Add gcc 9.1
v4 cover Add gcc 9.1
Before adding gcc 9.1 [1] to Buildroot, add some fixes for issues discovered by toolchain-builder [2] that are trigged by this new compiler version. (disable powerpc Werror, uClibc-ng asm sparc and ARC glibc).
Then add gcc 9.1 for internal and extenal toolchain backend and switch gcc 8.x as default compiler.
Thanks to Yann E. MORIN for providing a patch series that add new ARM cores added by gcc 9.1.
Each new ARM core has been build tested for aarch64:
https://gitlab.com/kubu93/buildroot/pipelines/60318953
Finally, enable the new ork1 (openrisc) that will replace in the following series the old gcc fork (gcc 5.x).
[1] https://gcc.gnu.org/gcc-9/changes.html
[2] https://gitlab.com/kubu93/toolchains-builder/pipelines/59794237
Romain Naour (9):
package/glibc/arc: fix build issue with gcc-9.1
linux: disable Werror for powerpc kernels
package/uclibc: remove asm constraint
toolchain: add gcc 9 entry
package/gcc: add support for gcc 9.1
arch: add BR2_ARCH_NEEDS_GCC_AT_LEAST_9
toolchain-external: add gcc 9 entry
package/gcc: switch to gcc 8.x as the default
package/gcc: enable gcc 9.1 for ork1 (openrisc)
Yann E. MORIN (6):
arch/arm: move dependency on 64-bit down to individual cores arch/arm: some cores have a different name with gcc-9 arch/arm: saphira is in fact an armv8.4a
arch/arm: add two new 64-bit-only armv8a cores arch/arm: add two new cortex-based armv8.2a cores arch/arm: add two new non-cortex-based armv8.2a cores
arch/Config.in | 4 + arch/Config.in.arm | 95 +++++++++-- linux/linux.mk | 2 + package/gcc/Config.in.host | 8 +- package/gcc/gcc.hash | 2 + package/gcc/gcc.mk | 4 +- ...Add-support-for-GCC-9-attribute-copy.patch | 160 ++++++++++++++++++ ...arc-syscalls.h-remove-asm-constraint.patch | 46 +++++ toolchain/Config.in | 5 + .../Config.in.options | 6 + 10 files changed, 313 insertions(+), 19 deletions(-) create mode 100644 package/glibc/arc-2018.09-release/0001-Add-support-for-GCC-9-attribute-copy.patch create mode 100644 package/uclibc/0003-RFC-sparc-syscalls.h-remove-asm-constraint.patch
Signature · 1 line
2.17.1
PATCH v4 1/15 package/glibc/arc: fix build issue with gcc-9.1
From: Romain Naour <[email protected]>
Backport from upstream [1] to fix build issue with glibc arc-2018.09-release (based on glibc 2.28) and gcc 9.1 [2]]
Using glibc arc-2.29-2019.02.08 or the upcoming arc-2019.03 release include this patch.
The upstream patch [1] modify mostly x86_64 files, but also two generic headers where the new __attribute_copy__() macro is defined *and* used to complement existing attribute-related macros.
So, it is not much the x86_64 part we're interested in, but the generic two headers, which changes percolate to the various functions via the various attribute macros...
[1] https://sourceware.org/git/?p=glibc.git;a=commit;h=1626a1cfcd6ba1cc64721be1036c2873211d499b
[2] https://gitlab.com/kubu93/toolchains-builder/-/jobs/205435682
---
v2: remove x86_64 part of the backported patch (Yann).
commit log (Yann)
--- ...Add-support-for-GCC-9-attribute-copy.patch | 160 ++++++++++++++++++ 1 file changed, 160 insertions(+) create mode 100644 package/glibc/arc-2018.09-release/0001-Add-support-for-GCC-9-attribute-copy.patch
package/glibc/arc-2018.09-release/0001-Add-support-for-GCC-9-attribute-copy.patch new file +160 -0
@@ -0,0 +1,160 @@ +From 3e0c7985c9535304d87af3ad728f5f6d34557f0d Mon Sep 17 00:00:00 2001 +From: Martin Sebor <[email protected]> +Date: Fri, 9 Nov 2018 17:24:12 -0700 +Subject: [PATCH] Add support for GCC 9 attribute copy. + +GCC 9 has gained an enhancement to help detect attribute mismatches +between alias declarations and their targets. It consists of a new +warning, -Wattribute-alias, an enhancement to an existing warning, +-Wmissing-attributes, and a new attribute called copy. + +The purpose of the warnings is to help identify either possible bugs +(an alias declared with more restrictive attributes than its target +promises) or optimization or diagnostic opportunities (an alias target +missing some attributes that it could be declared with that might +benefit analysis and code generation). The purpose of the new +attribute is to easily apply (almost) the same set of attributes +to one declaration as those already present on another. + +As expected (and intended) the enhancement triggers warnings for +many alias declarations in Glibc code. This change, tested on +x86_64-linux, avoids all instances of the new warnings by making +use of the attribute where appropriate. To fully benefit from +the enhancement Glibc will need to be compiled with + -Wattribute-alias=2 and remaining warnings reviewed and dealt with +(there are a couple of thousand but most should be straightforward +to deal with). + +ChangeLog: + + * include/libc-symbols.h (__attribute_copy__): Define macro unless + it's already defined. + (_strong_alias): Use __attribute_copy__. + (_weak_alias, __hidden_ver1, __hidden_nolink2): Same. + * misc/sys/cdefs.h (__attribute_copy__): New macro. + * sysdeps/x86_64/multiarch/memchr.c (memchr): Use __attribute_copy__. + * sysdeps/x86_64/multiarch/memcmp.c (memcmp): Same. + * sysdeps/x86_64/multiarch/mempcpy.c (mempcpy): Same. + * sysdeps/x86_64/multiarch/memset.c (memset): Same. + * sysdeps/x86_64/multiarch/stpcpy.c (stpcpy): Same. + * sysdeps/x86_64/multiarch/strcat.c (strcat): Same. + * sysdeps/x86_64/multiarch/strchr.c (strchr): Same. + * sysdeps/x86_64/multiarch/strcmp.c (strcmp): Same. + * sysdeps/x86_64/multiarch/strcpy.c (strcpy): Same. + * sysdeps/x86_64/multiarch/strcspn.c (strcspn): Same. + * sysdeps/x86_64/multiarch/strlen.c (strlen): Same. + * sysdeps/x86_64/multiarch/strncmp.c (strncmp): Same. + * sysdeps/x86_64/multiarch/strncpy.c (strncpy): Same. + * sysdeps/x86_64/multiarch/strnlen.c (strnlen): Same. + * sysdeps/x86_64/multiarch/strpbrk.c (strpbrk): Same. + * sysdeps/x86_64/multiarch/strrchr.c (strrchr): Same. + * sysdeps/x86_64/multiarch/strspn.c (strspn): Same. + +Using glibc arc-2.29-2019.02.08 or the upcoming arc-2019.03 release include this patch. + +[1] https://sourceware.org/git/?p=glibc.git;a=commit;h=1626a1cfcd6ba1cc64721be1036c2873211d499b +[2] https://gitlab.com/kubu93/toolchains-builder/-/jobs/205435682 + +Signed-off-by: Romain Naour <[email protected]> +[Romain: + - Backport from upstream [1] to fix build issue with glibc arc-2018.09-release + (based on glibc 2.28) and gcc 9.1 [2] + - Also drop the x86-64 related hunks to avoid confusion. +] +Cc: Alexey Brodkin <[email protected]> +Cc: Evgeniy Didin <[email protected]> +Cc: Eugeniy Paltsev <[email protected]> +--- + include/libc-symbols.h | 20 +++++++++++++++----- + misc/sys/cdefs.h | 10 ++++++++++ + 2 files changed, 25 insertions(+), 5 deletions(-) + +diff --git a/include/libc-symbols.h b/include/libc-symbols.h +index 8b9273c13a..e71a479e84 100644 +--- a/include/libc-symbols.h ++++ b/include/libc-symbols.h +@@ -125,6 +125,11 @@ + # define ASM_LINE_SEP ; + #endif + ++#ifndef __attribute_copy__ ++/* Provide an empty definition when cdefs.h is not included. */ ++# define __attribute_copy__(arg) ++#endif ++ + #ifndef __ASSEMBLER__ + /* GCC understands weak symbols and aliases; use its interface where + possible, instead of embedded assembly language. */ +@@ -132,7 +137,8 @@ + /* Define ALIASNAME as a strong alias for NAME. */ + # define strong_alias(name, aliasname) _strong_alias(name, aliasname) + # define _strong_alias(name, aliasname) \ +- extern __typeof (name) aliasname __attribute__ ((alias (#name))); ++ extern __typeof (name) aliasname __attribute__ ((alias (#name))) \ ++ __attribute_copy__ (name); + + /* This comes between the return type and function name in + a function definition to make that definition weak. */ +@@ -143,14 +149,16 @@ + If weak aliases are not available, this defines a strong alias. */ + # define weak_alias(name, aliasname) _weak_alias (name, aliasname) + # define _weak_alias(name, aliasname) \ +- extern __typeof (name) aliasname __attribute__ ((weak, alias (#name))); ++ extern __typeof (name) aliasname __attribute__ ((weak, alias (#name))) \ ++ __attribute_copy__ (name); + + /* Same as WEAK_ALIAS, but mark symbol as hidden. */ + # define weak_hidden_alias(name, aliasname) \ + _weak_hidden_alias (name, aliasname) + # define _weak_hidden_alias(name, aliasname) \ + extern __typeof (name) aliasname \ +- __attribute__ ((weak, alias (#name), __visibility__ ("hidden"))); ++ __attribute__ ((weak, alias (#name), __visibility__ ("hidden"))) \ ++ __attribute_copy__ (name); + + /* Declare SYMBOL as weak undefined symbol (resolved to 0 if not defined). */ + # define weak_extern(symbol) _weak_extern (weak symbol) +@@ -532,7 +540,8 @@ for linking") + # define __hidden_ver1(local, internal, name) \ + extern __typeof (name) __EI_##name __asm__(__hidden_asmname (#internal)); \ + extern __typeof (name) __EI_##name \ +- __attribute__((alias (__hidden_asmname (#local)))) ++ __attribute__((alias (__hidden_asmname (#local)))) \ ++ __attribute_copy__ (name) + # define hidden_ver(local, name) __hidden_ver1(local, __GI_##name, name); + # define hidden_data_ver(local, name) hidden_ver(local, name) + # define hidden_def(name) __hidden_ver1(__GI_##name, name, name); +@@ -545,7 +554,8 @@ for linking") + # define __hidden_nolink1(local, internal, name, version) \ + __hidden_nolink2 (local, internal, name, version) + # define __hidden_nolink2(local, internal, name, version) \ +- extern __typeof (name) internal __attribute__ ((alias (#local))); \ ++ extern __typeof (name) internal __attribute__ ((alias (#local))) \ ++ __attribute_copy__ (name); \ + __hidden_nolink3 (local, internal, #name "@" #version) + # define __hidden_nolink3(local, internal, vername) \ + __asm__ (".symver " #internal ", " vername); +diff --git a/misc/sys/cdefs.h b/misc/sys/cdefs.h +index 3f6fe3cc85..8d585680a6 100644 +--- a/misc/sys/cdefs.h ++++ b/misc/sys/cdefs.h +@@ -431,6 +431,16 @@ + # define __attribute_nonstring__ + #endif + ++/* Undefine (also defined in libc-symbols.h). */ ++#undef __attribute_copy__ ++#if __GNUC_PREREQ (9, 0) ++/* Copies attributes from the declaration or type referenced by ++ the argument. */ ++# define __attribute_copy__(arg) __attribute__ ((__copy__ (arg))) ++#else ++# define __attribute_copy__(arg) ++#endif ++ + #if (!defined _Static_assert && !defined __cplusplus \ + && (defined __STDC_VERSION__ ? __STDC_VERSION__ : 0) < 201112 \ + && (!__GNUC_PREREQ (4, 6) || defined __STRICT_ANSI__)) +-- +2.14.5 +
Signature · 1 line
2.17.1
PATCH v4 1/15 package/glibc/arc: fix build issue with gcc-9.1
Quoted 7 lines · On Thu, 20 Jun 2019 12:07:11 +0200 Giulio Benetti…
From: Romain Naour <[email protected]>
Backport from upstream [1] to fix build issue with glibc arc-2018.09-release (based on glibc 2.28) and gcc 9.1 [2]]
Using glibc arc-2.29-2019.02.08 or the upcoming arc-2019.03 release include this patch.
Since then, we have indeed updated ARC glibc's version to 2019.03, so this patch is no longer necessary if I understood correctly. I marked it as "Not Applicable" in patchwork.
Thanks,
Thomas
Signature · 3 lines
Thomas Petazzoni, CTO, Bootlin Embedded Linux and Kernel engineering https://bootlin.com
PATCH v4 2/15 linux: disable Werror for powerpc kernels
In reply to Giulio BenettiFrom: Romain Naour <[email protected]>
From patch [1] included in kernel >= 5.0:
"The upcoming GCC 9 release extends the -Wmissing-attributes warnings (enabled by -Wall) to C and aliases: it warns when particular function attributes are missing in the aliases but not in their target.
In particular, it triggers for all the init/cleanup_module aliases in the kernel (defined by the module_init/exit macros), ending up being very noisy.
These aliases point to the __init/__exit functions of a module, which are defined as __cold (among other attributes). However, the aliases themselves do not have the __cold attribute.
Since the compiler behaves differently when compiling a __cold function as well as when compiling paths leading to calls to __cold functions, the warning is trying to point out the possibly-forgotten attribute in the alias."
Werror is set by default while building ppc kernel [2], but some warning can be introduced while building current kernel with newer compiler (for example building kernel 4.19 with gcc 9.1).
For the same reason why we remove Werror in packages's compiler flags. Building with Werror is not bulletproof when we start using a newer compiler that introduce new warnings. This is the case here.
Also this option is a bit strange since it's specific to ppc kernels:
"The intention is to make it harder for people to inadvertantly introduce warnings in the arch/powerpc code."
Other kernel developers on other arch may be interested by a similar/more generic option.
So, It's clearly intended for kernel developers.
Instead of backporting this patch [1] to kernel 4.19, select unconditionally the Kconfig option CONFIG_PPC_DISABLE_WERROR that allow to disable Werror.
Fixes:
https://gitlab.com/kubu93/toolchains-builder/-/jobs/205435741
[1] https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/commit/?h=a6e60d84989fa0e91db7f236eda40453b0e44afa
[2] https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/commit/?h=ba55bd74360ea4b8b95e73ed79474d37ff482b36
[3] https://gitlab.com/bootlin/toolchains-builder
---
v2: add a shared fragment in board/fragments/linux and update
qemu ppc* defconfigs. (Arnout)
v3: commit log (Yann)
enable the option from linux.mk instead of the defconfig.
--- linux/linux.mk | 2 ++ 1 file changed, 2 insertions(+)
linux/linux.mk +2 -0
@@ -316,6 +316,8 @@ define LINUX_KCONFIG_FIXUP_CMDS $(LINUX_FIXUP_CONFIG_ENDIANNESS) $(if $(BR2_arm)$(BR2_armeb), $(call KCONFIG_ENABLE_OPT,CONFIG_AEABI,$(@D)/.config)) + $(if $(BR2_powerpc)$(BR2_powerpc64)$(BR2_powerpc64le), + $(call KCONFIG_ENABLE_OPT,CONFIG_PPC_DISABLE_WERROR,$(@D)/.config)) $(if $(BR2_TARGET_ROOTFS_CPIO), $(call KCONFIG_ENABLE_OPT,CONFIG_BLK_DEV_INITRD,$(@D)/.config)) # As the kernel gets compiled before root filesystems are
Signature · 1 line
2.17.1
Thomas Petazzoni
AppliedPATCH v4 2/15 linux: disable Werror for powerpc kernels
Quoted 48 lines · On Thu, 20 Jun 2019 12:07:12 +0200 Giulio Benetti…
From: Romain Naour <[email protected]>
From patch [1] included in kernel >= 5.0:
"The upcoming GCC 9 release extends the -Wmissing-attributes warnings (enabled by -Wall) to C and aliases: it warns when particular function attributes are missing in the aliases but not in their target.In particular, it triggers for all the
init/cleanup_modulealiases in the kernel (defined by the module_init/exit macros), ending up being very noisy.These aliases point to the __init/__exit functions of a module, which are defined as
__cold(among other attributes). However, the aliases themselves do not have the__coldattribute.Since the compiler behaves differently when compiling a
__coldfunction as well as when compiling paths leading to calls to__coldfunctions, the warning is trying to point out the possibly-forgotten attribute in the alias."Werror is set by default while building ppc kernel [2], but some warning can be introduced while building current kernel with newer compiler (for example building kernel 4.19 with gcc 9.1).
For the same reason why we remove Werror in packages's compiler flags. Building with Werror is not bulletproof when we start using a newer compiler that introduce new warnings. This is the case here.
Also this option is a bit strange since it's specific to ppc kernels:
"The intention is to make it harder for people to inadvertantly introduce warnings in thearch/powerpccode."
Other kernel developers on other arch may be interested by a similar/more generic option.So, It's clearly intended for kernel developers.
Instead of backporting this patch [1] to kernel 4.19, select unconditionally the
KconfigoptionCONFIG_PPC_DISABLE_WERRORthat allow to disable Werror.Fixes:
https://gitlab.com/kubu93/toolchains-builder/-/jobs/205435741[1] https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/commit/?h=a6e60d84989fa0e91db7f236eda40453b0e44afa
[2] https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/commit/?h=ba55bd74360ea4b8b95e73ed79474d37ff482b36
[3] https://gitlab.com/bootlin/toolchains-builder
Quoted 6 lines
Fix-suggested-by Yann E. MORIN <[email protected]>Signed-off-by Romain Naour <[email protected]>Signed-off-by Giulio Benetti <[email protected]>---
v2: add a shared fragment in board/fragments/linux and update
qemu ppc* defconfigs. (Arnout)
Applied to master, thanks.
Thomas
Signature · 3 lines
Thomas Petazzoni, CTO, Bootlin Embedded Linux and Kernel engineering https://bootlin.com
PATCH v4 2/15 linux: disable Werror for powerpc kernels
In reply to Giulio BenettiQuoted 52 lines
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
"Giulio" == Giulio Benetti <[email protected]> writes:
From: Romain Naour <[email protected]>
From patch [1] included in kernel >= 5.0:
"The upcoming GCC 9 release extends the -Wmissing-attributes warnings (enabled by -Wall) to C and aliases: it warns when particular function attributes are missing in the aliases but not in their target.In particular, it triggers for all the
init/cleanup_modulealiases in the kernel (defined by the module_init/exit macros), ending up being very noisy.These aliases point to the __init/__exit functions of a module, which are defined as
__cold(among other attributes). However, the aliases themselves do not have the__coldattribute.Since the compiler behaves differently when compiling a
__coldfunction as well as when compiling paths leading to calls to__coldfunctions, the warning is trying to point out the possibly-forgotten attribute in the alias."Werror is set by default while building ppc kernel [2], but some warning can be introduced while building current kernel with newer compiler (for example building kernel 4.19 with gcc 9.1).
For the same reason why we remove Werror in packages's compiler flags. Building with Werror is not bulletproof when we start using a newer compiler that introduce new warnings. This is the case here.
Also this option is a bit strange since it's specific to ppc kernels:
"The intention is to make it harder for people to inadvertantly introduce warnings in thearch/powerpccode."
Other kernel developers on other arch may be interested by a similar/more generic option.So, It's clearly intended for kernel developers.
Instead of backporting this patch [1] to kernel 4.19, select unconditionally the
KconfigoptionCONFIG_PPC_DISABLE_WERRORthat allow to disable Werror.Fixes:
https://gitlab.com/kubu93/toolchains-builder/-/jobs/205435741[1] https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/commit/?h=a6e60d84989fa0e91db7f236eda40453b0e44afa
[2] https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/commit/?h=ba55bd74360ea4b8b95e73ed79474d37ff482b36
[3] https://gitlab.com/bootlin/toolchains-builderFix-suggested-by Yann E. MORIN <[email protected]>Signed-off-by Romain Naour <[email protected]>Signed-off-by Giulio Benetti <[email protected]>
Quoted 6 lines
---
v2: add a shared fragment in board/fragments/linux and update
qemu ppc* defconfigs. (Arnout)v3: commit log (Yann)
enable the option from linux.mk instead of the defconfig.
Committed to 2019.02.x and 2019.05.x, thanks.
Signature · 1 line
Bye, Peter Korsgaard
PATCH v4 2/15 linux: disable Werror for powerpc kernels
Hi Peter,
Quoted 56 lines · Peter Korsgaard
Quoted 55 lines
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
"Giulio" == Giulio Benetti <[email protected]> writes:
From: Romain Naour <[email protected]>
From patch [1] included in kernel >= 5.0:
"The upcoming GCC 9 release extends the -Wmissing-attributes warnings (enabled by -Wall) to C and aliases: it warns when particular function attributes are missing in the aliases but not in their target.In particular, it triggers for all the
init/cleanup_modulealiases in the kernel (defined by the module_init/exit macros), ending up being very noisy.These aliases point to the __init/__exit functions of a module, which are defined as
__cold(among other attributes). However, the aliases themselves do not have the__coldattribute.Since the compiler behaves differently when compiling a
__coldfunction as well as when compiling paths leading to calls to__coldfunctions, the warning is trying to point out the possibly-forgotten attribute in the alias."Werror is set by default while building ppc kernel [2], but some warning can be introduced while building current kernel with newer compiler (for example building kernel 4.19 with gcc 9.1).
For the same reason why we remove Werror in packages's compiler flags. Building with Werror is not bulletproof when we start using a newer compiler that introduce new warnings. This is the case here.
Also this option is a bit strange since it's specific to ppc kernels:
"The intention is to make it harder for people to inadvertantly introduce warnings in thearch/powerpccode."
Other kernel developers on other arch may be interested by a similar/more generic option.So, It's clearly intended for kernel developers.
Instead of backporting this patch [1] to kernel 4.19, select unconditionally the
KconfigoptionCONFIG_PPC_DISABLE_WERRORthat allow to disable Werror.Fixes:
https://gitlab.com/kubu93/toolchains-builder/-/jobs/205435741[1] https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/commit/?h=a6e60d84989fa0e91db7f236eda40453b0e44afa
[2] https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/commit/?h=ba55bd74360ea4b8b95e73ed79474d37ff482b36
[3] https://gitlab.com/bootlin/toolchains-builderFix-suggested-by Yann E. MORIN <[email protected]>Signed-off-by Romain Naour <[email protected]>Signed-off-by Giulio Benetti <[email protected]>---
v2: add a shared fragment in board/fragments/linux and update
qemu ppc* defconfigs. (Arnout)
Quoted 4 lines
Quoted 2 lines
v3: commit log (Yann)
enable the option from linux.mk instead of the defconfig.Committed to 2019.02.x and 2019.05.x, thanks.
It is in 2019.02.x, but not in 2019.05.x as of commit 1818dae1e0.
baruch
Signature · 3 lines
http://baruch.siach.name/blog/ ~. .~ Tk Open Systems
=}------------------------------------------------ooO--U--Ooo------------{=
- baruch at tkos.co.il - tel: +972.52.368.4656, http://www.tkos.co.il -PATCH v4 2/15 linux: disable Werror for powerpc kernels
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
"Baruch" == Baruch Siach <[email protected]> writes:
Hi,
Quoted 3 lines
Quoted 1 line
Committed to 2019.02.x and 2019.05.x, thanks.
It is in 2019.02.x, but not in 2019.05.x as of commit
1818dae1e0.
Ups, indeed - Fixed now, thanks!
Signature · 1 line
Bye, Peter Korsgaard
PATCH v4 3/15 package/uclibc: remove asm constraint
In reply to Giulio BenettiFrom: Romain Naour <[email protected]>
uClibc-ng don't build with gcc 9.1 [1] due to a new check that "catch illegal asm constraint usage" [2].
gcc 9.1 print this error:
"invalid hard register usage between earlyclobber operand and input operand"
The asm constraint is present in uClibc since it support sparc (back in 2002)[3].
Note: There is no such constraint is Glibc counterpart code [4].
[1] https://gitlab.com/kubu93/toolchains-builder/-/jobs/205435757
[2] https://github.com/gcc-mirror/gcc/commit/b782636f28f5c378897c238081d28d7a4a6ca578
[3] https://cgit.uclibc-ng.org/cgi/cgit/uclibc-ng.git/commit/?id=3b6d086531102b6d09ce852feb1e370d5dca3ce9
[4] https://sourceware.org/git/?p=glibc.git;a=blob;f=sysdeps/unix/sysv/linux/sparc/sysdep.h;h=981b2a26b7a91093f821c97876e55bc4be2d9f8a;hb=HEAD
---
I did not fully analyze the asm code to understand why the constraint is present. I lightly tested using qemu_sparc_ss10_defconfig.
Any help is welcome(Romain)
--- ...arc-syscalls.h-remove-asm-constraint.patch | 46 +++++++++++++++++++ 1 file changed, 46 insertions(+) create mode 100644 package/uclibc/0003-RFC-sparc-syscalls.h-remove-asm-constraint.patch
package/uclibc/0003-RFC-sparc-syscalls.h-remove-asm-constraint.patch new file +46 -0
@@ -0,0 +1,46 @@ +From 1854c332efa0ed5b9140d8df2a983625a1e2d9ef Mon Sep 17 00:00:00 2001 +From: Romain Naour <[email protected]> +Date: Sun, 5 May 2019 10:56:30 +0200 +Subject: [PATCH] [RFC] sparc: syscalls.h: remove asm constraint + +uClibc-ng don't build with gcc 9.1 [1] due to a new check that +"catch illegal asm constraint usage" [2]. + +gcc 9.1 print this error: +"invalid hard register usage between earlyclobber operand and input operand" + +The asm constraint is present in uClibc since it support sparc (back in 2002)[3]. + +Note: There is no such constraint is Glibc counterpart code [4]. + +[1] https://gitlab.com/kubu93/toolchains-builder/-/jobs/205435757 +[2] https://github.com/gcc-mirror/gcc/commit/b782636f28f5c378897c238081d28d7a4a6ca578 +[3] https://cgit.uclibc-ng.org/cgi/cgit/uclibc-ng.git/commit/?id=3b6d086531102b6d09ce852feb1e370d5dca3ce9 +[4] https://sourceware.org/git/?p=glibc.git;a=blob;f=sysdeps/unix/sysv/linux/sparc/sysdep.h;h=981b2a26b7a91093f821c97876e55bc4be2d9f8a;hb=HEAD + +Signed-off-by: Romain Naour <[email protected]> +Cc: Waldemar Brodkorb <[email protected]> +--- +I did not fully analyze the asm code to understand why the constraint is present. +I lightly tested using qemu_sparc_ss10_defconfig. +Any help is welcome :) +--- + libc/sysdeps/linux/sparc/bits/syscalls.h | 2 +- + 1 file changed, 1 insertion(+), 1 deletion(-) + +diff --git a/libc/sysdeps/linux/sparc/bits/syscalls.h b/libc/sysdeps/linux/sparc/bits/syscalls.h +index 75af7a1573..e52ab98466 100644 +--- a/libc/sysdeps/linux/sparc/bits/syscalls.h ++++ b/libc/sysdeps/linux/sparc/bits/syscalls.h +@@ -33,7 +33,7 @@ + register long __g1 __asm__("g1") = sys_num; \ + LOAD_ARGS_##nr(args) \ + __asm__ __volatile__( __SYSCALL_STRING \ +- : "=r" (__res), "=&r" (__o0) \ ++ : "=r" (__res), "=r" (__o0) \ + : "1" (__o0) ASM_ARGS_##nr, "r" (__g1) \ + : __SYSCALL_CLOBBERS ); \ + } \ +-- +2.14.5 +
Signature · 1 line
2.17.1
PATCH v4 3/15 package/uclibc: remove asm constraint
Quoted 20 lines · On Thu, 20 Jun 2019 12:07:13 +0200 Giulio Benetti…
From: Romain Naour <[email protected]>
uClibc-ng don't build with gcc 9.1 [1] due to a new check that "catch illegal asm constraint usage" [2].
gcc 9.1 print this error:
"invalid hard register usage between earlyclobber operand and input operand"The asm constraint is present in uClibc since it support sparc (back in 2002)[3].
Note: There is no such constraint is Glibc counterpart code [4].
[1] https://gitlab.com/kubu93/toolchains-builder/-/jobs/205435757
[2] https://github.com/gcc-mirror/gcc/commit/b782636f28f5c378897c238081d28d7a4a6ca578
[3] https://cgit.uclibc-ng.org/cgi/cgit/uclibc-ng.git/commit/?id=3b6d086531102b6d09ce852feb1e370d5dca3ce9
[4] https://sourceware.org/git/?p=glibc.git;a=blob;f=sysdeps/unix/sysv/linux/sparc/sysdep.h;h=981b2a26b7a91093f821c97876e55bc4be2d9f8a;hb=HEADSigned-off-by Romain Naour <[email protected]>Cc Waldemar Brodkorb <[email protected]>Signed-off-by Giulio Benetti <[email protected]>
Quoted 8 lines
---
I did not fully analyze the asm code to understand why the constraint is present. I lightly tested usingqemu_sparc_ss10_defconfig.
Any help is welcome(Romain)--- ...arc-syscalls.h-remove-asm-constraint.patch | 46 +++++++++++++++++++ 1 file changed, 46 insertions(+) create mode 100644 package/uclibc/0003-RFC-sparc-syscalls.h-remove-asm-constraint.patch
Since the patch was applied upstream in uclibc-ng, I backported the uclibc-ng patch. Of course the code is the same, but the file name was a bit different. Also, I updated the commit log to indicate that we're talking about a sparc issue.
Applied with those changes. Thanks!
Thomas
Signature · 3 lines
Thomas Petazzoni, CTO, Bootlin Embedded Linux and Kernel engineering https://bootlin.com
PATCH v4 4/15 toolchain: add gcc 9 entry
In reply to Giulio BenettiFrom: Romain Naour <[email protected]>
In order to add gcc 9 support for internal and external toolchain in follow-up commits, introduce BR2_TOOLCHAIN_GCC_AT_LEAST_9 symbol.
--- toolchain/Config.in | 5 +++++ 1 file changed, 5 insertions(+)
toolchain/Config.in +5 -0
@@ -507,10 +507,15 @@ config BR2_TOOLCHAIN_GCC_AT_LEAST_8 bool select BR2_TOOLCHAIN_GCC_AT_LEAST_7 +config BR2_TOOLCHAIN_GCC_AT_LEAST_9 + bool + select BR2_TOOLCHAIN_GCC_AT_LEAST_8 + # This order guarantees that the highest version is set, as kconfig # stops affecting a value on the first matching default. config BR2_TOOLCHAIN_GCC_AT_LEAST string + default "9" if BR2_TOOLCHAIN_GCC_AT_LEAST_9 default "8" if BR2_TOOLCHAIN_GCC_AT_LEAST_8 default "7" if BR2_TOOLCHAIN_GCC_AT_LEAST_7 default "6" if BR2_TOOLCHAIN_GCC_AT_LEAST_6
Signature · 1 line
2.17.1
Thomas Petazzoni
AppliedPATCH v4 4/15 toolchain: add gcc 9 entry
Quoted 10 lines · On Thu, 20 Jun 2019 12:07:14 +0200 Giulio Benetti…
From: Romain Naour <[email protected]>
In order to add gcc 9 support for internal and external toolchain in follow-up commits, introduce
BR2_TOOLCHAIN_GCC_AT_LEAST_9symbol.Signed-off-by Romain Naour <[email protected]>Signed-off-by Giulio Benetti <[email protected]>--- toolchain/Config.in | 5 +++++ 1 file changed, 5 insertions(+)
Applied to master, thanks.
Thomas
Signature · 3 lines
Thomas Petazzoni, CTO, Bootlin Embedded Linux and Kernel engineering https://bootlin.com
PATCH v4 5/15 package/gcc: add support for gcc 9.1
In reply to Giulio BenettiFrom: Romain Naour <[email protected]>
---
v2: remove !BR2_ARCH_NEEDS_GCC_AT_LEAST_9 from gcc 8.x
This symbol will be introduced by the next patch (Yann)
v3: rebase patch on top of master branch sha1:
04efb17c863606cdbc2405d01c3d48d6868c5245 (Giulio)
v4: rebase patch on top of master branch sha1:
b34ead55c06cad5b6443da358cd63e9c71265d69 (Giulio)
---
package/gcc/Config.in.host | 7 +++++++
package/gcc/gcc.hash | 2 ++
2 files changed, 9 insertions(+)package/gcc/Config.in.host +7 -0
@@ -83,6 +83,12 @@ config BR2_GCC_VERSION_8_X depends on !BR2_powerpc_SPE select BR2_TOOLCHAIN_GCC_AT_LEAST_8 +config BR2_GCC_VERSION_9_X + bool "gcc 9.x" + # Broken or unsupported architectures + depends on !BR2_or1k + select BR2_TOOLCHAIN_GCC_AT_LEAST_9 + endchoice # libcilkrts was introduced in gcc 4.9 (oldest gcc version we
@@ -98,6 +104,7 @@ config BR2_GCC_VERSION default "6.5.0" if BR2_GCC_VERSION_6_X default "7.4.0" if BR2_GCC_VERSION_7_X default "8.3.0" if BR2_GCC_VERSION_8_X + default "9.1.0" if BR2_GCC_VERSION_9_X default "arc-2019.03-release" if BR2_GCC_VERSION_ARC default "or1k-musl-5.4.0-20170218" if BR2_GCC_VERSION_OR1K default "48152afb96c59733d5bc79e3399bb7b3d4b44266" if BR2_GCC_VERSION_CSKY
package/gcc/gcc.hash +2 -0
@@ -8,6 +8,8 @@ sha512 ce046f9a50050fd54b870aab764f7db187fe7ea92eb4aaffb7c3689ca623755604e231f2 sha512 8864d8e4b97c2e1a4f17422f6e68120172ebefeab97b1757734f7185ca68a6b9a89011c6833c03fa454c17b0ac35b15e1d284881e6971035948ac6100f3aa45e gcc-7.4.0.tar.xz # From ftp://gcc.gnu.org/pub/gcc/releases/gcc-8.3.0/sha512.sum sha512 1811337ae3add9680cec64968a2509d085b6dc5b6783fc1e8c295e3e47416196fd1a3ad8dfe7e10be2276b4f62c357659ce2902f239f60a8648548231b4b5802 gcc-8.3.0.tar.xz +# From ftp://gcc.gnu.org/pub/gcc/releases/gcc-9.1.0/sha512.sum +sha512 b6134df027e734cee5395afd739fcfa4ea319a6017d662e54e89df927dea19d3fff7a6e35d676685383034e3db01c9d0b653f63574c274eeb15a2cb0bc7a1f28 gcc-9.1.0.tar.xz # Locally calculated (fetched from Github) sha512 c3f63fb601140f92d2dbef96b2dcc025ef754d2f049bfcc3d33e746c7b218b5138f2d8768cd580134858a8d347d234cf3ead276638f6de409f8f07c986e136b6 gcc-arc-2019.03-release.tar.gz
Signature · 1 line
2.17.1
PATCH v4 5/15 package/gcc: add support for gcc 9.1
package/gcc/Config.in.host config BR2_GCC_VERSION_8_X · line 89
Quoted 4 lines · On Thu, 20 Jun 2019 12:07:15 +0200 Giulio Benetti…
package/gcc/Config.in.host +4
-0+config BR2_GCC_VERSION_9_X + bool "gcc 9.x" + # Broken or unsupported architectures + depends on !BR2_or1k
gcc 8.x also had a depends on !BR2_csky, but I see that C-SKY support has been merged in gcc 9.x, so I guess this one is not needed.
However, the depends !BR2_powerpc_spe that gcc 8.x has seemed necessary, so I've added it for gcc 9.x as well.
Applied with this change. Thanks!
Thomas
Signature · 3 lines
Thomas Petazzoni, CTO, Bootlin Embedded Linux and Kernel engineering https://bootlin.com
PATCH v4 6/15 arch: add BR2_ARCH_NEEDS_GCC_AT_LEAST_9
In reply to Giulio BenettiFrom: Romain Naour <[email protected]>
This new symbol will be used by architectures introduced with gcc 9 and by external toolchains based on gcc 9.
[1] https://gcc.gnu.org/gcc-9/changes.html
---
v3: rebase patch on top of master branch sha1:
04efb17c863606cdbc2405d01c3d48d6868c5245 (Giulio) --- arch/Config.in | 4 ++++ package/gcc/Config.in.host | 1 + 2 files changed, 5 insertions(+)
arch/Config.in +4 -0
@@ -301,6 +301,10 @@ config BR2_ARCH_NEEDS_GCC_AT_LEAST_8 bool select BR2_ARCH_NEEDS_GCC_AT_LEAST_7 +config BR2_ARCH_NEEDS_GCC_AT_LEAST_9 + bool + select BR2_ARCH_NEEDS_GCC_AT_LEAST_8 + # The following string values are defined by the individual # Config.in.$ARCH files config BR2_ARCH
package/gcc/Config.in.host +1 -0
@@ -75,6 +75,7 @@ config BR2_GCC_VERSION_7_X config BR2_GCC_VERSION_8_X bool "gcc 8.x" + depends on !BR2_ARCH_NEEDS_GCC_AT_LEAST_9 # Broken or unsupported architectures depends on !BR2_csky depends on !BR2_or1k
Signature · 1 line
2.17.1
PATCH v4 6/15 arch: add BR2_ARCH_NEEDS_GCC_AT_LEAST_9
Quoted 9 lines · On Thu, 20 Jun 2019 12:07:16 +0200 Giulio Benetti…
From: Romain Naour <[email protected]>
This new symbol will be used by architectures introduced with gcc 9 and by external toolchains based on gcc 9.
[1] https://gcc.gnu.org/gcc-9/changes.html
Signed-off-by Romain Naour <[email protected]>Signed-off-by Giulio Benetti <[email protected]>
Quoted 7 lines
---
v3: rebase patch on top of master branch sha1:04efb17c863606cdbc2405d01c3d48d6868c5245 (Giulio) --- arch/Config.in | 4 ++++ package/gcc/Config.in.host | 1 + 2 files changed, 5 insertions(+)
Applied to master, thanks. Shouldn't we be using this a bit more ?
I believe C-SKY could select BR2_ARCH_NEEDS_GCC_AT_LEAST_9, and ARC could select BR2_ARCH_NEEDS_GCC_AT_LEAST_8, this would allow to remove a few "depends !" from package/gcc/Config.in.host.
Best regards,
Thomas
Signature · 3 lines
Thomas Petazzoni, CTO, Bootlin Embedded Linux and Kernel engineering https://bootlin.com
PATCH v4 6/15 arch: add BR2_ARCH_NEEDS_GCC_AT_LEAST_9
Quoted 7 lines
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
"Thomas" == Thomas Petazzoni <[email protected]> writes:
Applied to master, thanks. Shouldn't we be using this a bit more ?
I believe C-SKY could select
BR2_ARCH_NEEDS_GCC_AT_LEAST_9, and ARC could selectBR2_ARCH_NEEDS_GCC_AT_LEAST_8, this would allow to remove a few "depends !" from package/gcc/Config.in.host.
Yes, that could be a nice cleanup - Even though we don't have a lot of such conditionals.
Signature · 1 line
Bye, Peter Korsgaard
PATCH v4 7/15 toolchain-external: add gcc 9 entry
In reply to Giulio BenettiFrom: Romain Naour <[email protected]>
This patch allows to use an external toolchain based on gcc 9.
--- .../toolchain-external-custom/Config.in.options | 6 ++++++ 1 file changed, 6 insertions(+)
toolchain/toolchain-external/toolchain-external-custom/Config.in.options +6 -0
@@ -27,6 +27,7 @@ config BR2_TOOLCHAIN_EXTERNAL_CUSTOM_PREFIX choice bool "External toolchain gcc version" + default BR2_TOOLCHAIN_EXTERNAL_GCC_9 if BR2_ARCH_NEEDS_GCC_AT_LEAST_9 default BR2_TOOLCHAIN_EXTERNAL_GCC_8 if BR2_ARCH_NEEDS_GCC_AT_LEAST_8 default BR2_TOOLCHAIN_EXTERNAL_GCC_7 if BR2_ARCH_NEEDS_GCC_AT_LEAST_7 default BR2_TOOLCHAIN_EXTERNAL_GCC_6 if BR2_ARCH_NEEDS_GCC_AT_LEAST_6
@@ -38,8 +39,13 @@ choice Set to the gcc version that is used by your external toolchain. +config BR2_TOOLCHAIN_EXTERNAL_GCC_9 + bool "9.x" + select BR2_TOOLCHAIN_GCC_AT_LEAST_9 + config BR2_TOOLCHAIN_EXTERNAL_GCC_8 bool "8.x" + depends on !BR2_ARCH_NEEDS_GCC_AT_LEAST_9 select BR2_TOOLCHAIN_GCC_AT_LEAST_8 config BR2_TOOLCHAIN_EXTERNAL_GCC_7
Signature · 1 line
2.17.1
Thomas Petazzoni
AppliedPATCH v4 7/15 toolchain-external: add gcc 9 entry
Quoted 9 lines · On Thu, 20 Jun 2019 12:07:17 +0200 Giulio Benetti…
From: Romain Naour <[email protected]>
This patch allows to use an external toolchain based on gcc 9.
Signed-off-by Romain Naour <[email protected]>Signed-off-by Giulio Benetti <[email protected]>--- .../toolchain-external-custom/Config.in.options | 6 ++++++ 1 file changed, 6 insertions(+)
Applied to master, thanks.
Thomas
Signature · 3 lines
Thomas Petazzoni, CTO, Bootlin Embedded Linux and Kernel engineering https://bootlin.com
PATCH v4 8/15 package/gcc: switch to gcc 8.x as the default
In reply to Giulio BenettiFrom: Romain Naour <[email protected]>
Even if gcc 7 is still maintained for some time (gcc 7.5 is pending), switch to gcc 8.x since it has been released since 2018-05-02 and gcc 9.x is available since 2019-05-03.
We have been having toolchains in the autobuilders with gcc 8.x for a while, so the vast majority of the problems should have already been solved.
---
v3: rebase patch on top of master branch sha1:
04efb17c863606cdbc2405d01c3d48d6868c5245 (Giulio) --- package/gcc/Config.in.host | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
package/gcc/Config.in.host +1 -1
@@ -5,7 +5,7 @@ choice default BR2_GCC_VERSION_ARC if BR2_arc default BR2_GCC_VERSION_CSKY if BR2_csky default BR2_GCC_VERSION_OR1K if BR2_or1k - default BR2_GCC_VERSION_7_X + default BR2_GCC_VERSION_8_X help Select the version of gcc you wish to use.
Signature · 1 line
2.17.1
Thomas Petazzoni
AppliedPATCH v4 8/15 package/gcc: switch to gcc 8.x as the default
Quoted 10 lines · On Thu, 20 Jun 2019 12:07:18 +0200 Giulio Benetti…
From: Romain Naour <[email protected]>
Even if gcc 7 is still maintained for some time (gcc 7.5 is pending), switch to gcc 8.x since it has been released since 2018-05-02 and gcc 9.x is available since 2019-05-03.
We have been having toolchains in the autobuilders with gcc 8.x for a while, so the vast majority of the problems should have already been solved.
Quoted 8 lines
Signed-off-by Romain Naour <[email protected]>Signed-off-by Giulio Benetti <[email protected]>---
v3: rebase patch on top of master branch sha1:04efb17c863606cdbc2405d01c3d48d6868c5245 (Giulio) --- package/gcc/Config.in.host | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
Applied to master, thanks.
What do we do with the older gcc versions ? We now have 4.9, 5, 6, 7, 8 and 9, which is quite a bit more than the normal number of 3 versions we want to support.
Arnout: do you still think gcc 4.9 is needed at this point ? If it is the case, should we keep gcc 4.9, but drop gcc 5 and 6 ?
Best regards,
Thomas
Signature · 3 lines
Thomas Petazzoni, CTO, Bootlin Embedded Linux and Kernel engineering https://bootlin.com
PATCH v4 8/15 package/gcc: switch to gcc 8.x as the default
Hi Thomas,
Le 22/06/2019 ? 21:44, Thomas Petazzoni a ?crit?:
Quoted 22 lines
Quoted 10 lines · On Thu, 20 Jun 2019 12:07:18 +0200 Giulio Benetti…
From: Romain Naour <[email protected]>
Even if gcc 7 is still maintained for some time (gcc 7.5 is pending), switch to gcc 8.x since it has been released since 2018-05-02 and gcc 9.x is available since 2019-05-03.
We have been having toolchains in the autobuilders with gcc 8.x for a while, so the vast majority of the problems should have already been solved.
Quoted 8 lines
Signed-off-by Romain Naour <[email protected]>Signed-off-by Giulio Benetti <[email protected]>---
v3: rebase patch on top of master branch sha1:04efb17c863606cdbc2405d01c3d48d6868c5245 (Giulio) --- package/gcc/Config.in.host | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
Quoted 1 line
Applied to master, thanks.
Thanks for merging this series :)
Quoted 7 lines
What do we do with the older gcc versions ? We now have 4.9, 5, 6, 7, 8 and 9, which is quite a bit more than the normal number of 3 versions we want to support.
Arnout: do you still think gcc 4.9 is needed at this point ? If it is the case, should we keep gcc 4.9, but drop gcc 5 and 6 ?
Currently glibc >= 2.29 needs gcc >= 5 on all architectures.
So we can only build a toolchain with gcc 4.9 with uClibc and musl. The upcoming release of glibc will require at least gcc 6.2 on all architectures [1].
If a toolchain with gcc 4.9 is really needed, it can be build with the 2019.02 LTS version.
Thomas, I have the same question about gdb 7.12.x.
[1]
https://sourceware.org/git/?p=glibc.git;a=commit;h=4dcbbc3b28aaeafe23e1a30db84055aa6f6fa987
Best regards,
Romain
Quoted 5 lines
Best regards,
Thomas
PATCH v4 8/15 package/gcc: switch to gcc 8.x as the default
Hello,
Quoted 10 lines · On Sun, 23 Jun 2019 00:35:38 +0200 Romain Naour <[email protected]> wrote:
Quoted 2 lines
Arnout: do you still think gcc 4.9 is needed at this point ? If it is the case, should we keep gcc 4.9, but drop gcc 5 and 6 ?
Currently glibc >= 2.29 needs gcc >= 5 on all architectures.
So we can only build a toolchain with gcc 4.9 with uClibc and musl. The upcoming release of glibc will require at least gcc 6.2 on all architectures [1].If a toolchain with gcc 4.9 is really needed, it can be build with the 2019.02 LTS version.
So, Peter said for the next LTS (2020.02), we probably want to have removed the oldest gcc versions (4.9, 5 and 6, most likely).
Quoted 1 line
Thomas, I have the same question about gdb 7.12.x.
I think we should also drop it. I was keeping it because it was the last gdb version that could be built with a toolchain that doesn't have C++11 support. But well, C++11 is there since gcc 4.8, so let's assume most sane people have moved their distribution to a version that has gcc 4.8 at least.
So on my side, green light to drop gdb 7.12.
Best regards,
Thomas
Signature · 3 lines
Thomas Petazzoni, CTO, Bootlin Embedded Linux and Kernel engineering https://bootlin.com
PATCH v4 8/15 package/gcc: switch to gcc 8.x as the default
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
"Thomas" == Thomas Petazzoni <[email protected]> writes:
Hi,
Quoted 9 lines
Quoted 1 line
Thomas, I have the same question about gdb 7.12.x.
I think we should also drop it. I was keeping it because it was the last gdb version that could be built with a toolchain that doesn't have C++11 support. But well, C++11 is there since gcc 4.8, so let's assume most sane people have moved their distribution to a version that has gcc 4.8 at least.
So on my side, green light to drop gdb 7.12.
Same from me!
Signature · 1 line
Bye, Peter Korsgaard
PATCH v4 8/15 package/gcc: switch to gcc 8.x as the default
In reply to Thomas PetazzoniQuoted 13 lines · Thomas Petazzoni
Hello,
Quoted 8 lines · On Sun, 23 Jun 2019 00:35:38 +0200 Romain Naour <[email protected]> wrote:
Quoted 2 lines
Arnout: do you still think gcc 4.9 is needed at this point ? If it is the case, should we keep gcc 4.9, but drop gcc 5 and 6 ?
Currently glibc >= 2.29 needs gcc >= 5 on all architectures.
So we can only build a toolchain with gcc 4.9 with uClibc and musl. The upcoming release of glibc will require at least gcc 6.2 on all architectures [1].
Quoted 5 lines
Quoted 2 lines
If a toolchain with gcc 4.9 is really needed, it can be build with the 2019.02 LTS version.
So, Peter said for the next LTS (2020.02), we probably want to have removed the oldest gcc versions (4.9, 5 and 6, most likely).
Yes indeed, we can't keep 4.9 around indefinitely, so let's drop it and 5 and 6
as well, leaving just the three most recent ones: 7, 8, 9.
Quoted 9 lines
Quoted 1 line
Thomas, I have the same question about gdb 7.12.x.
I think we should also drop it. I was keeping it because it was the last gdb version that could be built with a toolchain that doesn't have C++11 support. But well, C++11 is there since gcc 4.8, so let's assume most sane people have moved their distribution to a version that has gcc 4.8 at least.
So on my side, green light to drop gdb 7.12.
For gdb, does it actually make sense to keep multiple versions?
Regards,
Arnout
PATCH v4 8/15 package/gcc: switch to gcc 8.x as the default
Hi Arnout, All,
Le 26/06/2019 ? 00:25, Arnout Vandecappelle a ?crit?:
Quoted 16 lines · Thomas Petazzoni
Quoted 13 lines
Hello,
Quoted 8 lines · On Sun, 23 Jun 2019 00:35:38 +0200 Romain Naour <[email protected]> wrote:
Quoted 2 lines
Arnout: do you still think gcc 4.9 is needed at this point ? If it is the case, should we keep gcc 4.9, but drop gcc 5 and 6 ?
Currently glibc >= 2.29 needs gcc >= 5 on all architectures.
So we can only build a toolchain with gcc 4.9 with uClibc and musl. The upcoming release of glibc will require at least gcc 6.2 on all architectures [1].
Quoted 8 lines
Quoted 5 lines
Quoted 2 lines
If a toolchain with gcc 4.9 is really needed, it can be build with the 2019.02 LTS version.
So, Peter said for the next LTS (2020.02), we probably want to have removed the oldest gcc versions (4.9, 5 and 6, most likely).
Yes indeed, we can't keep 4.9 around indefinitely, so let's drop it and 5 and 6
as well, leaving just the three most recent ones: 7, 8, 9.
Note: gcc 5 is still used by beaglebone_qt5_defconfig in order to test the SGX PowerVR GPU using TI's sgx demo package.
This defconfig doesn't seems maintained for a long time (compared to beaglebone_defconfig). I'm adding Lothar Felten in Cc.
Lothar: it would be great if you can have a look at this defconfig and try to use gcc 8 (or at least gcc 7).
Quoted 10 lines
Quoted 10 lines
Quoted 3 lines
Thomas, I have the same question about gdb 7.12.x.
I think we should also drop it. I was keeping it because it was the last gdb version that could be built with a toolchain that doesn't have C++11 support. But well, C++11 is there since gcc 4.8, so let's assume most sane people have moved their distribution to a version that has gcc 4.8 at least.
Quoted 3 lines
Quoted 1 line
So on my side, green light to drop gdb 7.12.
For gdb, does it actually make sense to keep multiple versions?
It would make the maintenance easier...
Best regards,
Romain
Quoted 9 lines
Regards,
Arnout_______________________________________________
buildroot mailing list
buildroot at busybox.net
http://lists.busybox.net/mailman/listinfo/buildroot
PATCH v4 8/15 package/gcc: switch to gcc 8.x as the default
Quoted 7 lines · On Wed, 26 Jun 2019 08:55:00 +0200 Romain Naour <[email protected]> wrote:
Note: gcc 5 is still used by
beaglebone_qt5_defconfigin order to test the SGX PowerVR GPU using TI's sgx demo package.
This defconfig doesn't seems maintained for a long time (compared tobeaglebone_defconfig). I'm adding Lothar Felten in Cc.Lothar: it would be great if you can have a look at this defconfig and try to use gcc 8 (or at least gcc 7).
Lothar did quite a bit of work on this, but we had the issue of the GBM API, where I tried to do a virtual package, but it didn't work well, Lothar did further experiments with Qt with GBM, and it worked, but I never had the chance to have a look.
So, I wouldn't blame Lothar for not updating the defconfig here :-)
Best regards,
Thomas
Signature · 3 lines
Thomas Petazzoni, CTO, Bootlin Embedded Linux and Kernel engineering https://bootlin.com
PATCH v4 8/15 package/gcc: switch to gcc 8.x as the default
Hi Thomas,
Hi Thomas,
Le 26/06/2019 ? 09:23, Thomas Petazzoni a ?crit?:
Quoted 8 lines
Quoted 5 lines · On Wed, 26 Jun 2019 08:55:00 +0200 Romain Naour <[email protected]> wrote:
Note: gcc 5 is still used by
beaglebone_qt5_defconfigin order to test the SGX PowerVR GPU using TI's sgx demo package.
This defconfig doesn't seems maintained for a long time (compared tobeaglebone_defconfig). I'm adding Lothar Felten in Cc.
Quoted 7 lines
Quoted 2 lines
Lothar: it would be great if you can have a look at this defconfig and try to use gcc 8 (or at least gcc 7).
Lothar did quite a bit of work on this, but we had the issue of the GBM API, where I tried to do a virtual package, but it didn't work well, Lothar did further experiments with Qt with GBM, and it worked, but I never had the chance to have a look.
I used git log for having the history on this defconfig, the drawback of this method is I'm missing all work that is not merged yet.
I wasn't aware of Lothar's work, sorry.
Quoted 2 lines
So, I wouldn't blame Lothar for not updating the defconfig here :-)
I don't blame Lothar for not updating this defconfig, I understand working on TI's SGX packages is difficult :)
My intention is to notify Lothar that we are going to remove gcc 5.x.
Best regards,
Romain
Quoted 5 lines
Best regards,
Thomas
PATCH v4 8/15 package/gcc: switch to gcc 8.x as the default
In reply to Arnout VandecappelleQuoted 3 lines · On Wed, 26 Jun 2019 00:25:59 +0200 Arnout Vandecappelle <[email protected]> wrote:
Quoted 1 line
So on my side, green light to drop gdb 7.12.
For gdb, does it actually make sense to keep multiple versions?
I think it made sense for gdb 7.12 vs. gdb >= 8.x, due to this C++11 requirement on the host.
But if we are going to remove 7.12, then indeed, it probably no longer makes sense to have a version choice.
Best regards,
Thomas
Signature · 3 lines
Thomas Petazzoni, CTO, Bootlin Embedded Linux and Kernel engineering https://bootlin.com
PATCH v4 8/15 package/gcc: switch to gcc 8.x as the default
In reply to Thomas PetazzoniQuoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
"Thomas" == Thomas Petazzoni <[email protected]> writes:
Hi,
Quoted 6 lines
What do we do with the older gcc versions ? We now have 4.9, 5, 6, 7, 8 and 9, which is quite a bit more than the normal number of 3 versions we want to support.
Arnout: do you still think gcc 4.9 is needed at this point ? If it is the case, should we keep gcc 4.9, but drop gcc 5 and 6 ?
I know it is painful for people stuck on E.G. old bootloader/kernel versions, but I think it makes sense to slowly remove these old gcc versions, so E.G. by the time we release 2020.02 that we only support gcc 7/8/9.
Signature · 1 line
Bye, Peter Korsgaard
PATCH v4 9/15 arch/arm: move dependency on 64-bit down to individual cores
In reply to Giulio BenettiFrom: "Yann E. MORIN" <[email protected]>
It will make it easier to introduce new variants anywhere in the list, when those variants have different bitness requirements.
--- arch/Config.in.arm | 22 ++++++++++++++-------- 1 file changed, 14 insertions(+), 8 deletions(-)
arch/Config.in.arm +14 -8
@@ -376,50 +376,55 @@ config BR2_exynos_m1 select BR2_ARM_CPU_ARMV8A select BR2_ARCH_HAS_MMU_OPTIONAL select BR2_ARCH_NEEDS_GCC_AT_LEAST_5 -if BR2_ARCH_IS_64 config BR2_falkor bool "falkor" + depends on BR2_ARCH_IS_64 select BR2_ARM_CPU_HAS_FP_ARMV8 select BR2_ARM_CPU_ARMV8A select BR2_ARCH_HAS_MMU_OPTIONAL select BR2_ARCH_NEEDS_GCC_AT_LEAST_7 config BR2_qdf24xx bool "qdf24xx" + depends on BR2_ARCH_IS_64 select BR2_ARM_CPU_HAS_FP_ARMV8 select BR2_ARM_CPU_ARMV8A select BR2_ARCH_HAS_MMU_OPTIONAL select BR2_ARCH_NEEDS_GCC_AT_LEAST_6 config BR2_thunderx bool "thunderx" + depends on BR2_ARCH_IS_64 select BR2_ARM_CPU_HAS_FP_ARMV8 select BR2_ARM_CPU_ARMV8A select BR2_ARCH_HAS_MMU_OPTIONAL select BR2_ARCH_NEEDS_GCC_AT_LEAST_5 config BR2_thunderxt81 bool "thunderxt81" + depends on BR2_ARCH_IS_64 select BR2_ARM_CPU_HAS_FP_ARMV8 select BR2_ARM_CPU_ARMV8A select BR2_ARCH_HAS_MMU_OPTIONAL select BR2_ARCH_NEEDS_GCC_AT_LEAST_7 config BR2_thunderxt83 bool "thunderxt83" + depends on BR2_ARCH_IS_64 select BR2_ARM_CPU_HAS_FP_ARMV8 select BR2_ARM_CPU_ARMV8A select BR2_ARCH_HAS_MMU_OPTIONAL select BR2_ARCH_NEEDS_GCC_AT_LEAST_7 config BR2_thunderxt88 bool "thunderxt88" + depends on BR2_ARCH_IS_64 select BR2_ARM_CPU_HAS_FP_ARMV8 select BR2_ARM_CPU_ARMV8A select BR2_ARCH_HAS_MMU_OPTIONAL select BR2_ARCH_NEEDS_GCC_AT_LEAST_7 config BR2_thunderxt88p1 bool "thunderxt88p1" + depends on BR2_ARCH_IS_64 select BR2_ARM_CPU_HAS_FP_ARMV8 select BR2_ARM_CPU_ARMV8A select BR2_ARCH_HAS_MMU_OPTIONAL select BR2_ARCH_NEEDS_GCC_AT_LEAST_7 -endif # BR2_ARCH_IS_64 config BR2_xgene1 bool "xgene1" select BR2_ARM_CPU_HAS_ARM if !BR2_ARCH_IS_64
@@ -430,59 +435,60 @@ config BR2_xgene1 select BR2_ARCH_HAS_MMU_OPTIONAL select BR2_ARCH_NEEDS_GCC_AT_LEAST_5 -if BR2_ARCH_IS_64 comment "armv8.1a cores" config BR2_thunderx2t99 bool "thunderx2t99" + depends on BR2_ARCH_IS_64 select BR2_ARM_CPU_HAS_FP_ARMV8 select BR2_ARM_CPU_ARMV8A select BR2_ARCH_HAS_MMU_OPTIONAL select BR2_ARCH_NEEDS_GCC_AT_LEAST_7 config BR2_thunderx2t99p1 bool "thunderx2t99p1" + depends on BR2_ARCH_IS_64 select BR2_ARM_CPU_HAS_FP_ARMV8 select BR2_ARM_CPU_ARMV8A select BR2_ARCH_HAS_MMU_OPTIONAL select BR2_ARCH_NEEDS_GCC_AT_LEAST_7 config BR2_vulcan bool "vulcan" + depends on BR2_ARCH_IS_64 select BR2_ARM_CPU_HAS_FP_ARMV8 select BR2_ARM_CPU_ARMV8A select BR2_ARCH_HAS_MMU_OPTIONAL select BR2_ARCH_NEEDS_GCC_AT_LEAST_7 -endif # BR2_ARCH_IS_64 -if BR2_ARCH_IS_64 comment "armv8.2a cores" config BR2_cortex_a55 bool "cortex-A55" + depends on BR2_ARCH_IS_64 select BR2_ARM_CPU_HAS_FP_ARMV8 select BR2_ARM_CPU_ARMV8A select BR2_ARCH_HAS_MMU_OPTIONAL select BR2_ARCH_NEEDS_GCC_AT_LEAST_8 config BR2_cortex_a75 bool "cortex-A75" + depends on BR2_ARCH_IS_64 select BR2_ARM_CPU_HAS_FP_ARMV8 select BR2_ARM_CPU_ARMV8A select BR2_ARCH_HAS_MMU_OPTIONAL select BR2_ARCH_NEEDS_GCC_AT_LEAST_8 config BR2_cortex_a75_a55 bool "cortex-A75/A55 big.LITTLE" + depends on BR2_ARCH_IS_64 select BR2_ARM_CPU_HAS_FP_ARMV8 select BR2_ARM_CPU_ARMV8A select BR2_ARCH_HAS_MMU_OPTIONAL select BR2_ARCH_NEEDS_GCC_AT_LEAST_8 -endif # BR2_ARCH_IS_64 -if BR2_ARCH_IS_64 comment "armv8.3a cores" config BR2_saphira bool "saphira" + depends on BR2_ARCH_IS_64 select BR2_ARM_CPU_HAS_FP_ARMV8 select BR2_ARM_CPU_ARMV8A select BR2_ARCH_HAS_MMU_OPTIONAL select BR2_ARCH_NEEDS_GCC_AT_LEAST_8 -endif # BR2_ARCH_IS_64 endchoice config BR2_ARM_ENABLE_NEON
Signature · 1 line
2.17.1
Arnout Vandecappelle
AppliedPATCH v4 9/15 arch/arm: move dependency on 64-bit down to individual cores
Quoted 9 lines · Giulio Benetti
From: "Yann E. MORIN" <[email protected]>
It will make it easier to introduce new variants anywhere in the list, when those variants have different bitness requirements.
Signed-off-by "Yann E. MORIN" <[email protected]>Cc Thomas Petazzoni <[email protected]>Signed-off-by Romain Naour <[email protected]>Signed-off-by Giulio Benetti <[email protected]>
I've applied this and the rest of the series to master, thanks. With some small
modifications, noted in each individual patch.
Regards,
Arnout
PATCH v4 10/15 arch/arm: some cores have a different name with gcc-9
In reply to Giulio BenettiFrom: "Yann E. MORIN" <[email protected]>
In gcc-9, some cores from the ThunderX familly have been renamed to their marketting names, i.e. OcteonTX. Subsequently, new core names have been added to gcc, with the old names still being around.
Update the prompts with the new names as alternative to the existing names. We still keep the kconfig options as-is, so that we do not need to add legacy handling.
However, since there is no guarantee for how long gcc will retain compatibility for the older names, we readily switch over to using the new names when using a gcc 9-or-later, but keep using the older names with gcc older than 9.
The ordering of the 'default' clauses in kconfig is important, with the first default which conditions is satisfied, to be used, so we just add the new core names under a more liberal condition than the older names.
--- arch/Config.in.arm | 15 +++++++++------ 1 file changed, 9 insertions(+), 6 deletions(-)
arch/Config.in.arm +9 -6
@@ -391,21 +391,21 @@ config BR2_qdf24xx select BR2_ARCH_HAS_MMU_OPTIONAL select BR2_ARCH_NEEDS_GCC_AT_LEAST_6 config BR2_thunderx - bool "thunderx" + bool "thunderx (aka octeontx)" depends on BR2_ARCH_IS_64 select BR2_ARM_CPU_HAS_FP_ARMV8 select BR2_ARM_CPU_ARMV8A select BR2_ARCH_HAS_MMU_OPTIONAL select BR2_ARCH_NEEDS_GCC_AT_LEAST_5 config BR2_thunderxt81 - bool "thunderxt81" + bool "thunderxt81 (aka octeontx81)" depends on BR2_ARCH_IS_64 select BR2_ARM_CPU_HAS_FP_ARMV8 select BR2_ARM_CPU_ARMV8A select BR2_ARCH_HAS_MMU_OPTIONAL select BR2_ARCH_NEEDS_GCC_AT_LEAST_7 config BR2_thunderxt83 - bool "thunderxt83" + bool "thunderxt83 (aka octeontx83)" depends on BR2_ARCH_IS_64 select BR2_ARM_CPU_HAS_FP_ARMV8 select BR2_ARM_CPU_ARMV8A
@@ -824,9 +824,12 @@ config BR2_GCC_TARGET_CPU default "exynos-m1" if BR2_exynos_m1 default "falkor" if BR2_falkor default "qdf24xx" if BR2_qdf24xx - default "thunderx" if BR2_thunderx - default "thunderxt81" if BR2_thunderxt81 - default "thunderxt83" if BR2_thunderxt83 + default "thunderx" if BR2_thunderx && !BR2_TOOLCHAIN_GCC_AT_LEAST_9 + default "octeontx" if BR2_thunderx + default "thunderxt81" if BR2_thunderxt81 && !BR2_TOOLCHAIN_GCC_AT_LEAST_9 + default "octeontx81" if BR2_thunderxt81 + default "thunderxt83" if BR2_thunderxt83 && !BR2_TOOLCHAIN_GCC_AT_LEAST_9 + default "octeontx83" if BR2_thunderxt83 default "thunderxt88" if BR2_thunderxt88 default "thunderxt88p1" if BR2_thunderxt88p1 default "xgene1" if BR2_xgene1
Signature · 1 line
2.17.1
PATCH v4 10/15 arch/arm: some cores have a different name with gcc-9
Quoted 1 line · Giulio Benetti
From: "Yann E. MORIN" <[email protected]>
[snip]
Quoted 3 lines
The ordering of the 'default' clauses in kconfig is important, with the first default which conditions is satisfied, to be used, so we just add the new core names under a more liberal condition than the older names.
This is not so readable / readily understandable, IMO. Therefore, I've changed
this into:
[snip]
arch/Config.in.arm config BR2_GCC_TARGET_CPU · line 828
Quoted 2 lines
arch/Config.in.arm +2
-0+ default "thunderx" if BR2_thunderx && !BR2_TOOLCHAIN_GCC_AT_LEAST_9 + default "octeontx" if BR2_thunderx
default "thunderx" if BR2_thunderx && !BR2_TOOLCHAIN_GCC_AT_LEAST_9 default "octeontx" if BR2_thunderx && BR2_TOOLCHAIN_GCC_AT_LEAST_9
etc.
Regards,
Arnout
Quoted 8 lines
arch/Config.in.arm +4
-0+ default "thunderxt81" if BR2_thunderxt81 && !BR2_TOOLCHAIN_GCC_AT_LEAST_9 + default "octeontx81" if BR2_thunderxt81 + default "thunderxt83" if BR2_thunderxt83 && !BR2_TOOLCHAIN_GCC_AT_LEAST_9 + default "octeontx83" if BR2_thunderxt83 default "thunderxt88" if BR2_thunderxt88 default "thunderxt88p1" if BR2_thunderxt88p1 default "xgene1" if BR2_xgene1
PATCH v4 10/15 arch/arm: some cores have a different name with gcc-9
Arnout, All,
Thanks for eventually applying the series. :-)
On 2019-07-13 23:13 +0200, Arnout Vandecappelle spake thusly:
Quoted 9 lines · Giulio Benetti
Quoted 1 line
From: "Yann E. MORIN" <[email protected]>
[snip]
Quoted 3 lines
The ordering of the 'default' clauses in kconfig is important, with the first default which conditions is satisfied, to be used, so we just add the new core names under a more liberal condition than the older names.
This is not so readable / readily understandable, IMO. Therefore, I've changed
this into:
We depend on the default ordering in other places as well.
Quoted 6 lines
[snip]
arch/Config.in.arm config BR2_GCC_TARGET_CPU · line 828
Quoted 2 lines
arch/Config.in.arm +2
-0+ default "thunderx" if BR2_thunderx && !BR2_TOOLCHAIN_GCC_AT_LEAST_9 + default "octeontx" if BR2_thunderxdefault "thunderx" if BR2_thunderx && !BR2_TOOLCHAIN_GCC_AT_LEAST_9 default "octeontx" if BR2_thunderx && BR2_TOOLCHAIN_GCC_AT_LEAST_9
You and I reallt have a differnt way toi see things, because I find this less readable now...
But oh well...
Thanks for applying anyway. :-)
Regards,
Yann E. MORIN.
Quoted 13 lines
etc.
Regards,
ArnoutQuoted 8 lines
arch/Config.in.arm +4
-0+ default "thunderxt81" if BR2_thunderxt81 && !BR2_TOOLCHAIN_GCC_AT_LEAST_9 + default "octeontx81" if BR2_thunderxt81 + default "thunderxt83" if BR2_thunderxt83 && !BR2_TOOLCHAIN_GCC_AT_LEAST_9 + default "octeontx83" if BR2_thunderxt83 default "thunderxt88" if BR2_thunderxt88 default "thunderxt88p1" if BR2_thunderxt88p1 default "xgene1" if BR2_xgene1
Signature · 6 lines
.-----------------.--------------------.------------------.--------------------. | Yann E. MORIN | Real-Time Embedded | /"\ ASCII RIBBON | Erics' conspiracy: | | +33 662 376 056 | Software Designer | \ / CAMPAIGN | ___ | | +33 561 099 427 `------------.-------: X AGAINST | \e/ There is no | | http://ymorin.is-a-geek.org/ | _/*\_ | / \ HTML MAIL | v conspiracy. | '------------------------------^-------^------------------^--------------------'
PATCH v4 11/15 arch/arm: saphira is in fact an armv8.4a
In reply to Giulio BenettiFrom: "Yann E. MORIN" <[email protected]>
... and not an armv8.3a like previously supposed:
https://gcc.gnu.org/git/?p=gcc.git;a=commit;h=886d991373e4dc5a746d0a33de64f1b36e61eed9
So, change the correspoding labels and comments.
--- arch/Config.in.arm | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
arch/Config.in.arm +2 -2
@@ -481,7 +481,7 @@ config BR2_cortex_a75_a55 select BR2_ARCH_HAS_MMU_OPTIONAL select BR2_ARCH_NEEDS_GCC_AT_LEAST_8 -comment "armv8.3a cores" +comment "armv8.4a cores" config BR2_saphira bool "saphira" depends on BR2_ARCH_IS_64
@@ -841,7 +841,7 @@ config BR2_GCC_TARGET_CPU default "cortex-a55" if BR2_cortex_a55 default "cortex-a75" if BR2_cortex_a75 default "cortex-a75.cortex-a55" if BR2_cortex_a75_a55 - # armv8.3a + # armv8.4a default "saphira" if BR2_saphira config BR2_GCC_TARGET_ABI
Signature · 1 line
2.17.1
PATCH v4 12/15 arch/arm: add two new 64-bit-only armv8a cores
In reply to Giulio BenettiFrom: "Yann E. MORIN" <[email protected]>
Build tested:
https://gitlab.com/kubu93/buildroot/pipelines/60318953
--- arch/Config.in.arm | 16 ++++++++++++++++ 1 file changed, 16 insertions(+)
arch/Config.in.arm +16 -0
@@ -367,6 +367,13 @@ config BR2_cortex_a73_a53 select BR2_ARM_CPU_ARMV8A select BR2_ARCH_HAS_MMU_OPTIONAL select BR2_ARCH_NEEDS_GCC_AT_LEAST_7 +config BR2_emag + bool "emag" + depends on BR2_ARCH_IS_64 + select BR2_ARM_CPU_HAS_FP_ARMV8 + select BR2_ARM_CPU_ARMV8A + select BR2_ARCH_HAS_MMU_OPTIONAL + select BR2_ARCH_NEEDS_GCC_AT_LEAST_9 config BR2_exynos_m1 bool "exynos-m1" select BR2_ARM_CPU_HAS_ARM if !BR2_ARCH_IS_64
@@ -383,6 +390,13 @@ config BR2_falkor select BR2_ARM_CPU_ARMV8A select BR2_ARCH_HAS_MMU_OPTIONAL select BR2_ARCH_NEEDS_GCC_AT_LEAST_7 +config BR2_phecda + bool "phecda" + depends on BR2_ARCH_IS_64 + select BR2_ARM_CPU_HAS_FP_ARMV8 + select BR2_ARM_CPU_ARMV8A + select BR2_ARCH_HAS_MMU_OPTIONAL + select BR2_ARCH_NEEDS_GCC_AT_LEAST_9 config BR2_qdf24xx bool "qdf24xx" depends on BR2_ARCH_IS_64
@@ -821,8 +835,10 @@ config BR2_GCC_TARGET_CPU default "cortex-a73" if BR2_cortex_a73 default "cortex-a73.cortex-a35" if BR2_cortex_a73_a35 default "cortex-a73.cortex-a53" if BR2_cortex_a73_a53 + default "emag" if BR2_emag default "exynos-m1" if BR2_exynos_m1 default "falkor" if BR2_falkor + default "phecda" if BR2_phecda default "qdf24xx" if BR2_qdf24xx default "thunderx" if BR2_thunderx && !BR2_TOOLCHAIN_GCC_AT_LEAST_9 default "octeontx" if BR2_thunderx
Signature · 1 line
2.17.1
PATCH v4 13/15 arch/arm: add two new cortex-based armv8.2a cores
In reply to Giulio BenettiFrom: "Yann E. MORIN" <[email protected]>
The cortex-a76 implements the full amrv8.2a extensions, and some optional extensions from the armv8.3a, armv8.4a, and armv8.5a sets, but none of their mandatory extensions, which means that it does not qualify for better than an armv8.2a:
https://developer.arm.com/products/processors/cortex-a/cortex-a76 http://infocenter.arm.com/help/topic/com.arm.doc.100798_0301_00_en/giq1479805174793.html http://infocenter.arm.com/help/topic/com.arm.doc.100798_0301_00_en/fjv1477559794375.html
Also, gcc fits it in the armv8.2a category, too:
https://gcc.gnu.org/git/?p=gcc.git;a=blob;f=gcc/config/aarch64/aarch64-cores.def;h=67ce42fb8aacd4c246295f32151a03b1f318ae44;hb=HEAD#l97
Build tested:
https://gitlab.com/kubu93/buildroot/pipelines/60318953
--- arch/Config.in.arm | 20 ++++++++++++++++++++ 1 file changed, 20 insertions(+)
arch/Config.in.arm +20 -0
@@ -494,6 +494,24 @@ config BR2_cortex_a75_a55 select BR2_ARM_CPU_ARMV8A select BR2_ARCH_HAS_MMU_OPTIONAL select BR2_ARCH_NEEDS_GCC_AT_LEAST_8 +config BR2_cortex_a76 + bool "cortex-A76" + select BR2_ARM_CPU_HAS_ARM if !BR2_ARCH_IS_64 + select BR2_ARM_CPU_HAS_NEON if !BR2_ARCH_IS_64 + select BR2_ARM_CPU_HAS_THUMB2 if !BR2_ARCH_IS_64 + select BR2_ARM_CPU_HAS_FP_ARMV8 + select BR2_ARM_CPU_ARMV8A + select BR2_ARCH_HAS_MMU_OPTIONAL + select BR2_ARCH_NEEDS_GCC_AT_LEAST_9 +config BR2_cortex_a76_a55 + bool "cortex-A76/A55 big.LITTLE" + select BR2_ARM_CPU_HAS_ARM if !BR2_ARCH_IS_64 + select BR2_ARM_CPU_HAS_NEON if !BR2_ARCH_IS_64 + select BR2_ARM_CPU_HAS_THUMB2 if !BR2_ARCH_IS_64 + select BR2_ARM_CPU_HAS_FP_ARMV8 + select BR2_ARM_CPU_ARMV8A + select BR2_ARCH_HAS_MMU_OPTIONAL + select BR2_ARCH_NEEDS_GCC_AT_LEAST_9 comment "armv8.4a cores" config BR2_saphira
@@ -857,6 +875,8 @@ config BR2_GCC_TARGET_CPU default "cortex-a55" if BR2_cortex_a55 default "cortex-a75" if BR2_cortex_a75 default "cortex-a75.cortex-a55" if BR2_cortex_a75_a55 + default "cortex-a76" if BR2_cortex_a76 + default "cortex-a76.cortex-a55" if BR2_cortex_a76_a55 # armv8.4a default "saphira" if BR2_saphira
Signature · 1 line
2.17.1
PATCH v4 14/15 arch/arm: add two new non-cortex-based armv8.2a cores
In reply to Giulio BenettiFrom: "Yann E. MORIN" <[email protected]>
The Neoverse N1 CPU was supported in GCC earlier through the codename Ares [1].
[1] https://gcc.gnu.org/git/?p=gcc.git;a=commit;h=37cf0ddecfd1eb5c6852a44135af5a92e5103931
Build tested:
https://gitlab.com/kubu93/buildroot/pipelines/60318953
---
v2: add neoverse-e1 (Yann)
reorder alphabetically after renaming ares to neoverse-n1.
Note: The Neoverse E1 CPU needs at least binutils 2.32 to support Speculative Store Bypass Safe instruction (SSBS) which is part of ARMv8.5-A architecture extensions. Postpone the support for this new ARM core.
See: https://sourceware.org/git/gitweb.cgi?p=binutils-gdb.git;h=104fefeebb544b7745bb353b63110afa46119647
--- arch/Config.in.arm | 18 ++++++++++++++++++ 1 file changed, 18 insertions(+)
arch/Config.in.arm +18 -0
@@ -512,6 +512,22 @@ config BR2_cortex_a76_a55 select BR2_ARM_CPU_ARMV8A select BR2_ARCH_HAS_MMU_OPTIONAL select BR2_ARCH_NEEDS_GCC_AT_LEAST_9 +config BR2_neoverse_n1 + bool "neoverse-N1 (alias ares)" + select BR2_ARM_CPU_HAS_ARM if !BR2_ARCH_IS_64 + select BR2_ARM_CPU_HAS_NEON if !BR2_ARCH_IS_64 + select BR2_ARM_CPU_HAS_THUMB2 if !BR2_ARCH_IS_64 + select BR2_ARM_CPU_HAS_FP_ARMV8 + select BR2_ARM_CPU_ARMV8A + select BR2_ARCH_HAS_MMU_OPTIONAL + select BR2_ARCH_NEEDS_GCC_AT_LEAST_9 +config BR2_tsv110 + bool "tsv110" + depends on BR2_ARCH_IS_64 + select BR2_ARM_CPU_HAS_FP_ARMV8 + select BR2_ARM_CPU_ARMV8A + select BR2_ARCH_HAS_MMU_OPTIONAL + select BR2_ARCH_NEEDS_GCC_AT_LEAST_9 comment "armv8.4a cores" config BR2_saphira
@@ -877,6 +893,8 @@ config BR2_GCC_TARGET_CPU default "cortex-a75.cortex-a55" if BR2_cortex_a75_a55 default "cortex-a76" if BR2_cortex_a76 default "cortex-a76.cortex-a55" if BR2_cortex_a76_a55 + default "neoverse-n1" if BR2_neoverse_n1 + default "tsv110" if BR2_tsv110 # armv8.4a default "saphira" if BR2_saphira
Signature · 1 line
2.17.1
PATCH v4 14/15 arch/arm: add two new non-cortex-based armv8.2a cores
Quoted 31 lines · Giulio Benetti
From: "Yann E. MORIN" <[email protected]>
The Neoverse N1 CPU was supported in GCC earlier through the codename Ares [1].
[1] https://gcc.gnu.org/git/?p=gcc.git;a=commit;h=37cf0ddecfd1eb5c6852a44135af5a92e5103931
Build tested:
https://gitlab.com/kubu93/buildroot/pipelines/60318953Signed-off-by "Yann E. MORIN" <[email protected]>[Romain: rename BR2_ares to BR2_neoverse_n1]Signed-off-by Romain Naour <[email protected]>Cc Thomas Petazzoni <[email protected]>Signed-off-by Giulio Benetti <[email protected]>---
v2: add neoverse-e1 (Yann)
reorder alphabetically after renaming ares to neoverse-n1.Note: The Neoverse E1 CPU needs at least binutils 2.32 to support Speculative Store Bypass Safe instruction (SSBS) which is part of ARMv8.5-A architecture extensions. Postpone the support for this new ARM core.
See: https://sourceware.org/git/gitweb.cgi?p=binutils-gdb.git;h=104fefeebb544b7745bb353b63110afa46119647--- arch/Config.in.arm | 18 ++++++++++++++++++ 1 file changed, 18 insertions(+)arch/Config.in.arm +0
-0
arch/Config.in.arm config BR2_cortex_a76_a55 · line 516
Quoted 5 lines
arch/Config.in.arm +2
-0@@ -512,6 +512,22 @@ config BR2_cortex_a76_a55 select BR2_ARM_CPU_ARMV8A select BR2_ARCH_HAS_MMU_OPTIONAL select BR2_ARCH_NEEDS_GCC_AT_LEAST_9 +config BR2_neoverse_n1 + bool "neoverse-N1 (alias ares)"
In an earlier patch, you used 'aka', so I did the same here.
arch/Config.in.arm config BR2_cortex_a76_a55 · line 529
Quoted 13 lines
arch/Config.in.arm +13
-0+ select BR2_ARM_CPU_HAS_ARM if !BR2_ARCH_IS_64 + select BR2_ARM_CPU_HAS_NEON if !BR2_ARCH_IS_64 + select BR2_ARM_CPU_HAS_THUMB2 if !BR2_ARCH_IS_64 + select BR2_ARM_CPU_HAS_FP_ARMV8 + select BR2_ARM_CPU_ARMV8A + select BR2_ARCH_HAS_MMU_OPTIONAL + select BR2_ARCH_NEEDS_GCC_AT_LEAST_9 +config BR2_tsv110 + bool "tsv110" + depends on BR2_ARCH_IS_64 + select BR2_ARM_CPU_HAS_FP_ARMV8 + select BR2_ARM_CPU_ARMV8A + select BR2_ARCH_HAS_MMU_OPTIONAL
Somewhat unrelated to this patch, but: is this really correct? I thought the
MMU was mandatory on 64-bit ARM. At least, the kernel forces MMU on for arm64.
Regards,
Arnout
Quoted 14 lines
arch/Config.in.arm +3
-0+ select BR2_ARCH_NEEDS_GCC_AT_LEAST_9 comment "armv8.4a cores" config BR2_saphira@@ -877,6 +893,8 @@ config BR2_GCC_TARGET_CPU default "cortex-a75.cortex-a55" if BR2_cortex_a75_a55 default "cortex-a76" if BR2_cortex_a76 default "cortex-a76.cortex-a55" if BR2_cortex_a76_a55 + default "neoverse-n1" if BR2_neoverse_n1 + default "tsv110" if BR2_tsv110 # armv8.4a default "saphira" if BR2_saphira
PATCH v4 14/15 arch/arm: add two new non-cortex-based armv8.2a cores
Arnout, All,
On 2019-07-13 23:22 +0200, Arnout Vandecappelle spake thusly:
Quoted 2 lines · Giulio Benetti
Quoted 1 line
From: "Yann E. MORIN" <[email protected]>
[--SNIP--]
Quoted 8 lines
arch/Config.in.arm config BR2_cortex_a76_a55 · line 529
Quoted 6 lines
arch/Config.in.arm +6
-0+config BR2_tsv110 + bool "tsv110" + depends on BR2_ARCH_IS_64 + select BR2_ARM_CPU_HAS_FP_ARMV8 + select BR2_ARM_CPU_ARMV8A + select BR2_ARCH_HAS_MMU_OPTIONALSomewhat unrelated to this patch, but: is this really correct? I thought the
MMU was mandatory on 64-bit ARM. At least, the kernel forces MMU on for arm64.
TBH, I don't know. I stupidly replicated the settings from the other existing AArch64 cores we already had in the list.
Thanks!
Regards,
Yann E. MORIN.
Signature · 6 lines
.-----------------.--------------------.------------------.--------------------. | Yann E. MORIN | Real-Time Embedded | /"\ ASCII RIBBON | Erics' conspiracy: | | +33 662 376 056 | Software Designer | \ / CAMPAIGN | ___ | | +33 561 099 427 `------------.-------: X AGAINST | \e/ There is no | | http://ymorin.is-a-geek.org/ | _/*\_ | / \ HTML MAIL | v conspiracy. | '------------------------------^-------^------------------^--------------------'
PATCH v4 15/15 package/gcc: enable gcc 9.1 for ork1 (openrisc)
In reply to Giulio BenettiFrom: Romain Naour <[email protected]>
openrisc support has been added with gcc 9.1.
Keep for now the old gcc 5 fork for ork1.
https://gcc.gnu.org/gcc-9/changes.html
Tested using qemu_or1k_defconfig.
---
v3: rebase patch on top of master branch sha1:
04efb17c863606cdbc2405d01c3d48d6868c5245 (Giulio) --- package/gcc/Config.in.host | 2 -- package/gcc/gcc.mk | 4 ++-- 2 files changed, 2 insertions(+), 4 deletions(-)
package/gcc/Config.in.host +0 -2
@@ -86,8 +86,6 @@ config BR2_GCC_VERSION_8_X config BR2_GCC_VERSION_9_X bool "gcc 9.x" - # Broken or unsupported architectures - depends on !BR2_or1k select BR2_TOOLCHAIN_GCC_AT_LEAST_9 endchoice
package/gcc/gcc.mk +2 -2
@@ -13,7 +13,7 @@ GCC_VERSION = $(call qstrip,$(BR2_GCC_VERSION)) ifeq ($(BR2_GCC_VERSION_ARC),y) GCC_SITE = $(call github,foss-for-synopsys-dwc-arc-processors,gcc,$(GCC_VERSION)) GCC_SOURCE = gcc-$(GCC_VERSION).tar.gz -else ifeq ($(BR2_or1k),y) +else ifeq ($(BR2_GCC_VERSION_OR1K),y) GCC_SITE = $(call github,openrisc,or1k-gcc,$(GCC_VERSION)) GCC_SOURCE = gcc-$(GCC_VERSION).tar.gz else ifeq ($(BR2_csky),y)
@@ -180,7 +180,7 @@ else HOST_GCC_COMMON_CONF_OPTS += --without-isl --without-cloog endif -ifeq ($(BR2_arc)$(BR2_or1k),y) +ifeq ($(BR2_arc)$(BR2_GCC_VERSION_OR1K),y) HOST_GCC_COMMON_DEPENDENCIES += host-flex host-bison endif
Signature · 1 line
2.17.1
PATCH v4 15/15 package/gcc: enable gcc 9.1 for ork1 (openrisc)
Quoted 2 lines · Giulio Benetti
openrisc support has been added with gcc 9.1.
Keep for now the old gcc 5 fork for ork1.
So, when will we remove this?
Same for ARC (GCC 8.x fork) and C-SKY (GCC 6.x fork): they are available in
GCC9, do we need to keep those old forks?
Regards,
Arnout
PATCH v4 15/15 package/gcc: enable gcc 9.1 for ork1 (openrisc)
Quoted 11 lines · Arnout Vandecappelle
Quoted 2 lines · Giulio Benetti
openrisc support has been added with gcc 9.1.
Keep for now the old gcc 5 fork for ork1.So, when will we remove this?
Same for ARC (GCC 8.x fork) and C-SKY (GCC 6.x fork): they are available in
GCC9, do we need to keep those old forks?
Keep the fork for C-SKY, the upstream of csky backend in gcc9 only support abiv2.
Quoted 7 lines
Regards,
Arnout
_______________________________________________
buildroot mailing list
buildroot at busybox.net
http://lists.busybox.net/mailman/listinfo/buildroot
Signature · 4 lines
Best Regards Guo Ren ML: https://lore.kernel.org/linux-csky/
PATCH v4 15/15 package/gcc: enable gcc 9.1 for ork1 (openrisc)
Guo, All,
On 2019-07-14 07:57 +0800, Guo Ren spake thusly:
Quoted 9 lines · Arnout Vandecappelle
Quoted 6 lines · Giulio Benetti
Quoted 2 lines
openrisc support has been added with gcc 9.1.
Keep for now the old gcc 5 fork for ork1.So, when will we remove this?
Same for ARC (GCC 8.x fork) and C-SKY (GCC 6.x fork): they are available in
GCC9, do we need to keep those old forks?Keep the fork for C-SKY, the upstream of csky backend in gcc9 only support abiv2.
As far as I can see, we have no way to choose the ABI in Buildroot. So, what would we loose by keeping only gcc-9?
Regards,
Yann E. MORIN.
Signature · 6 lines
.-----------------.--------------------.------------------.--------------------. | Yann E. MORIN | Real-Time Embedded | /"\ ASCII RIBBON | Erics' conspiracy: | | +33 662 376 056 | Software Designer | \ / CAMPAIGN | ___ | | +33 561 099 427 `------------.-------: X AGAINST | \e/ There is no | | http://ymorin.is-a-geek.org/ | _/*\_ | / \ HTML MAIL | v conspiracy. | '------------------------------^-------^------------------^--------------------'
PATCH v4 15/15 package/gcc: enable gcc 9.1 for ork1 (openrisc)
Quoted 8 lines · Yann E. MORIN
Guo, All,
On 2019-07-14 07:57 +0800, Guo Ren spake thusly:
Quoted 4 lines · Arnout Vandecappelle
Quoted 3 lines · Giulio Benetti
Quoted 2 lines
openrisc support has been added with gcc 9.1.
Keep for now the old gcc 5 fork for ork1.
Quoted 8 lines
Quoted 5 lines
Quoted 3 lines
So, when will we remove this?
Same for ARC (GCC 8.x fork) and C-SKY (GCC 6.x fork): they are available in
GCC9, do we need to keep those old forks?Keep the fork for C-SKY, the upstream of csky backend in gcc9 only support abiv2.
As far as I can see, we have no way to choose the ABI in Buildroot. So, what would we loose by keeping only gcc-9?
See package/Makefile.in
# For C-SKY abiv1 & abiv2
ifeq ($(BR2_csky),y)
ifeq ($(BR2_ck610),y)
ABI = abiv1
else
ABI = abiv2
endif
endif
gcc-9 don't support 610.
Signature · 4 lines
Best Regards Guo Ren ML: https://lore.kernel.org/linux-csky/
PATCH v4 15/15 package/gcc: enable gcc 9.1 for ork1 (openrisc)
Guo, All,
Peter, Thomas, Arnout, question below...
On 2019-07-14 17:18 +0800, Guo Ren spake thusly:
Quoted 15 lines · Yann E. MORIN
Quoted 5 lines
On 2019-07-14 07:57 +0800, Guo Ren spake thusly:
Quoted 4 lines · Arnout Vandecappelle
Quoted 3 lines · Giulio Benetti
Quoted 2 lines
openrisc support has been added with gcc 9.1.
Keep for now the old gcc 5 fork for ork1.Quoted 7 lines
Quoted 5 lines
Quoted 3 lines
So, when will we remove this?
Same for ARC (GCC 8.x fork) and C-SKY (GCC 6.x fork): they are available in
GCC9, do we need to keep those old forks?Keep the fork for C-SKY, the upstream of csky backend in gcc9 only support abiv2.
As far as I can see, we have no way to choose the ABI in Buildroot. So, what would we loose by keeping only gcc-9?
See package/Makefile.in
Quoted 8 lines
# For C-SKY abiv1 & abiv2
ifeq ($(BR2_csky),y)
ifeq ($(BR2_ck610),y)
ABI = abiv1
else
ABI = abiv2
endif
endif
Ah, thanks for pointing this out. I only had a look at arch/arch.mk.csky and arch/Config.in.csky.
This makes me wonder: should we move all this out of package/Makefile.in and into the per-arch .mk files instead?
Quoted 1 line
gcc-9 don't support 610.
There is currently nothign in Buildroot that prevents selecting gcc9 for the ck610...
Quoted 6 lines
Signature · 5 lines
-- Best Regards Guo Ren ML: https://lore.kernel.org/linux-csky/
Signature · 6 lines
.-----------------.--------------------.------------------.--------------------. | Yann E. MORIN | Real-Time Embedded | /"\ ASCII RIBBON | Erics' conspiracy: | | +33 662 376 056 | Software Designer | \ / CAMPAIGN | ___ | | +33 561 099 427 `------------.-------: X AGAINST | \e/ There is no | | http://ymorin.is-a-geek.org/ | _/*\_ | / \ HTML MAIL | v conspiracy. | '------------------------------^-------^------------------^--------------------'
PATCH v4 15/15 package/gcc: enable gcc 9.1 for ork1 (openrisc)
Quoted 9 lines · On Sun, 14 Jul 2019 11:39:16 +0200 "Yann E. MORIN" <[email protected]>…
Quoted 8 lines
# For C-SKY abiv1 & abiv2
ifeq ($(BR2_csky),y)
ifeq ($(BR2_ck610),y)
ABI = abiv1
else
ABI = abiv2
endif
endif
Quoted 5 lines
Ah, thanks for pointing this out. I only had a look at
arch/arch.mk.cskyandarch/Config.in.csky.This makes me wonder: should we move all this out of package/Makefile.in and into the per-arch .mk files instead?
Yes, I also thought about that when I merged Guo's patches for C-SKY. But other architectures already have stuff in package/Makefile.in, so it required a larger rework, which I didn't want to put as a pre-requisite to merging Guo's patches.
Best regards,
Thomas
Signature · 3 lines
Thomas Petazzoni, CTO, Bootlin Embedded Linux and Kernel engineering https://bootlin.com
PATCH v4 15/15 package/gcc: enable gcc 9.1 for ork1 (openrisc)
Thomas, All,
On 2019-07-14 12:00 +0200, Thomas Petazzoni spake thusly:
Quoted 2 lines
On Sun, 14 Jul 2019 11:39:16 +0200
"Yann E. MORIN" <[email protected]> wrote:
[--SNIP--]
Quoted 6 lines
Quoted 2 lines
This makes me wonder: should we move all this out of package/Makefile.in and into the per-arch .mk files instead?
Yes, I also thought about that when I merged Guo's patches for C-SKY. But other architectures already have stuff in package/Makefile.in, so it required a larger rework, which I didn't want to put as a pre-requisite to merging Guo's patches.
Exactly, it's OK that it was merged as-is.
But now that we have seen The Light, should we folow Its path and move arch-specific stuff into arch/arch.mk.* instead?
Especially with the still-pending mini-series for ARC:
https://patchwork.ozlabs.org/patch/1108865/ https://patchwork.ozlabs.org/patch/1108864/
Regards,
Yann E. MORIN.
Signature · 6 lines
.-----------------.--------------------.------------------.--------------------. | Yann E. MORIN | Real-Time Embedded | /"\ ASCII RIBBON | Erics' conspiracy: | | +33 662 376 056 | Software Designer | \ / CAMPAIGN | ___ | | +33 561 099 427 `------------.-------: X AGAINST | \e/ There is no | | http://ymorin.is-a-geek.org/ | _/*\_ | / \ HTML MAIL | v conspiracy. | '------------------------------^-------^------------------^--------------------'
PATCH v4 15/15 package/gcc: enable gcc 9.1 for ork1 (openrisc)
In reply to Arnout VandecappelleHi Arnout, All,
Le 13/07/2019 ? 23:24, Arnout Vandecappelle a ?crit?:
Quoted 7 lines · Giulio Benetti
Quoted 2 lines
openrisc support has been added with gcc 9.1.
Keep for now the old gcc 5 fork for ork1.So, when will we remove this?
Adding Waldemar in Cc.
Waldemar, I tested gcc 9.1 ork1 backend using qemu_or1k_defconfig, it's ok to remove the old gcc 5 fork ?
At least, we can remove this fork while removing gcc 5 for the internal toolchain.
Quoted 3 lines
Same for ARC (GCC 8.x fork) and C-SKY (GCC 6.x fork): they are available in
GCC9, do we need to keep those old forks?
For ARC and C-SKY we have architectures maintainer, I'll let them remove these gcc fork.
Best regards,
Romain
Quoted 4 lines
Regards,
Arnout
PATCH v4 15/15 package/gcc: enable gcc 9.1 for ork1 (openrisc)
In reply to Arnout VandecappelleHi Arnout, all,
Quoted 12 lines
-----Original Message-----
From: Arnout Vandecappelle <[email protected]>
Sent: Sunday, July 14, 2019 12:24 AM
To: Giulio Benetti <[email protected]>; buildroot at buildroot.org; arc- buildroot at synopsys.com; Guo Ren <[email protected]>
Cc: Yann E . MORIN <[email protected]>; Thomas Petazzoni <[email protected]>; Romain Naour <[email protected]>; Thomas De Schampheleire <[email protected]>
Subject: Re: [arc-buildroot] [Buildroot] [PATCH v4 15/15] package/gcc: enable gcc 9.1 for ork1 (openrisc)
Quoted 8 lines
Quoted 2 lines · Giulio Benetti
openrisc support has been added with gcc 9.1.
Keep for now the old gcc 5 fork for ork1.So, when will we remove this?
Same for ARC (GCC 8.x fork) and C-SKY (GCC 6.x fork): they are available in
GCC9, do we need to keep those old forks?
ARC support in GCC 9.x & Binutils 2.32 is very solid so we may even try to use it by default for ARC.
As for arc-xxxx.yy forks I'd like to keep them for a couple of future BR releases and if we see that vanilla GCC/Binutils work really good then forks may go avay.
Make sense?
-Alexey