| Message ID | alpine.DEB.2.21.9999.1908061929230.19468@viisi.sifive.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | riscv: kbuild: drop CONFIG_RISCV_ISA_C | expand |
On Tue, Aug 06, 2019 at 07:30:24PM -0700, Paul Walmsley wrote: > > The baseline ISA support requirement for the RISC-V Linux kernel > mandates compressed instructions, so it doesn't make sense for > compressed instruction support to be configurable. Looks good, Reviewed-by: Christoph Hellwig <hch@lst.de>
On Wed, Aug 7, 2019 at 10:30 AM Paul Walmsley <paul.walmsley@sifive.com> wrote: > > > The baseline ISA support requirement for the RISC-V Linux kernel > mandates compressed instructions, so it doesn't make sense for > compressed instruction support to be configurable. > > Signed-off-by: Paul Walmsley <paul.walmsley@sifive.com> > Cc: Atish Patra <atish.patra@wdc.com> > > --- > arch/riscv/Kconfig | 10 ---------- > arch/riscv/Makefile | 2 +- > 2 files changed, 1 insertion(+), 11 deletions(-) > Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Please do not drop it. Compressed instruction extension has some specific overhead in small RISC-V FPGA softcore, especialy in the ones which can't implement the register file read in a asynchronous manner because of the FPGA technology. What are reasons to enforce RVC ? On Wed, Aug 7, 2019 at 2:29 PM Bin Meng <bmeng.cn@gmail.com> wrote: > > On Wed, Aug 7, 2019 at 10:30 AM Paul Walmsley <paul.walmsley@sifive.com> wrote: > > > > > > The baseline ISA support requirement for the RISC-V Linux kernel > > mandates compressed instructions, so it doesn't make sense for > > compressed instruction support to be configurable. > > > > Signed-off-by: Paul Walmsley <paul.walmsley@sifive.com> > > Cc: Atish Patra <atish.patra@wdc.com> > > > > --- > > arch/riscv/Kconfig | 10 ---------- > > arch/riscv/Makefile | 2 +- > > 2 files changed, 1 insertion(+), 11 deletions(-) > > > > Reviewed-by: Bin Meng <bmeng.cn@gmail.com> > > _______________________________________________ > linux-riscv mailing list > linux-riscv@lists.infradead.org > http://lists.infradead.org/mailman/listinfo/linux-riscv
On Thu, Aug 08, 2019 at 02:18:53PM +0200, Charles Papon wrote: > Please do not drop it. > > Compressed instruction extension has some specific overhead in small > RISC-V FPGA softcore, especialy in the ones which can't implement the > register file read in a asynchronous manner because of the FPGA > technology. > What are reasons to enforce RVC ? Because it it the unix platform baseline as stated in the patch.
On 8/12/19 11:03 AM, Christoph Hellwig wrote: > On Thu, Aug 08, 2019 at 02:18:53PM +0200, Charles Papon wrote: >> Please do not drop it. >> >> Compressed instruction extension has some specific overhead in small >> RISC-V FPGA softcore, especialy in the ones which can't implement the >> register file read in a asynchronous manner because of the FPGA >> technology. >> What are reasons to enforce RVC ? > > Because it it the unix platform baseline as stated in the patch. > The same argument could be made for an FPU or MMU, yet there are options to disable those.
> Because it it the unix platform baseline as stated in the patch. I know that, but i'm looking for arguments why RVC could't be kept as an option, especialy it is only an optimisation option without behavioral/code changes. That baseline make sense for heavy linux distributions, where you expect everybody to compile with a baseline set of ISA extentions, to make binary exchanges easier. But for smaller systems, i do not see advantages having RVC forced. On Mon, Aug 12, 2019 at 5:03 PM Christoph Hellwig <hch@infradead.org> wrote: > > On Thu, Aug 08, 2019 at 02:18:53PM +0200, Charles Papon wrote: > > Please do not drop it. > > > > Compressed instruction extension has some specific overhead in small > > RISC-V FPGA softcore, especialy in the ones which can't implement the > > register file read in a asynchronous manner because of the FPGA > > technology. > > What are reasons to enforce RVC ? > > Because it it the unix platform baseline as stated in the patch.
On Tue, Aug 13, 2019 at 12:18:22AM +0200, Charles Papon wrote: > > Because it it the unix platform baseline as stated in the patch. > I know that, but i'm looking for arguments why RVC could't be kept as > an option, especialy it is only an optimisation option without > behavioral/code changes. > > That baseline make sense for heavy linux distributions, where you > expect everybody to compile with a baseline set of ISA extentions, to > make binary exchanges easier. > But for smaller systems, i do not see advantages having RVC forced. I don't fully agree with the benefits, but then again how little impact using the C extension has on the kernel build I'm now convinced that keeping it should be ok.
Hi Charles, On Tue, 13 Aug 2019, Charles Papon wrote: > > Because it it the unix platform baseline as stated in the patch. > I know that, but i'm looking for arguments why RVC could't be kept as > an option, especialy it is only an optimisation option without > behavioral/code changes. > > That baseline make sense for heavy linux distributions, where you > expect everybody to compile with a baseline set of ISA extentions, to > make binary exchanges easier. > But for smaller systems, i do not see advantages having RVC forced. OK - I agree with you. Still, I think it would be good if we made this option depend on other more general kernel configuration parameters for smaller systems. Will think about this further. Thanks for commenting on this, and am looking forward to adding a VexRiscv system to our kernel tests - - Paul
diff --git a/arch/riscv/Kconfig b/arch/riscv/Kconfig index 59a4727ecd6c..8c5b9329ec46 100644 --- a/arch/riscv/Kconfig +++ b/arch/riscv/Kconfig @@ -197,16 +197,6 @@ config TUNE_GENERIC endchoice -config RISCV_ISA_C - bool "Emit compressed instructions when building Linux" - default y - help - Adds "C" to the ISA subsets that the toolchain is allowed to emit - when building Linux, which results in compressed instructions in the - Linux binary. - - If you don't know what to do here, say Y. - menu "supported PMU type" depends on PERF_EVENTS diff --git a/arch/riscv/Makefile b/arch/riscv/Makefile index 7a117be8297c..e23e066c55e2 100644 --- a/arch/riscv/Makefile +++ b/arch/riscv/Makefile @@ -38,7 +38,7 @@ endif riscv-march-$(CONFIG_ARCH_RV32I) := rv32ima riscv-march-$(CONFIG_ARCH_RV64I) := rv64ima riscv-march-$(CONFIG_FPU) := $(riscv-march-y)fd -riscv-march-$(CONFIG_RISCV_ISA_C) := $(riscv-march-y)c +riscv-march-y := $(riscv-march-y)c KBUILD_CFLAGS += -march=$(subst fd,,$(riscv-march-y)) KBUILD_AFLAGS += -march=$(riscv-march-y)
The baseline ISA support requirement for the RISC-V Linux kernel mandates compressed instructions, so it doesn't make sense for compressed instruction support to be configurable. Signed-off-by: Paul Walmsley <paul.walmsley@sifive.com> Cc: Atish Patra <atish.patra@wdc.com> --- arch/riscv/Kconfig | 10 ---------- arch/riscv/Makefile | 2 +- 2 files changed, 1 insertion(+), 11 deletions(-)