20.0.0: Difference between revisions

Line 324:

The anti-downgrade fuses were [[Fuses#Anti-downgrade|updated]].

==== Kernel ====

* No/minimal compiler update

* Inverted boolean(s) in system control/targetsystem logic

** All KTargetSystem fields have inverted meaning

** KTargetSystem "is present" bool is now inverted

** MersenneTwister "is initialized" bool is now inverted

* KPageTable::ChangeProperties now does a data synchronization barrier after calling the set-attrs lambda

* KPageTableImpl::MergePages now takes an argument for a callback to call after updating page table PTE entries.

** All callers on NX are KPageTable::NoteUpdated

** This causes TLB entries to be flushed/etc every time an entry is written instead of once at the end.

** The callback is passed the disable-merge software-reserved bits from the entry, but this appears unused on NX.

*** This is probably for something in the Switch2 kernel/ifdef'd out.

* KAddressSpaceInfo::GetBegin now takes a size as a third argument; this is unused on NX.

* KAddressSpaceInfo::GetSize now performs more complicated logic based on the input type.

** If address space not CreateProcessFlag_AddressSpace32BitWithoutAlias then the size is returned directly.

** Otherwise:

*** If the type is Heap, it returns the requested size + the alias size.

*** If the type is Alias, it returns 0.

** KProcess::InitializeByParam now sets m_max_process_memory to KAddressSpaceInfo::GetSize(..., Type_Heap) instead of getting the page table's heap size.

** KPageTableBase::InitializeForProcess has simpler calculation of region extents for 32BitWithoutAlias now.

* KPageTableBase::MapPageGroup_ now checks if the input permission has the execute bit set, and invalidates instruction cache if so.

** This fixes a correctness bug; this was used to map pages as ReadExecute by JIT svcs, so stale instruction cache could result before.

** KPageTableBase::UnmapPageGroup also now invalidates instruction cache after unmapping, if the memory had the execute bit set.

* KSecureSystemResource now has better checking for the case where resource limit is nullptr.

** This is impossible on NX.

** This includes checks in GetUsed/TotalUserPhysicalMemorySize

** KSecureSystemResource::Destroy now checks before calling ReleaseLimit

* KPageTableBase::SetProcessMemoryPermission changes:

** Function now uses PageTableOperation_ChangePermsAndRefresh instead of _ChangePermsAndRefreshAndFlushDataCache when the execute permission is set.

** Data cache store + instruction cache invalidate is now done before operating instead of after

** instruction cache invalidate is done after memory block manager is updated.

* El0SynchronousExceptionHandler now always does tlbi vae1 instead of doing that or tlbi aside1 depending on status bits.

* El1SynchronousExceptionHandler now always does tlbi vaae1 instead of doing that or tlbi vmalle1 depending on status bits.

** NOTE: The checks Nintendo did previously have been bugged for many years and did not work.

** These checks were fixed to actually work now.

* UserspaceAccess changes:

** UserspaceAccess functions are now all called through helper functions, which directly call UserspaceAccess::*

*** This is probably "UserspaceAccessChecked", and likely for parity with the supervisor mode access function they added in 19.0.0.

** UserspaceAccess functions which previously had no callers have been deleted.

*** This includes UserspaceAccess::ClearUserMemory*.

* HandleException changes:

** After checking for thread termination, fixup is done on esr_el1/ec values.

*** If esr_ec is 0x20 or 0x24 (DataAbortEl0/InstructionAbortEl0):

**** If esr_el1 & 0x43F == 0x410 then pc is treated as 0

**** Otherwise, if pc is a kernel address, then esr_el1 has the ISFC bits set to hardcoded-value 4.

** This is also done in ReturnFromException

* Kernel::InitializeResourceManagers now initializes the managers in a different order.

* KPageTableBase::FindFreeArea now uses a helper to compute the start and end of the guarded region to generate an address within.

** The same logic is used (without helper, possibly inlined) in KMemoryBlockManager::FindFreeArea.

==See Also==

@@ Line 324: / Line 324: @@
 The anti-downgrade fuses were [[Fuses#Anti-downgrade|updated]].
+==== Kernel ====
+* No/minimal compiler update
+* Inverted boolean(s) in system control/targetsystem logic
+** All KTargetSystem fields have inverted meaning
+** KTargetSystem "is present" bool is now inverted
+** MersenneTwister "is initialized" bool is now inverted
+* KPageTable::ChangeProperties now does a data synchronization barrier after calling the set-attrs lambda
+* KPageTableImpl::MergePages now takes an argument for a callback to call after updating page table PTE entries.
+** All callers on NX are KPageTable::NoteUpdated
+** This causes TLB entries to be flushed/etc every time an entry is written instead of once at the end.
+** The callback is passed the disable-merge software-reserved bits from the entry, but this appears unused on NX.
+*** This is probably for something in the Switch2 kernel/ifdef'd out.
+* KAddressSpaceInfo::GetBegin now takes a size as a third argument; this is unused on NX.
+* KAddressSpaceInfo::GetSize now performs more complicated logic based on the input type.
+** If address space not CreateProcessFlag_AddressSpace32BitWithoutAlias then the size is returned directly.
+** Otherwise:
+*** If the type is Heap, it returns the requested size + the alias size.
+*** If the type is Alias, it returns 0.
+** KProcess::InitializeByParam now sets m_max_process_memory to KAddressSpaceInfo::GetSize(..., Type_Heap) instead of getting the page table's heap size.
+** KPageTableBase::InitializeForProcess has simpler calculation of region extents for 32BitWithoutAlias now.
+* KPageTableBase::MapPageGroup_ now checks if the input permission has the execute bit set, and invalidates instruction cache if so.
+** This fixes a correctness bug; this was used to map pages as ReadExecute by JIT svcs, so stale instruction cache could result before.
+** KPageTableBase::UnmapPageGroup also now invalidates instruction cache after unmapping, if the memory had the execute bit set.
+* KSecureSystemResource now has better checking for the case where resource limit is nullptr.
+** This is impossible on NX.
+** This includes checks in GetUsed/TotalUserPhysicalMemorySize
+** KSecureSystemResource::Destroy now checks before calling ReleaseLimit
+* KPageTableBase::SetProcessMemoryPermission changes:
+** Function now uses PageTableOperation_ChangePermsAndRefresh instead of _ChangePermsAndRefreshAndFlushDataCache when the execute permission is set.
+** Data cache store + instruction cache invalidate is now done before operating instead of after
+** instruction cache invalidate is done after memory block manager is updated.
+* El0SynchronousExceptionHandler now always does tlbi vae1 instead of doing that or tlbi aside1 depending on status bits.
+* El1SynchronousExceptionHandler now always does tlbi vaae1 instead of doing that or tlbi vmalle1 depending on status bits.
+** NOTE: The checks Nintendo did previously have been bugged for many years and did not work.
+** These checks were fixed to actually work now.
+* UserspaceAccess changes:
+** UserspaceAccess functions are now all called through helper functions, which directly call UserspaceAccess::*
+*** This is probably "UserspaceAccessChecked", and likely for parity with the supervisor mode access function they added in 19.0.0.
+** UserspaceAccess functions which previously had no callers have been deleted.
+*** This includes UserspaceAccess::ClearUserMemory*.
+* HandleException changes:
+** After checking for thread termination, fixup is done on esr_el1/ec values.
+*** If esr_ec is 0x20 or 0x24 (DataAbortEl0/InstructionAbortEl0):
+**** If esr_el1 & 0x43F == 0x410 then pc is treated as 0
+**** Otherwise, if pc is a kernel address, then esr_el1 has the ISFC bits set to hardcoded-value 4.
+** This is also done in ReturnFromException
+* Kernel::InitializeResourceManagers now initializes the managers in a different order.
+* KPageTableBase::FindFreeArea now uses a helper to compute the start and end of the guarded region to generate an address within.
+** The same logic is used (without helper, possibly inlined) in KMemoryBlockManager::FindFreeArea.
 ==See Also==