Date: Mon, 3 Nov 2014 13:14:34 +0000 (UTC) From: Mateusz Guzik <mjg@FreeBSD.org> To: src-committers@freebsd.org, svn-src-all@freebsd.org, svn-src-head@freebsd.org Subject: svn commit: r274048 - head/sys/sys Message-ID: <201411031314.sA3DEYX7096436@svn.freebsd.org>
next in thread | raw e-mail | index | archive | help
Author: mjg Date: Mon Nov 3 13:14:34 2014 New Revision: 274048 URL: https://svnweb.freebsd.org/changeset/base/274048 Log: Fix misplaced read memory barrier in seq. Impact on capability races was small: it was possible to get a spurious ENOTCAPABLE (early return), but it was not possible to bypass checks. Tidy up some comments. Modified: head/sys/sys/seq.h Modified: head/sys/sys/seq.h ============================================================================== --- head/sys/sys/seq.h Mon Nov 3 13:02:58 2014 (r274047) +++ head/sys/sys/seq.h Mon Nov 3 13:14:34 2014 (r274048) @@ -70,16 +70,16 @@ typedef uint32_t seq_t; #include <machine/cpu.h> /* - * This is a temporary hack until memory barriers are cleaned up. + * Stuff below is going away when we gain suitable memory barriers. * * atomic_load_acq_int at least on amd64 provides a full memory barrier, - * in a way which affects perforance. + * in a way which affects performance. * * Hack below covers all architectures and avoids most of the penalty at least - * on amd64. + * on amd64 but still has unnecessary cost. */ static __inline int -atomic_load_acq_rmb_int(volatile u_int *p) +atomic_load_rmb_int(volatile u_int *p) { volatile u_int v; @@ -88,6 +88,16 @@ atomic_load_acq_rmb_int(volatile u_int * return (v); } +static __inline int +atomic_rmb_load_int(volatile u_int *p) +{ + volatile u_int v = 0; + + atomic_load_acq_int(&v); + v = *p; + return (v); +} + static __inline bool seq_in_modify(seq_t seqp) { @@ -117,7 +127,7 @@ seq_read(seq_t *seqp) seq_t ret; for (;;) { - ret = atomic_load_acq_rmb_int(seqp); + ret = atomic_load_rmb_int(seqp); if (seq_in_modify(ret)) { cpu_spinwait(); continue; @@ -132,7 +142,7 @@ static __inline seq_t seq_consistent(seq_t *seqp, seq_t oldseq) { - return (atomic_load_acq_rmb_int(seqp) == oldseq); + return (atomic_rmb_load_int(seqp) == oldseq); } static __inline seq_t
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?201411031314.sA3DEYX7096436>