thermosphere: guest mem rewrite

This commit is contained in:
TuxSH 2020-03-05 02:01:03 +00:00
parent fc8a596409
commit 2986967f2a
7 changed files with 263 additions and 314 deletions

View File

@ -24,7 +24,7 @@
#include "hvisor_gdb_defines_internal.hpp"
#include "hvisor_gdb_packet_data.hpp"
#include "../guest_memory.h"
#include "../hvisor_guest_memory.hpp"
namespace ams::hvisor::gdb {
@ -40,7 +40,7 @@ namespace ams::hvisor::gdb {
return prefixLen == 0 ? ReplyErrno(ENOMEM) : -1;
}
size_t total = guestReadMemory(addr, len, membuf);
size_t total = GuestReadMemory(addr, len, membuf);
if (total == 0) {
return prefixLen == 0 ? ReplyErrno(EFAULT) : -EFAULT;
@ -73,7 +73,7 @@ namespace ams::hvisor::gdb {
return ReplyErrno(EILSEQ);
}
size_t total = guestWriteMemory(addr, len, workbuf);
size_t total = GuestWriteMemory(addr, len, workbuf);
return total == len ? ReplyOk() : ReplyErrno(EFAULT);
}

View File

@ -1,292 +0,0 @@
/*
* Copyright (c) 2019 Atmosphère-NX
*
* This program is free software; you can redistribute it and/or modify it
* under the terms and conditions of the GNU General Public License,
* version 2, as published by the Free Software Foundation.
*
* This program is distributed in the hope it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
* more details.
*
* You should have received a copy of the GNU General Public License
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
#include <string.h>
#include "guest_memory.h"
#include "memory_map.h"
#include "mmu.h"
#include "spinlock.h"
#include "core_ctx.h"
#include "sysreg.h"
#include "vgic.h"
#include "irq.h"
#include "caches.h"
static size_t guestReadWriteGicd(size_t offset, size_t size, void *readBuf, const void *writeBuf)
{
recursiveSpinlockLock(&g_irqManager.lock);
if (readBuf != NULL) {
size_t readOffset = 0;
size_t rem = size;
while (rem > 0) {
if ((offset + readOffset) % 4 == 0 && rem >= 4) {
// All accesses of this kind are valid
*(u32 *)((uintptr_t)readBuf + readOffset) = vgicReadGicdRegister(offset + readOffset, 4);
readOffset += 4;
rem -= 4;
} else if ((offset + readOffset) % 2 == 0 && rem >= 2) {
// All accesses of this kind would be translated to ldrh and are thus invalid. Abort.
size = readOffset;
goto end;
} else if (vgicValidateGicdRegisterAccess(offset + readOffset, 1)) {
// Valid byte access
*(u8 *)((uintptr_t)readBuf + readOffset) = vgicReadGicdRegister(offset + readOffset, 1);
readOffset += 1;
rem -= 1;
} else {
// Invalid byte access
size = readOffset;
goto end;
}
}
}
if (writeBuf != NULL) {
size_t writeOffset = 0;
size_t rem = size;
while (rem > 0) {
if ((offset + writeOffset) % 4 == 0 && rem >= 4) {
// All accesses of this kind are valid
vgicWriteGicdRegister(*(u32 *)((uintptr_t)writeBuf + writeOffset), offset + writeOffset, 4);
writeOffset += 4;
rem -= 4;
} else if ((offset + writeOffset) % 2 == 0 && rem >= 2) {
// All accesses of this kind would be translated to ldrh and are thus invalid. Abort.
size = writeOffset;
goto end;
} else if (vgicValidateGicdRegisterAccess(offset + writeOffset, 1)) {
// Valid byte access
vgicWriteGicdRegister(*(u32 *)((uintptr_t)writeBuf + writeOffset), offset + writeOffset, 1);
writeOffset += 1;
rem -= 1;
} else {
// Invalid byte access
size = writeOffset;
goto end;
}
}
}
end:
recursiveSpinlockUnlock(&g_irqManager.lock);
return size;
}
static size_t guestReadWriteDeviceMemory(void *addr, size_t size, void *readBuf, const void *writeBuf)
{
// We might trigger bus errors... ignore the exception and return early if that's the case
CoreCtx *curCtxBackup = currentCoreCtx;
__compiler_barrier();
currentCoreCtx = NULL;
__compiler_barrier();
uintptr_t addri = (uintptr_t)addr;
if (readBuf != NULL) {
size_t readOffset = 0;
size_t rem = size;
while (rem > 0 && (__compiler_barrier(), currentCoreCtx == NULL)) {
if ((addri + readOffset) % 4 == 0 && rem >= 4) {
*(vu32 *)((uintptr_t)readBuf + readOffset) = *(vu32 *)(addri + readOffset);
readOffset += 4;
rem -= 4;
} else if (readOffset % 2 == 0 && rem >= 2) {
*(vu16 *)((uintptr_t)readBuf + readOffset) = *(vu16 *)(addri + readOffset);
readOffset += 2;
rem -= 2;
} else {
*(vu8 *)((uintptr_t)readBuf + readOffset) = *(vu8 *)(addri + readOffset);
readOffset += 1;
rem -= 1;
}
}
if (rem != 0) {
size = readOffset;
goto end;
}
}
if (writeBuf != NULL) {
size_t writeOffset = 0;
size_t rem = size;
while (rem > 0 && (__compiler_barrier(), currentCoreCtx == NULL)) {
if ((addri + writeOffset) % 4 == 0 && rem >= 4) {
*(vu32 *)(addri + writeOffset) = *(vu32 *)((uintptr_t)writeBuf + writeOffset);
writeOffset += 4;
rem -= 4;
} else if (writeOffset % 2 == 0 && rem >= 2) {
*(vu16 *)(addri + writeOffset) = *(vu16 *)((uintptr_t)writeBuf + writeOffset);
writeOffset += 2;
rem -= 2;
} else {
*(vu8 *)(addri + writeOffset) = *(vu8 *)((uintptr_t)writeBuf + writeOffset);
writeOffset += 1;
rem -= 1;
}
}
if (rem != 0) {
size = writeOffset;
goto end;
}
}
end:
__compiler_barrier();
currentCoreCtx = curCtxBackup;
__compiler_barrier();
return size;
}
static size_t guestReadWriteNormalMemory(void *addr, size_t size, void *readBuf, const void *writeBuf)
{
if (readBuf != NULL) {
memcpy(readBuf, addr, size);
}
if (writeBuf != NULL) {
memcpy(addr, writeBuf, size);
// We may have written to executable memory or to translation tables...
// & the page may have various aliases.
// We need to ensure cache & TLB coherency.
cacheCleanDataCacheRangePoU(addr, size);
u32 policy = cacheGetInstructionCachePolicy();
if (policy == 1 || policy == 2) {
// AVIVT, VIVT
cacheInvalidateInstructionCache();
} else {
// VPIPT, PIPT
// Ez coherency, just do range operations...
cacheInvalidateInstructionCacheRangePoU(addr, size);
}
__tlb_invalidate_el1();
__dsb();
__isb();
}
return size;
}
static size_t guestReadWriteMemoryPage(uintptr_t addr, size_t size, void *readBuf, const void *writeBuf)
{
u64 irqFlags = maskIrq();
size_t offset = addr & 0xFFFull;
// Translate the VA, stages 1&2
__asm__ __volatile__ ("at s12e1r, %0" :: "r"(addr) : "memory");
u64 par = GET_SYSREG(par_el1);
if (par & PAR_F) {
// The translation failed. Why?
if (par & PAR_S) {
// Stage 2 fault. Could be an attempt to access the GICD, let's see what the IPA is...
__asm__ __volatile__ ("at s1e1r, %0" :: "r"(addr) : "memory");
par = GET_SYSREG(par_el1);
if ((par & PAR_F) != 0 || (par & PAR_PA_MASK) != MEMORY_MAP_VA_GICD) {
// The guest doesn't have access to it...
// Read as 0, write ignored
if (readBuf != NULL) {
memset(readBuf, 0, size);
}
} else {
// GICD mmio
size = guestReadWriteGicd(offset, size, readBuf, writeBuf);
}
} else {
// Oops, couldn't read/write anything (stage 1 fault)
size = 0;
}
} else {
/*
Translation didn't fail.
To avoid "B2.8 Mismatched memory attributes" we must use the same effective
attributes & shareability as the guest.
Note that par_el1 reports the effective shareablity of device and noncacheable memory as inner shareable.
In fact, the VMSAv8-64 section in the Armv8 ARM reads:
"The shareability field is only relevant if the memory is a Normal Cacheable memory type. All Device and Normal
Non-cacheable memory regions are always treated as Outer Shareable, regardless of the translation table
shareability attributes."
There's one corner case where we can't avoid it: another core is running,
changes the attributes (other than permissions) of the page, and issues
a broadcasting TLB maintenance instructions and/or accesses the page with the altered
attribute itself. We don't handle this corner case -- just don't read/write that kind of memory...
*/
u64 memAttribs = (par >> PAR_ATTR_SHIFT) & PAR_ATTR_MASK;
u32 shrb = (par >> PAR_SH_SHIFT) & PAR_SH_MASK;
uintptr_t pa = par & PAR_PA_MASK;
uintptr_t va = MEMORY_MAP_VA_GUEST_MEM + 0x2000 * currentCoreCtx->coreId;
u64 mair = GET_SYSREG(mair_el2);
mair |= memAttribs << (8 * MEMORY_MAP_MEMTYPE_NORMAL_GUEST_SLOT);
SET_SYSREG(mair_el2, mair);
__isb();
u64 attribs = MMU_PTE_BLOCK_XN | MMU_PTE_BLOCK_SH(shrb) | MMU_PTE_BLOCK_MEMTYPE(MEMORY_MAP_MEMTYPE_NORMAL_GUEST_SLOT);
mmu_map_page((uintptr_t *)MEMORY_MAP_VA_TTBL, va, pa, attribs);
// Note: no need to broadcast here
__tlb_invalidate_el2_page_local(pa);
__dsb_local();
void *vaddr = (void *)(va + offset);
if (memAttribs & 0xF0) {
// Normal memory, or unpredictable
size = guestReadWriteNormalMemory(vaddr, size, readBuf, writeBuf);
} else {
// Device memory, or unpredictable
size = guestReadWriteDeviceMemory(vaddr, size, readBuf, writeBuf);
}
__dsb_local();
__isb();
mmu_unmap_page((uintptr_t *)MEMORY_MAP_VA_TTBL, va);
// Note: no need to broadcast here
__tlb_invalidate_el2_page_local(pa);
__dsb_local();
mair &= ~(0xFFul << (8 * MEMORY_MAP_MEMTYPE_NORMAL_GUEST_SLOT));
SET_SYSREG(mair_el2, mair);
__isb();
}
restoreInterruptFlags(irqFlags);
return size;
}
size_t guestReadWriteMemory(uintptr_t addr, size_t size, void *readBuf, const void *writeBuf)
{
uintptr_t curAddr = addr;
size_t remainingAmount = size;
u8 *rb8 = (u8 *)readBuf;
const u8 *wb8 = (const u8*)writeBuf;
while (remainingAmount > 0) {
size_t expectedAmount = ((curAddr & ~0xFFFul) + 0x1000) - curAddr;
expectedAmount = expectedAmount > remainingAmount ? remainingAmount : expectedAmount;
size_t actualAmount = guestReadWriteMemoryPage(curAddr, expectedAmount, rb8, wb8);
curAddr += actualAmount;
rb8 += actualAmount;
wb8 += actualAmount;
remainingAmount -= actualAmount;
if (actualAmount != expectedAmount) {
break;
}
}
return curAddr - addr;
}

View File

@ -0,0 +1,240 @@
/*
* Copyright (c) 2019-2020 Atmosphère-NX
*
* This program is free software; you can redistribute it and/or modify it
* under the terms and conditions of the GNU General Public License,
* version 2, as published by the Free Software Foundation.
*
* This program is distributed in the hope it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
* more details.
*
* You should have received a copy of the GNU General Public License
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
#include "hvisor_virtual_gic.hpp"
#include "hvisor_safe_io_copy.hpp"
#include "cpu/hvisor_cpu_caches.hpp"
#include "cpu/hvisor_cpu_interrupt_mask_guard.hpp"
using namespace ams::hvisor;
using namespace ams::hvisor::cpu;
namespace {
template<typename T>
T ReadBufferValue(const void *buf, size_t off)
{
static_assert(std::is_unsigned_v<T> && sizeof(T) <= 4);
T ret;
std::memcpy(&ret, reinterpret_cast<const u8 *>(buf) + off, sizeof(T));
return ret;
}
template<typename T>
void WriteBufferValue(void *buf, size_t off, T val)
{
static_assert(std::is_unsigned_v<T> && sizeof(T) <= 4);
std::memcpy(reinterpret_cast<u8 *>(buf) + off, T, sizeof(T));
}
size_t GuestReadWriteGicd(size_t offset, size_t size, void *readBuf, const void *writeBuf)
{
auto &vgic = VirtualGic::GetInstance();
if (readBuf != nullptr) {
size_t readOffset = 0;
size_t rem = size;
while (rem > 0) {
if ((offset + readOffset) % 4 == 0 && rem >= 4) {
// All accesses of this kind are valid
WriteBufferValue<u32>(readBuf, readOffset, vgic.ReadGicdRegister(offset + readOffset, 4));
readOffset += 4;
rem -= 4;
} else if ((offset + readOffset) % 2 == 0 && rem >= 2) {
// All accesses of this kind would be translated to ldrh and are thus invalid. Abort.
return readOffset;
} else if (VirtualGic::ValidateGicdRegisterAccess(offset + readOffset, 1)) {
// Valid byte access
WriteBufferValue<u8>(readBuf, readOffset, vgic.ReadGicdRegister(offset + readOffset, 1));
readOffset += 1;
rem -= 1;
} else {
// Invalid byte access
return readOffset;
}
}
}
if (writeBuf != nullptr) {
size_t writeOffset = 0;
size_t rem = size;
while (rem > 0) {
if ((offset + writeOffset) % 4 == 0 && rem >= 4) {
// All accesses of this kind are valid
vgic.WriteGicdRegister(ReadBufferValue<u32>(writeBuf, writeOffset), offset + writeOffset, 4);
writeOffset += 4;
rem -= 4;
} else if ((offset + writeOffset) % 2 == 0 && rem >= 2) {
// All accesses of this kind would be translated to ldrh and are thus invalid. Abort.
return writeOffset;
} else if (VirtualGic::ValidateGicdRegisterAccess(offset + writeOffset, 1)) {
// Valid byte access
vgic.WriteGicdRegister(ReadBufferValue<u8>(writeBuf, writeOffset), offset + writeOffset, 1);
writeOffset += 1;
rem -= 1;
} else {
// Invalid byte access
return writeOffset;
}
}
}
return size;
}
size_t GuestReadWriteDeviceMemory(void *addr, size_t size, void *readBuf, const void *writeBuf)
{
if (readBuf != nullptr) {
size_t sz = SafeIoCopy(readBuf, addr, size);
if (sz < size) {
return sz;
}
}
if (writeBuf != nullptr) {
size_t sz = SafeIoCopy(addr, writeBuf, size);
if (sz < size) {
return sz;
}
}
// Translation tables must be on Normal memory & Device memory isn't cacheable, so we don't have
// that kind of thing to handle...
return size;
}
size_t GuestReadWriteNormalMemory(void *addr, size_t size, void *readBuf, const void *writeBuf)
{
if (readBuf != nullptr) {
std::memcpy(readBuf, addr, size);
}
if (writeBuf != nullptr) {
std::memcpy(addr, writeBuf, size);
// We may have written to executable memory or to translation tables...
// & the page may have various aliases.
// We need to ensure cache & TLB coherency.
CleanDataCacheRangePoU(addr, size);
u32 policy = GetInstructionCachePolicy();
if (policy == 1 || policy == 2) {
// AVIVT, VIVT
InvalidateInstructionCache();
} else {
// VPIPT, PIPT
// Ez coherency, just do range operations...
InvalidateInstructionCacheRangePoU(addr, size);
}
TlbInvalidateEl1();
dsb();
isb();
}
return size;
}
size_t GuestReadWriteMemoryPage(uintptr_t addr, size_t size, void *readBuf, const void *writeBuf)
{
InterruptMaskGuard ig{};
size_t offset = addr & 0xFFFul;
// Translate the VA, stages 1&2
__asm__ __volatile__ ("at s12e1r, %0" :: "r"(addr) : "memory");
u64 par = THERMOSPHERE_GET_SYSREG(par_el1);
if (par & PAR_F) {
// The translation failed. Why?
if (par & PAR_S) {
// Stage 2 fault. Could be an attempt to access the GICD, let's see what the IPA is...
__asm__ __volatile__ ("at s1e1r, %0" :: "r"(addr) : "memory");
par = THERMOSPHERE_GET_SYSREG(par_el1);
if ((par & PAR_F) != 0 || (par & PAR_PA_MASK) != VirtualGic::gicdPhysicalAddress) {
// The guest doesn't have access to it...
// Read as 0, write ignored
if (readBuf != NULL) {
std::memset(readBuf, 0, size);
}
} else {
// GICD mmio
size = GuestReadWriteGicd(offset, size, readBuf, writeBuf);
}
} else {
// Oops, couldn't read/write anything (stage 1 fault)
size = 0;
}
} else {
/*
Translation didn't fail.
To avoid "B2.8 Mismatched memory attributes" we must use the same effective
attributes & shareability as the guest.
Note that par_el1 reports the effective shareablity of device and noncacheable memory as inner shareable.
In fact, the VMSAv8-64 section in the Armv8 ARM reads:
"The shareability field is only relevant if the memory is a Normal Cacheable memory type. All Device and Normal
Non-cacheable memory regions are always treated as Outer Shareable, regardless of the translation table
shareability attributes."
There's one corner case where we can't avoid it: another core is running,
changes the attributes (other than permissions) of the page, and issues
a broadcasting TLB maintenance instructions and/or accesses the page with the altered
attribute itself. We don't handle this corner case -- just don't read/write that kind of memory...
*/
u64 memAttribs = (par >> PAR_ATTR_SHIFT) & PAR_ATTR_MASK;
u64 shrb = (par >> PAR_SH_SHIFT) & PAR_SH_MASK;
uintptr_t pa = par & PAR_PA_MASK;
uintptr_t va = MemoryMap::MapGuestPage(pa, memAttribs, shrb);
void *vaddr = reinterpret_cast<void *>(va + offset);
if (memAttribs & 0xF0) {
// Normal memory, or unpredictable
size = GuestReadWriteNormalMemory(vaddr, size, readBuf, writeBuf);
} else {
// Device memory, or unpredictable
size = GuestReadWriteDeviceMemory(vaddr, size, readBuf, writeBuf);
}
MemoryMap::UnmapGuestPage();
}
return size;
}
}
namespace ams::hvisor {
size_t GuestReadWriteMemory(uintptr_t addr, size_t size, void *readBuf, const void *writeBuf)
{
uintptr_t curAddr = addr;
size_t remainingAmount = size;
u8 *rb8 = reinterpret_cast<u8 *>(readBuf);
const u8 *wb8 = reinterpret_cast<const u8 *>(writeBuf);
while (remainingAmount > 0) {
size_t expectedAmount = ((curAddr & ~0xFFFul) + 0x1000) - curAddr;
expectedAmount = expectedAmount > remainingAmount ? remainingAmount : expectedAmount;
size_t actualAmount = GuestReadWriteMemoryPage(curAddr, expectedAmount, rb8, wb8);
curAddr += actualAmount;
rb8 += actualAmount;
wb8 += actualAmount;
remainingAmount -= actualAmount;
if (actualAmount != expectedAmount) {
break;
}
}
return curAddr - addr;
}
}

View File

@ -1,5 +1,5 @@
/*
* Copyright (c) 2019 Atmosphère-NX
* Copyright (c) 2019-2020 Atmosphère-NX
*
* This program is free software; you can redistribute it and/or modify it
* under the terms and conditions of the GNU General Public License,
@ -16,16 +16,20 @@
#pragma once
#include "utils.h"
#include "defines.hpp"
size_t guestReadWriteMemory(uintptr_t addr, size_t size, void *readBuf, const void *writeBuf);
namespace ams::hvisor {
size_t GuestReadWriteMemory(uintptr_t addr, size_t size, void *readBuf, const void *writeBuf);
inline size_t GuestReadMemory(uintptr_t addr, size_t size, void *buf)
{
return GuestReadWriteMemory(addr, size, buf, NULL);
}
inline size_t GuestWriteMemory(uintptr_t addr, size_t size, const void *buf)
{
return GuestReadWriteMemory(addr, size, NULL, buf);
}
static inline size_t guestReadMemory(uintptr_t addr, size_t size, void *buf)
{
return guestReadWriteMemory(addr, size, buf, NULL);
}
static inline size_t guestWriteMemory(uintptr_t addr, size_t size, const void *buf)
{
return guestReadWriteMemory(addr, size, NULL, buf);
}

View File

@ -21,8 +21,6 @@
#include "cpu/hvisor_cpu_mmu.hpp"
#include "cpu/hvisor_cpu_instructions.hpp"
#include "platform/interrupt_config.h" // TODO remove
namespace ams::hvisor {
uintptr_t MemoryMap::currentPlatformMmioPage = MemoryMap::mmioPlatBaseVa;
@ -176,7 +174,7 @@ namespace ams::hvisor {
isb();
}
uintptr_t MemoryMap::UnmapGuestPage()
void MemoryMap::UnmapGuestPage()
{
using namespace cpu;
using Builder = MmuTableBuilder<3, addressSpaceSize, true>;

View File

@ -85,7 +85,7 @@ namespace ams::hvisor {
// Caller is expected to disable interrupts, etc, etc.
static uintptr_t MapGuestPage(uintptr_t pa, u64 memAttribs, u64 shareability);
static uintptr_t UnmapGuestPage();
static void UnmapGuestPage();
public:
constexpr MemoryMap() = delete;

View File

@ -16,13 +16,12 @@
#include "hvisor_sw_breakpoint_manager.hpp"
#include "hvisor_core_context.hpp"
#include "hvisor_guest_memory.hpp"
#include "cpu/hvisor_cpu_instructions.hpp"
#include "cpu/hvisor_cpu_interrupt_mask_guard.hpp"
#include <mutex>
#include "guest_memory.h"
#define _REENT_ONLY
#include <cerrno>
@ -59,7 +58,7 @@ namespace ams::hvisor {
Breakpoint &bp = m_breakpoints[id];
u32 brkInst = 0xD4200000 | (bp.uid << 5);
size_t sz = guestReadWriteMemory(bp.address, 4, &bp.savedInstruction, &brkInst);
size_t sz = GuestReadWriteMemory(bp.address, 4, &bp.savedInstruction, &brkInst);
bp.applied = sz == 4;
return sz == 4;
}
@ -67,7 +66,7 @@ namespace ams::hvisor {
bool SwBreakpointManager::DoRevert(size_t id)
{
Breakpoint &bp = m_breakpoints[id];
size_t sz = guestWriteMemory(bp.address, 4, &bp.savedInstruction);
size_t sz = GuestWriteMemory(bp.address, 4, &bp.savedInstruction);
bp.applied = sz != 4;
return sz == 4;
}