There are two related issues that appear in certain combinations with
clang and GNU binutils.
The first occurs when a version of clang that supports zicsr or zifencei
via '-march=' [1] (i.e, >= 17.x) is used in combination with a version
of GNU binutils that do not recognize zicsr and zifencei in the
'-march=' value (i.e., < 2.36):
riscv64-linux-gnu-ld: -march=rv64i2p0_m2p0_a2p0_c2p0_zicsr2p0_zifencei2p0: Invalid or unknown z ISA extension: 'zifencei'
riscv64-linux-gnu-ld: failed to merge target specific data of file fs/efivarfs/file.o
riscv64-linux-gnu-ld: -march=rv64i2p0_m2p0_a2p0_c2p0_zicsr2p0_zifencei2p0: Invalid or unknown z ISA extension: 'zifencei'
riscv64-linux-gnu-ld: failed to merge target specific data of file fs/efivarfs/super.o
The second occurs when a version of clang that does not support zicsr or
zifencei via '-march=' (i.e., <= 16.x) is used in combination with a
version of GNU as that defaults to a newer ISA base spec, which requires
specifying zicsr and zifencei in the '-march=' value explicitly (i.e, >=
2.38):
../arch/riscv/kernel/kexec_relocate.S: Assembler messages:
../arch/riscv/kernel/kexec_relocate.S:147: Error: unrecognized opcode `fence.i', extension `zifencei' required
clang-12: error: assembler command failed with exit code 1 (use -v to see invocation)
This is the same issue addressed by commit 6df2a016c0c8 ("riscv: fix
build with binutils 2.38") (see [2] for additional information) but
older versions of clang miss out on it because the cc-option check
fails:
clang-12: error: invalid arch name 'rv64imac_zicsr_zifencei', unsupported standard user-level extension 'zicsr'
clang-12: error: invalid arch name 'rv64imac_zicsr_zifencei', unsupported standard user-level extension 'zicsr'
To resolve the first issue, only attempt to add zicsr and zifencei to
the march string when using the GNU assembler 2.38 or newer, which is
when the default ISA spec was updated, requiring these extensions to be
specified explicitly. LLVM implements an older version of the base
specification for all currently released versions, so these instructions
are available as part of the 'i' extension. If LLVM's implementation is
updated in the future, a CONFIG_AS_IS_LLVM condition can be added to
CONFIG_TOOLCHAIN_NEEDS_EXPLICIT_ZICSR_ZIFENCEI.
To resolve the second issue, use version 2.2 of the base ISA spec when
using an older version of clang that does not support zicsr or zifencei
via '-march=', as that is the spec version most compatible with the one
clang/LLVM implements and avoids the need to specify zicsr and zifencei
explicitly due to still being a part of 'i'.
[1]: 22e199e6af
[2]: https://lore.kernel.org/ZAxT7T9Xy1Fo3d5W@aurel32.net/
Cc: stable@vger.kernel.org
Link: https://github.com/ClangBuiltLinux/linux/issues/1808
Co-developed-by: Conor Dooley <conor.dooley@microchip.com>
Signed-off-by: Conor Dooley <conor.dooley@microchip.com>
Signed-off-by: Nathan Chancellor <nathan@kernel.org>
Acked-by: Conor Dooley <conor.dooley@microchip.com>
Link: https://lore.kernel.org/r/20230313-riscv-zicsr-zifencei-fiasco-v1-1-dd1b7840a551@kernel.org
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
After selecting ARCH_USE_CMPXCHG_LOCKREF, one straight futher
optimization is implementing the arch_cmpxchg64_relaxed() because the
lockref code does not need the cmpxchg to have barrier semantics. At
the same time, implement arch_cmpxchg64_acquire and
arch_cmpxchg64_release as well.
However, on both TH1520 and JH7110 platforms, I didn't see obvious
performance improvement with Linus' test case [1]. IMHO, this may
be related with the fence and lr.d/sc.d hw implementations. In theory,
lr/sc without fence could give performance improvement over lr/sc plus
fence, so add the code here to leave performance improvement room on
newer HW platforms.
Link: http://marc.info/?l=linux-fsdevel&m=137782380714721&w=4 [1]
Signed-off-by: Jisheng Zhang <jszhang@kernel.org>
Select ARCH_USE_CMPXCHG_LOCKREF to enable the cmpxchg-based lockless lockref
implementation for riscv.
Using Linus' test case[1] on TH1520 platform, I see a 11.2% improvement.
On JH7110 platform, I see 12.0% improvement.
Link: http://marc.info/?l=linux-fsdevel&m=137782380714721&w=4 [1]
Signed-off-by: Jisheng Zhang <jszhang@kernel.org>
Currently, riscv linux requires at least IMA, so all platforms have a
multiplier. And I assume the 'mul' efficiency is comparable or better
than a sequence of five or so register-dependent arithmetic
instructions. Select ARCH_HAS_FAST_MULTIPLIER to get slightly nicer
codegen. Refer to commit f9b4192923fa ("[PATCH] bitops: hweight()
speedup") for more details.
In a simple benchmark test calling hweight64() in a loop, it got:
about 14% preformance improvement on JH7110, tested on Milkv Mars.
about 23% performance improvement on TH1520 and SG2042, tested on
Sipeed LPI4A and SG2042 platform.
a slight performance drop on CV1800B, tested on milkv duo. Among all
riscv platforms in my hands, this is the only one which sees a slight
performance drop. It means the 'mul' isn't quick enough. However, the
situation exists on x86 too, for example, P4 doesn't have fast
integer multiplies as said in the above commit, x86 also selects
ARCH_HAS_FAST_MULTIPLIER. So let's select ARCH_HAS_FAST_MULTIPLIER
which can benefit almost riscv platforms.
Signed-off-by: Jisheng Zhang <jszhang@kernel.org>
Reviewed-by: Samuel Holland <samuel.holland@sifive.com>
Tested-by: Samuel Holland <samuel.holland@sifive.com>
