set_thread_area(2) System Calls Manual set_thread_area(2)

get_thread_area, set_thread_area - manipulate thread-local storage information

Standard C library (libc, -lc)

#include <sys/syscall.h>     /* Definition of SYS_* constants */
#include <unistd.h>
#if defined __i386__ || defined __x86_64__
# include <asm/ldt.h>        /* Definition of struct user_desc */
int syscall(SYS_get_thread_area, struct user_desc *u_info);
int syscall(SYS_set_thread_area, struct user_desc *u_info);
#elif defined __m68k__
int syscall(SYS_get_thread_area);
int syscall(SYS_set_thread_area, unsigned long tp);
#elif defined __mips__ || defined __csky__
int syscall(SYS_set_thread_area, unsigned long addr);
#endif

Note: glibc provides no wrappers for these system calls, necessitating the use of syscall(2).

These calls provide architecture-specific support for a thread-local storage implementation. At the moment, set_thread_area() is available on m68k, MIPS, C-SKY, and x86 (both 32-bit and 64-bit variants); get_thread_area() is available on m68k and x86.

On m68k, MIPS and C-SKY, set_thread_area() allows storing an arbitrary pointer (provided in the tp argument on m68k and in the addr argument on MIPS and C-SKY) in the kernel data structure associated with the calling thread; this pointer can later be retrieved using get_thread_area() (see also NOTES for information regarding obtaining the thread pointer on MIPS).

On x86, Linux dedicates three global descriptor table (GDT) entries for thread-local storage. For more information about the GDT, see the Intel Software Developer's Manual or the AMD Architecture Programming Manual.

Both of these system calls take an argument that is a pointer to a structure of the following type:


struct user_desc {
    unsigned int  entry_number;
    unsigned int  base_addr;
    unsigned int  limit;
    unsigned int  seg_32bit:1;
    unsigned int  contents:2;
    unsigned int  read_exec_only:1;
    unsigned int  limit_in_pages:1;
    unsigned int  seg_not_present:1;
    unsigned int  useable:1;
#ifdef __x86_64__
    unsigned int  lm:1;
#endif
};

get_thread_area() reads the GDT entry indicated by u_info->entry_number and fills in the rest of the fields in u_info.

set_thread_area() sets a TLS entry in the GDT.

The TLS array entry set by set_thread_area() corresponds to the value of u_info->entry_number passed in by the user. If this value is in bounds, set_thread_area() writes the TLS descriptor pointed to by u_info into the thread's TLS array.

When set_thread_area() is passed an entry_number of -1, it searches for a free TLS entry. If set_thread_area() finds a free TLS entry, the value of u_info->entry_number is set upon return to show which entry was changed.

A user_desc is considered "empty" if read_exec_only and seg_not_present are set to 1 and all of the other fields are 0. If an "empty" descriptor is passed to set_thread_area(), the corresponding TLS entry will be cleared. See BUGS for additional details.

Since Linux 3.19, set_thread_area() cannot be used to write non-present segments, 16-bit segments, or code segments, although clearing a segment is still acceptable.

On x86, these system calls return 0 on success, and -1 on failure, with errno set to indicate the error.

On C-SKY, MIPS and m68k, set_thread_area() always returns 0. On m68k, get_thread_area() returns the thread area pointer value (previously set via set_thread_area()).

u_info is an invalid pointer.
u_info->entry_number is out of bounds.
get_thread_area() or set_thread_area() was invoked as a 64-bit system call.
(set_thread_area()) A free TLS entry could not be located.

Linux.

Linux 2.5.29.
Linux 2.5.32.

These system calls are generally intended for use only by threading libraries.

arch_prctl(2) can interfere with set_thread_area() on x86. See arch_prctl(2) for more details. This is not normally a problem, as arch_prctl(2) is normally used only by 64-bit programs.

On MIPS, the current value of the thread area pointer can be obtained using the instruction:


rdhwr dest, $29

This instruction traps and is handled by kernel.

On 64-bit kernels before Linux 3.19, one of the padding bits in user_desc, if set, would prevent the descriptor from being considered empty (see modify_ldt(2)). As a result, the only reliable way to clear a TLS entry is to use memset(3) to zero the entire user_desc structure, including padding bits, and then to set the read_exec_only and seg_not_present bits. On Linux 3.19, a user_desc consisting entirely of zeros except for entry_number will also be interpreted as a request to clear a TLS entry, but this behaved differently on older kernels.

Prior to Linux 3.19, the DS and ES segment registers must not reference TLS entries.

arch_prctl(2), modify_ldt(2), ptrace(2) (PTRACE_GET_THREAD_AREA and PTRACE_SET_THREAD_AREA)

2024-05-02 Linux man-pages 6.9.1