Date: Mon, 28 Dec 2015 13:01:26 -0700 From: Warner Losh <imp@bsdimp.com> To: John Baldwin <jhb@freebsd.org> Cc: src-committers <src-committers@freebsd.org>, svn-src-head@freebsd.org, svn-src-all@freebsd.org, Warner Losh <imp@freebsd.org> Subject: Re: svn commit: r292809 - head/lib/libc/stdio Message-ID: <CANCZdfq4sMAEzJ-wxNnybJiiTnJV3k5NZgcQACnchHCgpKQxrQ@mail.gmail.com> In-Reply-To: <41508412.yspAtSoPCD@ralph.baldwin.cx> References: <201512272304.tBRN4C5D034464@repo.freebsd.org> <41508412.yspAtSoPCD@ralph.baldwin.cx>
next in thread | previous in thread | raw e-mail | index | archive | help
I'll look at that, but I don't think posix_memalign is the right way to go. The alignment of FILE is more strict than posix_memalign will return. Ian's idea of __alignof__ is the way to go. We allocate them in one block on purpose for performance, and posix_memalign would be a one at a time affair. Warner On Dec 28, 2015 12:42 PM, "John Baldwin" <jhb@freebsd.org> wrote: > On Sunday, December 27, 2015 11:04:12 PM Warner Losh wrote: > > Author: imp > > Date: Sun Dec 27 23:04:11 2015 > > New Revision: 292809 > > URL: https://svnweb.freebsd.org/changeset/base/292809 > > > > Log: > > The FILE structure has a mbstate_t in it. This structure needs to be > > aligned on a int64_t boundary. However, when we allocate the array of > > these structures, we use ALIGNBYTES which defaults to sizeof(int) on > > arm, i386 and others. The i386 stuff can handle unaligned accesses > > seemlessly. However, arm cannot. Take this into account when creating > > the array of FILEs, and add some comments about why. > > > > Differential Revision: https://reviews.freebsd.org/D4708 > > > > Modified: > > head/lib/libc/stdio/findfp.c > > > > Modified: head/lib/libc/stdio/findfp.c > > > ============================================================================== > > --- head/lib/libc/stdio/findfp.c Sun Dec 27 23:04:10 2015 > (r292808) > > +++ head/lib/libc/stdio/findfp.c Sun Dec 27 23:04:11 2015 > (r292809) > > @@ -41,6 +41,7 @@ __FBSDID("$FreeBSD$"); > > #include <unistd.h> > > #include <stdio.h> > > #include <stdlib.h> > > +#include <stdint.h> > > #include <string.h> > > > > #include <spinlock.h> > > @@ -96,11 +97,22 @@ moreglue(int n) > > struct glue *g; > > static FILE empty = { ._fl_mutex = PTHREAD_MUTEX_INITIALIZER }; > > FILE *p; > > + size_t align; > > > > - g = (struct glue *)malloc(sizeof(*g) + ALIGNBYTES + n * > sizeof(FILE)); > > + /* > > + * FILE has a mbstate_t variable. This variable tries to be int64_t > > + * aligned through its definition. int64_t may be larger than void > *, > > + * which is the size traditionally used for ALIGNBYTES. So, use > our own > > + * rounding instead of the MI ALIGN macros. If for some reason > > + * ALIGNBYTES is larger than int64_t, respect that too. There > appears to > > + * be no portable way to ask for FILE's alignment requirements > other > > + * than just knowing here. > > + */ > > + align = MAX(ALIGNBYTES, sizeof(int64_t)); > > + g = (struct glue *)malloc(sizeof(*g) + align + n * sizeof(FILE)); > > if (g == NULL) > > return (NULL); > > - p = (FILE *)ALIGN(g + 1); > > + p = (FILE *)roundup((uintptr_t)(g + 1), align); > > Can this use posix_memalign() rather than doing the alignment by hand? > > -- > John Baldwin >
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?CANCZdfq4sMAEzJ-wxNnybJiiTnJV3k5NZgcQACnchHCgpKQxrQ>