ryujinx/Ryujinx.Cpu/MemoryManager.cs
riperiperi b4d8d893a4
Memory Read/Write Tracking using Region Handles (#1272)
* WIP Range Tracking

- Texture invalidation seems to have large problems
- Buffer/Pool invalidation may have problems
- Mirror memory tracking puts an additional `add` in compiled code, we likely just want to make HLE access slower if this is the final solution.
- Native project is in the messiest possible location.
- [HACK] JIT memory access always uses native "fast" path
- [HACK] Trying some things with texture invalidation and views.

It works :)

Still a few hacks, messy things, slow things

More work in progress stuff (also move to memory project)

Quite a bit faster now.
- Unmapping GPU VA and CPU VA will now correctly update write tracking regions, and invalidate textures for the former.
- The Virtual range list is now non-overlapping like the physical one.
- Fixed some bugs where regions could leak.
- Introduced a weird bug that I still need to track down (consistent invalid buffer in MK8 ribbon road)

Move some stuff.

I think we'll eventually just put the dll and so for this in a nuget package.

Fix rebase.

[WIP] MultiRegionHandle variable size ranges

- Avoid reprotecting regions that change often (needs some tweaking)
- There's still a bug in buffers, somehow.
- Might want different api for minimum granularity

Fix rebase issue

Commit everything needed for software only tracking.

Remove native components.

Remove more native stuff.

Cleanup

Use a separate window for the background context, update opentk. (fixes linux)

Some experimental changes

Should get things working up to scratch - still need to try some things with flush/modification and res scale.

Include address with the region action.

Initial work to make range tracking work

Still a ton of bugs

Fix some issues with the new stuff.

* Fix texture flush instability

There's still some weird behaviour, but it's much improved without this. (textures with cpu modified data were flushing over it)

* Find the destination texture for Buffer->Texture full copy

Greatly improves performance for nvdec videos (with range tracking)

* Further improve texture tracking

* Disable Memory Tracking for view parents

This is a temporary approach to better match behaviour on master (where invalidations would be soaked up by views, rather than trigger twice)

The assumption is that when views are created to a texture, they will cover all of its data anyways. Of course, this can easily be improved in future.

* Introduce some tracking tests.

WIP

* Complete base tests.

* Add more tests for multiregion, fix existing test.

* Cleanup Part 1

* Remove unnecessary code from memory tracking

* Fix some inconsistencies with 3D texture rule.

* Add dispose tests.

* Use a background thread for the background context.

Rather than setting and unsetting a context as current, doing the work on a dedicated thread with signals seems to be a bit faster.

Also nerf the multithreading test a bit.

* Copy to texture with matching alignment

This extends the copy to work for some videos with unusual size, such as tutorial videos in SMO. It will only occur if the destination texture already exists at XCount size.

* Track reads for buffer copies. Synchronize new buffers before copying overlaps.

* Remove old texture flushing mechanisms.

Range tracking all the way, baby.

* Wake the background thread when disposing.

Avoids a deadlock when games are closed.

* Address Feedback 1

* Separate TextureCopy instance for background thread

Also `BackgroundContextWorker.InBackground` for a more sensible idenfifier for if we're in a background thread.

* Add missing XML docs.

* Address Feedback

* Maybe I should start drinking coffee.

* Some more feedback.

* Remove flush warning, Refocus window after making background context
2020-10-16 17:18:35 -03:00

629 lines
22 KiB
C#

using ARMeilleure.Memory;
using Ryujinx.Cpu.Tracking;
using Ryujinx.Memory;
using Ryujinx.Memory.Tracking;
using System;
using System.Collections.Generic;
using System.Runtime.CompilerServices;
using System.Runtime.InteropServices;
using System.Threading;
namespace Ryujinx.Cpu
{
/// <summary>
/// Represents a CPU memory manager.
/// </summary>
public sealed class MemoryManager : IMemoryManager, IDisposable, IVirtualMemoryManager
{
public const int PageBits = 12;
public const int PageSize = 1 << PageBits;
public const int PageMask = PageSize - 1;
private const int PteSize = 8;
private readonly InvalidAccessHandler _invalidAccessHandler;
/// <summary>
/// Address space width in bits.
/// </summary>
public int AddressSpaceBits { get; }
private readonly ulong _addressSpaceSize;
private readonly MemoryBlock _backingMemory;
private readonly MemoryBlock _pageTable;
/// <summary>
/// Page table base pointer.
/// </summary>
public IntPtr PageTablePointer => _pageTable.Pointer;
public MemoryTracking Tracking { get; }
/// <summary>
/// Creates a new instance of the memory manager.
/// </summary>
/// <param name="backingMemory">Physical backing memory where virtual memory will be mapped to</param>
/// <param name="addressSpaceSize">Size of the address space</param>
/// <param name="invalidAccessHandler">Optional function to handle invalid memory accesses</param>
public MemoryManager(MemoryBlock backingMemory, ulong addressSpaceSize, InvalidAccessHandler invalidAccessHandler = null)
{
_invalidAccessHandler = invalidAccessHandler;
ulong asSize = PageSize;
int asBits = PageBits;
while (asSize < addressSpaceSize)
{
asSize <<= 1;
asBits++;
}
AddressSpaceBits = asBits;
_addressSpaceSize = asSize;
_backingMemory = backingMemory;
_pageTable = new MemoryBlock((asSize / PageSize) * PteSize);
Tracking = new MemoryTracking(this, backingMemory, PageSize);
Tracking.EnablePhysicalProtection = false; // Disabled for now, as protection is done in software.
}
/// <summary>
/// Maps a virtual memory range into a physical memory range.
/// </summary>
/// <remarks>
/// Addresses and size must be page aligned.
/// </remarks>
/// <param name="va">Virtual memory address</param>
/// <param name="pa">Physical memory address</param>
/// <param name="size">Size to be mapped</param>
public void Map(ulong va, ulong pa, ulong size)
{
ulong remainingSize = size;
ulong oVa = va;
ulong oPa = pa;
while (remainingSize != 0)
{
_pageTable.Write((va / PageSize) * PteSize, PaToPte(pa));
va += PageSize;
pa += PageSize;
remainingSize -= PageSize;
}
Tracking.Map(oVa, oPa, size);
}
/// <summary>
/// Unmaps a previously mapped range of virtual memory.
/// </summary>
/// <param name="va">Virtual address of the range to be unmapped</param>
/// <param name="size">Size of the range to be unmapped</param>
public void Unmap(ulong va, ulong size)
{
ulong remainingSize = size;
ulong oVa = va;
while (remainingSize != 0)
{
_pageTable.Write((va / PageSize) * PteSize, 0UL);
va += PageSize;
remainingSize -= PageSize;
}
Tracking.Unmap(oVa, size);
}
/// <summary>
/// Reads data from CPU mapped memory.
/// </summary>
/// <typeparam name="T">Type of the data being read</typeparam>
/// <param name="va">Virtual address of the data in memory</param>
/// <returns>The data</returns>
/// <exception cref="InvalidMemoryRegionException">Throw for unhandled invalid or unmapped memory accesses</exception>
public T Read<T>(ulong va) where T : unmanaged
{
return MemoryMarshal.Cast<byte, T>(GetSpan(va, Unsafe.SizeOf<T>()))[0];
}
/// <summary>
/// Reads data from CPU mapped memory, with read tracking
/// </summary>
/// <typeparam name="T">Type of the data being read</typeparam>
/// <param name="va">Virtual address of the data in memory</param>
/// <returns>The data</returns>
public T ReadTracked<T>(ulong va) where T : unmanaged
{
SignalMemoryTracking(va, (ulong)Unsafe.SizeOf<T>(), false);
return MemoryMarshal.Cast<byte, T>(GetSpan(va, Unsafe.SizeOf<T>()))[0];
}
/// <summary>
/// Reads data from CPU mapped memory.
/// </summary>
/// <param name="va">Virtual address of the data in memory</param>
/// <param name="data">Span to store the data being read into</param>
/// <exception cref="InvalidMemoryRegionException">Throw for unhandled invalid or unmapped memory accesses</exception>
public void Read(ulong va, Span<byte> data)
{
ReadImpl(va, data);
}
/// <summary>
/// Writes data to CPU mapped memory.
/// </summary>
/// <typeparam name="T">Type of the data being written</typeparam>
/// <param name="va">Virtual address to write the data into</param>
/// <param name="value">Data to be written</param>
/// <exception cref="InvalidMemoryRegionException">Throw for unhandled invalid or unmapped memory accesses</exception>
public void Write<T>(ulong va, T value) where T : unmanaged
{
Write(va, MemoryMarshal.Cast<T, byte>(MemoryMarshal.CreateSpan(ref value, 1)));
}
/// <summary>
/// Writes data to CPU mapped memory, with write tracking.
/// </summary>
/// <param name="va">Virtual address to write the data into</param>
/// <param name="data">Data to be written</param>
/// <exception cref="InvalidMemoryRegionException">Throw for unhandled invalid or unmapped memory accesses</exception>
public void Write(ulong va, ReadOnlySpan<byte> data)
{
if (data.Length == 0)
{
return;
}
SignalMemoryTracking(va, (ulong)data.Length, true);
WriteImpl(va, data);
}
/// <summary>
/// Writes data to CPU mapped memory, without write tracking.
/// </summary>
/// <param name="va">Virtual address to write the data into</param>
/// <param name="data">Data to be written</param>
public void WriteUntracked(ulong va, ReadOnlySpan<byte> data)
{
if (data.Length == 0)
{
return;
}
WriteImpl(va, data);
}
[MethodImpl(MethodImplOptions.AggressiveInlining)]
/// <summary>
/// Writes data to CPU mapped memory.
/// </summary>
/// <param name="va">Virtual address to write the data into</param>
/// <param name="data">Data to be written</param>
private void WriteImpl(ulong va, ReadOnlySpan<byte> data)
{
try
{
if (IsContiguousAndMapped(va, data.Length))
{
data.CopyTo(_backingMemory.GetSpan(GetPhysicalAddressInternal(va), data.Length));
}
else
{
int offset = 0, size;
if ((va & PageMask) != 0)
{
ulong pa = GetPhysicalAddressInternal(va);
size = Math.Min(data.Length, PageSize - (int)(va & PageMask));
data.Slice(0, size).CopyTo(_backingMemory.GetSpan(pa, size));
offset += size;
}
for (; offset < data.Length; offset += size)
{
ulong pa = GetPhysicalAddressInternal(va + (ulong)offset);
size = Math.Min(data.Length - offset, PageSize);
data.Slice(offset, size).CopyTo(_backingMemory.GetSpan(pa, size));
}
}
}
catch (InvalidMemoryRegionException)
{
if (_invalidAccessHandler == null || !_invalidAccessHandler(va))
{
throw;
}
}
}
/// <summary>
/// Gets a read-only span of data from CPU mapped memory.
/// </summary>
/// <remarks>
/// This may perform a allocation if the data is not contiguous in memory.
/// For this reason, the span is read-only, you can't modify the data.
/// </remarks>
/// <param name="va">Virtual address of the data</param>
/// <param name="size">Size of the data</param>
/// <param name="tracked">True if read tracking is triggered on the span</param>
/// <returns>A read-only span of the data</returns>
/// <exception cref="InvalidMemoryRegionException">Throw for unhandled invalid or unmapped memory accesses</exception>
public ReadOnlySpan<byte> GetSpan(ulong va, int size, bool tracked = false)
{
if (size == 0)
{
return ReadOnlySpan<byte>.Empty;
}
if (tracked)
{
SignalMemoryTracking(va, (ulong)size, false);
}
if (IsContiguousAndMapped(va, size))
{
return _backingMemory.GetSpan(GetPhysicalAddressInternal(va), size);
}
else
{
Span<byte> data = new byte[size];
ReadImpl(va, data);
return data;
}
}
/// <summary>
/// Gets a region of memory that can be written to.
/// </summary>
/// <remarks>
/// If the requested region is not contiguous in physical memory,
/// this will perform an allocation, and flush the data (writing it
/// back to guest memory) on disposal.
/// </remarks>
/// <param name="va">Virtual address of the data</param>
/// <param name="size">Size of the data</param>
/// <returns>A writable region of memory containing the data</returns>
/// <exception cref="InvalidMemoryRegionException">Throw for unhandled invalid or unmapped memory accesses</exception>
public WritableRegion GetWritableRegion(ulong va, int size)
{
if (size == 0)
{
return new WritableRegion(null, va, Memory<byte>.Empty);
}
if (IsContiguousAndMapped(va, size))
{
return new WritableRegion(null, va, _backingMemory.GetMemory(GetPhysicalAddressInternal(va), size));
}
else
{
Memory<byte> memory = new byte[size];
GetSpan(va, size).CopyTo(memory.Span);
return new WritableRegion(this, va, memory);
}
}
/// <summary>
/// Gets a reference for the given type at the specified virtual memory address.
/// </summary>
/// <remarks>
/// The data must be located at a contiguous memory region.
/// </remarks>
/// <typeparam name="T">Type of the data to get the reference</typeparam>
/// <param name="va">Virtual address of the data</param>
/// <returns>A reference to the data in memory</returns>
/// <exception cref="MemoryNotContiguousException">Throw if the specified memory region is not contiguous in physical memory</exception>
public ref T GetRef<T>(ulong va) where T : unmanaged
{
if (!IsContiguous(va, Unsafe.SizeOf<T>()))
{
ThrowMemoryNotContiguous();
}
SignalMemoryTracking(va, (ulong)Unsafe.SizeOf<T>(), true);
return ref _backingMemory.GetRef<T>(GetPhysicalAddressInternal(va));
}
private void ThrowMemoryNotContiguous() => throw new MemoryNotContiguousException();
[MethodImpl(MethodImplOptions.AggressiveInlining)]
private bool IsContiguousAndMapped(ulong va, int size) => IsContiguous(va, size) && IsMapped(va);
[MethodImpl(MethodImplOptions.AggressiveInlining)]
private bool IsContiguous(ulong va, int size)
{
if (!ValidateAddress(va))
{
return false;
}
ulong endVa = (va + (ulong)size + PageMask) & ~(ulong)PageMask;
va &= ~(ulong)PageMask;
int pages = (int)((endVa - va) / PageSize);
for (int page = 0; page < pages - 1; page++)
{
if (!ValidateAddress(va + PageSize))
{
return false;
}
if (GetPhysicalAddressInternal(va) + PageSize != GetPhysicalAddressInternal(va + PageSize))
{
return false;
}
va += PageSize;
}
return true;
}
/// <summary>
/// Gets the physical regions that make up the given virtual address region.
/// If any part of the virtual region is unmapped, null is returned.
/// </summary>
/// <param name="va">Virtual address of the range</param>
/// <param name="size">Size of the range</param>
/// <returns>Array of physical regions</returns>
public (ulong address, ulong size)[] GetPhysicalRegions(ulong va, ulong size)
{
if (!ValidateAddress(va))
{
return null;
}
ulong endVa = (va + size + PageMask) & ~(ulong)PageMask;
va &= ~(ulong)PageMask;
int pages = (int)((endVa - va) / PageSize);
List<(ulong, ulong)> regions = new List<(ulong, ulong)>();
ulong regionStart = GetPhysicalAddressInternal(va);
ulong regionSize = PageSize;
for (int page = 0; page < pages - 1; page++)
{
if (!ValidateAddress(va + PageSize))
{
return null;
}
ulong newPa = GetPhysicalAddressInternal(va + PageSize);
if (GetPhysicalAddressInternal(va) + PageSize != newPa)
{
regions.Add((regionStart, regionSize));
regionStart = newPa;
regionSize = 0;
}
va += PageSize;
regionSize += PageSize;
}
regions.Add((regionStart, regionSize));
return regions.ToArray();
}
private void ReadImpl(ulong va, Span<byte> data)
{
if (data.Length == 0)
{
return;
}
try
{
int offset = 0, size;
if ((va & PageMask) != 0)
{
ulong pa = GetPhysicalAddressInternal(va);
size = Math.Min(data.Length, PageSize - (int)(va & PageMask));
_backingMemory.GetSpan(pa, size).CopyTo(data.Slice(0, size));
offset += size;
}
for (; offset < data.Length; offset += size)
{
ulong pa = GetPhysicalAddressInternal(va + (ulong)offset);
size = Math.Min(data.Length - offset, PageSize);
_backingMemory.GetSpan(pa, size).CopyTo(data.Slice(offset, size));
}
}
catch (InvalidMemoryRegionException)
{
if (_invalidAccessHandler == null || !_invalidAccessHandler(va))
{
throw;
}
}
}
/// <summary>
/// Checks if the page at a given CPU virtual address.
/// </summary>
/// <param name="va">Virtual address to check</param>
/// <returns>True if the address is mapped, false otherwise</returns>
[MethodImpl(MethodImplOptions.AggressiveInlining)]
public bool IsMapped(ulong va)
{
if (!ValidateAddress(va))
{
return false;
}
return _pageTable.Read<ulong>((va / PageSize) * PteSize) != 0;
}
private bool ValidateAddress(ulong va)
{
return va < _addressSpaceSize;
}
/// <summary>
/// Performs address translation of the address inside a CPU mapped memory range.
/// </summary>
/// <remarks>
/// If the address is invalid or unmapped, -1 will be returned.
/// </remarks>
/// <param name="va">Virtual address to be translated</param>
/// <returns>The physical address</returns>
public ulong GetPhysicalAddress(ulong va)
{
// We return -1L if the virtual address is invalid or unmapped.
if (!ValidateAddress(va) || !IsMapped(va))
{
return ulong.MaxValue;
}
return GetPhysicalAddressInternal(va);
}
private ulong GetPhysicalAddressInternal(ulong va)
{
return PteToPa(_pageTable.Read<ulong>((va / PageSize) * PteSize) & ~(0xffffUL << 48)) + (va & PageMask);
}
/// <summary>
/// Reprotect a region of virtual memory for tracking. Sets software protection bits.
/// </summary>
/// <param name="va">Virtual address base</param>
/// <param name="size">Size of the region to protect</param>
/// <param name="protection">Memory protection to set</param>
public void TrackingReprotect(ulong va, ulong size, MemoryPermission protection)
{
// Protection is inverted on software pages, since the default value is 0.
protection = (~protection) & MemoryPermission.ReadAndWrite;
long tag = (long)protection << 48;
if (tag > 0)
{
tag |= long.MinValue; // If any protection is present, the whole pte is negative.
}
ulong endVa = (va + size + PageMask) & ~(ulong)PageMask;
long invTagMask = ~(0xffffL << 48);
while (va < endVa)
{
ref long pageRef = ref _pageTable.GetRef<long>((va >> PageBits) * PteSize);
long pte;
do
{
pte = Volatile.Read(ref pageRef);
}
while (Interlocked.CompareExchange(ref pageRef, (pte & invTagMask) | tag, pte) != pte);
va += PageSize;
}
}
/// <summary>
/// Obtains a memory tracking handle for the given virtual region. This should be disposed when finished with.
/// </summary>
/// <param name="address">CPU virtual address of the region</param>
/// <param name="size">Size of the region</param>
/// <returns>The memory tracking handle</returns>
public CpuRegionHandle BeginTracking(ulong address, ulong size)
{
return new CpuRegionHandle(Tracking.BeginTracking(address, size));
}
/// <summary>
/// Obtains a memory tracking handle for the given virtual region, with a specified granularity. This should be disposed when finished with.
/// </summary>
/// <param name="address">CPU virtual address of the region</param>
/// <param name="size">Size of the region</param>
/// <param name="granularity">Desired granularity of write tracking</param>
/// <returns>The memory tracking handle</returns>
public CpuMultiRegionHandle BeginGranularTracking(ulong address, ulong size, ulong granularity)
{
return new CpuMultiRegionHandle(Tracking.BeginGranularTracking(address, size, granularity));
}
/// <summary>
/// Obtains a smart memory tracking handle for the given virtual region, with a specified granularity. This should be disposed when finished with.
/// </summary>
/// <param name="address">CPU virtual address of the region</param>
/// <param name="size">Size of the region</param>
/// <param name="granularity">Desired granularity of write tracking</param>
/// <returns>The memory tracking handle</returns>
public CpuSmartMultiRegionHandle BeginSmartGranularTracking(ulong address, ulong size, ulong granularity)
{
return new CpuSmartMultiRegionHandle(Tracking.BeginSmartGranularTracking(address, size, granularity));
}
/// <summary>
/// Alerts the memory tracking that a given region has been read from or written to.
/// This should be called before read/write is performed.
/// </summary>
/// <param name="va">Virtual address of the region</param>
/// <param name="size">Size of the region</param>
public void SignalMemoryTracking(ulong va, ulong size, bool write)
{
// We emulate guard pages for software memory access. This makes for an easy transition to
// tracking using host guard pages in future, but also supporting platforms where this is not possible.
// Write tag includes read protection, since we don't have any read actions that aren't performed before write too.
long tag = (write ? 3L : 1L) << 48;
ulong endVa = (va + size + PageMask) & ~(ulong)PageMask;
while (va < endVa)
{
ref long pageRef = ref _pageTable.GetRef<long>((va >> PageBits) * PteSize);
long pte;
pte = Volatile.Read(ref pageRef);
if ((pte & tag) != 0)
{
Tracking.VirtualMemoryEvent(va, size, write);
break;
}
va += PageSize;
}
}
private ulong PaToPte(ulong pa)
{
return (ulong)_backingMemory.GetPointer(pa, PageSize).ToInt64();
}
private ulong PteToPa(ulong pte)
{
return (ulong)((long)pte - _backingMemory.Pointer.ToInt64());
}
/// <summary>
/// Disposes of resources used by the memory manager.
/// </summary>
public void Dispose() => _pageTable.Dispose();
}
}