I noticed a pretty significant delay (of ~5s on my machine) after
executing `make`.
This was because TARGET was currently defined as a recursively expanded
variable. This means that the command:
rustc -Z unstable-options --print target-spec-json | grep llvm-target | cut -d '"' -f4
..will get run everytime TARGET is mentioned. This patch fixes the above
by defining the variable as an immediately expanded value. This will
ensure the command only get run once and still skips if predefined.
Signed-off-by: Anhad Singh <andypython@protonmail.com>
In the next big refactor (next PR), all of the platform functionality
used by both relibc and ld.so will be moved into a `platform`/`sysdeps`
crate and then ld.so would be moved out of relibc and not link with it.
I think doing it in a seperate PR would make it more managable, as when
I did half of it, the diff was pretty huge and that way it would be
easier to review too :)
Signed-off-by: Anhad Singh <andypython@protonmail.com>
Fix CI break.
man pthread_barrier_wait(3)
> ...the constant PTHREAD_BARRIER_SERIAL_THREAD shall be returned to
> one *unspecified* thread and zero shall be returned to each of the remaining
> threads.
Signed-off-by: Anhad Singh <andypython@protonmail.com>
This header is more or less the same across libc implementations.
`musl` uses its `_Noreturn` macro to detect which `noreturn`
should be used: C11 or GCC's extension as a fallback.
`glibc` simply defines `noreturn` as `_Noreturn`.
This implementation is based off of `musl`'s.
`_Noreturn` is deprecated as of C23.
Memory chunk is allocated with `malloc` and used as the `dest` buffer
for `strncat`. The `dest` argument in `strncat` has to be NUL terminated,
however it was not.
This commit fixes this issue in mk{fifo,nod,noat}.c.
Almost all dynamic tests pass now.
Signed-off-by: Anhad Singh <andypython@protonmail.com>
Closes: #200
The spec doesn't define which error code to set if `len` is too large.
`ENOMEM` is the closest defined condition that fits:
>> [...] the range [addr,addr+len) exceeds that allowed for the address
space of a process [...]
Logically, `len` would lead to `addr` exceeding the address space of a
process if rounding it up to the next page size causes an overflow.
From man dlopen(3)
> According to the ISO C standard, casting between function
> pointers and 'void *', as done above, produces undefined results.
> POSIX.1-2001 and POSIX.1-2008 accepted this state of affairs and
> proposed the following workaround:
>
> *(void **) (&cosine) = dlsym(handle, "cos");
>
> This (clumsy) cast conforms with the ISO C standard and will
> avoid any compiler warnings.
>
> The 2013 Technical Corrigendum 1 to POSIX.1-2008 improved matters
> by requiring that conforming implementations support casting
> 'void *' to a function pointer. Nevertheless, some compilers
> (e.g., gcc with the '-pedantic' option) may complain about the
> cast used in this program.
Signed-off-by: Anhad Singh <andypython@protonmail.com>
The bug is described for `x86_64` and Linux but is the same for other architectures and on Redox.
`Tcb::current()` is used to retrieve the current TCB, which is done by by reading `fs:[0x10]`. The TCB layout describes that at offset `0x10`, there is `tcb_ptr: *mut GenericTcb<...>`, which is nothing more but a pointer to itself.
This is fine as otherwise a system call would be required to get the TCB (`arch_prctl(ARCH_GET_FS)` on Linux).
However, this is problematic as the function may be called when the FS base is not set, and in that case the expected output of the function should be [`None`], but we don't currently handle that.
To fix this, any code paths that maybe call this function on an uninitialized TCB are be switched to call `current_slow()`. Which just uses `arch_prctl(ARCH_GET_FS)` to get the FS base and check if it's non-zero.
Signed-off-by: Anhad Singh <andypython@protonmail.com>
Anhad Singh