Linux On IBM Z "s390" To See ~11% Higher Syscall Entry Performance On Linux 6.8
For those interested in Linux on IBM Z / s390, there's a small change yielding measurable benefits to the s390 system call entry performance with the forthcoming Linux 6.8 kernel.
The change moves various syscall entry functions from a C file to the kernel's respective common entry header file and is set to always-inline them. With s390 being one of the architectures that uses the generic entry code, this always-inlining of the syscall enter/exit functions is providing a significant boost to the performance.
Sven Schnelle of IBM's Linux team noted that by always inlining the generic syscall enter/exit functions, with a small test case he's found that the system call time dropped by around 11%. A very nice improvement to the syscall performance by just shifting around about 130 lines of code for at least s390.
The three patches moving around the syscall enter/exit to user-mode functions were merged as part of the generic syscall entry code changes in Linux 6.8 Git as of last week.
Quite nice seeing a number of small but significant performance optimizations hitting the Linux kernel recently in different areas.
The change moves various syscall entry functions from a C file to the kernel's respective common entry header file and is set to always-inline them. With s390 being one of the architectures that uses the generic entry code, this always-inlining of the syscall enter/exit functions is providing a significant boost to the performance.
Sven Schnelle of IBM's Linux team noted that by always inlining the generic syscall enter/exit functions, with a small test case he's found that the system call time dropped by around 11%. A very nice improvement to the syscall performance by just shifting around about 130 lines of code for at least s390.
The three patches moving around the syscall enter/exit to user-mode functions were merged as part of the generic syscall entry code changes in Linux 6.8 Git as of last week.
Quite nice seeing a number of small but significant performance optimizations hitting the Linux kernel recently in different areas.
21 Comments