v8/include/v8-internal.h

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

930 lines
38 KiB
C
Raw Permalink Normal View History

// Copyright 2018 the V8 project authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.
#ifndef INCLUDE_V8_INTERNAL_H_
#define INCLUDE_V8_INTERNAL_H_
#include <stddef.h>
#include <stdint.h>
#include <string.h>
Reland "Reland "[sandbox] Sandboxify EmbedderDataSlots"" This is a reland of commit eca383c947da0c4faa890f72e4f89c6c8f592062 More DCHECK failures have been fixed with https://crrev.com/c/3857423 Original change's description: > Reland "[sandbox] Sandboxify EmbedderDataSlots" > > This is a reland of commit e1f585ed94d799555793a4150281ce2baf22a666 > > ExternalPointerTable issues have been fixed in > https://crrev.com/c/3849650 and https://crrev.com/c/3849376 > > Original change's description: > > [sandbox] Sandboxify EmbedderDataSlots > > > > Bug: v8:10391 > > Change-Id: If85a308a6f6ed1b17d86f87b4911c82d2327ea72 > > Cq-Include-Trybots: luci.v8.try:v8_linux64_heap_sandbox_dbg_ng,v8_linux_arm64_sim_heap_sandbox_dbg_ng > > Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3757341 > > Reviewed-by: Igor Sheludko <ishell@chromium.org> > > Commit-Queue: Samuel Groß <saelo@chromium.org> > > Cr-Commit-Position: refs/heads/main@{#82623} > > Bug: v8:10391 > Change-Id: If77f6c10e81c30c2dfa6b33c788bc4a36e4da135 > Cq-Include-Trybots: luci.v8.try:v8_linux64_heap_sandbox_dbg_ng,v8_linux_arm64_sim_heap_sandbox_dbg_ng > Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3852602 > Commit-Queue: Samuel Groß <saelo@chromium.org> > Reviewed-by: Igor Sheludko <ishell@chromium.org> > Cr-Commit-Position: refs/heads/main@{#82686} Bug: v8:10391 Change-Id: Id982c022e50004e903851b160d30fc8767280e5b Cq-Include-Trybots: luci.v8.try:v8_linux64_heap_sandbox_dbg_ng,v8_linux_arm64_sim_heap_sandbox_dbg_ng Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3854679 Reviewed-by: Igor Sheludko <ishell@chromium.org> Commit-Queue: Samuel Groß <saelo@chromium.org> Cr-Commit-Position: refs/heads/main@{#82767}
2022-08-25 12:24:48 +00:00
#include <atomic>
#include <type_traits>
#include "v8-version.h" // NOLINT(build/include_directory)
#include "v8config.h" // NOLINT(build/include_directory)
namespace v8 {
enum CpuFeature {
#if V8_TARGET_ARCH_IA32 || V8_TARGET_ARCH_X64
SSE4_2,
SSE4_1,
SSSE3,
SSE3,
SAHF,
AVX,
AVX2,
FMA3,
BMI1,
BMI2,
LZCNT,
POPCNT,
INTEL_ATOM,
CETSS,
#elif V8_TARGET_ARCH_ARM
// - Standard configurations. The baseline is ARMv6+VFPv2.
ARMv7, // ARMv7-A + VFPv3-D32 + NEON
ARMv7_SUDIV, // ARMv7-A + VFPv4-D32 + NEON + SUDIV
ARMv8, // ARMv8-A (+ all of the above)
// ARM feature aliases (based on the standard configurations above).
VFPv3 = ARMv7,
NEON = ARMv7,
VFP32DREGS = ARMv7,
SUDIV = ARMv7_SUDIV,
#elif V8_TARGET_ARCH_ARM64
JSCVT,
#elif V8_TARGET_ARCH_MIPS64
FPU,
FP64FPU,
MIPSr1,
MIPSr2,
MIPSr6,
MIPS_SIMD, // MSA instructions
#elif V8_TARGET_ARCH_LOONG64
FPU,
#elif V8_TARGET_ARCH_PPC || V8_TARGET_ARCH_PPC64
PPC_6_PLUS,
PPC_7_PLUS,
PPC_8_PLUS,
PPC_9_PLUS,
PPC_10_PLUS,
#elif V8_TARGET_ARCH_S390X
FPU,
DISTINCT_OPS,
GENERAL_INSTR_EXT,
FLOATING_POINT_EXT,
VECTOR_FACILITY,
VECTOR_ENHANCE_FACILITY_1,
VECTOR_ENHANCE_FACILITY_2,
MISC_INSTR_EXT2,
#elif V8_TARGET_ARCH_RISCV64
FPU,
FP64FPU,
RISCV_SIMD,
#elif V8_TARGET_ARCH_RISCV32
FPU,
FP64FPU,
RISCV_SIMD,
#endif
NUMBER_OF_CPU_FEATURES
};
class Array;
class Context;
class Data;
class Isolate;
template <typename T>
class Local;
namespace internal {
class Isolate;
typedef uintptr_t Address;
static const Address kNullAddress = 0;
[sandbox] Implement GC for the external pointer table The external pointer table is now managed by the GC, which marks entries that are alive during major GC, then sweeps the table afterwards to free all dead entries and build a free list from them. For now, only major GCs are supported, Scavenger GCs do not interact with the external pointer table. In more detail, garbage collection of the external pointer table works as follows: 1. The external pointer table now reserves a large region of virtual address space for its backing buffer and is then never reallocated, only grown in place until the maximum size is reached. 2. When the GC's marking visitor marks a HeapObject with an external pointer as alive, it also marks the corresponding external pointer table entry as alive. This can happen on a background thread. 3. For that, it uses the MSB of each entry in the table to indicate whether the entry has been marked or not. This works because the MSB is always cleared during the AND-based type check performed when accessing an external pointer. 4. After marking, the external pointer table is swept while the mutator is stopped. This builds an inline, singly-linked freelist of all newly-dead and previously-free entries. 5. When allocating an entry from the table, the first entry on the freelist is used. If the freelist is empty, the table grows, populating the freelist with the new entries. 6. Every newly-allocated entry is marked as alive, and every store to an existing entry also automatically marks that entry as alive (by also setting the MSB). This simplifies the design of the table GC with regards to concurrency (See ExternalPointerTable::Mark). Bug: v8:10391 Change-Id: I8877fdf5576af3761bde65298951bb09e601bd14 Cq-Include-Trybots: luci.v8.try:v8_linux64_heap_sandbox_dbg_ng,v8_linux_arm64_sim_heap_sandbox_dbg_ng Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3359625 Reviewed-by: Michael Lippautz <mlippautz@chromium.org> Reviewed-by: Maya Lekova <mslekova@chromium.org> Reviewed-by: Igor Sheludko <ishell@chromium.org> Commit-Queue: Samuel Groß <saelo@chromium.org> Cr-Commit-Position: refs/heads/main@{#78708}
2022-01-20 16:01:41 +00:00
constexpr int KB = 1024;
constexpr int MB = KB * 1024;
constexpr int GB = MB * 1024;
#ifdef V8_TARGET_ARCH_X64
constexpr size_t TB = size_t{GB} * 1024;
#endif
/**
* Configuration of tagging scheme.
*/
const int kApiSystemPointerSize = sizeof(void*);
const int kApiDoubleSize = sizeof(double);
const int kApiInt32Size = sizeof(int32_t);
const int kApiInt64Size = sizeof(int64_t);
const int kApiSizetSize = sizeof(size_t);
// Tag information for HeapObject.
const int kHeapObjectTag = 1;
const int kWeakHeapObjectTag = 3;
const int kHeapObjectTagSize = 2;
const intptr_t kHeapObjectTagMask = (1 << kHeapObjectTagSize) - 1;
const intptr_t kHeapObjectReferenceTagMask = 1 << (kHeapObjectTagSize - 1);
Allowing map word to be used for other state in GC header. This CL adds features to pack/unpack map words. Currently V8 cannot store extra metadata in object headers -- because V8 objects do not have a proper header, but only a map pointer at the start of the object. To store per-object metadata like marking data, a side table is required as the per-object metadata storage. This CL enables V8 to use higher unused bits in a 64-bit map word as per-object metadata storage. Map pointer stores come with an extra step to encode the metadata into the pointer (we call it "map packing"). Map pointer loads will also remove the metadata bits as well (we call it "map packing"). Since the map word is no longer a valid pointer after packing, we also change the tag of the packed map word to make it looks like a Smi. This helps various GC and barrier code to correctly skip them instead of blindly dereferencing this invalid pointer. A ninja flag `v8_enable_map_packing` is provided to turn this map-packing feature on and off. It is disabled by default. * Only works on x64 platform, with `v8_enable_pointer_compression` set to `false` Bug: v8:11624 Change-Id: Ia2bdf79553945e5fc0b0874c87803d2cc733e073 Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2247561 Reviewed-by: Ulan Degenbaev <ulan@chromium.org> Reviewed-by: Nico Hartmann <nicohartmann@chromium.org> Reviewed-by: Toon Verwaest <verwaest@chromium.org> Reviewed-by: Georg Neis <neis@chromium.org> Commit-Queue: Ulan Degenbaev <ulan@chromium.org> Cr-Commit-Position: refs/heads/master@{#73915}
2021-04-06 12:01:44 +00:00
// Tag information for fowarding pointers stored in object headers.
// 0b00 at the lowest 2 bits in the header indicates that the map word is a
// forwarding pointer.
const int kForwardingTag = 0;
const int kForwardingTagSize = 2;
const intptr_t kForwardingTagMask = (1 << kForwardingTagSize) - 1;
// Tag information for Smi.
const int kSmiTag = 0;
const int kSmiTagSize = 1;
const intptr_t kSmiTagMask = (1 << kSmiTagSize) - 1;
template <size_t tagged_ptr_size>
struct SmiTagging;
constexpr intptr_t kIntptrAllBitsSet = intptr_t{-1};
constexpr uintptr_t kUintptrAllBitsSet =
static_cast<uintptr_t>(kIntptrAllBitsSet);
// Smi constants for systems where tagged pointer is a 32-bit value.
template <>
struct SmiTagging<4> {
enum { kSmiShiftSize = 0, kSmiValueSize = 31 };
static constexpr intptr_t kSmiMinValue =
static_cast<intptr_t>(kUintptrAllBitsSet << (kSmiValueSize - 1));
static constexpr intptr_t kSmiMaxValue = -(kSmiMinValue + 1);
V8_INLINE static int SmiToInt(const internal::Address value) {
int shift_bits = kSmiTagSize + kSmiShiftSize;
// Truncate and shift down (requires >> to be sign extending).
return static_cast<int32_t>(static_cast<uint32_t>(value)) >> shift_bits;
}
V8_INLINE static constexpr bool IsValidSmi(intptr_t value) {
// Is value in range [kSmiMinValue, kSmiMaxValue].
// Use unsigned operations in order to avoid undefined behaviour in case of
// signed integer overflow.
return (static_cast<uintptr_t>(value) -
static_cast<uintptr_t>(kSmiMinValue)) <=
(static_cast<uintptr_t>(kSmiMaxValue) -
static_cast<uintptr_t>(kSmiMinValue));
}
};
// Smi constants for systems where tagged pointer is a 64-bit value.
template <>
struct SmiTagging<8> {
enum { kSmiShiftSize = 31, kSmiValueSize = 32 };
static constexpr intptr_t kSmiMinValue =
static_cast<intptr_t>(kUintptrAllBitsSet << (kSmiValueSize - 1));
static constexpr intptr_t kSmiMaxValue = -(kSmiMinValue + 1);
V8_INLINE static int SmiToInt(const internal::Address value) {
int shift_bits = kSmiTagSize + kSmiShiftSize;
// Shift down and throw away top 32 bits.
return static_cast<int>(static_cast<intptr_t>(value) >> shift_bits);
}
V8_INLINE static constexpr bool IsValidSmi(intptr_t value) {
// To be representable as a long smi, the value must be a 32-bit integer.
return (value == static_cast<int32_t>(value));
}
};
#ifdef V8_COMPRESS_POINTERS
[sandbox] Implement GC for the external pointer table The external pointer table is now managed by the GC, which marks entries that are alive during major GC, then sweeps the table afterwards to free all dead entries and build a free list from them. For now, only major GCs are supported, Scavenger GCs do not interact with the external pointer table. In more detail, garbage collection of the external pointer table works as follows: 1. The external pointer table now reserves a large region of virtual address space for its backing buffer and is then never reallocated, only grown in place until the maximum size is reached. 2. When the GC's marking visitor marks a HeapObject with an external pointer as alive, it also marks the corresponding external pointer table entry as alive. This can happen on a background thread. 3. For that, it uses the MSB of each entry in the table to indicate whether the entry has been marked or not. This works because the MSB is always cleared during the AND-based type check performed when accessing an external pointer. 4. After marking, the external pointer table is swept while the mutator is stopped. This builds an inline, singly-linked freelist of all newly-dead and previously-free entries. 5. When allocating an entry from the table, the first entry on the freelist is used. If the freelist is empty, the table grows, populating the freelist with the new entries. 6. Every newly-allocated entry is marked as alive, and every store to an existing entry also automatically marks that entry as alive (by also setting the MSB). This simplifies the design of the table GC with regards to concurrency (See ExternalPointerTable::Mark). Bug: v8:10391 Change-Id: I8877fdf5576af3761bde65298951bb09e601bd14 Cq-Include-Trybots: luci.v8.try:v8_linux64_heap_sandbox_dbg_ng,v8_linux_arm64_sim_heap_sandbox_dbg_ng Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3359625 Reviewed-by: Michael Lippautz <mlippautz@chromium.org> Reviewed-by: Maya Lekova <mslekova@chromium.org> Reviewed-by: Igor Sheludko <ishell@chromium.org> Commit-Queue: Samuel Groß <saelo@chromium.org> Cr-Commit-Position: refs/heads/main@{#78708}
2022-01-20 16:01:41 +00:00
// See v8:7703 or src/common/ptr-compr-inl.h for details about pointer
// compression.
constexpr size_t kPtrComprCageReservationSize = size_t{1} << 32;
constexpr size_t kPtrComprCageBaseAlignment = size_t{1} << 32;
static_assert(
kApiSystemPointerSize == kApiInt64Size,
"Pointer compression can be enabled only for 64-bit architectures");
const int kApiTaggedSize = kApiInt32Size;
#else
const int kApiTaggedSize = kApiSystemPointerSize;
#endif
constexpr bool PointerCompressionIsEnabled() {
return kApiTaggedSize != kApiSystemPointerSize;
}
[sandbox] Implement GC for the external pointer table The external pointer table is now managed by the GC, which marks entries that are alive during major GC, then sweeps the table afterwards to free all dead entries and build a free list from them. For now, only major GCs are supported, Scavenger GCs do not interact with the external pointer table. In more detail, garbage collection of the external pointer table works as follows: 1. The external pointer table now reserves a large region of virtual address space for its backing buffer and is then never reallocated, only grown in place until the maximum size is reached. 2. When the GC's marking visitor marks a HeapObject with an external pointer as alive, it also marks the corresponding external pointer table entry as alive. This can happen on a background thread. 3. For that, it uses the MSB of each entry in the table to indicate whether the entry has been marked or not. This works because the MSB is always cleared during the AND-based type check performed when accessing an external pointer. 4. After marking, the external pointer table is swept while the mutator is stopped. This builds an inline, singly-linked freelist of all newly-dead and previously-free entries. 5. When allocating an entry from the table, the first entry on the freelist is used. If the freelist is empty, the table grows, populating the freelist with the new entries. 6. Every newly-allocated entry is marked as alive, and every store to an existing entry also automatically marks that entry as alive (by also setting the MSB). This simplifies the design of the table GC with regards to concurrency (See ExternalPointerTable::Mark). Bug: v8:10391 Change-Id: I8877fdf5576af3761bde65298951bb09e601bd14 Cq-Include-Trybots: luci.v8.try:v8_linux64_heap_sandbox_dbg_ng,v8_linux_arm64_sim_heap_sandbox_dbg_ng Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3359625 Reviewed-by: Michael Lippautz <mlippautz@chromium.org> Reviewed-by: Maya Lekova <mslekova@chromium.org> Reviewed-by: Igor Sheludko <ishell@chromium.org> Commit-Queue: Samuel Groß <saelo@chromium.org> Cr-Commit-Position: refs/heads/main@{#78708}
2022-01-20 16:01:41 +00:00
#ifdef V8_31BIT_SMIS_ON_64BIT_ARCH
using PlatformSmiTagging = SmiTagging<kApiInt32Size>;
#else
using PlatformSmiTagging = SmiTagging<kApiTaggedSize>;
#endif
// TODO(ishell): Consinder adding kSmiShiftBits = kSmiShiftSize + kSmiTagSize
// since it's used much more often than the inividual constants.
const int kSmiShiftSize = PlatformSmiTagging::kSmiShiftSize;
const int kSmiValueSize = PlatformSmiTagging::kSmiValueSize;
const int kSmiMinValue = static_cast<int>(PlatformSmiTagging::kSmiMinValue);
const int kSmiMaxValue = static_cast<int>(PlatformSmiTagging::kSmiMaxValue);
constexpr bool SmiValuesAre31Bits() { return kSmiValueSize == 31; }
constexpr bool SmiValuesAre32Bits() { return kSmiValueSize == 32; }
V8_INLINE static constexpr internal::Address IntToSmi(int value) {
return (static_cast<Address>(value) << (kSmiTagSize + kSmiShiftSize)) |
kSmiTag;
}
/*
* Sandbox related types, constants, and functions.
*/
constexpr bool SandboxIsEnabled() {
#ifdef V8_ENABLE_SANDBOX
[sandbox] Implement GC for the external pointer table The external pointer table is now managed by the GC, which marks entries that are alive during major GC, then sweeps the table afterwards to free all dead entries and build a free list from them. For now, only major GCs are supported, Scavenger GCs do not interact with the external pointer table. In more detail, garbage collection of the external pointer table works as follows: 1. The external pointer table now reserves a large region of virtual address space for its backing buffer and is then never reallocated, only grown in place until the maximum size is reached. 2. When the GC's marking visitor marks a HeapObject with an external pointer as alive, it also marks the corresponding external pointer table entry as alive. This can happen on a background thread. 3. For that, it uses the MSB of each entry in the table to indicate whether the entry has been marked or not. This works because the MSB is always cleared during the AND-based type check performed when accessing an external pointer. 4. After marking, the external pointer table is swept while the mutator is stopped. This builds an inline, singly-linked freelist of all newly-dead and previously-free entries. 5. When allocating an entry from the table, the first entry on the freelist is used. If the freelist is empty, the table grows, populating the freelist with the new entries. 6. Every newly-allocated entry is marked as alive, and every store to an existing entry also automatically marks that entry as alive (by also setting the MSB). This simplifies the design of the table GC with regards to concurrency (See ExternalPointerTable::Mark). Bug: v8:10391 Change-Id: I8877fdf5576af3761bde65298951bb09e601bd14 Cq-Include-Trybots: luci.v8.try:v8_linux64_heap_sandbox_dbg_ng,v8_linux_arm64_sim_heap_sandbox_dbg_ng Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3359625 Reviewed-by: Michael Lippautz <mlippautz@chromium.org> Reviewed-by: Maya Lekova <mslekova@chromium.org> Reviewed-by: Igor Sheludko <ishell@chromium.org> Commit-Queue: Samuel Groß <saelo@chromium.org> Cr-Commit-Position: refs/heads/main@{#78708}
2022-01-20 16:01:41 +00:00
return true;
#else
return false;
#endif
}
// SandboxedPointers are guaranteed to point into the sandbox. This is achieved
// for example by storing them as offset rather than as raw pointers.
using SandboxedPointer_t = Address;
#ifdef V8_ENABLE_SANDBOX
[sandbox] Implement GC for the external pointer table The external pointer table is now managed by the GC, which marks entries that are alive during major GC, then sweeps the table afterwards to free all dead entries and build a free list from them. For now, only major GCs are supported, Scavenger GCs do not interact with the external pointer table. In more detail, garbage collection of the external pointer table works as follows: 1. The external pointer table now reserves a large region of virtual address space for its backing buffer and is then never reallocated, only grown in place until the maximum size is reached. 2. When the GC's marking visitor marks a HeapObject with an external pointer as alive, it also marks the corresponding external pointer table entry as alive. This can happen on a background thread. 3. For that, it uses the MSB of each entry in the table to indicate whether the entry has been marked or not. This works because the MSB is always cleared during the AND-based type check performed when accessing an external pointer. 4. After marking, the external pointer table is swept while the mutator is stopped. This builds an inline, singly-linked freelist of all newly-dead and previously-free entries. 5. When allocating an entry from the table, the first entry on the freelist is used. If the freelist is empty, the table grows, populating the freelist with the new entries. 6. Every newly-allocated entry is marked as alive, and every store to an existing entry also automatically marks that entry as alive (by also setting the MSB). This simplifies the design of the table GC with regards to concurrency (See ExternalPointerTable::Mark). Bug: v8:10391 Change-Id: I8877fdf5576af3761bde65298951bb09e601bd14 Cq-Include-Trybots: luci.v8.try:v8_linux64_heap_sandbox_dbg_ng,v8_linux_arm64_sim_heap_sandbox_dbg_ng Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3359625 Reviewed-by: Michael Lippautz <mlippautz@chromium.org> Reviewed-by: Maya Lekova <mslekova@chromium.org> Reviewed-by: Igor Sheludko <ishell@chromium.org> Commit-Queue: Samuel Groß <saelo@chromium.org> Cr-Commit-Position: refs/heads/main@{#78708}
2022-01-20 16:01:41 +00:00
// Size of the sandbox, excluding the guard regions surrounding it.
#ifdef V8_TARGET_OS_ANDROID
// On Android, most 64-bit devices seem to be configured with only 39 bits of
// virtual address space for userspace. As such, limit the sandbox to 128GB (a
// quarter of the total available address space).
constexpr size_t kSandboxSizeLog2 = 37; // 128 GB
#else
// Everywhere else use a 1TB sandbox.
[sandbox] Implement GC for the external pointer table The external pointer table is now managed by the GC, which marks entries that are alive during major GC, then sweeps the table afterwards to free all dead entries and build a free list from them. For now, only major GCs are supported, Scavenger GCs do not interact with the external pointer table. In more detail, garbage collection of the external pointer table works as follows: 1. The external pointer table now reserves a large region of virtual address space for its backing buffer and is then never reallocated, only grown in place until the maximum size is reached. 2. When the GC's marking visitor marks a HeapObject with an external pointer as alive, it also marks the corresponding external pointer table entry as alive. This can happen on a background thread. 3. For that, it uses the MSB of each entry in the table to indicate whether the entry has been marked or not. This works because the MSB is always cleared during the AND-based type check performed when accessing an external pointer. 4. After marking, the external pointer table is swept while the mutator is stopped. This builds an inline, singly-linked freelist of all newly-dead and previously-free entries. 5. When allocating an entry from the table, the first entry on the freelist is used. If the freelist is empty, the table grows, populating the freelist with the new entries. 6. Every newly-allocated entry is marked as alive, and every store to an existing entry also automatically marks that entry as alive (by also setting the MSB). This simplifies the design of the table GC with regards to concurrency (See ExternalPointerTable::Mark). Bug: v8:10391 Change-Id: I8877fdf5576af3761bde65298951bb09e601bd14 Cq-Include-Trybots: luci.v8.try:v8_linux64_heap_sandbox_dbg_ng,v8_linux_arm64_sim_heap_sandbox_dbg_ng Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3359625 Reviewed-by: Michael Lippautz <mlippautz@chromium.org> Reviewed-by: Maya Lekova <mslekova@chromium.org> Reviewed-by: Igor Sheludko <ishell@chromium.org> Commit-Queue: Samuel Groß <saelo@chromium.org> Cr-Commit-Position: refs/heads/main@{#78708}
2022-01-20 16:01:41 +00:00
constexpr size_t kSandboxSizeLog2 = 40; // 1 TB
#endif // V8_TARGET_OS_ANDROID
[sandbox] Implement GC for the external pointer table The external pointer table is now managed by the GC, which marks entries that are alive during major GC, then sweeps the table afterwards to free all dead entries and build a free list from them. For now, only major GCs are supported, Scavenger GCs do not interact with the external pointer table. In more detail, garbage collection of the external pointer table works as follows: 1. The external pointer table now reserves a large region of virtual address space for its backing buffer and is then never reallocated, only grown in place until the maximum size is reached. 2. When the GC's marking visitor marks a HeapObject with an external pointer as alive, it also marks the corresponding external pointer table entry as alive. This can happen on a background thread. 3. For that, it uses the MSB of each entry in the table to indicate whether the entry has been marked or not. This works because the MSB is always cleared during the AND-based type check performed when accessing an external pointer. 4. After marking, the external pointer table is swept while the mutator is stopped. This builds an inline, singly-linked freelist of all newly-dead and previously-free entries. 5. When allocating an entry from the table, the first entry on the freelist is used. If the freelist is empty, the table grows, populating the freelist with the new entries. 6. Every newly-allocated entry is marked as alive, and every store to an existing entry also automatically marks that entry as alive (by also setting the MSB). This simplifies the design of the table GC with regards to concurrency (See ExternalPointerTable::Mark). Bug: v8:10391 Change-Id: I8877fdf5576af3761bde65298951bb09e601bd14 Cq-Include-Trybots: luci.v8.try:v8_linux64_heap_sandbox_dbg_ng,v8_linux_arm64_sim_heap_sandbox_dbg_ng Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3359625 Reviewed-by: Michael Lippautz <mlippautz@chromium.org> Reviewed-by: Maya Lekova <mslekova@chromium.org> Reviewed-by: Igor Sheludko <ishell@chromium.org> Commit-Queue: Samuel Groß <saelo@chromium.org> Cr-Commit-Position: refs/heads/main@{#78708}
2022-01-20 16:01:41 +00:00
constexpr size_t kSandboxSize = 1ULL << kSandboxSizeLog2;
// Required alignment of the sandbox. For simplicity, we require the
// size of the guard regions to be a multiple of this, so that this specifies
// the alignment of the sandbox including and excluding surrounding guard
// regions. The alignment requirement is due to the pointer compression cage
// being located at the start of the sandbox.
constexpr size_t kSandboxAlignment = kPtrComprCageBaseAlignment;
// Sandboxed pointers are stored inside the heap as offset from the sandbox
// base shifted to the left. This way, it is guaranteed that the offset is
// smaller than the sandbox size after shifting it to the right again. This
// constant specifies the shift amount.
constexpr uint64_t kSandboxedPointerShift = 64 - kSandboxSizeLog2;
// Size of the guard regions surrounding the sandbox. This assumes a worst-case
// scenario of a 32-bit unsigned index used to access an array of 64-bit
// values.
constexpr size_t kSandboxGuardRegionSize = 32ULL * GB;
static_assert((kSandboxGuardRegionSize % kSandboxAlignment) == 0,
"The size of the guard regions around the sandbox must be a "
"multiple of its required alignment.");
// On OSes where reserving virtual memory is too expensive to reserve the
// entire address space backing the sandbox, notably Windows pre 8.1, we create
// a partially reserved sandbox that doesn't actually reserve most of the
// memory, and so doesn't have the desired security properties as unrelated
// memory allocations could end up inside of it, but which still ensures that
// objects that should be located inside the sandbox are allocated within
// kSandboxSize bytes from the start of the sandbox. The minimum size of the
// region that is actually reserved for such a sandbox is specified by this
// constant and should be big enough to contain the pointer compression cage as
// well as the ArrayBuffer partition.
constexpr size_t kSandboxMinimumReservationSize = 8ULL * GB;
static_assert(kSandboxMinimumReservationSize > kPtrComprCageReservationSize,
"The minimum reservation size for a sandbox must be larger than "
"the pointer compression cage contained within it.");
// The maximum buffer size allowed inside the sandbox. This is mostly dependent
// on the size of the guard regions around the sandbox: an attacker must not be
// able to construct a buffer that appears larger than the guard regions and
// thereby "reach out of" the sandbox.
constexpr size_t kMaxSafeBufferSizeForSandbox = 32ULL * GB - 1;
static_assert(kMaxSafeBufferSizeForSandbox <= kSandboxGuardRegionSize,
"The maximum allowed buffer size must not be larger than the "
"sandbox's guard regions");
constexpr size_t kBoundedSizeShift = 29;
static_assert(1ULL << (64 - kBoundedSizeShift) ==
kMaxSafeBufferSizeForSandbox + 1,
"The maximum size of a BoundedSize must be synchronized with the "
"kMaxSafeBufferSizeForSandbox");
#endif // V8_ENABLE_SANDBOX
#ifdef V8_COMPRESS_POINTERS
#ifdef V8_TARGET_OS_ANDROID
// The size of the virtual memory reservation for an external pointer table.
// This determines the maximum number of entries in a table. Using a maximum
// size allows omitting bounds checks on table accesses if the indices are
// guaranteed (e.g. through shifting) to be below the maximum index. This
// value must be a power of two.
static const size_t kExternalPointerTableReservationSize = 512 * MB;
// The external pointer table indices stored in HeapObjects as external
// pointers are shifted to the left by this amount to guarantee that they are
// smaller than the maximum table size.
static const uint32_t kExternalPointerIndexShift = 6;
#else
static const size_t kExternalPointerTableReservationSize = 1024 * MB;
static const uint32_t kExternalPointerIndexShift = 5;
#endif // V8_TARGET_OS_ANDROID
// The maximum number of entries in an external pointer table.
static const size_t kMaxExternalPointers =
kExternalPointerTableReservationSize / kApiSystemPointerSize;
static_assert((1 << (32 - kExternalPointerIndexShift)) == kMaxExternalPointers,
"kExternalPointerTableReservationSize and "
"kExternalPointerIndexShift don't match");
#else // !V8_COMPRESS_POINTERS
// Needed for the V8.SandboxedExternalPointersCount histogram.
static const size_t kMaxExternalPointers = 0;
#endif // V8_COMPRESS_POINTERS
[sandbox] Implement GC for the external pointer table The external pointer table is now managed by the GC, which marks entries that are alive during major GC, then sweeps the table afterwards to free all dead entries and build a free list from them. For now, only major GCs are supported, Scavenger GCs do not interact with the external pointer table. In more detail, garbage collection of the external pointer table works as follows: 1. The external pointer table now reserves a large region of virtual address space for its backing buffer and is then never reallocated, only grown in place until the maximum size is reached. 2. When the GC's marking visitor marks a HeapObject with an external pointer as alive, it also marks the corresponding external pointer table entry as alive. This can happen on a background thread. 3. For that, it uses the MSB of each entry in the table to indicate whether the entry has been marked or not. This works because the MSB is always cleared during the AND-based type check performed when accessing an external pointer. 4. After marking, the external pointer table is swept while the mutator is stopped. This builds an inline, singly-linked freelist of all newly-dead and previously-free entries. 5. When allocating an entry from the table, the first entry on the freelist is used. If the freelist is empty, the table grows, populating the freelist with the new entries. 6. Every newly-allocated entry is marked as alive, and every store to an existing entry also automatically marks that entry as alive (by also setting the MSB). This simplifies the design of the table GC with regards to concurrency (See ExternalPointerTable::Mark). Bug: v8:10391 Change-Id: I8877fdf5576af3761bde65298951bb09e601bd14 Cq-Include-Trybots: luci.v8.try:v8_linux64_heap_sandbox_dbg_ng,v8_linux_arm64_sim_heap_sandbox_dbg_ng Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3359625 Reviewed-by: Michael Lippautz <mlippautz@chromium.org> Reviewed-by: Maya Lekova <mslekova@chromium.org> Reviewed-by: Igor Sheludko <ishell@chromium.org> Commit-Queue: Samuel Groß <saelo@chromium.org> Cr-Commit-Position: refs/heads/main@{#78708}
2022-01-20 16:01:41 +00:00
[sandbox] Prepare ExternalPointerTable rollout This CL does the following: - It enables (i.e. allocates and initializes) the per-Isolate ExternalPointerTable when the sandbox is enabled. - It refactors the list of external pointer tags to mark them as "sandboxed" or "unsandboxed". An unsandboxed external pointer has a null tag. - It changes V8_SANDBOXED_EXTERNAL_POINTERS to now essentially just enable sandboxing for all available tags. - It modifies all low-level external pointer accessors to perform the ExternalPointerLookup only if the tag is non-zero and otherwise treat the slot as containing a raw pointer. This now allows rolling out external pointer sandboxing incrementally (separately for each external pointer type), which will in turn allow for more precise performance measurements of the impact of the sandbox. Note: when an external pointer tag is now marked as sandboxed (and V8_SANDBOXED_EXTERNAL_POINTERS is not enabled), the underlying slots are still 64-bits in size. This simplifies the implementation as we would otherwise need to deal with variably-sized external pointer slots. Local benchmarking suggests that the benefits from 32-bit external pointer slots are insignificant on typical benchmarks, so this should be ok. Drive-by: rename kExternalPointerSize to kExternalPointerSlotSize to make it more clear what it refers to (the on-heap storage size). Also delete CodeStubAssembler::InitializeExternalPointerField as it is not currently used and the implementation is fairly inefficient. Bug: v8:10391 Change-Id: I7c38729c7e9048d737a1a8ced84749f5b1f7feab Cq-Include-Trybots: luci.v8.try:v8_linux64_heap_sandbox_dbg_ng,v8_linux_arm64_sim_heap_sandbox_dbg_ng Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3736447 Reviewed-by: Igor Sheludko <ishell@chromium.org> Reviewed-by: Michael Lippautz <mlippautz@chromium.org> Reviewed-by: Leszek Swirski <leszeks@chromium.org> Reviewed-by: Andreas Haas <ahaas@chromium.org> Commit-Queue: Samuel Groß <saelo@chromium.org> Reviewed-by: Nico Hartmann <nicohartmann@chromium.org> Cr-Commit-Position: refs/heads/main@{#81636}
2022-07-11 12:21:59 +00:00
// A ExternalPointerHandle represents a (opaque) reference to an external
// pointer that can be stored inside the sandbox. A ExternalPointerHandle has
// meaning only in combination with an (active) Isolate as it references an
// external pointer stored in the currently active Isolate's
// ExternalPointerTable. Internally, an ExternalPointerHandles is simply an
// index into an ExternalPointerTable that is shifted to the left to guarantee
// that it is smaller than the size of the table.
using ExternalPointerHandle = uint32_t;
// ExternalPointers point to objects located outside the sandbox. When
// sandboxed external pointers are enabled, these are stored on heap as
// ExternalPointerHandles, otherwise they are simply raw pointers.
Reland "[sandbox] Fold V8_SANDBOXED_EXTERNAL_POINTERS into V8_ENABLE_SANDBOX" This is a reland of commit 49c59678301fafcd7f70221cdd7936253a229093 The non-deterministic snapshot issue has been fixed by using the correct field size for CodeDataContainers in serializer.cc. Original change's description: > [sandbox] Fold V8_SANDBOXED_EXTERNAL_POINTERS into V8_ENABLE_SANDBOX > > Now that all external pointers have been sandboxed, > V8_SANDBOXED_EXTERNAL_POINTERS is no longer needed. This change also > shrinks external pointer slots to 32 bits when the sandbox is enabled. > > Bug: v8:10391 > Change-Id: Iccbef27ac107b988cb23fe9ef66da6fe0bae087a > Cq-Include-Trybots: luci.v8.try:v8_linux64_heap_sandbox_dbg_ng,v8_linux_arm64_sim_heap_sandbox_dbg_ng > Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3869269 > Reviewed-by: Leszek Swirski <leszeks@chromium.org> > Reviewed-by: Manos Koukoutos <manoskouk@chromium.org> > Reviewed-by: Nico Hartmann <nicohartmann@chromium.org> > Reviewed-by: Igor Sheludko <ishell@chromium.org> > Commit-Queue: Samuel Groß <saelo@chromium.org> > Cr-Commit-Position: refs/heads/main@{#83083} Bug: v8:10391 Change-Id: I29870404406902d99ba6016c570cc0c4d05c6c85 Cq-Include-Trybots: luci.v8.try:v8_linux64_heap_sandbox_dbg_ng,v8_linux_arm64_sim_heap_sandbox_dbg_ng Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3887899 Reviewed-by: Igor Sheludko <ishell@chromium.org> Reviewed-by: Leszek Swirski <leszeks@chromium.org> Reviewed-by: Nico Hartmann <nicohartmann@chromium.org> Reviewed-by: Manos Koukoutos <manoskouk@chromium.org> Commit-Queue: Samuel Groß <saelo@chromium.org> Cr-Commit-Position: refs/heads/main@{#83130}
2022-09-12 07:20:02 +00:00
#ifdef V8_ENABLE_SANDBOX
[sandbox] Prepare ExternalPointerTable rollout This CL does the following: - It enables (i.e. allocates and initializes) the per-Isolate ExternalPointerTable when the sandbox is enabled. - It refactors the list of external pointer tags to mark them as "sandboxed" or "unsandboxed". An unsandboxed external pointer has a null tag. - It changes V8_SANDBOXED_EXTERNAL_POINTERS to now essentially just enable sandboxing for all available tags. - It modifies all low-level external pointer accessors to perform the ExternalPointerLookup only if the tag is non-zero and otherwise treat the slot as containing a raw pointer. This now allows rolling out external pointer sandboxing incrementally (separately for each external pointer type), which will in turn allow for more precise performance measurements of the impact of the sandbox. Note: when an external pointer tag is now marked as sandboxed (and V8_SANDBOXED_EXTERNAL_POINTERS is not enabled), the underlying slots are still 64-bits in size. This simplifies the implementation as we would otherwise need to deal with variably-sized external pointer slots. Local benchmarking suggests that the benefits from 32-bit external pointer slots are insignificant on typical benchmarks, so this should be ok. Drive-by: rename kExternalPointerSize to kExternalPointerSlotSize to make it more clear what it refers to (the on-heap storage size). Also delete CodeStubAssembler::InitializeExternalPointerField as it is not currently used and the implementation is fairly inefficient. Bug: v8:10391 Change-Id: I7c38729c7e9048d737a1a8ced84749f5b1f7feab Cq-Include-Trybots: luci.v8.try:v8_linux64_heap_sandbox_dbg_ng,v8_linux_arm64_sim_heap_sandbox_dbg_ng Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3736447 Reviewed-by: Igor Sheludko <ishell@chromium.org> Reviewed-by: Michael Lippautz <mlippautz@chromium.org> Reviewed-by: Leszek Swirski <leszeks@chromium.org> Reviewed-by: Andreas Haas <ahaas@chromium.org> Commit-Queue: Samuel Groß <saelo@chromium.org> Reviewed-by: Nico Hartmann <nicohartmann@chromium.org> Cr-Commit-Position: refs/heads/main@{#81636}
2022-07-11 12:21:59 +00:00
using ExternalPointer_t = ExternalPointerHandle;
#else
using ExternalPointer_t = Address;
#endif
// When the sandbox is enabled, external pointers are stored in an external
// pointer table and are referenced from HeapObjects through an index (a
// "handle"). When stored in the table, the pointers are tagged with per-type
// tags to prevent type confusion attacks between different external objects.
// Besides type information bits, these tags also contain the GC marking bit
// which indicates whether the pointer table entry is currently alive. When a
[sandbox] Prepare ExternalPointerTable rollout This CL does the following: - It enables (i.e. allocates and initializes) the per-Isolate ExternalPointerTable when the sandbox is enabled. - It refactors the list of external pointer tags to mark them as "sandboxed" or "unsandboxed". An unsandboxed external pointer has a null tag. - It changes V8_SANDBOXED_EXTERNAL_POINTERS to now essentially just enable sandboxing for all available tags. - It modifies all low-level external pointer accessors to perform the ExternalPointerLookup only if the tag is non-zero and otherwise treat the slot as containing a raw pointer. This now allows rolling out external pointer sandboxing incrementally (separately for each external pointer type), which will in turn allow for more precise performance measurements of the impact of the sandbox. Note: when an external pointer tag is now marked as sandboxed (and V8_SANDBOXED_EXTERNAL_POINTERS is not enabled), the underlying slots are still 64-bits in size. This simplifies the implementation as we would otherwise need to deal with variably-sized external pointer slots. Local benchmarking suggests that the benefits from 32-bit external pointer slots are insignificant on typical benchmarks, so this should be ok. Drive-by: rename kExternalPointerSize to kExternalPointerSlotSize to make it more clear what it refers to (the on-heap storage size). Also delete CodeStubAssembler::InitializeExternalPointerField as it is not currently used and the implementation is fairly inefficient. Bug: v8:10391 Change-Id: I7c38729c7e9048d737a1a8ced84749f5b1f7feab Cq-Include-Trybots: luci.v8.try:v8_linux64_heap_sandbox_dbg_ng,v8_linux_arm64_sim_heap_sandbox_dbg_ng Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3736447 Reviewed-by: Igor Sheludko <ishell@chromium.org> Reviewed-by: Michael Lippautz <mlippautz@chromium.org> Reviewed-by: Leszek Swirski <leszeks@chromium.org> Reviewed-by: Andreas Haas <ahaas@chromium.org> Commit-Queue: Samuel Groß <saelo@chromium.org> Reviewed-by: Nico Hartmann <nicohartmann@chromium.org> Cr-Commit-Position: refs/heads/main@{#81636}
2022-07-11 12:21:59 +00:00
// pointer is written into the table, the tag is ORed into the top bits. When
// that pointer is later loaded from the table, it is ANDed with the inverse of
// the expected tag. If the expected and actual type differ, this will leave
// some of the top bits of the pointer set, rendering the pointer inaccessible.
// The AND operation also removes the GC marking bit from the pointer.
//
// The tags are constructed such that UNTAG(TAG(0, T1), T2) != 0 for any two
// (distinct) tags T1 and T2. In practice, this is achieved by generating tags
// that all have the same number of zeroes and ones but different bit patterns.
// With N type tag bits, this allows for (N choose N/2) possible type tags.
// Besides the type tag bits, the tags also have the GC marking bit set so that
// the marking bit is automatically set when a pointer is written into the
// external pointer table (in which case it is clearly alive) and is cleared
// when the pointer is loaded. The exception to this is the free entry tag,
// which doesn't have the mark bit set, as the entry is not alive. This
// construction allows performing the type check and removing GC marking bits
// from the pointer in one efficient operation (bitwise AND). The number of
// available bits is limited in the following way: on x64, bits [47, 64) are
// generally available for tagging (userspace has 47 address bits available).
// On Arm64, userspace typically has a 40 or 48 bit address space. However, due
// to top-byte ignore (TBI) and memory tagging (MTE), the top byte is unusable
// for type checks as type-check failures would go unnoticed or collide with
// MTE bits. Some bits of the top byte can, however, still be used for the GC
// marking bit. The bits available for the type tags are therefore limited to
// [48, 56), i.e. (8 choose 4) = 70 different types.
// The following options exist to increase the number of possible types:
// - Using multiple ExternalPointerTables since tags can safely be reused
// across different tables
// - Using "extended" type checks, where additional type information is stored
// either in an adjacent pointer table entry or at the pointed-to location
// - Using a different tagging scheme, for example based on XOR which would
// allow for 2**8 different tags but require a separate operation to remove
// the marking bit
//
// The external pointer sandboxing mechanism ensures that every access to an
// external pointer field will result in a valid pointer of the expected type
// even in the presence of an attacker able to corrupt memory inside the
// sandbox. However, if any data related to the external object is stored
// inside the sandbox it may still be corrupted and so must be validated before
// use or moved into the external object. Further, an attacker will always be
// able to substitute different external pointers of the same type for each
// other. Therefore, code using external pointers must be written in a
// "substitution-safe" way, i.e. it must always be possible to substitute
// external pointers of the same type without causing memory corruption outside
// of the sandbox. Generally this is achieved by referencing any group of
// related external objects through a single external pointer.
//
// Currently we use bit 62 for the marking bit which should always be unused as
// it's part of the non-canonical address range. When Arm's top-byte ignore
// (TBI) is enabled, this bit will be part of the ignored byte, and we assume
// that the Embedder is not using this byte (really only this one bit) for any
// other purpose. This bit also does not collide with the memory tagging
// extension (MTE) which would use bits [56, 60).
//
// External pointer tables are also available even when the sandbox is off but
// pointer compression is on. In that case, the mechanism can be used to easy
// alignment requirements as it turns unaligned 64-bit raw pointers into
// aligned 32-bit indices. To "opt-in" to the external pointer table mechanism
// for this purpose, instead of using the ExternalPointer accessors one needs to
// use ExternalPointerHandles directly and use them to access the pointers in an
// ExternalPointerTable.
constexpr uint64_t kExternalPointerMarkBit = 1ULL << 62;
constexpr uint64_t kExternalPointerTagMask = 0x40ff000000000000;
constexpr uint64_t kExternalPointerTagShift = 48;
// All possible 8-bit type tags.
// These are sorted so that tags can be grouped together and it can efficiently
// be checked if a tag belongs to a given group. See for example the
// IsSharedExternalPointerType routine.
constexpr uint64_t kAllExternalPointerTypeTags[] = {
0b00001111, 0b00010111, 0b00011011, 0b00011101, 0b00011110, 0b00100111,
0b00101011, 0b00101101, 0b00101110, 0b00110011, 0b00110101, 0b00110110,
0b00111001, 0b00111010, 0b00111100, 0b01000111, 0b01001011, 0b01001101,
0b01001110, 0b01010011, 0b01010101, 0b01010110, 0b01011001, 0b01011010,
0b01011100, 0b01100011, 0b01100101, 0b01100110, 0b01101001, 0b01101010,
0b01101100, 0b01110001, 0b01110010, 0b01110100, 0b01111000, 0b10000111,
0b10001011, 0b10001101, 0b10001110, 0b10010011, 0b10010101, 0b10010110,
0b10011001, 0b10011010, 0b10011100, 0b10100011, 0b10100101, 0b10100110,
0b10101001, 0b10101010, 0b10101100, 0b10110001, 0b10110010, 0b10110100,
0b10111000, 0b11000011, 0b11000101, 0b11000110, 0b11001001, 0b11001010,
0b11001100, 0b11010001, 0b11010010, 0b11010100, 0b11011000, 0b11100001,
0b11100010, 0b11100100, 0b11101000, 0b11110000};
#define TAG(i) \
((kAllExternalPointerTypeTags[i] << kExternalPointerTagShift) | \
kExternalPointerMarkBit)
[sandbox] Implement GC for the external pointer table The external pointer table is now managed by the GC, which marks entries that are alive during major GC, then sweeps the table afterwards to free all dead entries and build a free list from them. For now, only major GCs are supported, Scavenger GCs do not interact with the external pointer table. In more detail, garbage collection of the external pointer table works as follows: 1. The external pointer table now reserves a large region of virtual address space for its backing buffer and is then never reallocated, only grown in place until the maximum size is reached. 2. When the GC's marking visitor marks a HeapObject with an external pointer as alive, it also marks the corresponding external pointer table entry as alive. This can happen on a background thread. 3. For that, it uses the MSB of each entry in the table to indicate whether the entry has been marked or not. This works because the MSB is always cleared during the AND-based type check performed when accessing an external pointer. 4. After marking, the external pointer table is swept while the mutator is stopped. This builds an inline, singly-linked freelist of all newly-dead and previously-free entries. 5. When allocating an entry from the table, the first entry on the freelist is used. If the freelist is empty, the table grows, populating the freelist with the new entries. 6. Every newly-allocated entry is marked as alive, and every store to an existing entry also automatically marks that entry as alive (by also setting the MSB). This simplifies the design of the table GC with regards to concurrency (See ExternalPointerTable::Mark). Bug: v8:10391 Change-Id: I8877fdf5576af3761bde65298951bb09e601bd14 Cq-Include-Trybots: luci.v8.try:v8_linux64_heap_sandbox_dbg_ng,v8_linux_arm64_sim_heap_sandbox_dbg_ng Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3359625 Reviewed-by: Michael Lippautz <mlippautz@chromium.org> Reviewed-by: Maya Lekova <mslekova@chromium.org> Reviewed-by: Igor Sheludko <ishell@chromium.org> Commit-Queue: Samuel Groß <saelo@chromium.org> Cr-Commit-Position: refs/heads/main@{#78708}
2022-01-20 16:01:41 +00:00
// clang-format off
// When adding new tags, please ensure that the code using these tags is
// "substitution-safe", i.e. still operate safely if external pointers of the
// same type are swapped by an attacker. See comment above for more details.
[sandbox] Prepare ExternalPointerTable rollout This CL does the following: - It enables (i.e. allocates and initializes) the per-Isolate ExternalPointerTable when the sandbox is enabled. - It refactors the list of external pointer tags to mark them as "sandboxed" or "unsandboxed". An unsandboxed external pointer has a null tag. - It changes V8_SANDBOXED_EXTERNAL_POINTERS to now essentially just enable sandboxing for all available tags. - It modifies all low-level external pointer accessors to perform the ExternalPointerLookup only if the tag is non-zero and otherwise treat the slot as containing a raw pointer. This now allows rolling out external pointer sandboxing incrementally (separately for each external pointer type), which will in turn allow for more precise performance measurements of the impact of the sandbox. Note: when an external pointer tag is now marked as sandboxed (and V8_SANDBOXED_EXTERNAL_POINTERS is not enabled), the underlying slots are still 64-bits in size. This simplifies the implementation as we would otherwise need to deal with variably-sized external pointer slots. Local benchmarking suggests that the benefits from 32-bit external pointer slots are insignificant on typical benchmarks, so this should be ok. Drive-by: rename kExternalPointerSize to kExternalPointerSlotSize to make it more clear what it refers to (the on-heap storage size). Also delete CodeStubAssembler::InitializeExternalPointerField as it is not currently used and the implementation is fairly inefficient. Bug: v8:10391 Change-Id: I7c38729c7e9048d737a1a8ced84749f5b1f7feab Cq-Include-Trybots: luci.v8.try:v8_linux64_heap_sandbox_dbg_ng,v8_linux_arm64_sim_heap_sandbox_dbg_ng Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3736447 Reviewed-by: Igor Sheludko <ishell@chromium.org> Reviewed-by: Michael Lippautz <mlippautz@chromium.org> Reviewed-by: Leszek Swirski <leszeks@chromium.org> Reviewed-by: Andreas Haas <ahaas@chromium.org> Commit-Queue: Samuel Groß <saelo@chromium.org> Reviewed-by: Nico Hartmann <nicohartmann@chromium.org> Cr-Commit-Position: refs/heads/main@{#81636}
2022-07-11 12:21:59 +00:00
// Shared external pointers are owned by the shared Isolate and stored in the
// shared external pointer table associated with that Isolate, where they can
// be accessed from multiple threads at the same time. The objects referenced
// in this way must therefore always be thread-safe.
#define SHARED_EXTERNAL_POINTER_TAGS(V) \
V(kFirstSharedTag, TAG(0)) \
V(kWaiterQueueNodeTag, TAG(0)) \
V(kExternalStringResourceTag, TAG(1)) \
V(kExternalStringResourceDataTag, TAG(2)) \
V(kLastSharedTag, TAG(2))
// External pointers using these tags are kept in a per-Isolate external
// pointer table and can only be accessed when this Isolate is active.
#define PER_ISOLATE_EXTERNAL_POINTER_TAGS(V) \
V(kForeignForeignAddressTag, TAG(10)) \
V(kNativeContextMicrotaskQueueTag, TAG(11)) \
V(kEmbedderDataSlotPayloadTag, TAG(12)) \
/* This tag essentially stands for a `void*` pointer in the V8 API, and */ \
/* it is the Embedder's responsibility to ensure type safety (against */ \
/* substitution) and lifetime validity of these objects. */ \
V(kExternalObjectValueTag, TAG(13)) \
V(kCallHandlerInfoCallbackTag, TAG(14)) \
V(kAccessorInfoGetterTag, TAG(15)) \
V(kAccessorInfoSetterTag, TAG(16)) \
V(kWasmInternalFunctionCallTargetTag, TAG(17)) \
V(kWasmTypeInfoNativeTypeTag, TAG(18)) \
V(kWasmExportedFunctionDataSignatureTag, TAG(19)) \
V(kWasmContinuationJmpbufTag, TAG(20)) \
V(kArrayBufferExtensionTag, TAG(21))
// All external pointer tags.
#define ALL_EXTERNAL_POINTER_TAGS(V) \
SHARED_EXTERNAL_POINTER_TAGS(V) \
PER_ISOLATE_EXTERNAL_POINTER_TAGS(V)
#define EXTERNAL_POINTER_TAG_ENUM(Name, Tag) Name = Tag,
#define MAKE_TAG(HasMarkBit, TypeTag) \
((static_cast<uint64_t>(TypeTag) << kExternalPointerTagShift) | \
(HasMarkBit ? kExternalPointerMarkBit : 0))
enum ExternalPointerTag : uint64_t {
// Empty tag value. Mostly used as placeholder.
kExternalPointerNullTag = MAKE_TAG(0, 0b00000000),
// External pointer tag that will match any external pointer. Use with care!
kAnyExternalPointerTag = MAKE_TAG(1, 0b11111111),
// The free entry tag has all type bits set so every type check with a
// different type fails. It also doesn't have the mark bit set as free
// entries are (by definition) not alive.
kExternalPointerFreeEntryTag = MAKE_TAG(0, 0b11111111),
[sandbox] Implement external pointer table compaction The external pointer table is already to some degree self-compacting: as the freelist is sorted in ascending order, free entries at the start of the table should quickly fill up. However, any live entry at the end of the table makes it impossible to shrink the table, thereby causing higher memory consumption. To solve this, this CL implements a simple table compaction algorithm, used when the freelist has become sufficiently large (currently >= 10% of the table capacity): - The goal of the algorithm is to shrink the table by freelist_size/2 entries at the end of compaction (during sweeping). - At the start of the marking phase, the compaction area is computed as roughly [capacity - freelist_size/2, capacity). - When an entry is marked as alive that lies inside the compaction area, a new "relocation entry" is allocated for it from the freelist and the address of the handle for that entry is stored in that entry. If there are no more free entries before the compaction area, compaction is aborted. This is expected to happen rarely and is logged into a histogram. - During sweeping, all relocation entries are "resolved": the content of the old entry is copied into the new entry and the handle is updated to point to the new entry. - Finally, the table is shrunk and the last initial_freelist_size/2 entries are decommitted. See also the comments in the ExternalPointerTable class for more details. Bug: v8:10391 Change-Id: I28d475c3596590e860421f0a054e2ad4dbebd487 Cq-Include-Trybots: luci.v8.try:v8_linux64_heap_sandbox_dbg_ng,v8_linux_arm64_sim_heap_sandbox_dbg_ng Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3794645 Reviewed-by: Igor Sheludko <ishell@chromium.org> Reviewed-by: Michael Lippautz <mlippautz@chromium.org> Commit-Queue: Samuel Groß <saelo@chromium.org> Cr-Commit-Position: refs/heads/main@{#82484}
2022-08-12 20:08:47 +00:00
// Evacuation entries are used during external pointer table compaction.
kExternalPointerEvacuationEntryTag = MAKE_TAG(1, 0b11100111),
ALL_EXTERNAL_POINTER_TAGS(EXTERNAL_POINTER_TAG_ENUM)
};
#undef MAKE_TAG
#undef TAG
#undef EXTERNAL_POINTER_TAG_ENUM
// clang-format on
[sandbox] Prepare ExternalPointerTable rollout This CL does the following: - It enables (i.e. allocates and initializes) the per-Isolate ExternalPointerTable when the sandbox is enabled. - It refactors the list of external pointer tags to mark them as "sandboxed" or "unsandboxed". An unsandboxed external pointer has a null tag. - It changes V8_SANDBOXED_EXTERNAL_POINTERS to now essentially just enable sandboxing for all available tags. - It modifies all low-level external pointer accessors to perform the ExternalPointerLookup only if the tag is non-zero and otherwise treat the slot as containing a raw pointer. This now allows rolling out external pointer sandboxing incrementally (separately for each external pointer type), which will in turn allow for more precise performance measurements of the impact of the sandbox. Note: when an external pointer tag is now marked as sandboxed (and V8_SANDBOXED_EXTERNAL_POINTERS is not enabled), the underlying slots are still 64-bits in size. This simplifies the implementation as we would otherwise need to deal with variably-sized external pointer slots. Local benchmarking suggests that the benefits from 32-bit external pointer slots are insignificant on typical benchmarks, so this should be ok. Drive-by: rename kExternalPointerSize to kExternalPointerSlotSize to make it more clear what it refers to (the on-heap storage size). Also delete CodeStubAssembler::InitializeExternalPointerField as it is not currently used and the implementation is fairly inefficient. Bug: v8:10391 Change-Id: I7c38729c7e9048d737a1a8ced84749f5b1f7feab Cq-Include-Trybots: luci.v8.try:v8_linux64_heap_sandbox_dbg_ng,v8_linux_arm64_sim_heap_sandbox_dbg_ng Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3736447 Reviewed-by: Igor Sheludko <ishell@chromium.org> Reviewed-by: Michael Lippautz <mlippautz@chromium.org> Reviewed-by: Leszek Swirski <leszeks@chromium.org> Reviewed-by: Andreas Haas <ahaas@chromium.org> Commit-Queue: Samuel Groß <saelo@chromium.org> Reviewed-by: Nico Hartmann <nicohartmann@chromium.org> Cr-Commit-Position: refs/heads/main@{#81636}
2022-07-11 12:21:59 +00:00
// True if the external pointer must be accessed from the shared isolate's
// external pointer table.
V8_INLINE static constexpr bool IsSharedExternalPointerType(
ExternalPointerTag tag) {
return tag >= kFirstSharedTag && tag <= kLastSharedTag;
[sandbox] Prepare ExternalPointerTable rollout This CL does the following: - It enables (i.e. allocates and initializes) the per-Isolate ExternalPointerTable when the sandbox is enabled. - It refactors the list of external pointer tags to mark them as "sandboxed" or "unsandboxed". An unsandboxed external pointer has a null tag. - It changes V8_SANDBOXED_EXTERNAL_POINTERS to now essentially just enable sandboxing for all available tags. - It modifies all low-level external pointer accessors to perform the ExternalPointerLookup only if the tag is non-zero and otherwise treat the slot as containing a raw pointer. This now allows rolling out external pointer sandboxing incrementally (separately for each external pointer type), which will in turn allow for more precise performance measurements of the impact of the sandbox. Note: when an external pointer tag is now marked as sandboxed (and V8_SANDBOXED_EXTERNAL_POINTERS is not enabled), the underlying slots are still 64-bits in size. This simplifies the implementation as we would otherwise need to deal with variably-sized external pointer slots. Local benchmarking suggests that the benefits from 32-bit external pointer slots are insignificant on typical benchmarks, so this should be ok. Drive-by: rename kExternalPointerSize to kExternalPointerSlotSize to make it more clear what it refers to (the on-heap storage size). Also delete CodeStubAssembler::InitializeExternalPointerField as it is not currently used and the implementation is fairly inefficient. Bug: v8:10391 Change-Id: I7c38729c7e9048d737a1a8ced84749f5b1f7feab Cq-Include-Trybots: luci.v8.try:v8_linux64_heap_sandbox_dbg_ng,v8_linux_arm64_sim_heap_sandbox_dbg_ng Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3736447 Reviewed-by: Igor Sheludko <ishell@chromium.org> Reviewed-by: Michael Lippautz <mlippautz@chromium.org> Reviewed-by: Leszek Swirski <leszeks@chromium.org> Reviewed-by: Andreas Haas <ahaas@chromium.org> Commit-Queue: Samuel Groß <saelo@chromium.org> Reviewed-by: Nico Hartmann <nicohartmann@chromium.org> Cr-Commit-Position: refs/heads/main@{#81636}
2022-07-11 12:21:59 +00:00
}
// Sanity checks.
#define CHECK_SHARED_EXTERNAL_POINTER_TAGS(Tag, ...) \
static_assert(IsSharedExternalPointerType(Tag));
#define CHECK_NON_SHARED_EXTERNAL_POINTER_TAGS(Tag, ...) \
static_assert(!IsSharedExternalPointerType(Tag));
SHARED_EXTERNAL_POINTER_TAGS(CHECK_SHARED_EXTERNAL_POINTER_TAGS)
PER_ISOLATE_EXTERNAL_POINTER_TAGS(CHECK_NON_SHARED_EXTERNAL_POINTER_TAGS)
#undef CHECK_NON_SHARED_EXTERNAL_POINTER_TAGS
#undef CHECK_SHARED_EXTERNAL_POINTER_TAGS
#undef SHARED_EXTERNAL_POINTER_TAGS
#undef EXTERNAL_POINTER_TAGS
// {obj} must be the raw tagged pointer representation of a HeapObject
// that's guaranteed to never be in ReadOnlySpace.
V8_EXPORT internal::Isolate* IsolateFromNeverReadOnlySpaceObject(Address obj);
// Returns if we need to throw when an error occurs. This infers the language
// mode based on the current context and the closure. This returns true if the
// language mode is strict.
V8_EXPORT bool ShouldThrowOnError(v8::internal::Isolate* isolate);
/**
* This class exports constants and functionality from within v8 that
* is necessary to implement inline functions in the v8 api. Don't
* depend on functions and constants defined here.
*/
class Internals {
Allowing map word to be used for other state in GC header. This CL adds features to pack/unpack map words. Currently V8 cannot store extra metadata in object headers -- because V8 objects do not have a proper header, but only a map pointer at the start of the object. To store per-object metadata like marking data, a side table is required as the per-object metadata storage. This CL enables V8 to use higher unused bits in a 64-bit map word as per-object metadata storage. Map pointer stores come with an extra step to encode the metadata into the pointer (we call it "map packing"). Map pointer loads will also remove the metadata bits as well (we call it "map packing"). Since the map word is no longer a valid pointer after packing, we also change the tag of the packed map word to make it looks like a Smi. This helps various GC and barrier code to correctly skip them instead of blindly dereferencing this invalid pointer. A ninja flag `v8_enable_map_packing` is provided to turn this map-packing feature on and off. It is disabled by default. * Only works on x64 platform, with `v8_enable_pointer_compression` set to `false` Bug: v8:11624 Change-Id: Ia2bdf79553945e5fc0b0874c87803d2cc733e073 Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2247561 Reviewed-by: Ulan Degenbaev <ulan@chromium.org> Reviewed-by: Nico Hartmann <nicohartmann@chromium.org> Reviewed-by: Toon Verwaest <verwaest@chromium.org> Reviewed-by: Georg Neis <neis@chromium.org> Commit-Queue: Ulan Degenbaev <ulan@chromium.org> Cr-Commit-Position: refs/heads/master@{#73915}
2021-04-06 12:01:44 +00:00
#ifdef V8_MAP_PACKING
V8_INLINE static constexpr internal::Address UnpackMapWord(
internal::Address mapword) {
// TODO(wenyuzhao): Clear header metadata.
return mapword ^ kMapWordXorMask;
}
#endif
public:
// These values match non-compiler-dependent values defined within
// the implementation of v8.
static const int kHeapObjectMapOffset = 0;
static const int kMapInstanceTypeOffset = 1 * kApiTaggedSize + kApiInt32Size;
static const int kStringResourceOffset =
1 * kApiTaggedSize + 2 * kApiInt32Size;
static const int kOddballKindOffset = 4 * kApiTaggedSize + kApiDoubleSize;
static const int kJSObjectHeaderSize = 3 * kApiTaggedSize;
static const int kFixedArrayHeaderSize = 2 * kApiTaggedSize;
static const int kEmbedderDataArrayHeaderSize = 2 * kApiTaggedSize;
static const int kEmbedderDataSlotSize = kApiSystemPointerSize;
Reland "Reland "[sandbox] Sandboxify EmbedderDataSlots"" This is a reland of commit eca383c947da0c4faa890f72e4f89c6c8f592062 More DCHECK failures have been fixed with https://crrev.com/c/3857423 Original change's description: > Reland "[sandbox] Sandboxify EmbedderDataSlots" > > This is a reland of commit e1f585ed94d799555793a4150281ce2baf22a666 > > ExternalPointerTable issues have been fixed in > https://crrev.com/c/3849650 and https://crrev.com/c/3849376 > > Original change's description: > > [sandbox] Sandboxify EmbedderDataSlots > > > > Bug: v8:10391 > > Change-Id: If85a308a6f6ed1b17d86f87b4911c82d2327ea72 > > Cq-Include-Trybots: luci.v8.try:v8_linux64_heap_sandbox_dbg_ng,v8_linux_arm64_sim_heap_sandbox_dbg_ng > > Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3757341 > > Reviewed-by: Igor Sheludko <ishell@chromium.org> > > Commit-Queue: Samuel Groß <saelo@chromium.org> > > Cr-Commit-Position: refs/heads/main@{#82623} > > Bug: v8:10391 > Change-Id: If77f6c10e81c30c2dfa6b33c788bc4a36e4da135 > Cq-Include-Trybots: luci.v8.try:v8_linux64_heap_sandbox_dbg_ng,v8_linux_arm64_sim_heap_sandbox_dbg_ng > Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3852602 > Commit-Queue: Samuel Groß <saelo@chromium.org> > Reviewed-by: Igor Sheludko <ishell@chromium.org> > Cr-Commit-Position: refs/heads/main@{#82686} Bug: v8:10391 Change-Id: Id982c022e50004e903851b160d30fc8767280e5b Cq-Include-Trybots: luci.v8.try:v8_linux64_heap_sandbox_dbg_ng,v8_linux_arm64_sim_heap_sandbox_dbg_ng Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3854679 Reviewed-by: Igor Sheludko <ishell@chromium.org> Commit-Queue: Samuel Groß <saelo@chromium.org> Cr-Commit-Position: refs/heads/main@{#82767}
2022-08-25 12:24:48 +00:00
#ifdef V8_ENABLE_SANDBOX
[sandbox] Prepare ExternalPointerTable rollout This CL does the following: - It enables (i.e. allocates and initializes) the per-Isolate ExternalPointerTable when the sandbox is enabled. - It refactors the list of external pointer tags to mark them as "sandboxed" or "unsandboxed". An unsandboxed external pointer has a null tag. - It changes V8_SANDBOXED_EXTERNAL_POINTERS to now essentially just enable sandboxing for all available tags. - It modifies all low-level external pointer accessors to perform the ExternalPointerLookup only if the tag is non-zero and otherwise treat the slot as containing a raw pointer. This now allows rolling out external pointer sandboxing incrementally (separately for each external pointer type), which will in turn allow for more precise performance measurements of the impact of the sandbox. Note: when an external pointer tag is now marked as sandboxed (and V8_SANDBOXED_EXTERNAL_POINTERS is not enabled), the underlying slots are still 64-bits in size. This simplifies the implementation as we would otherwise need to deal with variably-sized external pointer slots. Local benchmarking suggests that the benefits from 32-bit external pointer slots are insignificant on typical benchmarks, so this should be ok. Drive-by: rename kExternalPointerSize to kExternalPointerSlotSize to make it more clear what it refers to (the on-heap storage size). Also delete CodeStubAssembler::InitializeExternalPointerField as it is not currently used and the implementation is fairly inefficient. Bug: v8:10391 Change-Id: I7c38729c7e9048d737a1a8ced84749f5b1f7feab Cq-Include-Trybots: luci.v8.try:v8_linux64_heap_sandbox_dbg_ng,v8_linux_arm64_sim_heap_sandbox_dbg_ng Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3736447 Reviewed-by: Igor Sheludko <ishell@chromium.org> Reviewed-by: Michael Lippautz <mlippautz@chromium.org> Reviewed-by: Leszek Swirski <leszeks@chromium.org> Reviewed-by: Andreas Haas <ahaas@chromium.org> Commit-Queue: Samuel Groß <saelo@chromium.org> Reviewed-by: Nico Hartmann <nicohartmann@chromium.org> Cr-Commit-Position: refs/heads/main@{#81636}
2022-07-11 12:21:59 +00:00
static const int kEmbedderDataSlotExternalPointerOffset = kApiTaggedSize;
#else
static const int kEmbedderDataSlotExternalPointerOffset = 0;
#endif
Reland^2 "[runtime] Move Context::native_context to the map" This is a reland of c7c47c68f2c0eb1155cfacab785aba54f61d9218. This makes TSAN happy in addition to: Previously I presumed that the context read from a frame in the profiler was a valid context. Turns out that on non-intel we're not guaranteed that the frame is properly set up. In the case we looked at, the profiler took a sample right before writing the frame marker indicating a builtin frame, causing the "context" pointer from that frame to be a bytecode array. Since we'll read random garbage on the stack as a possible context pointer, I made the code reading the native context from it a little more defensive. Bug: v8:9860 Tbr: ulan@chromium.org, neis@chromium.org, ishell@chromium.org Original change's description: > [runtime] Move Context::native_context to the map > > Remove the native context slot from contexts by making context maps > native-context-specific. Now we require 2 loads to go from a context to the > native context, but we have 1 field fewer to store when creating contexts. > > Change-Id: I3c0d7c50c94060c4129db684f46a567de6f30e8d > Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/1859629 > Commit-Queue: Toon Verwaest <verwaest@chromium.org> > Reviewed-by: Igor Sheludko <ishell@chromium.org> > Reviewed-by: Peter Marshall <petermarshall@chromium.org> > Reviewed-by: Maya Lekova <mslekova@chromium.org> > Reviewed-by: Georg Neis <neis@chromium.org> > Reviewed-by: Ulan Degenbaev <ulan@chromium.org> > Reviewed-by: Toon Verwaest <verwaest@chromium.org> > Cr-Commit-Position: refs/heads/master@{#64296} Change-Id: I4d0ab4cbbb23a9ae616407f17ef8f35a0b68ddb4 Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/1864654 Commit-Queue: Toon Verwaest <verwaest@chromium.org> Reviewed-by: Ulan Degenbaev <ulan@chromium.org> Reviewed-by: Georg Neis <neis@chromium.org> Reviewed-by: Igor Sheludko <ishell@chromium.org> Reviewed-by: Peter Marshall <petermarshall@chromium.org> Cr-Commit-Position: refs/heads/master@{#64360}
2019-10-17 15:58:38 +00:00
static const int kNativeContextEmbedderDataOffset = 6 * kApiTaggedSize;
static const int kStringRepresentationAndEncodingMask = 0x0f;
static const int kStringEncodingMask = 0x8;
static const int kExternalTwoByteRepresentationTag = 0x02;
static const int kExternalOneByteRepresentationTag = 0x0a;
static const uint32_t kNumIsolateDataSlots = 4;
static const int kStackGuardSize = 7 * kApiSystemPointerSize;
static const int kBuiltinTier0EntryTableSize = 7 * kApiSystemPointerSize;
static const int kBuiltinTier0TableSize = 7 * kApiSystemPointerSize;
// ExternalPointerTable layout guarantees.
static const int kExternalPointerTableBufferOffset = 0;
[sandbox] Implement external pointer table compaction The external pointer table is already to some degree self-compacting: as the freelist is sorted in ascending order, free entries at the start of the table should quickly fill up. However, any live entry at the end of the table makes it impossible to shrink the table, thereby causing higher memory consumption. To solve this, this CL implements a simple table compaction algorithm, used when the freelist has become sufficiently large (currently >= 10% of the table capacity): - The goal of the algorithm is to shrink the table by freelist_size/2 entries at the end of compaction (during sweeping). - At the start of the marking phase, the compaction area is computed as roughly [capacity - freelist_size/2, capacity). - When an entry is marked as alive that lies inside the compaction area, a new "relocation entry" is allocated for it from the freelist and the address of the handle for that entry is stored in that entry. If there are no more free entries before the compaction area, compaction is aborted. This is expected to happen rarely and is logged into a histogram. - During sweeping, all relocation entries are "resolved": the content of the old entry is copied into the new entry and the handle is updated to point to the new entry. - Finally, the table is shrunk and the last initial_freelist_size/2 entries are decommitted. See also the comments in the ExternalPointerTable class for more details. Bug: v8:10391 Change-Id: I28d475c3596590e860421f0a054e2ad4dbebd487 Cq-Include-Trybots: luci.v8.try:v8_linux64_heap_sandbox_dbg_ng,v8_linux_arm64_sim_heap_sandbox_dbg_ng Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3794645 Reviewed-by: Igor Sheludko <ishell@chromium.org> Reviewed-by: Michael Lippautz <mlippautz@chromium.org> Commit-Queue: Samuel Groß <saelo@chromium.org> Cr-Commit-Position: refs/heads/main@{#82484}
2022-08-12 20:08:47 +00:00
static const int kExternalPointerTableSize = 4 * kApiSystemPointerSize;
// IsolateData layout guarantees.
static const int kIsolateCageBaseOffset = 0;
static const int kIsolateStackGuardOffset =
kIsolateCageBaseOffset + kApiSystemPointerSize;
static const int kVariousBooleanFlagsOffset =
kIsolateStackGuardOffset + kStackGuardSize;
static const int kBuiltinTier0EntryTableOffset =
kVariousBooleanFlagsOffset + 8;
static const int kBuiltinTier0TableOffset =
kBuiltinTier0EntryTableOffset + kBuiltinTier0EntryTableSize;
static const int kIsolateEmbedderDataOffset =
kBuiltinTier0TableOffset + kBuiltinTier0TableSize;
static const int kIsolateFastCCallCallerFpOffset =
kIsolateEmbedderDataOffset + kNumIsolateDataSlots * kApiSystemPointerSize;
static const int kIsolateFastCCallCallerPcOffset =
kIsolateFastCCallCallerFpOffset + kApiSystemPointerSize;
static const int kIsolateFastApiCallTargetOffset =
kIsolateFastCCallCallerPcOffset + kApiSystemPointerSize;
static const int kIsolateLongTaskStatsCounterOffset =
kIsolateFastApiCallTargetOffset + kApiSystemPointerSize;
#ifdef V8_COMPRESS_POINTERS
static const int kIsolateExternalPointerTableOffset =
kIsolateLongTaskStatsCounterOffset + kApiSizetSize;
static const int kIsolateSharedExternalPointerTableAddressOffset =
kIsolateExternalPointerTableOffset + kExternalPointerTableSize;
static const int kIsolateRootsOffset =
kIsolateSharedExternalPointerTableAddressOffset + kApiSystemPointerSize;
#else
static const int kIsolateRootsOffset =
kIsolateLongTaskStatsCounterOffset + kApiSizetSize;
#endif
static const int kUndefinedValueRootIndex = 4;
static const int kTheHoleValueRootIndex = 5;
static const int kNullValueRootIndex = 6;
static const int kTrueValueRootIndex = 7;
static const int kFalseValueRootIndex = 8;
static const int kEmptyStringRootIndex = 9;
static const int kNodeClassIdOffset = 1 * kApiSystemPointerSize;
static const int kNodeFlagsOffset = 1 * kApiSystemPointerSize + 3;
static const int kNodeStateMask = 0x3;
static const int kNodeStateIsWeakValue = 2;
static const int kTracedNodeClassIdOffset = kApiSystemPointerSize;
static const int kFirstNonstringType = 0x80;
static const int kOddballType = 0x83;
static const int kForeignType = 0xcc;
static const int kJSSpecialApiObjectType = 0x410;
static const int kJSObjectType = 0x421;
static const int kFirstJSApiObjectType = 0x422;
static const int kLastJSApiObjectType = 0x80A;
static const int kUndefinedOddballKind = 5;
static const int kNullOddballKind = 3;
// Constants used by PropertyCallbackInfo to check if we should throw when an
// error occurs.
static const int kThrowOnError = 0;
static const int kDontThrow = 1;
static const int kInferShouldThrowMode = 2;
// Soft limit for AdjustAmountofExternalAllocatedMemory. Trigger an
// incremental GC once the external memory reaches this limit.
static constexpr int kExternalAllocationSoftLimit = 64 * 1024 * 1024;
Allowing map word to be used for other state in GC header. This CL adds features to pack/unpack map words. Currently V8 cannot store extra metadata in object headers -- because V8 objects do not have a proper header, but only a map pointer at the start of the object. To store per-object metadata like marking data, a side table is required as the per-object metadata storage. This CL enables V8 to use higher unused bits in a 64-bit map word as per-object metadata storage. Map pointer stores come with an extra step to encode the metadata into the pointer (we call it "map packing"). Map pointer loads will also remove the metadata bits as well (we call it "map packing"). Since the map word is no longer a valid pointer after packing, we also change the tag of the packed map word to make it looks like a Smi. This helps various GC and barrier code to correctly skip them instead of blindly dereferencing this invalid pointer. A ninja flag `v8_enable_map_packing` is provided to turn this map-packing feature on and off. It is disabled by default. * Only works on x64 platform, with `v8_enable_pointer_compression` set to `false` Bug: v8:11624 Change-Id: Ia2bdf79553945e5fc0b0874c87803d2cc733e073 Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2247561 Reviewed-by: Ulan Degenbaev <ulan@chromium.org> Reviewed-by: Nico Hartmann <nicohartmann@chromium.org> Reviewed-by: Toon Verwaest <verwaest@chromium.org> Reviewed-by: Georg Neis <neis@chromium.org> Commit-Queue: Ulan Degenbaev <ulan@chromium.org> Cr-Commit-Position: refs/heads/master@{#73915}
2021-04-06 12:01:44 +00:00
#ifdef V8_MAP_PACKING
static const uintptr_t kMapWordMetadataMask = 0xffffULL << 48;
// The lowest two bits of mapwords are always `0b10`
static const uintptr_t kMapWordSignature = 0b10;
// XORing a (non-compressed) map with this mask ensures that the two
// low-order bits are 0b10. The 0 at the end makes this look like a Smi,
// although real Smis have all lower 32 bits unset. We only rely on these
// values passing as Smis in very few places.
static const int kMapWordXorMask = 0b11;
#endif
V8_EXPORT static void CheckInitializedImpl(v8::Isolate* isolate);
V8_INLINE static void CheckInitialized(v8::Isolate* isolate) {
#ifdef V8_ENABLE_CHECKS
CheckInitializedImpl(isolate);
#endif
}
V8_INLINE static bool HasHeapObjectTag(const internal::Address value) {
return (value & kHeapObjectTagMask) == static_cast<Address>(kHeapObjectTag);
}
V8_INLINE static int SmiValue(const internal::Address value) {
return PlatformSmiTagging::SmiToInt(value);
}
V8_INLINE static constexpr internal::Address IntToSmi(int value) {
return internal::IntToSmi(value);
}
V8_INLINE static constexpr bool IsValidSmi(intptr_t value) {
return PlatformSmiTagging::IsValidSmi(value);
}
V8_INLINE static int GetInstanceType(const internal::Address obj) {
typedef internal::Address A;
A map = ReadTaggedPointerField(obj, kHeapObjectMapOffset);
Allowing map word to be used for other state in GC header. This CL adds features to pack/unpack map words. Currently V8 cannot store extra metadata in object headers -- because V8 objects do not have a proper header, but only a map pointer at the start of the object. To store per-object metadata like marking data, a side table is required as the per-object metadata storage. This CL enables V8 to use higher unused bits in a 64-bit map word as per-object metadata storage. Map pointer stores come with an extra step to encode the metadata into the pointer (we call it "map packing"). Map pointer loads will also remove the metadata bits as well (we call it "map packing"). Since the map word is no longer a valid pointer after packing, we also change the tag of the packed map word to make it looks like a Smi. This helps various GC and barrier code to correctly skip them instead of blindly dereferencing this invalid pointer. A ninja flag `v8_enable_map_packing` is provided to turn this map-packing feature on and off. It is disabled by default. * Only works on x64 platform, with `v8_enable_pointer_compression` set to `false` Bug: v8:11624 Change-Id: Ia2bdf79553945e5fc0b0874c87803d2cc733e073 Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2247561 Reviewed-by: Ulan Degenbaev <ulan@chromium.org> Reviewed-by: Nico Hartmann <nicohartmann@chromium.org> Reviewed-by: Toon Verwaest <verwaest@chromium.org> Reviewed-by: Georg Neis <neis@chromium.org> Commit-Queue: Ulan Degenbaev <ulan@chromium.org> Cr-Commit-Position: refs/heads/master@{#73915}
2021-04-06 12:01:44 +00:00
#ifdef V8_MAP_PACKING
map = UnpackMapWord(map);
#endif
return ReadRawField<uint16_t>(map, kMapInstanceTypeOffset);
}
V8_INLINE static int GetOddballKind(const internal::Address obj) {
return SmiValue(ReadTaggedSignedField(obj, kOddballKindOffset));
}
V8_INLINE static bool IsExternalTwoByteString(int instance_type) {
int representation = (instance_type & kStringRepresentationAndEncodingMask);
return representation == kExternalTwoByteRepresentationTag;
}
V8_INLINE static constexpr bool CanHaveInternalField(int instance_type) {
static_assert(kJSObjectType + 1 == kFirstJSApiObjectType);
static_assert(kJSObjectType < kLastJSApiObjectType);
static_assert(kFirstJSApiObjectType < kLastJSApiObjectType);
// Check for IsJSObject() || IsJSSpecialApiObject() || IsJSApiObject()
return instance_type == kJSSpecialApiObjectType ||
// inlined version of base::IsInRange
(static_cast<unsigned>(static_cast<unsigned>(instance_type) -
static_cast<unsigned>(kJSObjectType)) <=
static_cast<unsigned>(kLastJSApiObjectType - kJSObjectType));
}
V8_INLINE static uint8_t GetNodeFlag(internal::Address* obj, int shift) {
uint8_t* addr = reinterpret_cast<uint8_t*>(obj) + kNodeFlagsOffset;
return *addr & static_cast<uint8_t>(1U << shift);
}
V8_INLINE static void UpdateNodeFlag(internal::Address* obj, bool value,
int shift) {
uint8_t* addr = reinterpret_cast<uint8_t*>(obj) + kNodeFlagsOffset;
uint8_t mask = static_cast<uint8_t>(1U << shift);
*addr = static_cast<uint8_t>((*addr & ~mask) | (value << shift));
}
V8_INLINE static uint8_t GetNodeState(internal::Address* obj) {
uint8_t* addr = reinterpret_cast<uint8_t*>(obj) + kNodeFlagsOffset;
return *addr & kNodeStateMask;
}
V8_INLINE static void UpdateNodeState(internal::Address* obj, uint8_t value) {
uint8_t* addr = reinterpret_cast<uint8_t*>(obj) + kNodeFlagsOffset;
*addr = static_cast<uint8_t>((*addr & ~kNodeStateMask) | value);
}
V8_INLINE static void SetEmbedderData(v8::Isolate* isolate, uint32_t slot,
void* data) {
internal::Address addr = reinterpret_cast<internal::Address>(isolate) +
kIsolateEmbedderDataOffset +
slot * kApiSystemPointerSize;
*reinterpret_cast<void**>(addr) = data;
}
V8_INLINE static void* GetEmbedderData(const v8::Isolate* isolate,
uint32_t slot) {
internal::Address addr = reinterpret_cast<internal::Address>(isolate) +
kIsolateEmbedderDataOffset +
slot * kApiSystemPointerSize;
return *reinterpret_cast<void* const*>(addr);
}
V8_INLINE static void IncrementLongTasksStatsCounter(v8::Isolate* isolate) {
internal::Address addr = reinterpret_cast<internal::Address>(isolate) +
kIsolateLongTaskStatsCounterOffset;
++(*reinterpret_cast<size_t*>(addr));
}
V8_INLINE static internal::Address* GetRoot(v8::Isolate* isolate, int index) {
internal::Address addr = reinterpret_cast<internal::Address>(isolate) +
kIsolateRootsOffset +
index * kApiSystemPointerSize;
return reinterpret_cast<internal::Address*>(addr);
}
[sandbox] Prepare ExternalPointerTable rollout This CL does the following: - It enables (i.e. allocates and initializes) the per-Isolate ExternalPointerTable when the sandbox is enabled. - It refactors the list of external pointer tags to mark them as "sandboxed" or "unsandboxed". An unsandboxed external pointer has a null tag. - It changes V8_SANDBOXED_EXTERNAL_POINTERS to now essentially just enable sandboxing for all available tags. - It modifies all low-level external pointer accessors to perform the ExternalPointerLookup only if the tag is non-zero and otherwise treat the slot as containing a raw pointer. This now allows rolling out external pointer sandboxing incrementally (separately for each external pointer type), which will in turn allow for more precise performance measurements of the impact of the sandbox. Note: when an external pointer tag is now marked as sandboxed (and V8_SANDBOXED_EXTERNAL_POINTERS is not enabled), the underlying slots are still 64-bits in size. This simplifies the implementation as we would otherwise need to deal with variably-sized external pointer slots. Local benchmarking suggests that the benefits from 32-bit external pointer slots are insignificant on typical benchmarks, so this should be ok. Drive-by: rename kExternalPointerSize to kExternalPointerSlotSize to make it more clear what it refers to (the on-heap storage size). Also delete CodeStubAssembler::InitializeExternalPointerField as it is not currently used and the implementation is fairly inefficient. Bug: v8:10391 Change-Id: I7c38729c7e9048d737a1a8ced84749f5b1f7feab Cq-Include-Trybots: luci.v8.try:v8_linux64_heap_sandbox_dbg_ng,v8_linux_arm64_sim_heap_sandbox_dbg_ng Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3736447 Reviewed-by: Igor Sheludko <ishell@chromium.org> Reviewed-by: Michael Lippautz <mlippautz@chromium.org> Reviewed-by: Leszek Swirski <leszeks@chromium.org> Reviewed-by: Andreas Haas <ahaas@chromium.org> Commit-Queue: Samuel Groß <saelo@chromium.org> Reviewed-by: Nico Hartmann <nicohartmann@chromium.org> Cr-Commit-Position: refs/heads/main@{#81636}
2022-07-11 12:21:59 +00:00
#ifdef V8_ENABLE_SANDBOX
V8_INLINE static internal::Address* GetExternalPointerTableBase(
v8::Isolate* isolate) {
internal::Address addr = reinterpret_cast<internal::Address>(isolate) +
kIsolateExternalPointerTableOffset +
kExternalPointerTableBufferOffset;
return *reinterpret_cast<internal::Address**>(addr);
}
V8_INLINE static internal::Address* GetSharedExternalPointerTableBase(
v8::Isolate* isolate) {
internal::Address addr = reinterpret_cast<internal::Address>(isolate) +
kIsolateSharedExternalPointerTableAddressOffset;
addr = *reinterpret_cast<internal::Address*>(addr);
addr += kExternalPointerTableBufferOffset;
return *reinterpret_cast<internal::Address**>(addr);
}
#endif
template <typename T>
V8_INLINE static T ReadRawField(internal::Address heap_object_ptr,
int offset) {
internal::Address addr = heap_object_ptr + offset - kHeapObjectTag;
#ifdef V8_COMPRESS_POINTERS
if (sizeof(T) > kApiTaggedSize) {
// TODO(ishell, v8:8875): When pointer compression is enabled 8-byte size
// fields (external pointers, doubles and BigInt data) are only
// kTaggedSize aligned so we have to use unaligned pointer friendly way of
// accessing them in order to avoid undefined behavior in C++ code.
T r;
memcpy(&r, reinterpret_cast<void*>(addr), sizeof(T));
return r;
}
#endif
return *reinterpret_cast<const T*>(addr);
}
V8_INLINE static internal::Address ReadTaggedPointerField(
internal::Address heap_object_ptr, int offset) {
#ifdef V8_COMPRESS_POINTERS
uint32_t value = ReadRawField<uint32_t>(heap_object_ptr, offset);
Reland^2 "[ptr-cage] Rename IsolateRoot to PtrComprCageBase" This is a reland of e28dadc2070b202aab77b9e50a46e50be02f8f1c The original failure was due to a stale Win32 bot. The reland failure was due to idempotent task deduplication returning the exact same failure. See crbug/1196064 Original change's description: > [ptr-cage] Rename IsolateRoot to PtrComprCageBase > > Currently, IsolateRoot is both the address of the Isolate root and the > base address of the pointer compression reservation. This CL teases the > two uses apart by renaming IsolateRoot to PtrComprCageBase. > > - In addition to V8_COMPRESS_POINTERS, add a > V8_COMPRESS_POINTERS_IN_ISOLATE_CAGE (vs SHARED_CAGE). > > - Rename GetIsolate* helpers to GetPtrComprCageBase. When > V8_COMPRESS_POINTERS_IN_ISOLATE_CAGE is true, the helpers remain as > aliases to GetPtrComprCageBase. > > - Rename kPtrComprIsolateRootAlignment to kPtrComprCageBaseAlignment. > > Bug: v8:11460 > Change-Id: I1d715f678ce9a0b5731895612ca14f56579b1c48 > Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2783672 > Commit-Queue: Shu-yu Guo <syg@chromium.org> > Auto-Submit: Shu-yu Guo <syg@chromium.org> > Reviewed-by: Igor Sheludko <ishell@chromium.org> > Reviewed-by: Ross McIlroy <rmcilroy@chromium.org> > Cr-Commit-Position: refs/heads/master@{#73790} Bug: v8:11460 No-Try: true Tbr: ishell@chromium.org Tbr: rmcilroy@chromium.org Change-Id: Id69311cf3267ebe1297fff159de0be48b15b65a3 Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2806546 Reviewed-by: Shu-yu Guo <syg@chromium.org> Commit-Queue: Shu-yu Guo <syg@chromium.org> Cr-Commit-Position: refs/heads/master@{#73795}
2021-04-05 19:42:59 +00:00
internal::Address base =
GetPtrComprCageBaseFromOnHeapAddress(heap_object_ptr);
return base + static_cast<internal::Address>(static_cast<uintptr_t>(value));
#else
return ReadRawField<internal::Address>(heap_object_ptr, offset);
#endif
}
V8_INLINE static internal::Address ReadTaggedSignedField(
internal::Address heap_object_ptr, int offset) {
#ifdef V8_COMPRESS_POINTERS
uint32_t value = ReadRawField<uint32_t>(heap_object_ptr, offset);
return static_cast<internal::Address>(static_cast<uintptr_t>(value));
#else
return ReadRawField<internal::Address>(heap_object_ptr, offset);
#endif
}
V8_INLINE static v8::Isolate* GetIsolateForSandbox(internal::Address obj) {
[sandbox] Prepare ExternalPointerTable rollout This CL does the following: - It enables (i.e. allocates and initializes) the per-Isolate ExternalPointerTable when the sandbox is enabled. - It refactors the list of external pointer tags to mark them as "sandboxed" or "unsandboxed". An unsandboxed external pointer has a null tag. - It changes V8_SANDBOXED_EXTERNAL_POINTERS to now essentially just enable sandboxing for all available tags. - It modifies all low-level external pointer accessors to perform the ExternalPointerLookup only if the tag is non-zero and otherwise treat the slot as containing a raw pointer. This now allows rolling out external pointer sandboxing incrementally (separately for each external pointer type), which will in turn allow for more precise performance measurements of the impact of the sandbox. Note: when an external pointer tag is now marked as sandboxed (and V8_SANDBOXED_EXTERNAL_POINTERS is not enabled), the underlying slots are still 64-bits in size. This simplifies the implementation as we would otherwise need to deal with variably-sized external pointer slots. Local benchmarking suggests that the benefits from 32-bit external pointer slots are insignificant on typical benchmarks, so this should be ok. Drive-by: rename kExternalPointerSize to kExternalPointerSlotSize to make it more clear what it refers to (the on-heap storage size). Also delete CodeStubAssembler::InitializeExternalPointerField as it is not currently used and the implementation is fairly inefficient. Bug: v8:10391 Change-Id: I7c38729c7e9048d737a1a8ced84749f5b1f7feab Cq-Include-Trybots: luci.v8.try:v8_linux64_heap_sandbox_dbg_ng,v8_linux_arm64_sim_heap_sandbox_dbg_ng Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3736447 Reviewed-by: Igor Sheludko <ishell@chromium.org> Reviewed-by: Michael Lippautz <mlippautz@chromium.org> Reviewed-by: Leszek Swirski <leszeks@chromium.org> Reviewed-by: Andreas Haas <ahaas@chromium.org> Commit-Queue: Samuel Groß <saelo@chromium.org> Reviewed-by: Nico Hartmann <nicohartmann@chromium.org> Cr-Commit-Position: refs/heads/main@{#81636}
2022-07-11 12:21:59 +00:00
#ifdef V8_ENABLE_SANDBOX
return reinterpret_cast<v8::Isolate*>(
internal::IsolateFromNeverReadOnlySpaceObject(obj));
#else
// Not used in non-sandbox mode.
return nullptr;
#endif
}
[sandbox] Turn ExternalPointerTag into a template parameter The ExternalPointerTags are assumed to be compile-time constants in most cases, so turning them into template parameters enforces that. As decisions such as whether to use the per-isolate or the shared external pointer table are encoded into the tag values, forcing those to be compile-time constants guarantees that the compiler will be able to inline the correct logic when accessing an external pointer. With this, there are now two (high-level) ways of accessing external pointer fields from C++: the Read/WriteExternalPointerField methods which require the ExternalPointerTag to be a template parameter, and the ExternalPointerSlot class which takes the tag as an argument. The latter is for example used for snapshot deserialization and by the garbage collector (more generally, by the ObjectVisitor::VisitExternalPointer method), where the tag is not a compile-time constant. Finally, this CL also introduces a new ExternalPointerHandle type which represents the (opaque) on-heap representation of a reference to an entry in an ExternalPointerTable when sandboxing is enabled. Making this its own type makes the code a bit more readable. Bug: v8:10391 Change-Id: I867b8ce41d15d485f1dc66786f233c710c56afcb Cq-Include-Trybots: luci.v8.try:v8_linux64_heap_sandbox_dbg_ng,v8_linux_arm64_sim_heap_sandbox_dbg_ng Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3720641 Reviewed-by: Tobias Tebbi <tebbi@chromium.org> Reviewed-by: Leszek Swirski <leszeks@chromium.org> Commit-Queue: Samuel Groß <saelo@chromium.org> Reviewed-by: Igor Sheludko <ishell@chromium.org> Reviewed-by: Michael Lippautz <mlippautz@chromium.org> Cr-Commit-Position: refs/heads/main@{#81402}
2022-06-27 12:00:24 +00:00
template <ExternalPointerTag tag>
V8_INLINE static internal::Address ReadExternalPointerField(
[sandbox] Turn ExternalPointerTag into a template parameter The ExternalPointerTags are assumed to be compile-time constants in most cases, so turning them into template parameters enforces that. As decisions such as whether to use the per-isolate or the shared external pointer table are encoded into the tag values, forcing those to be compile-time constants guarantees that the compiler will be able to inline the correct logic when accessing an external pointer. With this, there are now two (high-level) ways of accessing external pointer fields from C++: the Read/WriteExternalPointerField methods which require the ExternalPointerTag to be a template parameter, and the ExternalPointerSlot class which takes the tag as an argument. The latter is for example used for snapshot deserialization and by the garbage collector (more generally, by the ObjectVisitor::VisitExternalPointer method), where the tag is not a compile-time constant. Finally, this CL also introduces a new ExternalPointerHandle type which represents the (opaque) on-heap representation of a reference to an entry in an ExternalPointerTable when sandboxing is enabled. Making this its own type makes the code a bit more readable. Bug: v8:10391 Change-Id: I867b8ce41d15d485f1dc66786f233c710c56afcb Cq-Include-Trybots: luci.v8.try:v8_linux64_heap_sandbox_dbg_ng,v8_linux_arm64_sim_heap_sandbox_dbg_ng Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3720641 Reviewed-by: Tobias Tebbi <tebbi@chromium.org> Reviewed-by: Leszek Swirski <leszeks@chromium.org> Commit-Queue: Samuel Groß <saelo@chromium.org> Reviewed-by: Igor Sheludko <ishell@chromium.org> Reviewed-by: Michael Lippautz <mlippautz@chromium.org> Cr-Commit-Position: refs/heads/main@{#81402}
2022-06-27 12:00:24 +00:00
v8::Isolate* isolate, internal::Address heap_object_ptr, int offset) {
[sandbox] Prepare ExternalPointerTable rollout This CL does the following: - It enables (i.e. allocates and initializes) the per-Isolate ExternalPointerTable when the sandbox is enabled. - It refactors the list of external pointer tags to mark them as "sandboxed" or "unsandboxed". An unsandboxed external pointer has a null tag. - It changes V8_SANDBOXED_EXTERNAL_POINTERS to now essentially just enable sandboxing for all available tags. - It modifies all low-level external pointer accessors to perform the ExternalPointerLookup only if the tag is non-zero and otherwise treat the slot as containing a raw pointer. This now allows rolling out external pointer sandboxing incrementally (separately for each external pointer type), which will in turn allow for more precise performance measurements of the impact of the sandbox. Note: when an external pointer tag is now marked as sandboxed (and V8_SANDBOXED_EXTERNAL_POINTERS is not enabled), the underlying slots are still 64-bits in size. This simplifies the implementation as we would otherwise need to deal with variably-sized external pointer slots. Local benchmarking suggests that the benefits from 32-bit external pointer slots are insignificant on typical benchmarks, so this should be ok. Drive-by: rename kExternalPointerSize to kExternalPointerSlotSize to make it more clear what it refers to (the on-heap storage size). Also delete CodeStubAssembler::InitializeExternalPointerField as it is not currently used and the implementation is fairly inefficient. Bug: v8:10391 Change-Id: I7c38729c7e9048d737a1a8ced84749f5b1f7feab Cq-Include-Trybots: luci.v8.try:v8_linux64_heap_sandbox_dbg_ng,v8_linux_arm64_sim_heap_sandbox_dbg_ng Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3736447 Reviewed-by: Igor Sheludko <ishell@chromium.org> Reviewed-by: Michael Lippautz <mlippautz@chromium.org> Reviewed-by: Leszek Swirski <leszeks@chromium.org> Reviewed-by: Andreas Haas <ahaas@chromium.org> Commit-Queue: Samuel Groß <saelo@chromium.org> Reviewed-by: Nico Hartmann <nicohartmann@chromium.org> Cr-Commit-Position: refs/heads/main@{#81636}
2022-07-11 12:21:59 +00:00
#ifdef V8_ENABLE_SANDBOX
static_assert(tag != kExternalPointerNullTag);
// See src/sandbox/external-pointer-table-inl.h. Logic duplicated here so
// it can be inlined and doesn't require an additional call.
internal::Address* table = IsSharedExternalPointerType(tag)
? GetSharedExternalPointerTableBase(isolate)
: GetExternalPointerTableBase(isolate);
internal::ExternalPointerHandle handle =
ReadRawField<ExternalPointerHandle>(heap_object_ptr, offset);
uint32_t index = handle >> kExternalPointerIndexShift;
std::atomic<internal::Address>* ptr =
reinterpret_cast<std::atomic<internal::Address>*>(&table[index]);
internal::Address entry =
std::atomic_load_explicit(ptr, std::memory_order_relaxed);
return entry & ~tag;
#else
[sandbox] Prepare ExternalPointerTable rollout This CL does the following: - It enables (i.e. allocates and initializes) the per-Isolate ExternalPointerTable when the sandbox is enabled. - It refactors the list of external pointer tags to mark them as "sandboxed" or "unsandboxed". An unsandboxed external pointer has a null tag. - It changes V8_SANDBOXED_EXTERNAL_POINTERS to now essentially just enable sandboxing for all available tags. - It modifies all low-level external pointer accessors to perform the ExternalPointerLookup only if the tag is non-zero and otherwise treat the slot as containing a raw pointer. This now allows rolling out external pointer sandboxing incrementally (separately for each external pointer type), which will in turn allow for more precise performance measurements of the impact of the sandbox. Note: when an external pointer tag is now marked as sandboxed (and V8_SANDBOXED_EXTERNAL_POINTERS is not enabled), the underlying slots are still 64-bits in size. This simplifies the implementation as we would otherwise need to deal with variably-sized external pointer slots. Local benchmarking suggests that the benefits from 32-bit external pointer slots are insignificant on typical benchmarks, so this should be ok. Drive-by: rename kExternalPointerSize to kExternalPointerSlotSize to make it more clear what it refers to (the on-heap storage size). Also delete CodeStubAssembler::InitializeExternalPointerField as it is not currently used and the implementation is fairly inefficient. Bug: v8:10391 Change-Id: I7c38729c7e9048d737a1a8ced84749f5b1f7feab Cq-Include-Trybots: luci.v8.try:v8_linux64_heap_sandbox_dbg_ng,v8_linux_arm64_sim_heap_sandbox_dbg_ng Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3736447 Reviewed-by: Igor Sheludko <ishell@chromium.org> Reviewed-by: Michael Lippautz <mlippautz@chromium.org> Reviewed-by: Leszek Swirski <leszeks@chromium.org> Reviewed-by: Andreas Haas <ahaas@chromium.org> Commit-Queue: Samuel Groß <saelo@chromium.org> Reviewed-by: Nico Hartmann <nicohartmann@chromium.org> Cr-Commit-Position: refs/heads/main@{#81636}
2022-07-11 12:21:59 +00:00
return ReadRawField<Address>(heap_object_ptr, offset);
#endif // V8_ENABLE_SANDBOX
}
#ifdef V8_COMPRESS_POINTERS
Reland^2 "[ptr-cage] Rename IsolateRoot to PtrComprCageBase" This is a reland of e28dadc2070b202aab77b9e50a46e50be02f8f1c The original failure was due to a stale Win32 bot. The reland failure was due to idempotent task deduplication returning the exact same failure. See crbug/1196064 Original change's description: > [ptr-cage] Rename IsolateRoot to PtrComprCageBase > > Currently, IsolateRoot is both the address of the Isolate root and the > base address of the pointer compression reservation. This CL teases the > two uses apart by renaming IsolateRoot to PtrComprCageBase. > > - In addition to V8_COMPRESS_POINTERS, add a > V8_COMPRESS_POINTERS_IN_ISOLATE_CAGE (vs SHARED_CAGE). > > - Rename GetIsolate* helpers to GetPtrComprCageBase. When > V8_COMPRESS_POINTERS_IN_ISOLATE_CAGE is true, the helpers remain as > aliases to GetPtrComprCageBase. > > - Rename kPtrComprIsolateRootAlignment to kPtrComprCageBaseAlignment. > > Bug: v8:11460 > Change-Id: I1d715f678ce9a0b5731895612ca14f56579b1c48 > Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2783672 > Commit-Queue: Shu-yu Guo <syg@chromium.org> > Auto-Submit: Shu-yu Guo <syg@chromium.org> > Reviewed-by: Igor Sheludko <ishell@chromium.org> > Reviewed-by: Ross McIlroy <rmcilroy@chromium.org> > Cr-Commit-Position: refs/heads/master@{#73790} Bug: v8:11460 No-Try: true Tbr: ishell@chromium.org Tbr: rmcilroy@chromium.org Change-Id: Id69311cf3267ebe1297fff159de0be48b15b65a3 Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2806546 Reviewed-by: Shu-yu Guo <syg@chromium.org> Commit-Queue: Shu-yu Guo <syg@chromium.org> Cr-Commit-Position: refs/heads/master@{#73795}
2021-04-05 19:42:59 +00:00
V8_INLINE static internal::Address GetPtrComprCageBaseFromOnHeapAddress(
internal::Address addr) {
Reland^2 "[ptr-cage] Rename IsolateRoot to PtrComprCageBase" This is a reland of e28dadc2070b202aab77b9e50a46e50be02f8f1c The original failure was due to a stale Win32 bot. The reland failure was due to idempotent task deduplication returning the exact same failure. See crbug/1196064 Original change's description: > [ptr-cage] Rename IsolateRoot to PtrComprCageBase > > Currently, IsolateRoot is both the address of the Isolate root and the > base address of the pointer compression reservation. This CL teases the > two uses apart by renaming IsolateRoot to PtrComprCageBase. > > - In addition to V8_COMPRESS_POINTERS, add a > V8_COMPRESS_POINTERS_IN_ISOLATE_CAGE (vs SHARED_CAGE). > > - Rename GetIsolate* helpers to GetPtrComprCageBase. When > V8_COMPRESS_POINTERS_IN_ISOLATE_CAGE is true, the helpers remain as > aliases to GetPtrComprCageBase. > > - Rename kPtrComprIsolateRootAlignment to kPtrComprCageBaseAlignment. > > Bug: v8:11460 > Change-Id: I1d715f678ce9a0b5731895612ca14f56579b1c48 > Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2783672 > Commit-Queue: Shu-yu Guo <syg@chromium.org> > Auto-Submit: Shu-yu Guo <syg@chromium.org> > Reviewed-by: Igor Sheludko <ishell@chromium.org> > Reviewed-by: Ross McIlroy <rmcilroy@chromium.org> > Cr-Commit-Position: refs/heads/master@{#73790} Bug: v8:11460 No-Try: true Tbr: ishell@chromium.org Tbr: rmcilroy@chromium.org Change-Id: Id69311cf3267ebe1297fff159de0be48b15b65a3 Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2806546 Reviewed-by: Shu-yu Guo <syg@chromium.org> Commit-Queue: Shu-yu Guo <syg@chromium.org> Cr-Commit-Position: refs/heads/master@{#73795}
2021-04-05 19:42:59 +00:00
return addr & -static_cast<intptr_t>(kPtrComprCageBaseAlignment);
}
V8_INLINE static internal::Address DecompressTaggedField(
internal::Address heap_object_ptr, uint32_t value) {
Reland^2 "[ptr-cage] Rename IsolateRoot to PtrComprCageBase" This is a reland of e28dadc2070b202aab77b9e50a46e50be02f8f1c The original failure was due to a stale Win32 bot. The reland failure was due to idempotent task deduplication returning the exact same failure. See crbug/1196064 Original change's description: > [ptr-cage] Rename IsolateRoot to PtrComprCageBase > > Currently, IsolateRoot is both the address of the Isolate root and the > base address of the pointer compression reservation. This CL teases the > two uses apart by renaming IsolateRoot to PtrComprCageBase. > > - In addition to V8_COMPRESS_POINTERS, add a > V8_COMPRESS_POINTERS_IN_ISOLATE_CAGE (vs SHARED_CAGE). > > - Rename GetIsolate* helpers to GetPtrComprCageBase. When > V8_COMPRESS_POINTERS_IN_ISOLATE_CAGE is true, the helpers remain as > aliases to GetPtrComprCageBase. > > - Rename kPtrComprIsolateRootAlignment to kPtrComprCageBaseAlignment. > > Bug: v8:11460 > Change-Id: I1d715f678ce9a0b5731895612ca14f56579b1c48 > Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2783672 > Commit-Queue: Shu-yu Guo <syg@chromium.org> > Auto-Submit: Shu-yu Guo <syg@chromium.org> > Reviewed-by: Igor Sheludko <ishell@chromium.org> > Reviewed-by: Ross McIlroy <rmcilroy@chromium.org> > Cr-Commit-Position: refs/heads/master@{#73790} Bug: v8:11460 No-Try: true Tbr: ishell@chromium.org Tbr: rmcilroy@chromium.org Change-Id: Id69311cf3267ebe1297fff159de0be48b15b65a3 Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2806546 Reviewed-by: Shu-yu Guo <syg@chromium.org> Commit-Queue: Shu-yu Guo <syg@chromium.org> Cr-Commit-Position: refs/heads/master@{#73795}
2021-04-05 19:42:59 +00:00
internal::Address base =
GetPtrComprCageBaseFromOnHeapAddress(heap_object_ptr);
return base + static_cast<internal::Address>(static_cast<uintptr_t>(value));
}
#endif // V8_COMPRESS_POINTERS
};
// Only perform cast check for types derived from v8::Data since
// other types do not implement the Cast method.
template <bool PerformCheck>
struct CastCheck {
template <class T>
static void Perform(T* data);
};
template <>
template <class T>
void CastCheck<true>::Perform(T* data) {
T::Cast(data);
}
template <>
template <class T>
void CastCheck<false>::Perform(T* data) {}
template <class T>
V8_INLINE void PerformCastCheck(T* data) {
CastCheck<std::is_base_of<Data, T>::value &&
!std::is_same<Data, std::remove_cv_t<T>>::value>::Perform(data);
}
// A base class for backing stores, which is needed due to vagaries of
// how static casts work with std::shared_ptr.
class BackingStoreBase {};
// The maximum value in enum GarbageCollectionReason, defined in heap.h.
// This is needed for histograms sampling garbage collection reasons.
constexpr int kGarbageCollectionReasonMaxValue = 27;
} // namespace internal
} // namespace v8
#endif // INCLUDE_V8_INTERNAL_H_