Date: Wed, 3 Jul 2013 17:39:19 +0000 (UTC) From: Attilio Rao <attilio@FreeBSD.org> To: src-committers@freebsd.org, svn-src-user@freebsd.org Subject: svn commit: r252578 - user/attilio/vmobj-readlock/sys/vm Message-ID: <201307031739.r63HdJfk052012@svn.freebsd.org>
next in thread | raw e-mail | index | archive | help
Author: attilio Date: Wed Jul 3 17:39:19 2013 New Revision: 252578 URL: http://svnweb.freebsd.org/changeset/base/252578 Log: Improve comments about busy_lock implementation and usage. Sponsored by: EMC / Isilon storage division Modified: user/attilio/vmobj-readlock/sys/vm/vm_page.h Modified: user/attilio/vmobj-readlock/sys/vm/vm_page.h ============================================================================== --- user/attilio/vmobj-readlock/sys/vm/vm_page.h Wed Jul 3 17:27:32 2013 (r252577) +++ user/attilio/vmobj-readlock/sys/vm/vm_page.h Wed Jul 3 17:39:19 2013 (r252578) @@ -112,14 +112,13 @@ * * Page content access is regulated (mostly) by the busy mechanism. * When the page content is changing (for example, during a page READ - * operation) the page owner must acquire an hard busy token. Similarly, + * operation) the page owner must acquire an write busy token. Similarly, * when the page content is just being accessed for reading purposes * (for example, during a page WRITE operation) the page owner must - * acquire a soft busy token. - * The hard busy mechanism is controlled using vm_page_busy() and - * vm_page_unbusy() interfaces. Likewise the soft busy mechanism is - * controlled through the usage of vm_page_io_start() and - * vm_page_io_finish(). + * acquire a read busy token. + * The busy mechanism is regulated by the vm_page_busy_*() family of + * primitives and it follows a reader/writer lock pattern as the names + * suggest. */ #if PAGE_SIZE == 4096 @@ -184,7 +183,11 @@ struct vm_page { #define VPO_NOSYNC 0x10 /* do not collect for syncer */ /* - * ARXXX: Insert comments for busy here. + * Busy lock implementation details. + * The algorithm is taken mostly by rwlock(9) and sx(9) locks implementation, + * even if the support for owner identity is removed because of size + * constraints. Checks on lock recursion are then not possible, while the + * lock assertions effectiveness is someway reduced. */ #define VPB_LOCK_READ 0x01 #define VPB_LOCK_WRITE 0x02
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?201307031739.r63HdJfk052012>