Re: [PATCH v3 RESEND] uapi: Revert "bitops: avoid integer overflow in GENMASK(_ULL)"

[Yury Norov <yury.norov@gmail.com>]

On Wed, Feb 26, 2025 at 02:56:23PM +0800, I Hsin Cheng wrote:
> This patch reverts 'commit c32ee3d9abd2("bitops: avoid integer overflow in
>  GENMASK(_ULL)")'.
> 
> The code generation can be shrink by over 1KB by reverting this commit.
> Originally the commit claimed that clang would emit warnings using the
> implementation at that time.
> 
> The patch was applied and tested against numerous compilers, including
> gcc-13, gcc-12, gcc-11 cross-compiler, clang-17, clang-18 and clang-19.
> Various warning levels were set (-W=0, -W=1, -W=2) and CONFIG_WERROR
> disabled to complete the compilation. The results show that no compilation
> errors or warnings were generated due to the patch.
> 
> The results of code size reduction are summarized in the following table.
> The code size changes for clang are all zero across different versions,
> so they're not listed in the table.
> 
> For NR_CPUS=64 on x86_64.
> ----------------------------------------------
> |	        |   gcc-13 |   gcc-12 |   gcc-11 |
> ----------------------------------------------
> |       old | 22438085 | 22453915 | 22302033 |
> ----------------------------------------------
> |       new | 22436816 | 22452913 | 22300826 |
> ----------------------------------------------
> | new - old |    -1269 |    -1002 |    -1207 |
> ----------------------------------------------
> 
> For NR_CPUS=1024 on x86_64.
> ----------------------------------------------
> |	        |   gcc-13 |   gcc-12 |   gcc-11 |
> ----------------------------------------------
> |       old | 22493682 | 22509812 | 22357661 |
> ----------------------------------------------
> |       new | 22493230 | 22509487 | 22357250 |
> ----------------------------------------------
> | new - old |     -452 |     -325 |     -411 |
> ----------------------------------------------
> 
> For arm64 architecture, gcc cross-compiler was used and QEMU was
> utilized to execute a VM for a CPU-heavy workload to ensure no
> side effects and that functionalities remained correct. The test
> even demonstrated a positive result in terms of code size reduction:
> * Before: 31660668
> * After: 31658724
> * Difference (After - Before): -1944
> 
> An analysis of multiple functions compiled with gcc-13 on x86_64 was
> performed. In summary, the patch elimates one negation in almost
> every use case. However, negative effects may occur in some cases,
> such as the generation of additional "mov" instruction or increased
> register usage. The use of "~_UL(0) << (l)" may even result in the
> allocations of "%r*" registers instead of "%e*" registers (which are
> 32-bit registers) because the compiler cannot assume that the higher
> bits are zero.
> 
> Signed-off-by: I Hsin Cheng <richard120310@gmail.com>

Applied in bitmap-for-next. Thanks for the work!