hppa: Wire up cacheflush syscall
PATCH 1/1 hppa: Wire up cacheflush syscall
The hppa Linux kernel supports the cacheflush() syscall since version 6.5. This adds the glibc syscall wrapper.
---
sysdeps/unix/sysv/linux/hppa/Makefile +5 -0
@@ -15,6 +15,11 @@ test-xfail-check-wx-segment = * endif # $(subdir) == elf +ifeq ($(subdir),misc) +sysdep_routines += cacheflush +sysdep_headers += sys/cachectl.h +endif + ifeq ($(subdir),debug) test-xfail-tst-ssp-1 = $(have-ssp) endif # $(subdir) == debug
sysdeps/unix/sysv/linux/hppa/Versions +3 -0
@@ -23,4 +23,7 @@ libc { GLIBC_2.19 { fanotify_mark; } + GLIBC_2.40 { + cacheflush; + } }
sysdeps/unix/sysv/linux/hppa/libc.abilist +1 -0
@@ -2821,6 +2821,7 @@ GLIBC_2.4 sys_errlist D 0x400 GLIBC_2.4 sys_nerr D 0x4 GLIBC_2.4 unlinkat F GLIBC_2.4 unshare F +GLIBC_2.40 cacheflush F GLIBC_2.5 __readlinkat_chk F GLIBC_2.5 inet6_opt_append F GLIBC_2.5 inet6_opt_find F
sysdeps/unix/sysv/linux/hppa/sys/cachectl.h new file +36 -0
@@ -0,0 +1,36 @@ +/* HPPA cache flushing interface. +Copyright (C) 2018-2024 Free Software Foundation, Inc. +This file is part of the GNU C Library. + +The GNU C Library is free software; you can redistribute it and/or +modify it under the terms of the GNU Lesser General Public +License as published by the Free Software Foundation; either +version 2.1 of the License, or (at your option) any later version. + +The GNU C Library is distributed in the hope that it will be useful, +but WITHOUT ANY WARRANTY; without even the implied warranty of +MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU +Lesser General Public License for more details. + +You should have received a copy of the GNU Lesser General Public +License along with the GNU C Library. If not, see +<https://www.gnu.org/licenses/>. */ + +#ifndef _SYS_CACHECTL_H +#define _SYS_CACHECTL_H 1 + +#include <features.h> + +/* Get the kernel definition for the op bits. */ +#include <asm/cachectl.h> + +__BEGIN_DECLS + +#ifdef __USE_MISC +extern int cacheflush (void *__addr, const int __nbytes, + const int __op) __THROW; +#endif + +__END_DECLS + +#endif /* sys/cachectl.h */
sysdeps/unix/sysv/linux/hppa/syscalls.list +1 -0
@@ -8,5 +8,6 @@ listen - listen i:ii __listen listen shutdown - shutdown i:ii __shutdown shutdown socket - socket i:iii __socket socket socketpair - socketpair i:iiif __socketpair socketpair +cacheflush - cacheflush i:pii _flush_cache cacheflush personality EXTRA personality Ei:i __personality personality
* Helge Deller:
Quoted 6 lines
sysdeps/unix/sysv/linux/hppa/sys/cachectl.h new file +0
-0
sysdeps/unix/sysv/linux/hppa/sys/cachectl.h line 25
Quoted 8 lines
sysdeps/unix/sysv/linux/hppa/sys/cachectl.h new file +7
-0@@ -0,0 +1,36 @@ +#ifndef _SYS_CACHECTL_H +#define _SYS_CACHECTL_H 1 + +#include <features.h> + +/* Get the kernel definition for the op bits. */ +#include <asm/cachectl.h>
This makes this header (<sys/cachectl.h>) unusable with older kernel headers. I think it also results in a test failure with older headers. Is this a problem?
In similar cases, we use LINUX_VERSION_CODE checks (for older compilers) and __has_include to support backports.
Thanks,
Florian
Quoted 9 lines · Florian Weimer
* Helge Deller:
Quoted 6 lines
sysdeps/unix/sysv/linux/hppa/sys/cachectl.h new file +0
-0@@ -0,0 +1,36 @@
Quoted 11 lines
sysdeps/unix/sysv/linux/hppa/sys/cachectl.h line 25
Quoted 7 lines
sysdeps/unix/sysv/linux/hppa/sys/cachectl.h new file +7
-0+#ifndef _SYS_CACHECTL_H +#define _SYS_CACHECTL_H 1 + +#include <features.h> + +/* Get the kernel definition for the op bits. */ +#include <asm/cachectl.h> This makes this header (<
sys/cachectl.h>) unusable with older kernel headers. I think it also results in a test failure with older headers. Is this a problem?
hppa lives in debian unstable, so basically you should always use the latest kernel & kernel headers when upgrading glibc. So, personally I think it's ok to ask people to upgrade. Dave, any opinion on this?
Quoted 2 lines
In similar cases, we use
LINUX_VERSION_CODEchecks (for older compilers) and__has_includeto support backports.
I'd prefer to not add additional checks.
Another (maybe easier) possibility is to simply backport the kernel asm/cachectl.h file (just the header, not the implementation, so the syscall will return -ENOSYS on older kernels):
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=c6d96328fecdda16e12f3b3c33f3677f4bcef89f
Helge
Am Montag, 8. Juli 2024, 10:58:35 CEST schrieb Helge Deller:
Quoted 22 lines · Florian Weimer
Quoted 9 lines
* Helge Deller:
Quoted 6 lines
sysdeps/unix/sysv/linux/hppa/sys/cachectl.h new file +0
-0@@ -0,0 +1,36 @@Quoted 11 lines
sysdeps/unix/sysv/linux/hppa/sys/cachectl.h line 25
Quoted 7 lines
sysdeps/unix/sysv/linux/hppa/sys/cachectl.h new file +7
-0+#ifndef _SYS_CACHECTL_H +#define _SYS_CACHECTL_H 1 + +#include <features.h> + +/* Get the kernel definition for the op bits. */ +#include <asm/cachectl.h> This makes this header (<
sys/cachectl.h>) unusable with older kernel headers. I think it also results in a test failure with older headers. Is this a problem?
Quoted 2 lines
hppa lives in debian unstable, so basically you should always use the latest kernel & kernel headers when upgrading glibc.
Ahem.
https://www.gentoo.org/downloads/#hppa
Quoted 14 lines
So, personally I think it's ok to ask people to upgrade. Dave, any opinion on this?
Quoted 2 lines
In similar cases, we use
LINUX_VERSION_CODEchecks (for older compilers) and__has_includeto support backports.I'd prefer to not add additional checks.
Another (maybe easier) possibility is to simply backport the kernelasm/cachectl.hfile (just the header, not the implementation, so the syscall will return-ENOSYSon older kernels):
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=c6d96328fecdda16e12f3b3c33f3677f4bcef89fHelge
Signature · 4 lines
Andreas K. Hüttel [email protected] Gentoo Linux developer (council, toolchain, base-system, perl, libreoffice)
Quoted 23 lines · Andreas K. Huettel
Am Montag, 8. Juli 2024, 10:58:35 CEST schrieb Helge Deller:
Quoted 22 lines · Florian Weimer
Quoted 20 lines
* Helge Deller:
Quoted 6 lines
sysdeps/unix/sysv/linux/hppa/sys/cachectl.h new file +0
-0sysdeps/unix/sysv/linux/hppa/sys/cachectl.h line 25
Quoted 8 lines
sysdeps/unix/sysv/linux/hppa/sys/cachectl.h new file +7
-0@@ -0,0 +1,36 @@ +#ifndef _SYS_CACHECTL_H +#define _SYS_CACHECTL_H 1 + +#include <features.h> + +/* Get the kernel definition for the op bits. */ +#include <asm/cachectl.h> This makes this header (<
sys/cachectl.h>) unusable with older kernel headers. I think it also results in a test failure with older headers. Is this a problem?
Quoted 5 lines
Quoted 2 lines
hppa lives in debian unstable, so basically you should always use the latest kernel & kernel headers when upgrading glibc.
Sure. But I didn't mentioned it, because Gentoo usually uses a more recent kernel than Debian, right?
Helge
Quoted 24 lines · Helge Deller
Quoted 23 lines · Andreas K. Huettel
Am Montag, 8. Juli 2024, 10:58:35 CEST schrieb Helge Deller:
Quoted 22 lines · Florian Weimer
Quoted 20 lines
* Helge Deller:
Quoted 6 lines
sysdeps/unix/sysv/linux/hppa/sys/cachectl.h new file +0
-0sysdeps/unix/sysv/linux/hppa/sys/cachectl.h line 25
Quoted 8 lines
sysdeps/unix/sysv/linux/hppa/sys/cachectl.h new file +7
-0@@ -0,0 +1,36 @@ +#ifndef _SYS_CACHECTL_H +#define _SYS_CACHECTL_H 1 + +#include <features.h> + +/* Get the kernel definition for the op bits. */ +#include <asm/cachectl.h> This makes this header (<
sys/cachectl.h>) unusable with older kernel headers. I think it also results in a test failure with older headers. Is this a problem?
Quoted 9 lines
Quoted 5 lines
Quoted 2 lines
hppa lives in debian unstable, so basically you should always use the latest kernel & kernel headers when upgrading glibc.
Sure. But I didn't mentioned it, because Gentoo usually uses a more recent kernel than Debian, right?
In this instance, I wonder if we should do the extra checks. The kernel instability because of the cache / TLB issues means at least one of our machines runs an older kernel for now. (Dave's latest patches seem to help a lot there, but you get the point.)
thanks,
sam
Quoted 4 lines · Sam James
In this instance, I wonder if we should do the extra checks. The kernel instability because of the cache / TLB issues means at least one of our machines runs an older kernel for now. (Dave's latest patches seem to help a lot there, but you get the point.)
Can you be more specific? Do you still experience instability due to cache / TLB issues on this machine?
Dave
Signature · 1 line
John David Anglin [email protected]
Quoted 7 lines · John David Anglin
Quoted 4 lines · Sam James
In this instance, I wonder if we should do the extra checks. The kernel instability because of the cache / TLB issues means at least one of our machines runs an older kernel for now. (Dave's latest patches seem to help a lot there, but you get the point.)
Can you be more specific? Do you still experience instability due to cache / TLB issues on this machine?
Sorry for the delay. Our main machine, 'muta', has been suffering instability for a while. It ended up also having bad RAM and failing disks. We've now replaced them last night.
We had a lot of kernel crashes since then but after upgrading to 6.10.11, everything seems okay so far. Fingers crossed, and then I can get back to the other bits (like testing the binutils patch).
(The panics were all in pte_offset_map_nolock.)
Quoted 2 lines
Dave
thanks,
sam
Quoted 10 lines · Sam James
Quoted 7 lines · John David Anglin
Quoted 4 lines · Sam James
In this instance, I wonder if we should do the extra checks. The kernel instability because of the cache / TLB issues means at least one of our machines runs an older kernel for now. (Dave's latest patches seem to help a lot there, but you get the point.)
Can you be more specific? Do you still experience instability due to cache / TLB issues on this machine?
Quoted 6 lines
Sorry for the delay. Our main machine, 'muta', has been suffering instability for a while. It ended up also having bad RAM and failing disks. We've now replaced them last night.
We had a lot of kernel crashes since then but after upgrading to 6.10.11, everything seems okay so far.
FYI, Dave's commit 72d95924ee35 ("parisc: Try to fix random segmentation faults in package builds") [1] was a game-changer for the parisc kernel. It went into kernel 6.10 and since then
our 64-bit debian builds run absolutely stable.