *)table_base + (selector >> 3);
>         v = get_desc_base(d);
>  #ifdef CONFIG_X86_64
> -       if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
> -               v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
> +       /*
> +        * Extend the virtual address if we have a system descriptor entry for
> +        * LDT or TSS (available or busy).
> +        */
> +       if (d->s == 0 && (d->type == DESC_LDT || d->type == DESC_TSS ||
> +                         d->type == 11/*Busy TSS */)) {
> +               high32 = ((struct ldttss_desc64 *)d)->base3;
> +               v |= (u64)high32 << 32;
> +       }
>  #endif
>         return v;
>  }
> --
> 2.11.0.483.g087da7b7c-goog
>

Re: [PATCH v3 4/4] KVM: VMX: Simplify segment_base

[Thomas Garnier]

On Fri, Feb 17, 2017 at 9:49 AM, Jim Mattson <jmattson@google.com> wrote:
>
> Can we use the read-only GDT here? When expanding the virtual address
> for 64-bit system descriptors, isn't it sufficient to check (d->s == 0
> && d->type != 0)?

We can use the readonly GDT but I think doesn't matter one or the
other here. We have to check specific types for LDT or TSS, other
values describe other entries (cf Intel volume 3, 3.5) (for example 14
& 15 on 64-bits are for trap & interrupt gates).

>
>
> On Tue, Feb 14, 2017 at 11:42 AM, Thomas Garnier <thgarnie@google.com> wrote:
> > The KVM segment_base function is confusing. This patch replaces integers
> > with appropriate flags, simplify constructs and add comments.
> >
> > Signed-off-by: Thomas Garnier <thgarnie@google.com>
> > ---
> > Based on next-20170213
> > ---
> >  arch/x86/kvm/vmx.c | 26 ++++++++++++++++++--------
> >  1 file changed, 18 insertions(+), 8 deletions(-)
> >
> > diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
> > index 99167f20bc34..edb8326108dd 100644
> > --- a/arch/x86/kvm/vmx.c
> > +++ b/arch/x86/kvm/vmx.c
> > @@ -2062,25 +2062,35 @@ static unsigned long segment_base(u16 selector)
> >         struct desc_struct *d;
> >         unsigned long table_base;
> >         unsigned long v;
> > +       u32 high32;
> >
> > -       if (!(selector & ~3))
> > +       if (!(selector & ~SEGMENT_RPL_MASK))
> >                 return 0;
> >
> > -       table_base = get_current_gdt_rw_vaddr();
> > -
> > -       if (selector & 4) {           /* from ldt */
> > +       /* LDT selector */
> > +       if ((selector & SEGMENT_TI_MASK) == SEGMENT_LDT) {
> >                 u16 ldt_selector = kvm_read_ldt();
> >
> > -               if (!(ldt_selector & ~3))
> > +               if (!(ldt_selector & ~SEGMENT_RPL_MASK))
> >                         return 0;
> >
> >                 table_base = segment_base(ldt_selector);
> > +       } else {
> > +               table_base = get_current_gdt_rw_vaddr();
> >         }
> > -       d = (struct desc_struct *)(table_base + (selector & ~7));
> > +
> > +       d = (struct desc_struct *)table_base + (selector >> 3);
> >         v = get_desc_base(d);
> >  #ifdef CONFIG_X86_64
> > -       if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
> > -               v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
> > +       /*
> > +        * Extend the virtual address if we have a system descriptor entry for
> > +        * LDT or TSS (available or busy).
> > +        */
> > +       if (d->s == 0 && (d->type == DESC_LDT || d->type == DESC_TSS ||
> > +                         d->type == 11/*Busy TSS */)) {
> > +               high32 = ((struct ldttss_desc64 *)d)->base3;
> > +               v |= (u64)high32 << 32;
> > +       }
> >  #endif
> >         return v;
> >  }
> > --
> > 2.11.0.483.g087da7b7c-goog
> >




-- 
Thomas

Re: [PATCH v3 4/4] KVM: VMX: Simplify segment_base

[Jim Mattson]

On Fri, Feb 17, 2017 at 12:11 PM, Thomas Garnier <thgarnie@google.com> wrote:
> On Fri, Feb 17, 2017 at 9:49 AM, Jim Mattson <jmattson@google.com> wrote:
>>
>> Can we use the read-only GDT here? When expanding the virtual address
>> for 64-bit system descriptors, isn't it sufficient to check (d->s == 0
>> && d->type != 0)?
>
> We can use the readonly GDT but I think doesn't matter one or the
> other here. We have to check specific types for LDT or TSS, other
> values describe other entries (cf Intel volume 3, 3.5) (for example 14
> & 15 on 64-bits are for trap & interrupt gates).

According to volume 3 of the SDM, section 3.5.2:

The following system descriptors expand to 16 bytes:
— Call gate descriptors (see Section 5.8.3.1, “IA-32e Mode Call Gates”)
— IDT gate descriptors (see Section 6.14.1, “64-Bit Mode IDT”)
— LDT and TSS descriptors (see Section 7.2.3, “TSS Descriptor in 64-bit mode”).

All legal system descriptor types (except for 0: Upper 8 bytes of an 16-byte
descriptor) should get the high 32 bits of the base address from the next 8-byte
descriptor.

>
>>
>>
>> On Tue, Feb 14, 2017 at 11:42 AM, Thomas Garnier <thgarnie@google.com> wrote:
>> > The KVM segment_base function is confusing. This patch replaces integers
>> > with appropriate flags, simplify constructs and add comments.
>> >
>> > Signed-off-by: Thomas Garnier <thgarnie@google.com>
>> > ---
>> > Based on next-20170213
>> > ---
>> >  arch/x86/kvm/vmx.c | 26 ++++++++++++++++++--------
>> >  1 file changed, 18 insertions(+), 8 deletions(-)
>> >
>> > diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
>> > index 99167f20bc34..edb8326108dd 100644
>> > --- a/arch/x86/kvm/vmx.c
>> > +++ b/arch/x86/kvm/vmx.c
>> > @@ -2062,25 +2062,35 @@ static unsigned long segment_base(u16 selector)
>> >         struct desc_struct *d;
>> >         unsigned long table_base;
>> >         unsigned long v;
>> > +       u32 high32;
>> >
>> > -       if (!(selector & ~3))
>> > +       if (!(selector & ~SEGMENT_RPL_MASK))
>> >                 return 0;
>> >
>> > -       table_base = get_current_gdt_rw_vaddr();
>> > -
>> > -       if (selector & 4) {           /* from ldt */
>> > +       /* LDT selector */
>> > +       if ((selector & SEGMENT_TI_MASK) == SEGMENT_LDT) {
>> >                 u16 ldt_selector = kvm_read_ldt();
>> >
>> > -               if (!(ldt_selector & ~3))
>> > +               if (!(ldt_selector & ~SEGMENT_RPL_MASK))
>> >                         return 0;
>> >
>> >                 table_base = segment_base(ldt_selector);
>> > +       } else {
>> > +               table_base = get_current_gdt_rw_vaddr();
>> >         }
>> > -       d = (struct desc_struct *)(table_base + (selector & ~7));
>> > +
>> > +       d = (struct desc_struct *)table_base + (selector >> 3);
>> >         v = get_desc_base(d);
>> >  #ifdef CONFIG_X86_64
>> > -       if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
>> > -               v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
>> > +       /*
>> > +        * Extend the virtual address if we have a system descriptor entry for
>> > +        * LDT or TSS (available or busy).
>> > +        */
>> > +       if (d->s == 0 && (d->type == DESC_LDT || d->type == DESC_TSS ||
>> > +                         d->type == 11/*Busy TSS */)) {
>> > +               high32 = ((struct ldttss_desc64 *)d)->base3;
>> > +               v |= (u64)high32 << 32;
>> > +       }
>> >  #endif
>> >         return v;
>> >  }
>> > --
>> > 2.11.0.483.g087da7b7c-goog
>> >
>
>
>
>
> --
> Thomas

Re: [PATCH v3 4/4] KVM: VMX: Simplify segment_base

[Thomas Garnier]

On Fri, Feb 17, 2017 at 1:00 PM, Jim Mattson <jmattson@google.com> wrote:
> On Fri, Feb 17, 2017 at 12:11 PM, Thomas Garnier <thgarnie@google.com> wrote:
>> On Fri, Feb 17, 2017 at 9:49 AM, Jim Mattson <jmattson@google.com> wrote:
>>>
>>> Can we use the read-only GDT here? When expanding the virtual address
>>> for 64-bit system descriptors, isn't it sufficient to check (d->s == 0
>>> && d->type != 0)?
>>
>> We can use the readonly GDT but I think doesn't matter one or the
>> other here. We have to check specific types for LDT or TSS, other
>> values describe other entries (cf Intel volume 3, 3.5) (for example 14
>> & 15 on 64-bits are for trap & interrupt gates).
>
> According to volume 3 of the SDM, section 3.5.2:
>
> The following system descriptors expand to 16 bytes:
> — Call gate descriptors (see Section 5.8.3.1, “IA-32e Mode Call Gates”)
> — IDT gate descriptors (see Section 6.14.1, “64-Bit Mode IDT”)
> — LDT and TSS descriptors (see Section 7.2.3, “TSS Descriptor in 64-bit mode”).
>
> All legal system descriptor types (except for 0: Upper 8 bytes of an 16-byte
> descriptor) should get the high 32 bits of the base address from the next 8-byte
> descriptor.
>

Ok, then I will test an updated version next week.

>>
>>>
>>>
>>> On Tue, Feb 14, 2017 at 11:42 AM, Thomas Garnier <thgarnie@google.com> wrote:
>>> > The KVM segment_base function is confusing. This patch replaces integers
>>> > with appropriate flags, simplify constructs and add comments.
>>> >
>>> > Signed-off-by: Thomas Garnier <thgarnie@google.com>
>>> > ---
>>> > Based on next-20170213
>>> > ---
>>> >  arch/x86/kvm/vmx.c | 26 ++++++++++++++++++--------
>>> >  1 file changed, 18 insertions(+), 8 deletions(-)
>>> >
>>> > diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
>>> > index 99167f20bc34..edb8326108dd 100644
>>> > --- a/arch/x86/kvm/vmx.c
>>> > +++ b/arch/x86/kvm/vmx.c
>>> > @@ -2062,25 +2062,35 @@ static unsigned long segment_base(u16 selector)
>>> >         struct desc_struct *d;
>>> >         unsigned long table_base;
>>> >         unsigned long v;
>>> > +       u32 high32;
>>> >
>>> > -       if (!(selector & ~3))
>>> > +       if (!(selector & ~SEGMENT_RPL_MASK))
>>> >                 return 0;
>>> >
>>> > -       table_base = get_current_gdt_rw_vaddr();
>>> > -
>>> > -       if (selector & 4) {           /* from ldt */
>>> > +       /* LDT selector */
>>> > +       if ((selector & SEGMENT_TI_MASK) == SEGMENT_LDT) {
>>> >                 u16 ldt_selector = kvm_read_ldt();
>>> >
>>> > -               if (!(ldt_selector & ~3))
>>> > +               if (!(ldt_selector & ~SEGMENT_RPL_MASK))
>>> >                         return 0;
>>> >
>>> >                 table_base = segment_base(ldt_selector);
>>> > +       } else {
>>> > +               table_base = get_current_gdt_rw_vaddr();
>>> >         }
>>> > -       d = (struct desc_struct *)(table_base + (selector & ~7));
>>> > +
>>> > +       d = (struct desc_struct *)table_base + (selector >> 3);
>>> >         v = get_desc_base(d);
>>> >  #ifdef CONFIG_X86_64
>>> > -       if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
>>> > -               v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
>>> > +       /*
>>> > +        * Extend the virtual address if we have a system descriptor entry for
>>> > +        * LDT or TSS (available or busy).
>>> > +        */
>>> > +       if (d->s == 0 && (d->type == DESC_LDT || d->type == DESC_TSS ||
>>> > +                         d->type == 11/*Busy TSS */)) {
>>> > +               high32 = ((struct ldttss_desc64 *)d)->base3;
>>> > +               v |= (u64)high32 << 32;
>>> > +       }
>>> >  #endif
>>> >         return v;
>>> >  }
>>> > --
>>> > 2.11.0.483.g087da7b7c-goog
>>> >
>>
>>
>>
>>
>> --
>> Thomas



-- 
Thomas

Re: [PATCH v3 4/4] KVM: VMX: Simplify segment_base

[Andy Lutomirski]

On Fri, Feb 17, 2017 at 2:01 PM, Thomas Garnier <thgarnie@google.com> wrote:
> On Fri, Feb 17, 2017 at 1:00 PM, Jim Mattson <jmattson@google.com> wrote:
>> On Fri, Feb 17, 2017 at 12:11 PM, Thomas Garnier <thgarnie@google.com> wrote:
>>> On Fri, Feb 17, 2017 at 9:49 AM, Jim Mattson <jmattson@google.com> wrote:
>>>>
>>>> Can we use the read-only GDT here? When expanding the virtual address
>>>> for 64-bit system descriptors, isn't it sufficient to check (d->s == 0
>>>> && d->type != 0)?
>>>
>>> We can use the readonly GDT but I think doesn't matter one or the
>>> other here. We have to check specific types for LDT or TSS, other
>>> values describe other entries (cf Intel volume 3, 3.5) (for example 14
>>> & 15 on 64-bits are for trap & interrupt gates).
>>
>> According to volume 3 of the SDM, section 3.5.2:
>>
>> The following system descriptors expand to 16 bytes:
>> — Call gate descriptors (see Section 5.8.3.1, “IA-32e Mode Call Gates”)
>> — IDT gate descriptors (see Section 6.14.1, “64-Bit Mode IDT”)
>> — LDT and TSS descriptors (see Section 7.2.3, “TSS Descriptor in 64-bit mode”).
>>
>> All legal system descriptor types (except for 0: Upper 8 bytes of an 16-byte
>> descriptor) should get the high 32 bits of the base address from the next 8-byte
>> descriptor.
>>
>
> Ok, then I will test an updated version next week.
>

I'm going to send out some preliminary patches that just get rid of
this problem entirely.

Re: [PATCH v3 4/4] KVM: VMX: Simplify segment_base

[Thomas Garnier]

On Mon, Feb 20, 2017 at 8:56 AM, Andy Lutomirski <luto@amacapital.net> wrote:
> On Fri, Feb 17, 2017 at 2:01 PM, Thomas Garnier <thgarnie@google.com> wrote:
>> On Fri, Feb 17, 2017 at 1:00 PM, Jim Mattson <jmattson@google.com> wrote:
>>> On Fri, Feb 17, 2017 at 12:11 PM, Thomas Garnier <thgarnie@google.com> wrote:
>>>> On Fri, Feb 17, 2017 at 9:49 AM, Jim Mattson <jmattson@google.com> wrote:
>>>>>
>>>>> Can we use the read-only GDT here? When expanding the virtual address
>>>>> for 64-bit system descriptors, isn't it sufficient to check (d->s == 0
>>>>> && d->type != 0)?
>>>>
>>>> We can use the readonly GDT but I think doesn't matter one or the
>>>> other here. We have to check specific types for LDT or TSS, other
>>>> values describe other entries (cf Intel volume 3, 3.5) (for example 14
>>>> & 15 on 64-bits are for trap & interrupt gates).
>>>
>>> According to volume 3 of the SDM, section 3.5.2:
>>>
>>> The following system descriptors expand to 16 bytes:
>>> — Call gate descriptors (see Section 5.8.3.1, “IA-32e Mode Call Gates”)
>>> — IDT gate descriptors (see Section 6.14.1, “64-Bit Mode IDT”)
>>> — LDT and TSS descriptors (see Section 7.2.3, “TSS Descriptor in 64-bit mode”).
>>>
>>> All legal system descriptor types (except for 0: Upper 8 bytes of an 16-byte
>>> descriptor) should get the high 32 bits of the base address from the next 8-byte
>>> descriptor.
>>>
>>
>> Ok, then I will test an updated version next week.
>>
>
> I'm going to send out some preliminary patches that just get rid of
> this problem entirely.

Okay, I guess I will have to wait for it to be integrated to
linux-next then. Or would you rather to it after this patch set is
added?

-- 
Thomas

Re: [PATCH v3 4/4] KVM: VMX: Simplify segment_base

[Thomas Garnier]

On Mon, Feb 20, 2017 at 9:28 AM, Thomas Garnier <thgarnie@google.com> wrote:
> On Mon, Feb 20, 2017 at 8:56 AM, Andy Lutomirski <luto@amacapital.net> wrote:
>> On Fri, Feb 17, 2017 at 2:01 PM, Thomas Garnier <thgarnie@google.com> wrote:
>>> On Fri, Feb 17, 2017 at 1:00 PM, Jim Mattson <jmattson@google.com> wrote:
>>>> On Fri, Feb 17, 2017 at 12:11 PM, Thomas Garnier <thgarnie@google.com> wrote:
>>>>> On Fri, Feb 17, 2017 at 9:49 AM, Jim Mattson <jmattson@google.com> wrote:
>>>>>>
>>>>>> Can we use the read-only GDT here? When expanding the virtual address
>>>>>> for 64-bit system descriptors, isn't it sufficient to check (d->s == 0
>>>>>> && d->type != 0)?
>>>>>
>>>>> We can use the readonly GDT but I think doesn't matter one or the
>>>>> other here. We have to check specific types for LDT or TSS, other
>>>>> values describe other entries (cf Intel volume 3, 3.5) (for example 14
>>>>> & 15 on 64-bits are for trap & interrupt gates).
>>>>
>>>> According to volume 3 of the SDM, section 3.5.2:
>>>>
>>>> The following system descriptors expand to 16 bytes:
>>>> — Call gate descriptors (see Section 5.8.3.1, “IA-32e Mode Call Gates”)
>>>> — IDT gate descriptors (see Section 6.14.1, “64-Bit Mode IDT”)
>>>> — LDT and TSS descriptors (see Section 7.2.3, “TSS Descriptor in 64-bit mode”).
>>>>
>>>> All legal system descriptor types (except for 0: Upper 8 bytes of an 16-byte
>>>> descriptor) should get the high 32 bits of the base address from the next 8-byte
>>>> descriptor.
>>>>
>>>
>>> Ok, then I will test an updated version next week.
>>>
>>
>> I'm going to send out some preliminary patches that just get rid of
>> this problem entirely.
>
> Okay, I guess I will have to wait for it to be integrated to
> linux-next then. Or would you rather to it after this patch set is
> added?
>

Read your summary for the patchset of KVM cleanup, I will wait for it
to reach linux-next to rebase and send the new iteration.

Thanks for working on the clean-up.

> --
> Thomas



-- 
Thomas

Re: [PATCH v3 4/4] KVM: VMX: Simplify segment_base

[Ingo Molnar]

* Thomas Garnier <thgarnie@google.com> wrote:

> > Okay, I guess I will have to wait for it to be integrated to
> > linux-next then. Or would you rather to it after this patch set is
> > added?
> 
> Read your summary for the patchset of KVM cleanup, I will wait for it to reach 
> linux-next to rebase and send the new iteration.

Note that to be able to apply your readonly-GDT series to the x86 tree I'll need a 
stable SHA1 to base it on.

Paolo, how stable, non-rebasing are the KVM tree commits? Once Andy's patches hit 
the KVM tree I could pull from you and order it so that I'd send it to Linus only 
after you sent your bits upstream. That would save us 3 months of patch 
propagation latency.

Or should we keep Andy's KVM patches together with the GDT patches? Either 
workflow works for me - it's your call as these are predominantly KVM changes.

Thanks,

	Ingo

Re: [PATCH v3 4/4] KVM: VMX: Simplify segment_base

[Paolo Bonzini]

> Paolo, how stable, non-rebasing are the KVM tree commits?

Whatever ends in linux-next is stable.  I have a separate rebasing branch,
but it's not part of linux-next by design.

> Or should we keep Andy's KVM patches together with the GDT patches? Either
> workflow works for me - it's your call as these are predominantly KVM
> changes.

I'll delay my pull request to Linus a couple days so that I can test
Andy's 6 patches. Then you can just base your branch on Linus's tree.

Paolo

Re: [PATCH v3 4/4] KVM: VMX: Simplify segment_base

[Ingo Molnar]

* Paolo Bonzini <pbonzini@redhat.com> wrote:

> > Paolo, how stable, non-rebasing are the KVM tree commits?
> 
> Whatever ends in linux-next is stable.  I have a separate rebasing branch,
> but it's not part of linux-next by design.

Ok, that's nice!

> > Or should we keep Andy's KVM patches together with the GDT patches? Either 
> > workflow works for me - it's your call as these are predominantly KVM changes.
> 
> I'll delay my pull request to Linus a couple days so that I can test Andy's 6 
> patches. Then you can just base your branch on Linus's tree.

Fantastic, thank you!

	Ingo

Re: [PATCH v3 1/4] x86/mm: Adapt MODULES_END based on Fixmap section size

[Borislav Petkov]

On Tue, Feb 14, 2017 at 11:42:56AM -0800, Thomas Garnier wrote:
> This patch aligns MODULES_END to the beginning of the Fixmap section.
> It optimizes the space available for both sections. The address is
> pre-computed based on the number of pages required by the Fixmap
> section.
> 
> It will allow GDT remapping in the Fixmap section. The current
> MODULES_END static address does not provide enough space for the kernel
> to support a large number of processors.
> 
> Signed-off-by: Thomas Garnier <thgarnie@google.com>
> ---
> Based on next-20170213
> ---
>  arch/x86/include/asm/fixmap.h           | 8 ++++++++
>  arch/x86/include/asm/pgtable_64_types.h | 3 ---
>  arch/x86/kernel/module.c                | 1 +
>  arch/x86/mm/dump_pagetables.c           | 1 +
>  arch/x86/mm/kasan_init_64.c             | 1 +
>  5 files changed, 11 insertions(+), 3 deletions(-)
> 
> diff --git a/arch/x86/include/asm/fixmap.h b/arch/x86/include/asm/fixmap.h
> index 8554f960e21b..20231189e0e3 100644
> --- a/arch/x86/include/asm/fixmap.h
> +++ b/arch/x86/include/asm/fixmap.h
> @@ -132,6 +132,14 @@ enum fixed_addresses {
>  
>  extern void reserve_top_address(unsigned long reserve);
>  
> +/* On 64-bit, the module sections ends with the start of the fixmap */
> +#ifdef CONFIG_X86_64
> +#define MODULES_VADDR    (__START_KERNEL_map + KERNEL_IMAGE_SIZE)
> +#define MODULES_END   __fix_to_virt(__end_of_fixed_addresses + 1)
> +#define MODULES_LEN   (MODULES_END - MODULES_VADDR)
> +#endif /* CONFIG_X86_64 */

JFYI: so there's another patchset which adds KERNEL_MAPPING_SIZE:

https://lkml.kernel.org/r/1486040077-3719-1-git-send-email-bhe@redhat.com

and makes it a 1G, i.e., the KASLR default. I guess the above will have
to be KERNEL_MAPPING_SIZE then.

And why are you moving those to fixmap.h? What's wrong with
including fixmap.h into pgtable_64_types.h so that you can get
__end_of_fixed_addresses?

FWIW, I didn't even have to add any includes with my .config, i.e., that builds:

---
diff --git a/arch/x86/include/asm/pgtable_64_types.h b/arch/x86/include/asm/pgtable_64_types.h
index 3a264200c62f..eda7fa856fa9 100644
--- a/arch/x86/include/asm/pgtable_64_types.h
+++ b/arch/x86/include/asm/pgtable_64_types.h
@@ -67,7 +67,7 @@ typedef struct { pteval_t pte; } pte_t;
 #endif /* CONFIG_RANDOMIZE_MEMORY */
 #define VMALLOC_END	(VMALLOC_START + _AC((VMALLOC_SIZE_TB << 40) - 1, UL))
 #define MODULES_VADDR    (__START_KERNEL_map + KERNEL_IMAGE_SIZE)
-#define MODULES_END      _AC(0xffffffffff000000, UL)
+#define MODULES_END   __fix_to_virt(__end_of_fixed_addresses + 1)
 #define MODULES_LEN   (MODULES_END - MODULES_VADDR)
 #define ESPFIX_PGD_ENTRY _AC(-2, UL)
 #define ESPFIX_BASE_ADDR (ESPFIX_PGD_ENTRY << PGDIR_SHIFT)
---

but I wouldn't be surprised if some strange configuration would need it.

>  #define FIXADDR_SIZE	(__end_of_permanent_fixed_addresses << PAGE_SHIFT)
>  #define FIXADDR_START		(FIXADDR_TOP - FIXADDR_SIZE)
>  
> diff --git a/arch/x86/include/asm/pgtable_64_types.h b/arch/x86/include/asm/pgtable_64_types.h
> index 3a264200c62f..de8bace10200 100644
> --- a/arch/x86/include/asm/pgtable_64_types.h
> +++ b/arch/x86/include/asm/pgtable_64_types.h
> @@ -66,9 +66,6 @@ typedef struct { pteval_t pte; } pte_t;
>  #define VMEMMAP_START	__VMEMMAP_BASE
>  #endif /* CONFIG_RANDOMIZE_MEMORY */
>  #define VMALLOC_END	(VMALLOC_START + _AC((VMALLOC_SIZE_TB << 40) - 1, UL))
> -#define MODULES_VADDR    (__START_KERNEL_map + KERNEL_IMAGE_SIZE)
> -#define MODULES_END      _AC(0xffffffffff000000, UL)

How much of an ABI breakage would that be? See
Documentation/x86/x86_64/mm.txt.

With my .config MODULES_END becomes 0xffffffffff1fe000 and it'll remain
dynamic depending on .config. No idea how much in userspace relies on
MODULES_END being static 0xffffffffff000000...

Hmm.

-- 
Regards/Gruss,
    Boris.

SUSE Linux GmbH, GF: Felix Imendörffer, Jane Smithard, Graham Norton, HRB 21284 (AG Nürnberg)
-- 

Re: [PATCH v3 1/4] x86/mm: Adapt MODULES_END based on Fixmap section size

[Thomas Garnier]

On Wed, Feb 15, 2017 at 5:58 AM, Borislav Petkov <bp@suse.de> wrote:
>
> On Tue, Feb 14, 2017 at 11:42:56AM -0800, Thomas Garnier wrote:
> > This patch aligns MODULES_END to the beginning of the Fixmap section.
> > It optimizes the space available for both sections. The address is
> > pre-computed based on the number of pages required by the Fixmap
> > section.
> >
> > It will allow GDT remapping in the Fixmap section. The current
> > MODULES_END static address does not provide enough space for the kernel
> > to support a large number of processors.
> >
> > Signed-off-by: Thomas Garnier <thgarnie@google.com>
> > ---
> > Based on next-20170213
> > ---
> >  arch/x86/include/asm/fixmap.h           | 8 ++++++++
> >  arch/x86/include/asm/pgtable_64_types.h | 3 ---
> >  arch/x86/kernel/module.c                | 1 +
> >  arch/x86/mm/dump_pagetables.c           | 1 +
> >  arch/x86/mm/kasan_init_64.c             | 1 +
> >  5 files changed, 11 insertions(+), 3 deletions(-)
> >
> > diff --git a/arch/x86/include/asm/fixmap.h b/arch/x86/include/asm/fixmap.h
> > index 8554f960e21b..20231189e0e3 100644
> > --- a/arch/x86/include/asm/fixmap.h
> > +++ b/arch/x86/include/asm/fixmap.h
> > @@ -132,6 +132,14 @@ enum fixed_addresses {
> >
> >  extern void reserve_top_address(unsigned long reserve);
> >
> > +/* On 64-bit, the module sections ends with the start of the fixmap */
> > +#ifdef CONFIG_X86_64
> > +#define MODULES_VADDR    (__START_KERNEL_map + KERNEL_IMAGE_SIZE)
> > +#define MODULES_END   __fix_to_virt(__end_of_fixed_addresses + 1)
> > +#define MODULES_LEN   (MODULES_END - MODULES_VADDR)
> > +#endif /* CONFIG_X86_64 */
>
> JFYI: so there's another patchset which adds KERNEL_MAPPING_SIZE:
>
> https://lkml.kernel.org/r/1486040077-3719-1-git-send-email-bhe@redhat.com
>
> and makes it a 1G, i.e., the KASLR default. I guess the above will have
> to be KERNEL_MAPPING_SIZE then.
>
> And why are you moving those to fixmap.h? What's wrong with
> including fixmap.h into pgtable_64_types.h so that you can get
> __end_of_fixed_addresses?
>
> FWIW, I didn't even have to add any includes with my .config, i.e., that builds:
>

I tried to add fixmap.h and I thought I tried without too but maybe
not, I will try to adapt it on v4.

> ---
> diff --git a/arch/x86/include/asm/pgtable_64_types.h b/arch/x86/include/asm/pgtable_64_types.h
> index 3a264200c62f..eda7fa856fa9 100644
> --- a/arch/x86/include/asm/pgtable_64_types.h
> +++ b/arch/x86/include/asm/pgtable_64_types.h
> @@ -67,7 +67,7 @@ typedef struct { pteval_t pte; } pte_t;
>  #endif /* CONFIG_RANDOMIZE_MEMORY */
>  #define VMALLOC_END    (VMALLOC_START + _AC((VMALLOC_SIZE_TB << 40) - 1, UL))
>  #define MODULES_VADDR    (__START_KERNEL_map + KERNEL_IMAGE_SIZE)
> -#define MODULES_END      _AC(0xffffffffff000000, UL)
> +#define MODULES_END   __fix_to_virt(__end_of_fixed_addresses + 1)
>  #define MODULES_LEN   (MODULES_END - MODULES_VADDR)
>  #define ESPFIX_PGD_ENTRY _AC(-2, UL)
>  #define ESPFIX_BASE_ADDR (ESPFIX_PGD_ENTRY << PGDIR_SHIFT)
> ---
>
> but I wouldn't be surprised if some strange configuration would need it.
>
> >  #define FIXADDR_SIZE (__end_of_permanent_fixed_addresses << PAGE_SHIFT)
> >  #define FIXADDR_START                (FIXADDR_TOP - FIXADDR_SIZE)
> >
> > diff --git a/arch/x86/include/asm/pgtable_64_types.h b/arch/x86/include/asm/pgtable_64_types.h
> > index 3a264200c62f..de8bace10200 100644
> > --- a/arch/x86/include/asm/pgtable_64_types.h
> > +++ b/arch/x86/include/asm/pgtable_64_types.h
> > @@ -66,9 +66,6 @@ typedef struct { pteval_t pte; } pte_t;
> >  #define VMEMMAP_START        __VMEMMAP_BASE
> >  #endif /* CONFIG_RANDOMIZE_MEMORY */
> >  #define VMALLOC_END  (VMALLOC_START + _AC((VMALLOC_SIZE_TB << 40) - 1, UL))
> > -#define MODULES_VADDR    (__START_KERNEL_map + KERNEL_IMAGE_SIZE)
> > -#define MODULES_END      _AC(0xffffffffff000000, UL)
>
> How much of an ABI breakage would that be? See
> Documentation/x86/x86_64/mm.txt.
>
> With my .config MODULES_END becomes 0xffffffffff1fe000 and it'll remain
> dynamic depending on .config. No idea how much in userspace relies on
> MODULES_END being static 0xffffffffff000000...
>
> Hmm.
>

Why do you think they rely on it being static? The VSYSCALL address is
not changed for example.

> --
> Regards/Gruss,
>     Boris.
>
> SUSE Linux GmbH, GF: Felix Imendörffer, Jane Smithard, Graham Norton, HRB 21284 (AG Nürnberg)
> --

Thanks for the feedback.


-- 
Thomas

Re: [PATCH v3 1/4] x86/mm: Adapt MODULES_END based on Fixmap section size

[Borislav Petkov]

On Wed, Feb 15, 2017 at 07:40:51AM -0800, Thomas Garnier wrote:
> Why do you think they rely on it being static? The VSYSCALL address is
> not changed for example.

I don't know, that's why I'm asking first. Userspace is known to pick
different aspects of the kernel and deciding to use them for whatever.
And once it relies on that, there's no changing it.

-- 
Regards/Gruss,
    Boris.

SUSE Linux GmbH, GF: Felix Imendörffer, Jane Smithard, Graham Norton, HRB 21284 (AG Nürnberg)
-- 

Re: [PATCH v3 1/4] x86/mm: Adapt MODULES_END based on Fixmap section size

[Borislav Petkov]

On Wed, Feb 15, 2017 at 04:49:21PM +0100, Borislav Petkov wrote:
> On Wed, Feb 15, 2017 at 07:40:51AM -0800, Thomas Garnier wrote:
> > Why do you think they rely on it being static? The VSYSCALL address is
> > not changed for example.
> 
> I don't know, that's why I'm asking first. Userspace is known to pick
> different aspects of the kernel and deciding to use them for whatever.
> And once it relies on that, there's no changing it.

Ok, so we should be fine here because those addresses are internal to
the kernel, I'm being told, so anything using them will break and gets
to keep the pieces even.

-- 
Regards/Gruss,
    Boris.

SUSE Linux GmbH, GF: Felix Imendörffer, Jane Smithard, Graham Norton, HRB 21284 (AG Nürnberg)
--