* fixes broken compilation after merging

* basic implementation for pure pascal programs to allocate tls space

* disable generic threadvar handling if section thread vars are available

Fix arm-android compilation by prepending syscall_nr_ prefix also to __ARM_NR_XXX arm specific linux syscalls

* properly initialize phdr and phnum

* do not call InitTLS when a compiler is used which does not include the abi tag section

Improve syscall number testing script:
+ Add support for parameters:
-v sets verbose variable to 1
VAR=value gets evaluated.
* moves to rtl/$os if called at diferent level.
+ check if there is a incude directive in sysnr.inc file.
+ Add compilation of C source code with the C variant of the syscall number.
+ Support 32 and 64-bit version of C compilation for systems
allowing 32 and 64-bit executable (like x86_64-linux).
+ Add 'preprocessing' using AWK to test {$ifdef CPUXXX} macro.
+ Add summary at end of script.
* Rename check_syscall_number to check_c_syscall_from_fpc_rtl
* Rename check_syscall_number_reverse to check_c_syscall_in_fpc_rtl
* Try to remove most intermediate files.
+ Generate add_missing_syscalls.inc file if numbers are missing.

Eliminate wrong line (bug report 35441)

* freebsd and openbsd also use sys/syscall.h header
+ Add AWK parser to removed parts of source not concerning current CPU
(only handles CPUXXX like CPUI386)

More improvements to script testing syscall numbers

* Also allow '-' in include file names
+ Add os_cpu and cpu_source variables
* Cope with aarch64 GNU C compiler that does not support -m64 option
+ Add support for arm test on aarch64 machine
* Better accept expressions instead of constants in C #define

Add script to use h2paschk on t_OS.h2paschk file for rtl tests for system header compliance

Add first linux RTL compliance test file t_linux.h2paschk

Also remove -m32 option for arm native GNU C compiler

* Improve handling of command line args setting variables (did not work if it contained spaces)
+ Group all recognized macros into list_defines in preproc.awk file
+ Convert Pascal style hexadecimal constants into C style

Disable 64-bit version of CPU_SOURCE is 32-bit

* properly return from the signal handler on ARM/Linux instead of calling
HandleErrorAddrFrame directly (fixes psabieh exception handling, and
generally is cleaner)

Fix compilation with release 3.0.4 ppcarm

sparc32 bit needs SA_SIGINFO

* cosmetics: indention

* improved software floating point exception handling in the rtl

prepare source to also support riscv32 code

Replace obsolete scall by ecall instruction

Add missing and needed syscall_nr_XXX

Add si_XXX.inc files, pointing to riscv64 counterparts

Also use syscall_nr_fstatat for FpStat function for riscv32 cpu

+ Enable FPU exception handling for m68k processor

* get paramstr(0) only if really needed

Examine all exception bits in m68k fpsr register to decide which into which run time error it is converted

* Avoid doing a syscall with a nil parameter in fileexists

+ xtensa-linux: OS specific system unit header files

Use same field type and name as in /ust/include/sys/stat.h

* Use off_t type for l_start and l_end fields instead of kernel_off_t
+ Distinguish rlim_t type depending on fs32bit macro being set or not.

* fix for Mantis 0036895: the new epoll_pwait was missing the size parameter for the sigmask parameter

* fix O_LARGEFILE for Aarch64, RISC-V 32-bit and RISC-V 64-bit

F_GetLk/SetLk have different values for sparc/sparc64

* rename local import of fpc_geteipasebx

* fpc_geteipasebx and fpc_geteipasecx are used as procedures, not functions, so declare them as such

* linux-m68k: depend on FPU defines to compile FPU code, instead of CPU68020. also, minor cosmetics

* linux: fix some formatting in ostypes.inc, no functional change

* linux: use ugetrlimit syscall on m68k for rlimits retrieval. this fixes fpgetrlimit() call, and stack checking on Linux/m68k (tstack.pp test), at least on anything newer than kernel 2.4+ (2001)

* Do_IsDevice treats pipes and sockets the same way as character devices

Do not set __USE_LARGEFILE64 macro for sparc CPU, as it does not seem to be used for regular linux kernels

Add handling of extra parameters, using eval if equal sign is found.
+ Add TMP_DIR variable and use it.
* Only remove temp files if verbose is not set.

Fix typo in script comment

Unix LocalToEpoch: fix typo

* remove the BaseUnix dependency added by mistake

Add some types and records to rtl/linux/t_linux.h2paschk and set macro _LARGEFILE64_SOURCE for __i386__ C code

Correct computation of operatingsystem_parameter_envp, which wrongly pointed to the NIL pointer separating argv from environment variables

* remove unused file

* cwstring: try to load the iconv library with .6 suffix as well

* cwstring: open iconv with the transliterate flag instead of setting it later with iconvctl because it may be unavailable

* revert 0047631: cwstring: open iconv with the transliterate flag instead of setting it later with iconvctl because it may be unavailable

* cwstring: open iconv with the transliterate flag if iconvctl is not available

+ linux implementation of System.GetCPUCount

* if the Linux target is compiled with FPC_USE_LIBC the initialization of TLS is not required (according to one comment by Florian), so disable all related code in that case (as for example the syscall numbers are not available then)

+ add declaration of FpSchedGetAffinity when the Linux target is compiled with FPC_USE_LIBC

Also set __stkptr inside cprt0.as startup assembly source

Also set __stkptr inside cprt0.as and dllprt0.as startup assembly sources

* x86_64-linux: assembler helpers syscalls do not need a stackframe

* FPC_VERSION => FPC_FULLVERSION

* another FPC_VERSION => FPC_FULLVERSIOn fix
* switch back to 32 Bit mode if no thumb code is generated

* arm-linux: sighnd.inc must check for CPUTHUMB and CPUTHUMB2

* fix compilation for CPUs without thumb

o RiscV64 glibc startup code fixed:
* there is no _init/_fini, call libc_csu_* instead
* fix loading of global_pointer

+ RiscV: set threadvarblocksize
* sanity check if threadvarblocksize is really set

* mips and xtensa compilation fixed

* RiscV: corrected setup of gp

+ RiscV: initial implementation of gprof support
* cleanup

* fix for Mantis 0037252: apply patch by Bi0T1N to implement TThread.NameThreadForDebugging for Mac OS X 10.6 and newer (older versions don't provide the required functions)

Fix compilation with 3.2.2 release for arm and i386 CPUs

+ added baud rate constants for arm as proposed by Piotr Kardasz in 0039101

* indention fixed

* check paramstr argument for negative values, resolves gitlab 0039410

LLVM: add Read/WritePort* implementations

Fixes Unix graph compilation on x86 with LLVM

* clean up of assembler files not used anymore

* powerpc64-linux: unused assembler loader files removed

+ additional skeleton files for mips64

Disable range and overflow checks inside detect_linuxvcs procedure

* make __lib_csu* weak symbols on m68k-linux

Darwin: use recent versions of stat(fs) records when available

When the RTL is built for Mac OS X 10.6 or newer, use the new versions
and the corresponding alternate system library functions.

Also extracted the Darwin stat record definition to reduce the
ifdef hell.

Darwin: fix compilation on macOS/AArch64 after 1f328cf3

No function suffixes there since there's no backward compatibility issue.

m68k-linux: try to be as ColdFire compatible in the startup code as possible, so explicitly use jsr, otherwise this might compile to a long bsr, that is not CF compatible (except cfv4)

FPU exception mask: generlised system unit interface

* Extra linux fcntl constants

Linux: fix pthread types

Remove legacy implementation-specific fields (they don't match the current
glibc fields anymore since a long time) and fix alignment fields (they are
union fields that overlap with the rest of the record).

All platforms now use the modern opaque array-based definitions.

Linux mutex: restored the __m_kind field

This particular field is at the same offset on all architectures with the same
bit width, and is guaranteed by glibc to remain at this offset for backward
compatibility with static initialisers.

Resolves gitlab 0040021

* Char -> AnsiChar

* Char -> AnsiChar

* string -> shortstring

* string -> shortstring

* string -> shortstring

* string -> shortstring

* String -> ansistring

* Remove use of internal conversion routines

* Some forgotten string -> shortstring instances

* cleanup of outdated defines

* fix of Erroneous reading of the value of the StackLength variable at program start based on issue report by Sergey Larin, resolves gitlab 0040211

Try to implement linux si_uc.inc for xtensa CPU

Attempt to use local _init and _fini procedures

Support use statx syscall in fpstat and fpfstat

fix when FPC_USE_LIBC is disabled not call PASCALMAIN

fix store stacklength but not its addr in info

Use prlimit64 in FpSetRLimit and FpSetRLimit

Set STACK_MARGIN_MAX to four times of pagesize on LoongArch64

It fixed a test named tstack. Actually, Linux user get the stacktop is
incorrect in current. The value of stack pointer when enter the procedure
is thought as it is in the highest stack page. But if the parameters is
enough large, lager than 16k, most of arch will also fail this test.
Since the LoongArch default pagesize is 16k, it fail often because
stack-layout-randomization causes the stackpointer underflow to next page.
I had tried fixed it in rtl/linux/[arch]/si_*. But because the codes is
unworthy and hardly the parameters is enough large, just expand
STACK_MARGIN_MAX for simplicity.

* Fix for i386 breakage after merge request !433

* Additional fix

Restore usage of fpexecve function inside ExecuteProcess
as set by commit 305d34a8f6f

Adapt USE_STATX code to FPC_DOTTEDUNITS

* Small improvement in inittls. Fixes issue 40408

Use frame register $r22 (aliased as $fp) inside SignalToRunerror for loongarch64 CPU to fix backtrace dumps

Modify references to old syslinux unit in comments

* Ensure null-terminated strings. Patch by modersohn, fixes issue gitlab 0040499

* Add pthread_mutex_timedlock, sem_timedwait may be needed in future

* GetEpochTime (not used though it seems) should return time_t

* ensure that the SysCall unit is also usable when FPC_USE_LIBC is defined

Fix compilation of syscall unit for Linux with -dFPC_USE_LIBC option

Fix syscall unit compilation for i386-linux using -dFPC_USE_LIBC

+ support for fputimens on non-linux OSes
* reworked fputimens support on linux
+ FileSetDate for all unix OSes

Avoid double declaration of seterrno

* fix bootstrapping with with 3.2.2 and -dFPC_USE_LIBC

* fix utimensat and utimensat if the rtl is compiled with -dFPC_USE_LIBC

* use TTimespecArr instead of tkernel_timespecs in sysutils

* cleaned up TTimespecArr declarations

* fix darwin compilation after TTimespecArr changes

* compilation on linux with FPC_USE_LIBC fixed

Use call to seterrno function inside FpSysCall functions for powerpc cpu on linux OS

* fix compilation for mipsel-android (hopefully)

* defines fixed

* Prevent file handle leak

* Set monitor structures in place

Re-implement (for debugging purposes) buildrtl.pp unit for Linux OS (not used in regular RTL compilation)

* Allow to disable monitor

Fix reading past the end of the string in Unix.FSearch.

Add CC_OPT for riscv processors

* Fix compilation under unicode rtl

Use '__global_pointer$' special linker symbol to set gp,
because its value can be different from __BSS_END__ - 0x800.

Details from binutils-2.40/ld/emulparams/elf32lriscv-defs.sh
// We must cover as much of sdata as possible if it exists. If sdata+bss is
// smaller than 0x1000 then we should start from bss end to cover as much of
// the program as possible. But we can't allow gp to cover any of rodata, as
// the address of variables in rodata may change during relaxation, so we start
// from data in that case.
OTHER_END_SYMBOLS="${CREATE_SHLIB-__BSS_END__ = .;
__global_pointer$ = MIN(__SDATA_BEGIN__ + 0x800,
MAX(__DATA_BEGIN__ + 0x800, __BSS_END__ - 0x800));}"

* upated syscalls
+ RiscV specific syscalls added

+ support for RiscV hwprobe syscall
+ test

Fix compilation of riscvXX-linux targets with -dFPC_USE_LIBC

Riscv64 linux also needs _STAT_VER_LINUX to be zero for correct libc rtl behavior

Fix _FPC_xtensa_exit first parameter passing

* fix rlim_t for i386 and other 32 bit CPUs supporting 64 bit fs calls

* Patch for inotify_event struct. Fixes issue gitlab 0040860

Avoid range check error in FindGetFileInfo function on AIX OS and change type of local variable LinAttr to match argument type of fpS_ISDIR function

+ constants for more RiscV extensions added

* Add intrinsics unit for ARM platforms

Replace unused AARCH macro by AARCH64

Only add intrinsics unit dependency for trunk branch

Add intrinsics unit dependency for powerpc and powerpc64 CPUs

files update

* Correct some fixes of typos

Use '$if declared(fpfutimens)' to set USE_FUTIMES macro

* Faster (and correct) TThread.GetSystemTimes. Fixes issue gitlab 0040597