arch/Config.in.riscv: add Zicsr and Zifencei standalone extensions
PATCH 1/11 arch/Config.in.riscv: add Zicsr and Zifencei standalone extensions
Since gcc 12, default Riscv ISA spec version was bump to 20191213 [1].
This introduce a major incompatibility issue is the csr read/write (csrr*/csrw*) instructions and fence.i instruction has separated from the "I" extension, become two standalone extensions: Zicsr and Zifencei; so you might get error messages like that: unrecognized opcode "csrr" (or "fence.i").
Indeed, without Zifencei we can't build opensbi bootloader [3]:
opensbi-1.0/lib/sbi/sbi_tlb.c: Assembler messages:opensbi-1.0/lib/sbi/sbi_tlb.c:190: Error: unrecognized opcode fence.i', extension zifencei' required
As a workaround, opensbi build system has been patched [4] to use -march=rv64imafdc_zicsr_zifencei when needed.
This workaround doesn't work in Buildroot due to the local patch 0001-Makefile-Don-t-specify-mabi-or-march.patch removing -march from CFLAGS.
Fix this issue by introducing two additional Kconfig option enabling Zicsr and Zifencei standalone extensions for gcc >= 12 as recommanded by [2].
Select Zicsr and Zifencei for General purpose (G) architecture variant (BR2_riscv_g) since theses extentions were implicitely enabled previously.
[1] https://gcc.gnu.org/gcc-12/changes.html
[2] https://groups.google.com/a/groups.riscv.org/g/sw-dev/c/aE1ZeHHCYf4
[3] https://github.com/riscv-software-src/opensbi/blob/v0.9/lib/sbi/sbi_tlb.c#L173
[4] https://github.com/riscv-software-src/opensbi/commit/5d53b55aa77ffeefd4012445dfa6ad3535e1ff2c
--- arch/Config.in.riscv | 21 ++++++++++++++++++++- arch/arch.mk.riscv | 6 ++++++ 2 files changed, 26 insertions(+), 1 deletion(-)
arch/Config.in.riscv +20 -1
@@ -18,6 +18,12 @@ config BR2_RISCV_ISA_RVD config BR2_RISCV_ISA_RVC bool +config BR2_RISCV_ISA_RVZicsr + bool + +config BR2_RISCV_ISA_RVZifencei + bool + choice prompt "Target Architecture Variant" default BR2_riscv_g
@@ -29,8 +35,11 @@ config BR2_riscv_g select BR2_RISCV_ISA_RVA select BR2_RISCV_ISA_RVF select BR2_RISCV_ISA_RVD + select BR2_RISCV_ISA_RVZicsr if BR2_TOOLCHAIN_GCC_AT_LEAST_12 + select BR2_RISCV_ISA_RVZifencei if BR2_TOOLCHAIN_GCC_AT_LEAST_12 help - General purpose (G) is equivalent to IMAFD. + General purpose (G) is equivalent to IMAFD + (with Zicsr and Zifencei since gcc >= 12). config BR2_riscv_custom bool "Custom architecture"
@@ -63,6 +72,16 @@ config BR2_RISCV_ISA_CUSTOM_RVD config BR2_RISCV_ISA_CUSTOM_RVC bool "Compressed Instructions (C)" select BR2_RISCV_ISA_RVC + +if BR2_TOOLCHAIN_GCC_AT_LEAST_12 +config BR2_RISCV_ISA_CUSTOM_RVZicsr + bool "Control and Status Register (CSR) Instructions" + select BR2_RISCV_ISA_RVZicsr + +config BR2_RISCV_ISA_CUSTOM_RVZifencei + bool "Instruction-Fetch Fence" + select BR2_RISCV_ISA_RVZifencei +endif endif choice
arch/arch.mk.riscv +6 -0
@@ -26,5 +26,11 @@ endif ifeq ($(BR2_RISCV_ISA_RVC),y) GCC_TARGET_ARCH := $(GCC_TARGET_ARCH)c endif +ifeq ($(BR2_RISCV_ISA_RVZicsr),y) +GCC_TARGET_ARCH := $(GCC_TARGET_ARCH)_zicsr +endif +ifeq ($(BR2_RISCV_ISA_RVZifencei),y) +GCC_TARGET_ARCH := $(GCC_TARGET_ARCH)_zifencei +endif endif
Signature · 6 lines
2.35.3 _______________________________________________ buildroot mailing list [email protected] https://lists.buildroot.org/mailman/listinfo/buildroot
PATCH 2/11 configs/qemu_riscv{32, 64}_virt: kernel bump version to 5.15.43
To build this defconfig with gcc 12, we need a patch from v5.15.24 [1] fixing the build with gcc 12.
---
The issue fixed by [1] can also be trigged with gcc 12 and binutils 2.37.
--- configs/qemu_riscv32_virt_defconfig | 2 +- configs/qemu_riscv64_virt_defconfig | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-)
configs/qemu_riscv32_virt_defconfig +1 -1
@@ -19,7 +19,7 @@ BR2_PACKAGE_HOST_LINUX_HEADERS_CUSTOM_5_15=y # Kernel BR2_LINUX_KERNEL=y BR2_LINUX_KERNEL_CUSTOM_VERSION=y -BR2_LINUX_KERNEL_CUSTOM_VERSION_VALUE="5.15.18" +BR2_LINUX_KERNEL_CUSTOM_VERSION_VALUE="5.15.43" BR2_LINUX_KERNEL_DEFCONFIG="rv32" BR2_LINUX_KERNEL_IMAGE=y
configs/qemu_riscv64_virt_defconfig +1 -1
@@ -19,7 +19,7 @@ BR2_PACKAGE_HOST_LINUX_HEADERS_CUSTOM_5_15=y # Kernel BR2_LINUX_KERNEL=y BR2_LINUX_KERNEL_CUSTOM_VERSION=y -BR2_LINUX_KERNEL_CUSTOM_VERSION_VALUE="5.15.18" +BR2_LINUX_KERNEL_CUSTOM_VERSION_VALUE="5.15.43" BR2_LINUX_KERNEL_USE_ARCH_DEFAULT_CONFIG=y BR2_LINUX_KERNEL_IMAGE=y
Signature · 6 lines
2.35.3 _______________________________________________ buildroot mailing list [email protected] https://lists.buildroot.org/mailman/listinfo/buildroot
Thomas Petazzoni via buildroot
AppliedRe: [Buildroot] [PATCH for-next 02/11] configs/qemu_riscv{32, 64}_virt: kernel bump version to 5.15.43
Quoted 5 lines · On Sun, 29 May 2022 15:18:02 +0200 Romain Naour <[email protected]> wrote:
To build this defconfig with gcc 12, we need a patch from v5.15.24 [1] fixing the build with gcc 12.
Quoted 8 lines
Signed-off-by Romain Naour <[email protected]>Cc Mark Corbin <[email protected]>---
The issue fixed by [1] can also be trigged with gcc 12 and binutils 2.37.--- configs/qemu_riscv32_virt_defconfig | 2 +- configs/qemu_riscv64_virt_defconfig | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-)
Applied to master, thanks.
Thomas
Signature · 7 lines
Thomas Petazzoni, CTO, Bootlin Embedded Linux and Kernel engineering https://bootlin.com _______________________________________________ buildroot mailing list [email protected] https://lists.buildroot.org/mailman/listinfo/buildroot
PATCH 3/11 package/gcc: disable libsanitizer for mips64{el} w/ n32 ABI
In reply to Romain Naourlibsanitizer has been enabled for mips64{el} in gcc 12 [1] but it fail to build when n32 ABI is used:
In file included from output/mips64el-buildroot-linux-gnu/sysroot/usr/include/bits/stat.h:25,
from output/mips64el-buildroot-linux-gnu/sysroot/usr/include/fcntl.h:78,
from ../../../../libsanitizer/sanitizer_common/sanitizer_linux.cpp:55:
output/mips64el-buildroot-linux-gnu/sysroot/usr/include/bits/struct_stat.h:190:8: error: redefinition of ‘struct stat64’
190 | struct stat64
| ^~~~~~In file included from ../../../../libsanitizer/sanitizer_common/sanitizer_linux.cpp:49:output/mips64el-buildroot-linux-gnu/sysroot/usr/include/asm/stat.h:52:8: note: previous definition of ‘struct stat64’
52 | struct stat64 {
| ^~~~~~Disable libsanitizer for mips64 with n32 ABI.
Note: Only glibc toolchains are affected since libsanitizer is disabled for musl and uClibc-ng toolchains [2].
Fixes:
https://gitlab.com/kubu93/toolchains-builder/-/jobs/2510178651
[1] https://gcc.gnu.org/git/?p=gcc.git;a=commitdiff;h=344e6f9f2abcff9b2bb4b26b693be4a599272f43
[2] https://git.buildroot.net/buildroot/commit/?id=5f4d658d888b539de9a6247ae5b1a0999de5d4ec
--- package/gcc/gcc.mk | 7 +++++++ 1 file changed, 7 insertions(+)
package/gcc/gcc.mk +7 -0
@@ -138,6 +138,13 @@ ifeq ($(BR2_sparc)$(BR2_sparc64),y) HOST_GCC_COMMON_CONF_OPTS += --disable-libsanitizer endif +# libsanitizer is available for mips64{el} since gcc 12 but fail to build +# with n32 ABI due to struct stat64 definition clash due to mixing +# kernel and user headers. +ifeq ($(BR2_mips64)$(BR2_mips64el)$(BR2_MIPS_NABI32),yy) +HOST_GCC_COMMON_CONF_OPTS += --disable-libsanitizer +endif + # The logic in libbacktrace/configure.ac to detect if __sync builtins # are available assumes they are as soon as target_subdir is not # empty, i.e when cross-compiling. However, some platforms do not have
Signature · 6 lines
2.35.3 _______________________________________________ buildroot mailing list [email protected] https://lists.buildroot.org/mailman/listinfo/buildroot
Re: [Buildroot] [PATCH for-next 03/11] package/gcc: disable libsanitizer for mips64{el} w/ n32 ABI
Quoted 17 lines · On Sun, 29 May 2022 15:18:03 +0200 Romain Naour <[email protected]> wrote:
libsanitizer has been enabled for mips64{el} in gcc 12 [1] but it fail to build when n32 ABI is used:
In file included from
output/mips64el-buildroot-linux-gnu/sysroot/usr/include/bits/stat.h:25,from output/mips64el-buildroot-linux-gnu/sysroot/usr/include/fcntl.h:78, from ../../../../libsanitizer/sanitizer_common/sanitizer_linux.cpp:55: output/mips64el-buildroot-linux-gnu/sysroot/usr/include/bits/struct_stat.h:190:8: error: redefinition of ‘struct stat64’ 190 | struct stat64 | ^~~~~~In file included from ../../../../libsanitizer/sanitizer_common/sanitizer_linux.cpp:49:
output/mips64el-buildroot-linux-gnu/sysroot/usr/include/asm/stat.h:52:8: note: previous definition of ‘struct stat64’52 | struct stat64 { | ^~~~~~Disable libsanitizer for mips64 with n32 ABI.
Quoted 8 lines
Note: Only glibc toolchains are affected since libsanitizer is disabled for musl and uClibc-ng toolchains [2].
Fixes:
https://gitlab.com/kubu93/toolchains-builder/-/jobs/2510178651[1] https://gcc.gnu.org/git/?p=gcc.git;a=commitdiff;h=344e6f9f2abcff9b2bb4b26b693be4a599272f43
[2] https://git.buildroot.net/buildroot/commit/?id=5f4d658d888b539de9a6247ae5b1a0999de5d4ec
Has this issue been reported to upstream gcc?
Quoted 8 lines
package/gcc/gcc.mk | 7 +++++++ 1 file changed, 7 insertions(+)package/gcc/gcc.mk +0
-0
package/gcc/gcc.mk ifeq ($(BR2_sparc)$(BR2_sparc64),y) · line 144
Quoted 7 lines
package/gcc/gcc.mk +4
-0@@ -138,6 +138,13 @@ ifeq ($(BR2_sparc)$(BR2_sparc64),y) HOST_GCC_COMMON_CONF_OPTS += --disable-libsanitizer endif +# libsanitizer is available for mips64{el} since gcc 12 but fail to build +# with n32 ABI due to struct stat64 definition clash due to mixing +# kernel and user headers. +ifeq ($(BR2_mips64)$(BR2_mips64el)$(BR2_MIPS_NABI32),yy)
Minor nit, perhaps:
ifeq ($(BR2_mips64)$(BR2_mips64el):$(BR2_MIPS_NABI32),y:y)
so it's absolutely clear how this works.
Thanks!
Thomas
Signature · 7 lines
Thomas Petazzoni, co-owner and CEO, Bootlin Embedded Linux and Kernel engineering and training https://bootlin.com _______________________________________________ buildroot mailing list [email protected] https://lists.buildroot.org/mailman/listinfo/buildroot
PATCH 4/11 package/gcc: disable libsanitizer for mips{el} and gcc > 12
In reply to Romain NaourSince [1], gcc contains a workaround for struct_kernel_stat_sz definition and apply a local patch when updating libsanitizer with upstream (llvm-project) [2].
Since gcc 12, the workaround is not enough and trigger the following error:
In file included from ../../../../libsanitizer/sanitizer_common/sanitizer_platform_limits_linux.cpp:21:
../../../../libsanitizer/sanitizer_common/sanitizer_platform_limits_linux.cpp:75:38: error: static assertion failed
75 | COMPILER_CHECK(struct_kernel_stat_sz == sizeof(struct stat));
| ~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~Disable libsanitizer for now.
Note: Only glibc toolchains are affected since libsanitizer is
disabled for musl and uClibc-ng toolchains [3].
Fixes:
https://gitlab.com/kubu93/toolchains-builder/-/jobs/2510178606
[1] http://gcc.gnu.org/ml/gcc/2018-03/msg00133.html
[2] https://gcc.gnu.org/git/?p=gcc.git;a=commitdiff;h=9f943b2446f2d0a345bbf9b4be3d3a4316372270
[3] https://git.buildroot.net/buildroot/commit/?id=5f4d658d888b539de9a6247ae5b1a0999de5d4ec
--- package/gcc/gcc.mk | 6 ++++++ 1 file changed, 6 insertions(+)
package/gcc/gcc.mk +6 -0
@@ -145,6 +145,12 @@ ifeq ($(BR2_mips64)$(BR2_mips64el)$(BR2_MIPS_NABI32),yy) HOST_GCC_COMMON_CONF_OPTS += --disable-libsanitizer endif +# libsanitizer bundled in gcc 12 fail to build for mips32 due to +# mixing kernel and user struct stat. +ifeq ($(BR2_TOOLCHAIN_GCC_AT_LEAST_12)$(BR2_mips)$(BR2_mipsel),yy) +HOST_GCC_COMMON_CONF_OPTS += --disable-libsanitizer +endif + # The logic in libbacktrace/configure.ac to detect if __sync builtins # are available assumes they are as soon as target_subdir is not # empty, i.e when cross-compiling. However, some platforms do not have
Signature · 6 lines
2.35.3 _______________________________________________ buildroot mailing list [email protected] https://lists.buildroot.org/mailman/listinfo/buildroot
PATCH 5/11 toolchain: enable libquadmath for PowerPC with VSX
In reply to Romain Naourfloat128 is available on PowerPC with VSX [1] but it requires libquadmath support.
[1] https://gcc.gnu.org/onlinedocs/gcc-9.1.0/gcc/Floating-Types.html
--- toolchain/Config.in | 2 ++ 1 file changed, 2 insertions(+)
toolchain/Config.in +2 -0
@@ -833,9 +833,11 @@ config BR2_TOOLCHAIN_HAS_ATOMIC # - At least, libquadmath is available on: # - i*86 # - x86_64 +# - PowerPC with vector scalar (VSX) instruction set # - When available, libquadmath requires wchar support. config BR2_TOOLCHAIN_HAS_LIBQUADMATH bool default y if BR2_i386 || BR2_x86_64 + default y if BR2_POWERPC_CPU_HAS_VSX endmenu
Signature · 6 lines
2.35.3 _______________________________________________ buildroot mailing list [email protected] https://lists.buildroot.org/mailman/listinfo/buildroot
Thomas Petazzoni via buildroot
AppliedRe: [Buildroot] [PATCH for-next 05/11] toolchain: enable libquadmath for PowerPC with VSX
Quoted 11 lines · On Sun, 29 May 2022 15:18:05 +0200 Romain Naour <[email protected]> wrote:
float128 is available on PowerPC with VSX [1] but it requires libquadmath support.
[1] https://gcc.gnu.org/onlinedocs/gcc-9.1.0/gcc/Floating-Types.html
Signed-off-by Romain Naour <[email protected]>Cc Cyril Bur <[email protected]>Cc Cédric Le Goater <[email protected]>--- toolchain/Config.in | 2 ++ 1 file changed, 2 insertions(+)
Applied to master, thanks.
Thomas
Signature · 7 lines
Thomas Petazzoni, CTO, Bootlin Embedded Linux and Kernel engineering https://bootlin.com _______________________________________________ buildroot mailing list [email protected] https://lists.buildroot.org/mailman/listinfo/buildroot
PATCH 6/11 package/gcc: add missing --enable-libquadmath-support option
In reply to Romain NaourWhen BR2_TOOLCHAIN_HAS_LIBQUADMATH is set, --enable-libquadmath-support option is missing. So the float128 support is not fully enabled in gcc.
This lead to a build issue with gcc 12 on PowerPC power8 due to missing M_2_SQRTPIq definition (provided by libquadmath.h).
../../../libgfortran/intrinsics/erfc_scaled.c: In function ‘erfc_scaled_r17’:../../../libgfortran/intrinsics/erfc_scaled.c:143:22: error: ‘M_2_SQRTPIq’ undeclared (first use in this function); did you mean ‘M_2_SQRTPIf’?
143 | # define _M_2_SQRTPI M_2_SQRTPIq
| ^~~~~~~~~~~
This is fixed by adding --enable-libquadmath-support (like crosstool-ng handling [1]).
Fixes:
https://gitlab.com/kubu93/toolchains-builder/-/jobs/2510178766
[1] https://github.com/crosstool-ng/crosstool-ng/blob/crosstool-ng-1.25.0/scripts/build/cc/gcc.sh#L370
--- package/gcc/gcc.mk | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
package/gcc/gcc.mk +1 -1
@@ -121,7 +121,7 @@ endif # quadmath support requires wchar ifeq ($(BR2_USE_WCHAR)$(BR2_TOOLCHAIN_HAS_LIBQUADMATH),yy) -HOST_GCC_COMMON_CONF_OPTS += --enable-libquadmath +HOST_GCC_COMMON_CONF_OPTS += --enable-libquadmath --enable-libquadmath-support else HOST_GCC_COMMON_CONF_OPTS += --disable-libquadmath --disable-libquadmath-support endif
Signature · 6 lines
2.35.3 _______________________________________________ buildroot mailing list [email protected] https://lists.buildroot.org/mailman/listinfo/buildroot
Thomas Petazzoni via buildroot
AppliedRe: [Buildroot] [PATCH for-next 06/11] package/gcc: add missing --enable-libquadmath-support option
Quoted 17 lines · On Sun, 29 May 2022 15:18:06 +0200 Romain Naour <[email protected]> wrote:
When
BR2_TOOLCHAIN_HAS_LIBQUADMATHis set, --enable-libquadmath-support option is missing. So the float128 support is not fully enabled in gcc.This lead to a build issue with gcc 12 on PowerPC power8 due to missing
M_2_SQRTPIqdefinition (provided bylibquadmath.h).
../../../libgfortran/intrinsics/erfc_scaled.c: In function ‘erfc_scaled_r17’:../../../libgfortran/intrinsics/erfc_scaled.c:143:22: error: ‘M_2_SQRTPIq’ undeclared (first use in this function); did you mean ‘M_2_SQRTPIf’?
143 | # define_M_2_SQRTPIM_2_SQRTPIq| ^~~~~~~~~~~This is fixed by adding --enable-libquadmath-support (like crosstool-ng handling [1]).
Fixes:
https://gitlab.com/kubu93/toolchains-builder/-/jobs/2510178766
Quoted 6 lines
[1] https://github.com/crosstool-ng/crosstool-ng/blob/crosstool-ng-1.25.0/scripts/build/cc/gcc.sh#L370
Signed-off-by Romain Naour <[email protected]>--- package/gcc/gcc.mk | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
Applied to master, thanks.
Thomas
Signature · 7 lines
Thomas Petazzoni, CTO, Bootlin Embedded Linux and Kernel engineering https://bootlin.com _______________________________________________ buildroot mailing list [email protected] https://lists.buildroot.org/mailman/listinfo/buildroot
PATCH 7/11 package/gcc: switch to https urls for archives hashes
In reply to Romain NaourFTP urls doesn't work anymore with Firefox [1] and Google Chrome [2], switch to https urls.
[1] https://blog.mozilla.org/addons/2020/04/13/what-to-expect-for-the-upcoming-deprecation-of-ftp-in-firefox/
[2] https://chromestatus.com/feature/6246151319715840
--- package/gcc/gcc.hash | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-)
package/gcc/gcc.hash +4 -4
@@ -1,10 +1,10 @@ -# From ftp://gcc.gnu.org/pub/gcc/releases/gcc-8.4.0/sha512.sum +# From https://gcc.gnu.org/pub/gcc/releases/gcc-8.4.0/sha512.sum sha512 6de904f552a02de33b11ef52312bb664396efd7e1ce3bbe37bfad5ef617f133095b3767b4804bc7fe78df335cb53bc83f1ac055baed40979ce4c2c3e46b70280 gcc-8.4.0.tar.xz -# From ftp://gcc.gnu.org/pub/gcc/releases/gcc-9.4.0/sha512.sum +# From https://gcc.gnu.org/pub/gcc/releases/gcc-9.4.0/sha512.sum sha512 dfd3500bf21784b8351a522d53463cf362ede66b0bc302edf350bb44e94418497a8b4b797b6af8ca9b2eeb746b3b115d9c3698381b989546e9151b4496415624 gcc-9.4.0.tar.xz -# From ftp://gcc.gnu.org/pub/gcc/releases/gcc-10.3.0/sha512.sum +# From https://gcc.gnu.org/pub/gcc/releases/gcc-10.3.0/sha512.sum sha512 2b2dd7453d48a398c29eaebd1422b70341001b8c90a62aee51e83344e7fdd8a8e45f82a4a9165bd7edc76dada912c932f4b6632c5636760fec4c5d7e402b3f86 gcc-10.3.0.tar.xz -# From ftp://gcc.gnu.org/pub/gcc/releases/gcc-11.3.0/sha512.sum +# From https://gcc.gnu.org/pub/gcc/releases/gcc-11.3.0/sha512.sum sha512 f0be5ad705c73b84477128a69c047f57dd47002f375eb60e1e842e08cf2009a509e92152bca345823926d550b7395ae6d4de7db51d1ee371c2dc37313881fca7 gcc-11.3.0.tar.xz # Locally calculated (fetched from Github)
Signature · 6 lines
2.35.3 _______________________________________________ buildroot mailing list [email protected] https://lists.buildroot.org/mailman/listinfo/buildroot
Thomas Petazzoni via buildroot
AppliedRe: [Buildroot] [PATCH for-next 07/11] package/gcc: switch to https urls for archives hashes
Quoted 10 lines · On Sun, 29 May 2022 15:18:07 +0200 Romain Naour <[email protected]> wrote:
FTP urls doesn't work anymore with Firefox [1] and Google Chrome [2], switch to https urls.
[1] https://blog.mozilla.org/addons/2020/04/13/what-to-expect-for-the-upcoming-deprecation-of-ftp-in-firefox/
[2] https://chromestatus.com/feature/6246151319715840Signed-off-by Romain Naour <[email protected]>--- package/gcc/gcc.hash | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-)
Applied to master, thanks.
Thomas
Signature · 7 lines
Thomas Petazzoni, CTO, Bootlin Embedded Linux and Kernel engineering https://bootlin.com _______________________________________________ buildroot mailing list [email protected] https://lists.buildroot.org/mailman/listinfo/buildroot
PATCH 8/11 package/gcc: add support for gcc 12
In reply to Romain Naourhttps://gcc.gnu.org/gcc-12/changes.html
https://gcc.gnu.org/gcc-12/porting_to.html
Tested with toolchain-builder:
https://gitlab.com/kubu93/toolchains-builder/-/pipelines/550404957
--- package/gcc/Config.in.host | 14 ++++++++++++++ package/gcc/gcc.hash | 2 ++ 2 files changed, 16 insertions(+)
package/gcc/Config.in.host +14 -0
@@ -62,6 +62,19 @@ config BR2_GCC_VERSION_11_X depends on !BR2_archs4x_rel31 select BR2_TOOLCHAIN_GCC_AT_LEAST_11 +config BR2_GCC_VERSION_12_X + bool "gcc 12.x" + # powerpc spe support has been deprecated since gcc 8.x. + # https://gcc.gnu.org/ml/gcc/2018-04/msg00102.html + depends on !BR2_powerpc_SPE + # uClibc-ng broken on sparc due to recent gcc changes + # that need to be reverted since gcc 8.4, 9.3 and 10.1. + # https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98784 + depends on !BR2_sparc + # ARC HS48 rel 31 only supported by gcc arc fork. + depends on !BR2_archs4x_rel31 + select BR2_TOOLCHAIN_GCC_AT_LEAST_12 + endchoice # libcilkrts was introduced in gcc 4.9 and removed in gcc 8.x
@@ -88,6 +101,7 @@ config BR2_GCC_VERSION default "9.4.0" if BR2_GCC_VERSION_9_X default "10.3.0" if BR2_GCC_VERSION_10_X default "11.3.0" if BR2_GCC_VERSION_11_X + default "12.1.0" if BR2_GCC_VERSION_12_X default "arc-2020.09-release" if BR2_GCC_VERSION_ARC config BR2_EXTRA_GCC_CONFIG_OPTIONS
package/gcc/gcc.hash +2 -0
@@ -6,6 +6,8 @@ sha512 dfd3500bf21784b8351a522d53463cf362ede66b0bc302edf350bb44e94418497a8b4b79 sha512 2b2dd7453d48a398c29eaebd1422b70341001b8c90a62aee51e83344e7fdd8a8e45f82a4a9165bd7edc76dada912c932f4b6632c5636760fec4c5d7e402b3f86 gcc-10.3.0.tar.xz # From https://gcc.gnu.org/pub/gcc/releases/gcc-11.3.0/sha512.sum sha512 f0be5ad705c73b84477128a69c047f57dd47002f375eb60e1e842e08cf2009a509e92152bca345823926d550b7395ae6d4de7db51d1ee371c2dc37313881fca7 gcc-11.3.0.tar.xz +# From https://gcc.gnu.org/pub/gcc/releases/gcc-12.1.0/sha512.sum +sha512 2121d295292814a6761edf1fba08c5f633ebe16f52b80e7b73a91050e71e1d2ed98bf17eebad263e191879561c02b48906c53faa4c4670c486a26fc75df23900 gcc-12.1.0.tar.xz # Locally calculated (fetched from Github) sha512 b0853e2b1c5998044392023fa653e399e74118c46e616504ac59e1a2cf27620f94434767ce06b6cf4ca3dfb57f81d6eda92752befaf095ea5e564a9181b4659c gcc-arc-2020.09-release.tar.gz
Signature · 6 lines
2.35.3 _______________________________________________ buildroot mailing list [email protected] https://lists.buildroot.org/mailman/listinfo/buildroot
Re: [Buildroot] [PATCH for-next 08/11] package/gcc: add support for gcc 12
Quoted 8 lines · Romain Naour
https://gcc.gnu.org/gcc-12/changes.html
https://gcc.gnu.org/gcc-12/porting_to.htmlTested with toolchain-builder:
https://gitlab.com/kubu93/toolchains-builder/-/pipelines/550404957Signed-off-by Romain Naour <[email protected]>
Quoted 53 lines
--- package/gcc/Config.in.host | 14 ++++++++++++++ package/gcc/gcc.hash | 2 ++ 2 files changed, 16 insertions(+)package/gcc/Config.in.host +14
-0@@ -62,6 +62,19 @@ config BR2_GCC_VERSION_11_X depends on !BR2_archs4x_rel31 select BR2_TOOLCHAIN_GCC_AT_LEAST_11 +config BR2_GCC_VERSION_12_X + bool "gcc 12.x" + # powerpc spe support has been deprecated since gcc 8.x. + # https://gcc.gnu.org/ml/gcc/2018-04/msg00102.html + depends on !BR2_powerpc_SPE + # uClibc-ng broken on sparc due to recent gcc changes + # that need to be reverted since gcc 8.4, 9.3 and 10.1. + # https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98784 + depends on !BR2_sparc + # ARC HS48 rel 31 only supported by gcc arc fork. + depends on !BR2_archs4x_rel31 + select BR2_TOOLCHAIN_GCC_AT_LEAST_12 + endchoice # libcilkrts was introduced in gcc 4.9 and removed in gcc 8.x@@ -88,6 +101,7 @@ config BR2_GCC_VERSION default "9.4.0" if BR2_GCC_VERSION_9_X default "10.3.0" if BR2_GCC_VERSION_10_X default "11.3.0" if BR2_GCC_VERSION_11_X + default "12.1.0" if BR2_GCC_VERSION_12_X default "arc-2020.09-release" if BR2_GCC_VERSION_ARC config BR2_EXTRA_GCC_CONFIG_OPTIONSpackage/gcc/gcc.hash +2
-1@@ -6,6 +6,8 @@ sha512 dfd3500bf21784b8351a522d53463cf362ede66b0bc302edf350bb44e94418497a8b4b79 sha512 2b2dd7453d48a398c29eaebd1422b70341001b8c90a62aee51e83344e7fdd8a8e45f82a4a9165bd7edc76dada912c932f4b6632c5636760fec4c5d7e402b3f86 gcc-10.3.0.tar.xz # From https://gcc.gnu.org/pub/gcc/releases/gcc-11.3.0/sha512.sum sha512 f0be5ad705c73b84477128a69c047f57dd47002f375eb60e1e842e08cf2009a509e92152bca345823926d550b7395ae6d4de7db51d1ee371c2dc37313881fca7 gcc-11.3.0.tar.xz +# From https://gcc.gnu.org/pub/gcc/releases/gcc-12.1.0/sha512.sum +sha512 2121d295292814a6761edf1fba08c5f633ebe16f52b80e7b73a91050e71e1d2ed98bf17eebad263e191879561c02b48906c53faa4c4670c486a26fc75df23900 gcc-12.1.0.tar.xz # Locally calculated (fetched from Github) sha512 b0853e2b1c5998044392023fa653e399e74118c46e616504ac59e1a2cf27620f94434767ce06b6cf4ca3dfb57f81d6eda92752befaf095ea5e564a9181b4659c gcc-arc-2020.09-release.tar.gz --2.35.3
Quoted 4 lines
_______________________________________________
buildroot mailing list
[email protected]
https://lists.buildroot.org/mailman/listinfo/buildroot
_______________________________________________
buildroot mailing list
[email protected]
https://lists.buildroot.org/mailman/listinfo/buildroot
Re: [Buildroot] [PATCH for-next 08/11] package/gcc: add support for gcc 12
Hello James,
Le 25/06/2022 à 08:45, James Hilliard a écrit :
Quoted 10 lines · Romain Naour
Quoted 8 lines
https://gcc.gnu.org/gcc-12/changes.html
https://gcc.gnu.org/gcc-12/porting_to.htmlTested with toolchain-builder:
https://gitlab.com/kubu93/toolchains-builder/-/pipelines/550404957Signed-off-by Romain Naour <[email protected]>Tested-by James Hilliard <[email protected]>
Thanks for testing but previous patches in this series seems to be holding back maintainers from merging gcc 12. If you have some time it would be nice if you can review them.
Best regards,
Romain
Quoted 53 lines
Quoted 53 lines
--- package/gcc/Config.in.host | 14 ++++++++++++++ package/gcc/gcc.hash | 2 ++ 2 files changed, 16 insertions(+)package/gcc/Config.in.host +14
-0@@ -62,6 +62,19 @@ config BR2_GCC_VERSION_11_X depends on !BR2_archs4x_rel31 select BR2_TOOLCHAIN_GCC_AT_LEAST_11 +config BR2_GCC_VERSION_12_X + bool "gcc 12.x" + # powerpc spe support has been deprecated since gcc 8.x. + # https://gcc.gnu.org/ml/gcc/2018-04/msg00102.html + depends on !BR2_powerpc_SPE + # uClibc-ng broken on sparc due to recent gcc changes + # that need to be reverted since gcc 8.4, 9.3 and 10.1. + # https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98784 + depends on !BR2_sparc + # ARC HS48 rel 31 only supported by gcc arc fork. + depends on !BR2_archs4x_rel31 + select BR2_TOOLCHAIN_GCC_AT_LEAST_12 + endchoice # libcilkrts was introduced in gcc 4.9 and removed in gcc 8.x@@ -88,6 +101,7 @@ config BR2_GCC_VERSION default "9.4.0" if BR2_GCC_VERSION_9_X default "10.3.0" if BR2_GCC_VERSION_10_X default "11.3.0" if BR2_GCC_VERSION_11_X + default "12.1.0" if BR2_GCC_VERSION_12_X default "arc-2020.09-release" if BR2_GCC_VERSION_ARC config BR2_EXTRA_GCC_CONFIG_OPTIONSpackage/gcc/gcc.hash +2
-1@@ -6,6 +6,8 @@ sha512 dfd3500bf21784b8351a522d53463cf362ede66b0bc302edf350bb44e94418497a8b4b79 sha512 2b2dd7453d48a398c29eaebd1422b70341001b8c90a62aee51e83344e7fdd8a8e45f82a4a9165bd7edc76dada912c932f4b6632c5636760fec4c5d7e402b3f86 gcc-10.3.0.tar.xz # From https://gcc.gnu.org/pub/gcc/releases/gcc-11.3.0/sha512.sum sha512 f0be5ad705c73b84477128a69c047f57dd47002f375eb60e1e842e08cf2009a509e92152bca345823926d550b7395ae6d4de7db51d1ee371c2dc37313881fca7 gcc-11.3.0.tar.xz +# From https://gcc.gnu.org/pub/gcc/releases/gcc-12.1.0/sha512.sum +sha512 2121d295292814a6761edf1fba08c5f633ebe16f52b80e7b73a91050e71e1d2ed98bf17eebad263e191879561c02b48906c53faa4c4670c486a26fc75df23900 gcc-12.1.0.tar.xz # Locally calculated (fetched from Github) sha512 b0853e2b1c5998044392023fa653e399e74118c46e616504ac59e1a2cf27620f94434767ce06b6cf4ca3dfb57f81d6eda92752befaf095ea5e564a9181b4659c gcc-arc-2020.09-release.tar.gz --2.35.3
Quoted 4 lines
Quoted 4 lines
_______________________________________________
buildroot mailing list
[email protected]
https://lists.buildroot.org/mailman/listinfo/buildroot
_______________________________________________
buildroot mailing list
[email protected]
https://lists.buildroot.org/mailman/listinfo/buildroot
PATCH 9/11 arch: add BR2_ARCH_NEEDS_GCC_AT_LEAST_12
In reply to Romain NaourThis new symbol will be used by architectures introduced with gcc 12.
--- arch/Config.in | 4 ++++ package/gcc/Config.in.host | 1 + 2 files changed, 5 insertions(+)
arch/Config.in +4 -0
@@ -321,6 +321,10 @@ config BR2_ARCH_NEEDS_GCC_AT_LEAST_11 bool select BR2_ARCH_NEEDS_GCC_AT_LEAST_10 +config BR2_ARCH_NEEDS_GCC_AT_LEAST_12 + bool + select BR2_ARCH_NEEDS_GCC_AT_LEAST_11 + # The following string values are defined by the individual # Config.in.$ARCH files config BR2_ARCH
package/gcc/Config.in.host +1 -0
@@ -51,6 +51,7 @@ config BR2_GCC_VERSION_10_X config BR2_GCC_VERSION_11_X bool "gcc 11.x" + depends on !BR2_ARCH_NEEDS_GCC_AT_LEAST_12 # powerpc spe support has been deprecated since gcc 8.x. # https://gcc.gnu.org/ml/gcc/2018-04/msg00102.html depends on !BR2_powerpc_SPE
Signature · 6 lines
2.35.3 _______________________________________________ buildroot mailing list [email protected] https://lists.buildroot.org/mailman/listinfo/buildroot
PATCH 10/11 package/gcc: switch to gcc 11.x as the default
In reply to Romain NaourEven if gcc 10.x is still maintained for some time, switch to gcc 11.x since it has been released since 2021-04-27 and gcc 12.x is available since "2022-05-10".
We have been having toolchains in the autobuilders with gcc 11.x since mid-June 2021, so the vast majority of the problems should have already been solved.
--- package/gcc/Config.in.host | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
package/gcc/Config.in.host +1 -1
@@ -4,7 +4,7 @@ choice prompt "GCC compiler Version" default BR2_GCC_VERSION_ARC if BR2_arc default BR2_GCC_VERSION_POWERPC_SPE if BR2_powerpc_SPE - default BR2_GCC_VERSION_10_X + default BR2_GCC_VERSION_11_X help Select the version of gcc you wish to use.
Signature · 6 lines
2.35.3 _______________________________________________ buildroot mailing list [email protected] https://lists.buildroot.org/mailman/listinfo/buildroot
PATCH 11/11 package/gcc: remove gcc 9.x
In reply to Romain Naourgcc 12.1 is around, gcc 11.3 is the default version, so drop 9.5 in order to reduce the gcc choice.
--- Config.in.legacy | 9 + ...-issue-with-set_got-clobbering-LR-r9.patch | 105 ------ ...TRANT-for-OpenRISC-when-pthread-is-p.patch | 31 -- ...get-92095-internal-error-with-O1-mcp.patch | 325 ------------------ ...-mcmodel-option-to-handle-large-GOTs.patch | 201 ----------- ...-cmodel-large-when-building-crtstuff.patch | 60 ---- package/gcc/Config.in.host | 15 - package/gcc/gcc.hash | 2 - 8 files changed, 9 insertions(+), 739 deletions(-) delete mode 100644 package/gcc/9.4.0/0001-or1k-Fix-issue-with-set_got-clobbering-LR-r9.patch delete mode 100644 package/gcc/9.4.0/0002-gcc-define-_REENTRANT-for-OpenRISC-when-pthread-is-p.patch delete mode 100644 package/gcc/9.4.0/0003-Revert-re-PR-target-92095-internal-error-with-O1-mcp.patch delete mode 100644 package/gcc/9.4.0/0004-or1k-Add-mcmodel-option-to-handle-large-GOTs.patch delete mode 100644 package/gcc/9.4.0/0005-or1k-Use-cmodel-large-when-building-crtstuff.patch
Config.in.legacy +9 -0
@@ -144,6 +144,15 @@ endif ############################################################################### +comment "Legacy options removed in 2022.08" + +config BR2_GCC_VERSION_9_X + bool "gcc 9.x support removed" + select BR2_LEGACY + help + Support for gcc version 9.x has been removed. The current + default version (11.x or later) has been selected instead. + comment "Legacy options removed in 2022.05" config BR2_KERNEL_HEADERS_5_16
package/gcc/9.4.0/0001-or1k-Fix-issue-with-set_got-clobbering-LR-r9.patch deleted +0 -105
105 lines not on this page · raw message
package/gcc/9.4.0/0002-gcc-define-_REENTRANT-for-OpenRISC-when-pthread-is-p.patch deleted +0 -31
@@ -1,31 +0,0 @@ -From f80e9941739fb3973b61fc6a5abddef5ad2faf73 Mon Sep 17 00:00:00 2001 -From: Bernd Kuhls <[email protected]> -Date: Fri, 27 Mar 2020 21:23:53 +0100 -Subject: [PATCH] gcc: define _REENTRANT for OpenRISC when -pthread is passed - -The detection of pthread support fails on OpenRISC unless _REENTRANT -is defined. Added the CPP_SPEC definition to correct this. - -Patch sent upstream: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94372 - -Signed-off-by: Bernd Kuhls <[email protected]> ---- - gcc/config/or1k/linux.h | 2 ++ - 1 file changed, 2 insertions(+) - -diff --git a/gcc/config/or1k/linux.h b/gcc/config/or1k/linux.h -index cbdc781418f..36303af892c 100644 ---- a/gcc/config/or1k/linux.h -+++ b/gcc/config/or1k/linux.h -@@ -32,6 +32,8 @@ - #undef MUSL_DYNAMIC_LINKER - #define MUSL_DYNAMIC_LINKER "/lib/ld-musl-or1k.so.1" - -+#define CPP_SPEC "%{pthread:-D_REENTRANT}" -+ - #undef LINK_SPEC - #define LINK_SPEC "%{h*} \ - %{static:-Bstatic} \ --- -2.31.1 -
package/gcc/9.4.0/0003-Revert-re-PR-target-92095-internal-error-with-O1-mcp.patch deleted +0 -325
325 lines not on this page · raw message
package/gcc/9.4.0/0004-or1k-Add-mcmodel-option-to-handle-large-GOTs.patch deleted +0 -201
201 lines not on this page · raw message
package/gcc/9.4.0/0005-or1k-Use-cmodel-large-when-building-crtstuff.patch deleted +0 -60
60 lines not on this page · raw message
package/gcc/Config.in.host +0 -15
@@ -21,20 +21,6 @@ config BR2_GCC_VERSION_POWERPC_SPE depends on BR2_powerpc_SPE select BR2_TOOLCHAIN_GCC_AT_LEAST_8 -config BR2_GCC_VERSION_9_X - bool "gcc 9.x" - depends on !BR2_ARCH_NEEDS_GCC_AT_LEAST_10 - # powerpc spe support has been deprecated since gcc 8.x. - # https://gcc.gnu.org/ml/gcc/2018-04/msg00102.html - depends on !BR2_powerpc_SPE - # C-SKY sk610 needs abiv1, which is not supported in - # upstream gcc. C-SKY gcc upstream support not tested - # with upstream binutils and glibc. - depends on !BR2_csky - # ARC HS48 rel 31 only supported by gcc arc fork. - depends on !BR2_archs4x_rel31 - select BR2_TOOLCHAIN_GCC_AT_LEAST_9 - config BR2_GCC_VERSION_10_X bool "gcc 10.x" depends on !BR2_ARCH_NEEDS_GCC_AT_LEAST_11
@@ -99,7 +85,6 @@ config BR2_GCC_SUPPORTS_DLANG config BR2_GCC_VERSION string default "8.4.0" if BR2_GCC_VERSION_POWERPC_SPE - default "9.4.0" if BR2_GCC_VERSION_9_X default "10.3.0" if BR2_GCC_VERSION_10_X default "11.3.0" if BR2_GCC_VERSION_11_X default "12.1.0" if BR2_GCC_VERSION_12_X
package/gcc/gcc.hash +0 -2
@@ -1,7 +1,5 @@ # From https://gcc.gnu.org/pub/gcc/releases/gcc-8.4.0/sha512.sum sha512 6de904f552a02de33b11ef52312bb664396efd7e1ce3bbe37bfad5ef617f133095b3767b4804bc7fe78df335cb53bc83f1ac055baed40979ce4c2c3e46b70280 gcc-8.4.0.tar.xz -# From https://gcc.gnu.org/pub/gcc/releases/gcc-9.4.0/sha512.sum -sha512 dfd3500bf21784b8351a522d53463cf362ede66b0bc302edf350bb44e94418497a8b4b797b6af8ca9b2eeb746b3b115d9c3698381b989546e9151b4496415624 gcc-9.4.0.tar.xz # From https://gcc.gnu.org/pub/gcc/releases/gcc-10.3.0/sha512.sum sha512 2b2dd7453d48a398c29eaebd1422b70341001b8c90a62aee51e83344e7fdd8a8e45f82a4a9165bd7edc76dada912c932f4b6632c5636760fec4c5d7e402b3f86 gcc-10.3.0.tar.xz # From https://gcc.gnu.org/pub/gcc/releases/gcc-11.3.0/sha512.sum
Signature · 6 lines
2.35.3 _______________________________________________ buildroot mailing list [email protected] https://lists.buildroot.org/mailman/listinfo/buildroot
Hello Romain,
Quoted 27 lines · On Sun, 29 May 2022 15:18:01 +0200 Romain Naour <[email protected]> wrote:
Since gcc 12, default Riscv ISA spec version was bump to 20191213 [1].
This introduce a major incompatibility issue is the csr read/write (csrr*/csrw*) instructions and fence.i instruction has separated from the "I" extension, become two standalone extensions: Zicsr and Zifencei; so you might get error messages like that: unrecognized opcode "csrr" (or "fence.i").
Indeed, without Zifencei we can't build opensbi bootloader [3]:
opensbi-1.0/lib/sbi/sbi_tlb.c: Assembler messages:opensbi-1.0/lib/sbi/sbi_tlb.c:190: Error: unrecognized opcodefence.i', extensionzifencei' requiredAs a workaround, opensbi build system has been patched [4] to use -march=rv64imafdc_zicsr_zifencei when needed.
This workaround doesn't work in Buildroot due to the local patch 0001-Makefile-Don-t-specify-mabi-or-march.patch removing -march from CFLAGS.Fix this issue by introducing two additional
Kconfigoption enabling Zicsr and Zifencei standalone extensions for gcc >= 12 as recommanded by [2].Select Zicsr and Zifencei for General purpose (G) architecture variant (
BR2_riscv_g) since theses extentions were implicitely enabled previously.
Quoted 7 lines
[1] https://gcc.gnu.org/gcc-12/changes.html
[2] https://groups.google.com/a/groups.riscv.org/g/sw-dev/c/aE1ZeHHCYf4
[3] https://github.com/riscv-software-src/opensbi/blob/v0.9/lib/sbi/sbi_tlb.c#L173
[4] https://github.com/riscv-software-src/opensbi/commit/5d53b55aa77ffeefd4012445dfa6ad3535e1ff2cSigned-off-by Romain Naour <[email protected]>Cc Mark Corbin <[email protected]>
Thanks for the patch and explanation, but I'm wondering if this isn't perhaps too complicated.
Quoted 2 lines
(quoted diff) +2
-0+ select BR2_RISCV_ISA_RVZicsr if BR2_TOOLCHAIN_GCC_AT_LEAST_12 + select BR2_RISCV_ISA_RVZifencei if BR2_TOOLCHAIN_GCC_AT_LEAST_12
[...]
Review config BR2_RISCV_ISA_CUSTOM_RVD · line 76
Quoted 13 lines
(quoted diff) +4
-1help - General purpose (G) is equivalent to IMAFD. + General purpose (G) is equivalent to IMAFD + (with Zicsr and Zifencei since gcc >= 12). config BR2_riscv_custom bool "Custom architecture"@@ -63,6 +72,16 @@ config BR2_RISCV_ISA_CUSTOM_RVD config BR2_RISCV_ISA_CUSTOM_RVC bool "Compressed Instructions (C)" select BR2_RISCV_ISA_RVC + +if BR2_TOOLCHAIN_GCC_AT_LEAST_12
The problem with this is that options in arch/ now depend on the gcc version that is selected, which is something we have tried to avoid.
Do a git grep GCC_AT_LEAST in arch/, you will see that we only use "select BR2_ARCH_NEEDS_GCC_AT_LEAST_xyz", i.e the choice of the architecture/variant/ABI is what drives the available gcc versions.
And here you are doing exactly the opposite: it's the selected gcc version that drives the architecture/variant/ABI options that are available.
At this point, I don't really see the need from a Buildroot perspective to separate icsr and ifencei from the base BR2_RISCV_ISA_RVI instruction set.
So what I would do is exactly what OpenSBI has done in
https://github.com/riscv-software-src/opensbi/commit/5d53b55aa77ffeefd4012445dfa6ad3535e1ff2c:
when gcc >= 12 is used, we simply use -march with a _zicsr_zifencei suffix, and that's it.
I.e something like this:
arch/arch.mk.riscv +4 -0
@@ -27,4 +27,8 @@ ifeq ($(BR2_RISCV_ISA_RVC),y) GCC_TARGET_ARCH := $(GCC_TARGET_ARCH)c endif +ifeq ($(BR2_TOOLCHAIN_GCC_AT_LEAST_12),y) +GCC_TARGET_ARCH := $(GCC_TARGET_ARCH)_zicsr_zifencei +endif + endif
I don't think we will see RISC-V cores running Linux that support the base I instruction, but not the csr and fence extensions... as even OpenSBI assumes the RISC-V core supports them!
What do you think?
Thomas
Signature · 7 lines
Thomas Petazzoni, co-owner and CEO, Bootlin Embedded Linux and Kernel engineering and training https://bootlin.com _______________________________________________ buildroot mailing list [email protected] https://lists.buildroot.org/mailman/listinfo/buildroot
Hello Thomas,
Le 23/07/2022 à 14:39, Thomas Petazzoni via buildroot a écrit :
Quoted 99 lines
Hello Romain,
Quoted 27 lines · On Sun, 29 May 2022 15:18:01 +0200 Romain Naour <[email protected]> wrote:
Since gcc 12, default Riscv ISA spec version was bump to 20191213 [1].
This introduce a major incompatibility issue is the csr read/write (csrr*/csrw*) instructions and fence.i instruction has separated from the "I" extension, become two standalone extensions: Zicsr and Zifencei; so you might get error messages like that: unrecognized opcode "csrr" (or "fence.i").
Indeed, without Zifencei we can't build opensbi bootloader [3]:
opensbi-1.0/lib/sbi/sbi_tlb.c: Assembler messages:opensbi-1.0/lib/sbi/sbi_tlb.c:190: Error: unrecognized opcodefence.i', extensionzifencei' requiredAs a workaround, opensbi build system has been patched [4] to use -march=rv64imafdc_zicsr_zifencei when needed.
This workaround doesn't work in Buildroot due to the local patch 0001-Makefile-Don-t-specify-mabi-or-march.patch removing -march from CFLAGS.Fix this issue by introducing two additional
Kconfigoption enabling Zicsr and Zifencei standalone extensions for gcc >= 12 as recommanded by [2].Select Zicsr and Zifencei for General purpose (G) architecture variant (
BR2_riscv_g) since theses extentions were implicitely enabled previously.Quoted 7 lines
[1] https://gcc.gnu.org/gcc-12/changes.html
[2] https://groups.google.com/a/groups.riscv.org/g/sw-dev/c/aE1ZeHHCYf4
[3] https://github.com/riscv-software-src/opensbi/blob/v0.9/lib/sbi/sbi_tlb.c#L173
[4] https://github.com/riscv-software-src/opensbi/commit/5d53b55aa77ffeefd4012445dfa6ad3535e1ff2cSigned-off-by Romain Naour <[email protected]>Cc Mark Corbin <[email protected]>Thanks for the patch and explanation, but I'm wondering if this isn't perhaps too complicated.
Quoted 2 lines
(quoted diff) +2
-0+ select BR2_RISCV_ISA_RVZicsr if BR2_TOOLCHAIN_GCC_AT_LEAST_12 + select BR2_RISCV_ISA_RVZifencei if BR2_TOOLCHAIN_GCC_AT_LEAST_12[...]
Review config BR2_RISCV_ISA_CUSTOM_RVD · line 76
Quoted 13 lines
(quoted diff) +4
-1help - General purpose (G) is equivalent to IMAFD. + General purpose (G) is equivalent to IMAFD + (with Zicsr and Zifencei since gcc >= 12). config BR2_riscv_custom bool "Custom architecture"@@ -63,6 +72,16 @@ config BR2_RISCV_ISA_CUSTOM_RVD config BR2_RISCV_ISA_CUSTOM_RVC bool "Compressed Instructions (C)" select BR2_RISCV_ISA_RVC + +if BR2_TOOLCHAIN_GCC_AT_LEAST_12The problem with this is that options in
arch/now depend on the gcc version that is selected, which is something we have tried to avoid.Do a git grep
GCC_AT_LEASTinarch/, you will see that we only use "selectBR2_ARCH_NEEDS_GCC_AT_LEAST_xyz", i.e the choice of the architecture/variant/ABI is what drives the available gcc versions.And here you are doing exactly the opposite: it's the selected gcc version that drives the architecture/variant/ABI options that are available.
At this point, I don't really see the need from a Buildroot perspective to separate icsr and ifencei from the base
BR2_RISCV_ISA_RVIinstruction set.So what I would do is exactly what OpenSBI has done in
https://github.com/riscv-software-src/opensbi/commit/5d53b55aa77ffeefd4012445dfa6ad3535e1ff2c:
when gcc >= 12 is used, we simply use -march with a_zicsr_zifenceisuffix, and that's it.I.e something like this:
arch/arch.mk.riscv +4
-0@@ -27,4 +27,8 @@ ifeq ($(BR2_RISCV_ISA_RVC),y) GCC_TARGET_ARCH := $(GCC_TARGET_ARCH)c endif +ifeq ($(BR2_TOOLCHAIN_GCC_AT_LEAST_12),y) +GCC_TARGET_ARCH := $(GCC_TARGET_ARCH)_zicsr_zifencei +endif + endif
Quoted 5 lines
I don't think we will see RISC-V cores running Linux that support the base I instruction, but not the csr and fence extensions... as even OpenSBI assumes the RISC-V core supports them!
What do you think?
Thanks for the review!
I'm not familiar with the riscv instruction set, so I was not sure if we should support the possibility to disable Zicsr and Zifencei standalone extensions with gcc 12. AFIU, they are unconditionally enabled in previous gcc releases.
Additionally we have received two patches to add "vector operations extension"
http://patchwork.ozlabs.org/project/buildroot/patch/[email protected]/
http://patchwork.ozlabs.org/project/buildroot/patch/[email protected]/
But this requires gcc-12
http://patchwork.ozlabs.org/project/buildroot/patch/[email protected]/#2933736
Do you think "Vector Instructions (V)" deserve an option ? (I would say yes).
Best regards,
Romain
Quoted 2 lines
Thomas
_______________________________________________
buildroot mailing list
[email protected]
https://lists.buildroot.org/mailman/listinfo/buildroot
Hello,
Quoted 3 lines · On Sat, 23 Jul 2022 15:00:40 +0200 Romain Naour <[email protected]> wrote:
I'm not familiar with the riscv instruction set, so I was not sure if we should support the possibility to disable Zicsr and Zifencei standalone extensions with gcc 12. AFIU, they are unconditionally enabled in previous gcc releases.
Correct, which is why I think they will always be present in Linux-capable RISC-V cores. The reason I believe they were split from the I part of the ISA is because I is the base part of the ISA, and some people wanted to make RISC-V designs without those instructions, but not RISC-V designs targeting Linux.
So for now, I think we should simply make the assumption that all RISC-V cores will have those instructions.
If that turns out to be wrong in the future, we can always revisit that.
Quoted 11 lines
Additionally we have received two patches to add "vector operations extension"
http://patchwork.ozlabs.org/project/buildroot/patch/[email protected]/
http://patchwork.ozlabs.org/project/buildroot/patch/[email protected]/
But this requires gcc-12
http://patchwork.ozlabs.org/project/buildroot/patch/[email protected]/#2933736
Do you think "Vector Instructions (V)" deserve an option ? (I would say yes).
Yes, V definitely needs an option, as there will clearly be Linux-capable RISC-V cores that do not implement the V extension.
The option for the V extension will have to select BR2_ARCH_NEEDS_GCC_AT_LEAST_12.
Best regards,
Thomas
Signature · 7 lines
Thomas Petazzoni, co-owner and CEO, Bootlin Embedded Linux and Kernel engineering and training https://bootlin.com _______________________________________________ buildroot mailing list [email protected] https://lists.buildroot.org/mailman/listinfo/buildroot