Changing bonnibel to respect the --arch flag to configure. This requires
some reworking of modules, mostly in the addition of the ModuleList
class instead of just a dict of modules.
This TLB shootdown implementation is pretty janky: When the kernel
unmaps a mapping, it sends an IPI to all other cores and doesn't even
wait to ensure they've finished handling it. Upon getting one of these
IPIs, the core just re-writes cr3 to flush all TLBs.
To make debugging easier, add names to processes. These are arbitrary
and supplied by the caller of process_create. The max is 31 characters
in debug configuration, and 0 in release.
Using a 0 address in vma_create_map or vma_map would run into issues if
VMAs had sizes that didn't end on page boundaries. Since any size that's
not a multiples of the page size is a lie, make vm_area's ctor enforce
page sizing.
There has been a random bug (that occurs frequently outside the debugger
but rarely inside the debugger, of course) that seemed to be caused by
inconsistent page mappings. Sometimes loading an ELF would work. Other
times loading that same ELF, the loader would complain of missing
sections or invalid headers. Worse, occasionally program execution would
jump off into random memory for no reason I could see by examining the
disassembly. This issue has been plauging me FOR A YEAR and I've been
pulling my hair out trying to find it.
https://stackoverflow.com/a/28384866
Eventually this stack overflow answer to a different question about
INVLPG gave me a hint that the 'accessed' flag of page table entries
might not be set on pages even if they end up in the TLB.
Good riddance to this damned bug.
This commit adds the 6s shell, and a bunch of supporting work for it.
Major changes include:
- New shell.yaml manifest to give 6s control of the TTY instead of
srv.logger
- Changes to mailbox syscalls to add max handles array size separate
from input size. Also reversed the meaning of the similar data size
argument in those syscalls. (Using the second arg as the max array
size and the first as the current valid size allows for the auto
verify code to verify handles properly, and simplifies user-side
code.)
- New util::unique_ptr smart pointer class similar to std::unique_ptr
- New ipc::message format that uses util::unique_ptr to manage ownership
and lifetimes and avoid extra copying.
- The service locator protocol now supports multiple handles per entry
- Channels got a major overhaul. They are now split into two VMAs, each
containing a mutex, a condition, and a util::bip_buffer. The order of
the VMAs determines which end of the pipe you're on. (ie, the creator
swaps them before handing them to the other thread.) Their API also
changed to be similar to that of util::bip_buffer, to avoid extra
copies.
- util::bip_buffer now keeps its state and its buffer together, so that
there are no pointers. This allows multiple processes to share them in
shared memory, like in channels.
- The UART driver changed from keeping buffers for the serial ports to
just keeping a channel, and the serial port objects read/write
directly from/to the channel.
Known issues:
- The shell doesn't actually do anything yet. It echos its input back to
the serial line and injects a prompt on new lines.
- The shell is one character behind in printing back to the serial line.
The enum_bitfields system never worked quite right, and always had edge cases where name
resolution for the SFINAE would fail. Move everything over to use util::bitset, which can
be constexpr and boils down to inline integer bitops in release mode.
Improved util::bitset itself, moving the array-backed base implementation into a new
util::sized_bitset, and making the single-inttype backed implementation the base case.
Also added a distinction between | or |= (which work with real bit values) and + or +=
(which work with bit indexes).
Added a release config, and fixed a few spots where optimizations broke things:
- Clang was generating incorrect code for run_ctor_list in libc's init.cpp (it
ignored a check for the end of the list)
- my rep movsb memcpy implementation used incorrect inline asm constraints, so
it was returning a pointer to the end of the copied range instead of the start.
Since this function was just inline asm anyway, I rewrote it in asm by hand in
a new memutils.s file.
User code can now set the log area and severity of log messages. This also updates the j6_log
syscall to take these parameters, but removes all calls to it except through j6::syslog().
Added a new IPC log category, and logging of mailbox calls in it. Also added
some logs in init's service locator to help track down the bug fixed in the
previous commit.
Passing a size of 0 in to vma_resize will now not attempt to alter the VMA size, but will
still put the size into the passed-in pointer. Using this allows querying the size of a
VMA without changing it.
While waiting for the main panicing CPU to finish, don't spin the CPU
without using `asm("pause")` to lower power consumption. Some panics can
get stuck in this state and oh man do my fans spin up.
This change allows the `vma_map` and `vma_create_map` syscalls to map to
addresses other than the one specified, and therefore makes the address
parameter to those syscalls `inout` in order to return the mapped
address.
Also add the `exact` flag for specifying that mapping needs to be done
at the exact address given. If the mapping collides with another, the
new `j6_err_collision` error is returned.
Bonnibel will now build dynamic libraries when they're dependencies for
non-statically linked modules. It will also copy those shared libraries
into the initrd image for programs being copied into the image.
Going back to letting mailboxes use variable-length data. Note that this
requires extra copies, so shared memory channels should be used for
anything in the hot path. But this allows better RPC over mailboxes and
other flexibility.
Other changes:
- added a j6::proto::sl::client class to act as a service locator
client, instead of duplicating that code in every program.
- moved protocol ids into j6/tables/protocols.inc so that C++ clients
can easily have their own API
This change allows for parameters in definition files to have the "list"
option while also needing the handle verification. The verify function
will now iterate through the list checking capabilities and types on
every valid handle in the list.
This is the first of two rather big changes to clean up includes
throughout the project. In this commit, the implicit semi-dependency on
libc that bonnibel adds to every module is removed. Previously, I was
sloppy with includes of libc headers and include directory order. Now,
the freestanding headers from libc are split out into libc_free, and an
implicit real dependency is added onto this module, unless `no_libc` is
set to `True`. The full libc needs to be explicitly specified as a
dependency to be used.
Several things needed to change in order to do this:
- Many places use `memset` or `memcpy` that cannot depend on libc. The
kernel has basic implementations of them itself for this reason. Now
those functions are moved into the lower-level `j6/memutils.h`, and
libc merely references them. Other modules are now free to reference
those functions from libj6 instead.
- The kernel's `assert.h` was renamed kassert.h (matching its `kassert`
function) so that the new `util/assert.h` can use `__has_include` to
detect it and make sure the `assert` macro is usable in libutil code.
- Several implementation header files under `__libj6/` also moved under
the new libc_free.
- A new `include_phase` property has been added to modules for Bonnibel,
which can be "normal" (default) or "late" which uses `-idirafter`
instead of `-I` for includes.
- Since `<utility>` and `<new>` are not freestanding, implementations of
`remove_reference`, `forward`, `move`, and `swap` were added to the
`util` namespace to replace those from `std`, and `util/new.h` was
added to declare `operator new` and `operator delete`.
In the heap allocator, new blocks allocated directly for an allocate
request (instead of indirectly as part of a block split) would only set
their order in the tracking map, not their free flag. This left
uninitialized data in the block info map, which thus meant it was marked
as free for looking up for merges. (Not for allocations, since the block
didn't actually appear in the free list.)
Now threads inherit their MXCSR (sans exception state bits) SIMD
settings from their creator. By default, all exceptions are masked, and
both "to zero" flags are set.
Initial support for XSAVE, but not XSAVEOPT or XSAVEC:
- Enable XSAVE and set up xcr0 for all CPUs
- Allocate XSAVE area for all non-kernel threads
- Call XSAVE and XRSTOR on task switch
This commit does a number of things to start the transition of channels
from kernel to user space:
- Remove channel objects / syscalls from the kernel
- Add mutex type in libj6
- Add condition type in libj6
- Add a `ring` type flag for VMA syscalls to create ring buffers
- Implement a rudimentary shared memory channel using all of the above
Add the syscalls j6_futex_wait and j6_futex_wake. Currently marking this
as WIP as they need more testing.
Added to support futexes:
- vm_area and vm_space support for looking up physical address for a
virtual address
- libj6 mutex implementation using futex system calls
There are some SSE instructions (moveaps, moveups) in userland code that
QEMU software emulation seems to be fine with but generate `#UD` on KVM.
So let's finally get floating-point support working. This is the first
step, just setting the control regs to try to fix that error.
This was kept in the kernel as a way to keep exercising the code, but it
doesn't belong there. This moves it to init, which doesn't do anything
but probe for devices currently - but at least it's executing the code
in userspace now.
Previously processes and threads would be deleted by the scheduler. Now,
only delete them based on refcounts - this allows joining an
already-exited thread, for instance.
Previously event tried to read its value in event::wake_observer, which
required jumping through some hoops in how wait_queue was designed, so
that a value wouldn't be wasted if the wait_queue was empty. Now, read
the event value in event::wait after returning from the thread::block
call instead, which simplifies the whole process and lets us simplify
the wait_queue API as well.
For the coming switch to cap/handle ref-counting being the main lifetime
determiner of objects, get rid of self handles for threads and processes
to avoid circular references. Instead, passing 0 to syscalls expecting a
thread or process handle signifies "this process/thread".
There was an inverted boolean logic in determining how many consecutive
pages were available.
Also adding some memory debugging tools I added to track down the recent
memory bugs:
- A direct debugcon::write call, for logging to the debugcon without the
possible page faults with the logger.
- A new vm_space::lock call, to make a page not fillable in memory
debugging mode
- A mode in heap_allocator to always alloc new pages, and lock freed
pages to cause page faults for use-after-free bugs.
- Logging in kobject on creation and deletion
- Page table cache structs are now page-sized for easy pointer math
Yet again burned by the fack that integer literals are assumed to be of
type int, so `1 << n` is 0 for any n >= 32. This burned me in the frame
allocator, but I also grepped for all instances of `1 <<` and fixed
those too.
Add a version of thread::block() that takes a lock and releases it after
marking the thread as unready, but before calling the scheduler.
Use this version of block() in the wait_queue.
