)]}' { "log": [ { "commit": "81b0c8713385ce1b1b9058e916edcf9561ad76d6", "tree": "4c5e8fde3d15503c609d5c5f74911f95fc528f03", "parents": [ "0686cd8fbe3e5fb1441ae84b9cbc813f9297b879" ], "author": { "name": "Andrew Morton", "email": "akpm@osdl.org", "time": "Thu Jun 29 02:24:26 2006 -0700" }, "committer": { "name": "Linus Torvalds", "email": "torvalds@g5.osdl.org", "time": "Thu Jun 29 10:26:20 2006 -0700" }, "message": "[PATCH] generic_file_buffered_write(): handle zero-length iovec segments\n\nThe recent generic_file_write() deadlock fix caused\ngeneric_file_buffered_write() to loop inifinitely when presented with a\nzero-length iovec segment. Fix.\n\nNote that this fix deliberately avoids calling -\u003eprepare_write(),\n-\u003ecommit_write() etc with a zero-length write. This is because I don\u0027t trust\nall filesystems to get that right.\n\nThis is a cautious approach, for 2.6.17.x. For 2.6.18 we should just go ahead\nand call -\u003eprepare_write() and -\u003ecommit_write() with the zero length and fix\nany broken filesystems. So I\u0027ll make that change once this code is stabilised\nand backported into 2.6.17.x.\n\nThe reason for preferring to call -\u003eprepare_write() and -\u003ecommit_write() with\nthe zero-length segment: a zero-length segment _should_ be sufficiently\nuncommon that this is the correct way of handling it. We don\u0027t want to\noptimise for poorly-written userspace at the expense of well-written\nuserspace.\n\nCc: \"Vladimir V. Saveliev\" \u003cvs@namesys.com\u003e\nCc: Neil Brown \u003cneilb@suse.de\u003e\nCc: Martin Schwidefsky \u003cschwidefsky@de.ibm.com\u003e\nCc: Chris Wright \u003cchrisw@sous-sol.org\u003e\nCc: Greg KH \u003cgreg@kroah.com\u003e\nCc: \u003cstable@kernel.org\u003e\nCc: walt \u003cwa1ter@myrealbox.com\u003e\nSigned-off-by: Andrew Morton \u003cakpm@osdl.org\u003e\nSigned-off-by: Linus Torvalds \u003ctorvalds@osdl.org\u003e\n" }, { "commit": "01408c4939479ec46c15aa7ef6e2406be50eeeca", "tree": "106ee144cc7214cc5cb78bc35a49fc654ef16fe9", "parents": [ "5f507d9e05b4dbfee34f3d967623ad3fbf0f28b3" ], "author": { "name": "NeilBrown", "email": "neilb@suse.de", "time": "Sun Jun 25 05:47:58 2006 -0700" }, "committer": { "name": "Linus Torvalds", "email": "torvalds@g5.osdl.org", "time": "Sun Jun 25 10:01:09 2006 -0700" }, "message": "[PATCH] Prepare for __copy_from_user_inatomic to not zero missed bytes\n\nThe problem is that when we write to a file, the copy from userspace to\npagecache is first done with preemption disabled, so if the source address is\nnot immediately available the copy fails *and* *zeros* *the* *destination*.\n\nThis is a problem because a concurrent read (which admittedly is an odd thing\nto do) might see zeros rather that was there before the write, or what was\nthere after, or some mixture of the two (any of these being a reasonable thing\nto see).\n\nIf the copy did fail, it will immediately be retried with preemption\nre-enabled so any transient problem with accessing the source won\u0027t cause an\nerror.\n\nThe first copying does not need to zero any uncopied bytes, and doing so\ncauses the problem. It uses copy_from_user_atomic rather than copy_from_user\nso the simple expedient is to change copy_from_user_atomic to *not* zero out\nbytes on failure.\n\nThe first of these two patches prepares for the change by fixing two places\nwhich assume copy_from_user_atomic does zero the tail. The two usages are\nvery similar pieces of code which copy from a userspace iovec into one or more\npage-cache pages. These are changed to remove the assumption.\n\nThe second patch changes __copy_from_user_inatomic* to not zero the tail.\nOnce these are accepted, I will look at similar patches of other architectures\nwhere this is important (ppc, mips and sparc being the ones I can find).\n\nThis patch:\n\nThere is a problem with __copy_from_user_inatomic zeroing the tail of the\nbuffer in the case of an error. As it is called in atomic context, the error\nmay be transient, so it results in zeros being written where maybe they\nshouldn\u0027t be.\n\nIn the usage in filemap, this opens a window for a well timed read to see data\n(zeros) which is not consistent with any ordering of reads and writes.\n\nMost cases where __copy_from_user_inatomic is called, a failure results in\n__copy_from_user being called immediately. As long as the latter zeros the\ntail, the former doesn\u0027t need to. However in *copy_from_user_iovec\nimplementations (in both filemap and ntfs/file), it is assumed that\ncopy_from_user_inatomic will zero the tail.\n\nThis patch removes that assumption, so that after this patch it will\nbe safe for copy_from_user_inatomic to not zero the tail.\n\nThis patch also adds some commentary to filemap.h and asm-i386/uaccess.h.\n\nAfter this patch, all architectures that might disable preempt when\nkmap_atomic is called need to have their __copy_from_user_inatomic* \"fixed\".\nThis includes\n - powerpc\n - i386\n - mips\n - sparc\n\nSigned-off-by: Neil Brown \u003cneilb@suse.de\u003e\nCc: David Howells \u003cdhowells@redhat.com\u003e\nCc: Anton Altaparmakov \u003caia21@cantab.net\u003e\nCc: Benjamin Herrenschmidt \u003cbenh@kernel.crashing.org\u003e\nCc: Paul Mackerras \u003cpaulus@samba.org\u003e\nCc: Ralf Baechle \u003cralf@linux-mips.org\u003e\nCc: William Lee Irwin III \u003cwli@holomorphy.com\u003e\nSigned-off-by: Andrew Morton \u003cakpm@osdl.org\u003e\nSigned-off-by: Linus Torvalds \u003ctorvalds@osdl.org\u003e\n" }, { "commit": "c22ce143d15eb288543fe9873e1c5ac1c01b69a1", "tree": "dc7d457b8952fc50dfc90df659b35de4117c61fc", "parents": [ "7dbdf43cfa635ddc3701cc8d1eab07597cd731c0" ], "author": { "name": "Hiro Yoshioka", "email": "hyoshiok@miraclelinux.com", "time": "Fri Jun 23 02:04:16 2006 -0700" }, "committer": { "name": "Linus Torvalds", "email": "torvalds@g5.osdl.org", "time": "Fri Jun 23 07:42:56 2006 -0700" }, "message": "[PATCH] x86: cache pollution aware __copy_from_user_ll()\n\nUse the x86 cache-bypassing copy instructions for copy_from_user().\n\nSome performance data are\n\nTotal of GLOBAL_POWER_EVENTS (CPU cycle samples)\n\n2.6.12.4.orig 1921587\n2.6.12.4.nt 1599424\n1599424/1921587\u003d83.23% (16.77% reduction)\n\nBSQ_CACHE_REFERENCE (L3 cache miss)\n2.6.12.4.orig 57427\n2.6.12.4.nt 20858\n20858/57427\u003d36.32% (63.7% reduction)\n\nL3 cache miss reduction of __copy_from_user_ll\nsamples %\n37408 65.1412 vmlinux __copy_from_user_ll\n23 0.1103 vmlinux __copy_user_zeroing_intel_nocache\n23/37408\u003d0.061% (99.94% reduction)\n\nTop 5 of 2.6.12.4.nt\nCounted GLOBAL_POWER_EVENTS events (time during which processor is not stopped) with a unit mask of 0x01 (mandatory) count 100000\nsamples % app name symbol name\n128392 8.0274 vmlinux __copy_user_zeroing_intel_nocache\n64206 4.0143 vmlinux journal_add_journal_head\n59746 3.7355 vmlinux do_get_write_access\n47674 2.9807 vmlinux journal_put_journal_head\n46021 2.8774 vmlinux journal_dirty_metadata\npattern9-0-cpu4-0-09011728/summary.out\n\nCounted BSQ_CACHE_REFERENCE events (cache references seen by the bus unit) with a unit mask of 0x3f (multiple flags) count 3000\nsamples % app name symbol name\n69755 4.2861 vmlinux __copy_user_zeroing_intel_nocache\n55685 3.4215 vmlinux journal_add_journal_head\n52371 3.2179 vmlinux __find_get_block\n45504 2.7960 vmlinux journal_put_journal_head\n36005 2.2123 vmlinux journal_stop\npattern9-0-cpu4-0-09011744/summary.out\n\nCounted BSQ_CACHE_REFERENCE events (cache references seen by the bus unit) with a unit mask of 0x200 (read 3rd level cache miss) count 3000\nsamples % app name symbol name\n1147 5.4994 vmlinux journal_add_journal_head\n881 4.2240 vmlinux journal_dirty_data\n872 4.1809 vmlinux blk_rq_map_sg\n734 3.5192 vmlinux journal_commit_transaction\n617 2.9582 vmlinux radix_tree_delete\npattern9-0-cpu4-0-09011731/summary.out\n\niozone results are\n\noriginal 2.6.12.4 CPU time \u003d 207.768 sec\ncache aware CPU time \u003d 184.783 sec\n(three times run)\n184.783/207.768\u003d88.94% (11.06% reduction)\n\noriginal:\npattern9-0-cpu4-0-08191720/iozone.out: CPU Utilization: Wall time 45.997 CPU time 64.527 CPU utilization 140.28 %\npattern9-0-cpu4-0-08191741/iozone.out: CPU Utilization: Wall time 46.878 CPU time 71.933 CPU utilization 153.45 %\npattern9-0-cpu4-0-08191743/iozone.out: CPU Utilization: Wall time 45.152 CPU time 71.308 CPU utilization 157.93 %\n\ncache awre:\npattern9-0-cpu4-0-09011728/iozone.out: CPU Utilization: Wall time 44.842 CPU time 62.465 CPU utilization 139.30 %\npattern9-0-cpu4-0-09011731/iozone.out: CPU Utilization: Wall time 44.718 CPU time 59.273 CPU utilization 132.55 %\npattern9-0-cpu4-0-09011744/iozone.out: CPU Utilization: Wall time 44.367 CPU time 63.045 CPU utilization 142.10 %\n\nSigned-off-by: Hiro Yoshioka \u003chyoshiok@miraclelinux.com\u003e\nSigned-off-by: Andrew Morton \u003cakpm@osdl.org\u003e\nSigned-off-by: Linus Torvalds \u003ctorvalds@osdl.org\u003e\n" }, { "commit": "eb6fe0c388e43b02e261f0fdee60e42f6298d7f7", "tree": "3924bfbbbb10afc19f3bcd4963af14121f2c0553", "parents": [ "6d79125bba55ee82701f1c7d4ebbc1aa20ecbe4e" ], "author": { "name": "Carsten Otte", "email": "cotte@de.ibm.com", "time": "Thu Jun 23 22:05:28 2005 -0700" }, "committer": { "name": "Linus Torvalds", "email": "torvalds@ppc970.osdl.org", "time": "Fri Jun 24 00:06:41 2005 -0700" }, "message": "[PATCH] xip: reduce code duplication\n\nThis patch reworks filemap_xip.c with the goal to reduce code duplication\nfrom mm/filemap.c. It applies agains 2.6.12-rc6-mm1. Instead of\nimplementing the aio functions, this one implements the synchronous\nread/write functions only. For readv and writev, the generic fallback is\nused. For aio, we rely on the application doing the fallback. Since our\n\"synchronous\" function does memcpy immediately anyway, there is no\nperformance difference between using the fallbacks or implementing each\noperation.\n\nSigned-off-by: Carsten Otte \u003ccotte@de.ibm.com\u003e\nSigned-off-by: Andrew Morton \u003cakpm@osdl.org\u003e\nSigned-off-by: Linus Torvalds \u003ctorvalds@osdl.org\u003e\n" }, { "commit": "ceffc078528befc008c6f2c2c4decda79eabd534", "tree": "a289e10162bdef0c0d9f6533f1a647b0fe1ed7a9", "parents": [ "420edbcc09008342c7b2665453f6b370739aadb0" ], "author": { "name": "Carsten Otte", "email": "cotte@de.ibm.com", "time": "Thu Jun 23 22:05:25 2005 -0700" }, "committer": { "name": "Linus Torvalds", "email": "torvalds@ppc970.osdl.org", "time": "Fri Jun 24 00:06:41 2005 -0700" }, "message": "[PATCH] xip: fs/mm: execute in place\n\n- generic_file* file operations do no longer have a xip/non-xip split\n- filemap_xip.c implements a new set of fops that require get_xip_page\n aop to work proper. all new fops are exported GPL-only (don\u0027t like to\n see whatever code use those except GPL modules)\n- __xip_unmap now uses page_check_address, which is no longer static\n in rmap.c, and defined in linux/rmap.h\n- mm/filemap.h is now much more clean, plainly having just Linus\u0027\n inline funcs moved here from filemap.c\n- fix includes in filemap_xip to make it build cleanly on i386\n\nSigned-off-by: Carsten Otte \u003ccotte@de.ibm.com\u003e\nSigned-off-by: Andrew Morton \u003cakpm@osdl.org\u003e\nSigned-off-by: Linus Torvalds \u003ctorvalds@osdl.org\u003e\n" } ] }