| Message ID | 57347BD4.8010105@huawei.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Thursday 12 May 2016 20:49:24 Zhangjian wrote: > Hi, > > On 2016/5/12 17:21, Arnd Bergmann wrote: > > On Thursday 12 May 2016 10:17:58 Catalin Marinas wrote: > >> On Wed, May 11, 2016 at 09:30:07PM +0200, Arnd Bergmann wrote: > >>> On Wednesday 11 May 2016 17:59:01 Catalin Marinas wrote: > >>> > >>> I don't think the shifts are a problem, the main downside would be > >>> the limit to 44 bits of file offsets (16TB files), but it's also > >>> unclear if that is a practical problem at all. If it is, we run > >>> into the same problem on all other 32-bit architectures too. > >> > >> I hope people are seriously thinking of moving to an LP64 ABI if they > >> have such large file offset needs. > > > > Good point. 44 bits of file size is certainly enough for mmap() > > on a 32-bit task: you would only be able to map a very small fraction > > of the file anyway, and if you want to map larger files, and should > > move to 64-bit tasks long before this becomes a limitation. > Hi, > > I apply the following patch in order to make use of the REAL mmmap2. LTP > test pass in litle endian. mmap16 successful with segfault in big endian. > > BTW, I saw the similar code in tile, mips, microblaze and s390 compat. Should > we merge these code into a common syscall wrapper? I think that's a good idea. The function used to be slightly different for each architecture, but now it seems we have a significant number of identical implementations that we could just merge them together into one. sys_mmap_pgoff was originally introduced as the common implementation and it reduced the amount of duplication a lot, but as its units are based on PAGE_SIZE rather than hardwired 4096 bytes, it's not as useful. > diff --git a/arch/arm64/kernel/sys_ilp32.c b/arch/arm64/kernel/sys_ilp32.c > index d85fe94..2cd72eb 100644 > --- a/arch/arm64/kernel/sys_ilp32.c > +++ b/arch/arm64/kernel/sys_ilp32.c > @@ -41,7 +41,16 @@ > #define compat_sys_sync_file_range sys_sync_file_range > #define compat_sys_truncate64 sys_truncate > #define sys_llseek sys_lseek > -#define sys_mmap2 sys_mmap > + > +SYSCALL_DEFINE6(mmap2, unsigned long, addr, unsigned long, len, > + unsigned long, prot, unsigned long, flags, unsigned long, fd, > + unsigned long, pgoff) > +{ > + if (pgoff & (~PAGE_MASK >> 12)) > + return -EINVAL; > + > + return sys_mmap_pgoff(addr, len, prot, flags, fd, pgoff >> (PAGE_SHIFT-12)); > +} > Looks good to me. Arnd
On Thu, May 12, 2016 at 03:06:39PM +0200, Arnd Bergmann wrote: > On Thursday 12 May 2016 20:49:24 Zhangjian wrote: > > Hi, > > > > On 2016/5/12 17:21, Arnd Bergmann wrote: > > > On Thursday 12 May 2016 10:17:58 Catalin Marinas wrote: > > >> On Wed, May 11, 2016 at 09:30:07PM +0200, Arnd Bergmann wrote: > > >>> On Wednesday 11 May 2016 17:59:01 Catalin Marinas wrote: > > >>> > > >>> I don't think the shifts are a problem, the main downside would be > > >>> the limit to 44 bits of file offsets (16TB files), but it's also > > >>> unclear if that is a practical problem at all. If it is, we run > > >>> into the same problem on all other 32-bit architectures too. > > >> > > >> I hope people are seriously thinking of moving to an LP64 ABI if they > > >> have such large file offset needs. > > > > > > Good point. 44 bits of file size is certainly enough for mmap() > > > on a 32-bit task: you would only be able to map a very small fraction > > > of the file anyway, and if you want to map larger files, and should > > > move to 64-bit tasks long before this becomes a limitation. > > Hi, > > > > I apply the following patch in order to make use of the REAL mmmap2. LTP > > test pass in litle endian. mmap16 successful with segfault in big endian. > > > > BTW, I saw the similar code in tile, mips, microblaze and s390 compat. Should > > we merge these code into a common syscall wrapper? > > I think that's a good idea. The function used to be slightly different > for each architecture, but now it seems we have a significant number > of identical implementations that we could just merge them together > into one. > > sys_mmap_pgoff was originally introduced as the common implementation > and it reduced the amount of duplication a lot, but as its units > are based on PAGE_SIZE rather than hardwired 4096 bytes, it's > not as useful. > microblaze and mips (twice) are doing like this. And aarh32 as well, in arch/arm64/kernel/entry32.S In previous submissions it was a patch that shares aarch32 code to ilp32. If we decided turn around again, I think, we'd pick up that patch. The other option is to make this hack generic, as so many arches use it.
diff --git a/arch/arm64/kernel/sys_ilp32.c b/arch/arm64/kernel/sys_ilp32.c index d85fe94..2cd72eb 100644 --- a/arch/arm64/kernel/sys_ilp32.c +++ b/arch/arm64/kernel/sys_ilp32.c @@ -41,7 +41,16 @@ #define compat_sys_sync_file_range sys_sync_file_range #define compat_sys_truncate64 sys_truncate #define sys_llseek sys_lseek -#define sys_mmap2 sys_mmap + +SYSCALL_DEFINE6(mmap2, unsigned long, addr, unsigned long, len, + unsigned long, prot, unsigned long, flags, unsigned long, fd, + unsigned long, pgoff) +{ + if (pgoff & (~PAGE_MASK >> 12)) + return -EINVAL; + + return sys_mmap_pgoff(addr, len, prot, flags, fd, pgoff >> (PAGE_SHIFT-12)); +} asmlinkage long ilp32_sys_rt_sigreturn_wrapper(void); #define compat_sys_rt_sigreturn ilp32_sys_rt_sigreturn_wrapper glibc: diff --git a/sysdeps/unix/sysv/linux/aarch64/ilp32/mmap.c b/sysdeps/unix/sysv/linux/aarch64/ilp32/mmap.c index e69de29..f75e251 100644 --- a/sysdeps/unix/sysv/linux/aarch64/ilp32/mmap.c +++ b/sysdeps/unix/sysv/linux/aarch64/ilp32/mmap.c @@ -0,0 +1,2 @@ +#include <sysdeps/unix/sysv/linux/generic/wordsize-32/mmap.c> + diff --git a/sysdeps/unix/sysv/linux/aarch64/ilp32/mmap64.c b/sysdeps/unix/sysv/linux/aarch64/ilp32/mmap64.c index c6c7f1d..6f1a141 100644 --- a/sysdeps/unix/sysv/linux/aarch64/ilp32/mmap64.c +++ b/sysdeps/unix/sysv/linux/aarch64/ilp32/mmap64.c @@ -1,29 +1 @@ -#include <errno.h> -#include <unistd.h> -#include <sys/mman.h> -#include <string.h> - -#include <sysdep.h> -#include <sys/syscall.h> - -/* mmap is provided by mmap as they are the same. */ -void *__mmap (void *__addr, size_t __len, int __prot, - int __flags, int __fd, __off_t __offset) -{ - void *result; - result = (void *) INLINE_SYSCALL (mmap2, 6, __addr, __len, __prot, __flags, - return result; -} -/* mmap64 is provided by mmap as they are the same. */ -void *__mmap64 (void *__addr, size_t __len, int __prot, - int __flags, int __fd, __off64_t __offset) -{ - void *result; - result = (void *) - INLINE_SYSCALL (mmap2, 6, __addr, - __len, __prot, __flags, __fd, __offset); - return result; -} -weak_alias (__mmap, mmap) -weak_alias (__mmap64, mmap64) +#include <sysdeps/unix/sysv/linux/mmap64.c>