Hardware tests show that trying to unmap an unmapped buffer already should always succeed. Hardware test was tested up to 32 iterations of attempting to unmap
When a destructor isn't defaulted into a cpp file, it can cause the use
of forward declarations to seemingly fail to compile for non-obvious
reasons. It also allows inlining of the construction/destruction logic
all over the place where a constructor or destructor is invoked, which
can lead to code bloat. This isn't so much a worry here, given the
services won't be created and destroyed frequently.
The cause of the above mentioned non-obvious errors can be demonstrated
as follows:
------- Demonstrative example, if you know how the described error happens, skip forwards -------
Assume we have the following in the header, which we'll call "thing.h":
\#include <memory>
// Forward declaration. For example purposes, assume the definition
// of Object is in some header named "object.h"
class Object;
class Thing {
public:
// assume no constructors or destructors are specified here,
// or the constructors/destructors are defined as:
//
// Thing() = default;
// ~Thing() = default;
//
// ... Some interface member functions would be defined here
private:
std::shared_ptr<Object> obj;
};
If this header is included in a cpp file, (which we'll call "main.cpp"),
this will result in a compilation error, because even though no
destructor is specified, the destructor will still need to be generated by
the compiler because std::shared_ptr's destructor is *not* trivial (in
other words, it does something other than nothing), as std::shared_ptr's
destructor needs to do two things:
1. Decrement the shared reference count of the object being pointed to,
and if the reference count decrements to zero,
2. Free the Object instance's memory (aka deallocate the memory it's
pointing to).
And so the compiler generates the code for the destructor doing this inside main.cpp.
Now, keep in mind, the Object forward declaration is not a complete type. All it
does is tell the compiler "a type named Object exists" and allows us to
use the name in certain situations to avoid a header dependency. So the
compiler needs to generate destruction code for Object, but the compiler
doesn't know *how* to destruct it. A forward declaration doesn't tell
the compiler anything about Object's constructor or destructor. So, the
compiler will issue an error in this case because it's undefined
behavior to try and deallocate (or construct) an incomplete type and
std::shared_ptr and std::unique_ptr make sure this isn't the case
internally.
Now, if we had defaulted the destructor in "thing.cpp", where we also
include "object.h", this would never be an issue, as the destructor
would only have its code generated in one place, and it would be in a
place where the full class definition of Object would be visible to the
compiler.
---------------------- End example ----------------------------
Given these service classes are more than certainly going to change in
the future, this defaults the constructors and destructors into the
relevant cpp files to make the construction and destruction of all of
the services consistent and unlikely to run into cases where forward
declarations are indirectly causing compilation errors. It also has the
plus of avoiding the need to rebuild several services if destruction
logic changes, since it would only be necessary to recompile the single
cpp file.
core.h is kind of a massive header in terms what it includes within
itself. It includes VFS utilities, kernel headers, file_sys header,
ARM-related headers, etc. This means that changing anything in the
headers included by core.h essentially requires you to rebuild almost
all of core.
Instead, we can modify the System class to use the PImpl idiom, which
allows us to move all of those headers to the cpp file and forward
declare the bulk of the types that would otherwise be included, reducing
compile times. This change specifically only performs the PImpl portion.
As means to pave the way for getting rid of global state within core,
This eliminates kernel global state by removing all globals. Instead
this introduces a KernelCore class which acts as a kernel instance. This
instance lives in the System class, which keeps its lifetime contained
to the lifetime of the System class.
This also forces the kernel types to actually interact with the main
kernel instance itself instead of having transient kernel state placed
all over several translation units, keeping everything together. It also
has a nice consequence of making dependencies much more explicit.
This also makes our initialization a tad bit more correct. Previously we
were creating a kernel process before the actual kernel was initialized,
which doesn't really make much sense.
The KernelCore class itself follows the PImpl idiom, which allows
keeping all the implementation details sealed away from everything else,
which forces the use of the exposed API and allows us to avoid any
unnecessary inclusions within the main kernel header.
Makes the class interface consistent and provides accessors for
obtaining a reference to the memory manager instance.
Given we also return references, this makes our more flimsy uses of
const apparent, given const doesn't propagate through pointers in the
way one would typically expect. This makes our mutable state more
apparent in some places.
All calling code assumes that the rasterizer will be in a valid state,
which is a totally fine assumption. The only way the rasterizer wouldn't
be is if initialization is done incorrectly or fails, which is checked
against in System::Init().
We move the initialization of the renderer to the core class, while
keeping the creation of it and any other specifics in video_core. This
way we can ensure that the renderer is initialized and doesn't give
unfettered access to the renderer. This also makes dependencies on types
more explicit.
For example, the GPU class doesn't need to depend on the
existence of a renderer, it only needs to care about whether or not it
has a rasterizer, but since it was accessing the global variable, it was
also making the renderer a part of its dependency chain. By adjusting
the interface, we can get rid of this dependency.
Avoids copies from being made, since the string is only ever used for
lookup, the data is never transfered anywhere.
Ideally, we'd use a std::string_view here, but devices is a
std::unordered_map, not a std::map, so we can't use heterogenous lookup
here.
Avoids unnecessary reference count increments and decrements.
In one case, we don't need to make a shared_ptr copy at all,
just to call a member function.
* add IoctlCommands with their params in nvidia_ctrl_gpu.h
* add function related to the changes done previously
* fix clang-format
* delete trailing whitespace
* correct mistake
* Add some IoctlCommand with their params to nvhost_gpu
* fix clang-format
* delete trailing whitespace
* fix some clang-format
* delete one other trailing whitespace
* last clang-format fix
* Implemented NVHOST_IOCTL_CHANNEL_GET_WAITBASE
struct + 4 seems to be hard coded at 0 and struct + 0 seems to be ignored?
* IocGetWaitbase -> IocChannelGetWaitbaseCommand
* Added super late fixes
This makes the formatting expectations more obvious (e.g. any zero padding specified
is padding that's entirely dedicated to the value being printed, not any pretty-printing
that also gets tacked on).
Also added a consistency check and a comment for the case when the object id is different than its handle. The real nvservices doesn't make a distinction between ids and handles, each object gets an unique handle which doubles as its id.
* FinishInitalize needed for 3.0.1+ games
* nvdrv:s and nvdrv:t both use NVDRV
* Most settings return 0 on hardware, disabled NV_MEMORY_PROFILER for now.
NVN_THROUGH_OPENGL & NVRM_GPU_PREVENT_USE are a few interesting settings to look at. Carefully choosing settings can help with drawing graphics later on
* Initial /dev/nvhost-gpu support
* ZCullBind
* Stubbed SetErrorNotifier
* Fixed SetErrorNotifier log, Added SetChannelPriority
* Allocate GPFIFO Ex2, Allocate Obj Ctx, Submit GPFIFO
* oops
* Fixed up naming/structs/enums. Used vector instead of array for "gpfifo_entry"
* Added missing fixes
* /dev/nvhost-ctrl-gpu
* unneeded struct
* Forgot u32 in enum class
* Automatic descriptor swapping for ioctls, fixed nvgpu_gpu_get_tpc_masks_args being incorrect size
* nvdrv#QueryEvent
* Renamed logs for nvdrv
* Refactor ioctl so nv_result isn't needed
* /dev/nvhost-as-gpu
* Fixed Log service naming, CtxObjects now u32, renamed all structs, added static_asserts to structs, used INSERT_PADDING_WORDS instead of u32s
* nvdevices now uses "Ioctl" union,
* IoctlGpfifoEntry now uses bit field
* final changes
* Added nvmemp, Added /dev/nvhost-ctrl, SetClientPID now stores pid
* used clang-format-3.9 instead
* lowercase pid
* Moved nvmemp handlers to cpp
* Removed unnecessary logging for NvOsGetConfigU32. Cleaned up log and changed to LOG_DEBUG
* using std::arrays instead of c arrays
* nvhost get config now uses std::array completely
* added pid logging back
* updated cmakelist
* missing includes
* added array, removed memcpy
* clang-format6.0
NVFlinger will call into the nvdisp_disp0 device to perform screen flips, bypassing the ioctl interface.
We now have the address of the framebuffer to draw, we just need to actually put it on the screen.
