1 commit 5651fadddd81ecbab1e4fa33878cc8d0b1ddae6c
2 Author: Oliver Pinter <oliver.pntr@gmail.com>
3 Date: Sat Mar 29 16:20:25 2008 +0100
7 Signed-off-by: Oliver Pinter <oliver.pntr@gmail.com>
9 commit 1ee0bb06edf100344d5b940f58439575863a6187
10 Author: andrew.patterson@hp.com <andrew.patterson@hp.com>
11 Date: Sat Mar 29 16:18:53 2008 +0100
13 MCA when shutting down tulip quad-NIC
15 Shutting down the network causes an MCA because of an IO TLB error when
16 a DEC quad 10/100 card is in any slot. This problem was originally seen
19 Acked-by: Olaf Kirch <okir@suse.de>
20 Signed-off-by: Oliver Pinter <oliver.pntr@gmail.com>
22 drivers/net/tulip/tulip_core.c | 4 ++++
23 1 file changed, 4 insertions(+)
25 commit b362b7edb1a3b9ba69eecb0cbd95fd4c4ba96994
26 Author: Joe Korty <joe.korty@ccur.com>
27 Date: Wed Mar 5 15:04:59 2008 -0800
29 slab: NUMA slab allocator migration bugfix
31 NUMA slab allocator cpu migration bugfix
33 The NUMA slab allocator (specifically, cache_alloc_refill)
34 is not refreshing its local copies of what cpu and what
35 numa node it is on, when it drops and reacquires the irq
36 block that it inherited from its caller. As a result
37 those values become invalid if an attempt to migrate the
38 process to another numa node occured while the irq block
41 The solution is to make cache_alloc_refill reload these
42 variables whenever it drops and reacquires the irq block.
44 The error is very difficult to hit. When it does occur,
45 one gets the following oops + stack traceback bits in
46 check_spinlock_acquired:
48 kernel BUG at mm/slab.c:2417
49 cache_alloc_refill+0xe6
53 This patch was developed against 2.6.23, ported to and
54 compiled-tested only against 2.6.25-rc4.
56 Signed-off-by: Joe Korty <joe.korty@ccur.com>
57 Signed-off-by: Christoph Lameter <clameter@sgi.com>
58 Signed-off-by: Oliver Pinter <oliver.pntr@gmail.com>
60 commit 078288fc83e496c7965fa10b694353125c349ef7
61 Author: NeilBrown <neilb@suse.de>
62 Date: Wed Feb 6 01:40:00 2008 -0800
64 md: fix an occasional deadlock in raid5
66 raid5's 'make_request' function calls generic_make_request on underlying
67 devices and if we run out of stripe heads, it could end up waiting for one of
68 those requests to complete. This is bad as recursive calls to
69 generic_make_request go on a queue and are not even attempted until
70 make_request completes.
72 So: don't make any generic_make_request calls in raid5 make_request until all
73 waiting has been done. We do this by simply setting STRIPE_HANDLE instead of
74 calling handle_stripe().
76 If we need more stripe_heads, raid5d will get called to process the pending
77 stripe_heads which will call generic_make_request from a
79 This change by itself causes a performance hit. So add a change so that
80 raid5_activate_delayed is only called at unplug time, never in raid5. This
81 seems to bring back the performance numbers. Calling it in raid5d was
86 How about we queue it for 2.6.25-rc1 and then about when -rc2 comes out,
87 we queue it for 2.6.24.y?
89 Acked-by: Dan Williams <dan.j.williams@intel.com>
90 Signed-off-by: Neil Brown <neilb@suse.de>
91 Tested-by: dean gaudet <dean@arctic.org>
92 Cc: <stable@kernel.org>
93 Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
94 Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
95 Signed-off-by: Oliver Pinter <oliver.pntr@gmail.com>
97 commit 70a7cd1e8fbe736b854f39efd98528b7dfd64ee1
98 Author: Suresh Jayaraman <sjayaraman@suse.de>
99 Date: Sat Mar 29 16:18:51 2008 +0100
101 Fix sign mount option and sign proc config setting
103 Patch mainline: 2.6.23-rc1
105 Backported the fix (2.6.23-rc1) from Steve French. The original patch removes
106 few commented functions (which are not required) as part of this fix, backport
107 also does the same to retain compatibility.
109 We were checking the wrong (old) global variable to determine
110 whether to override server and force signing on the SMB
113 Acked-by: Dave Kleikamp <shaggy@austin.ibm.com>
114 Signed-off-by: Steve French <sfrench@us.ibm.com>
115 Signed-off-by: Suresh Jayaraman <sjayaraman@suse.de>
116 Signed-off-by: Oliver Pinter <oliver.pntr@gmail.com>
118 commit 982b92285ec9f45e1658c95376f9e1a47ba47090
119 Author: Steve French <sfrench@us.ibm.com>
120 Date: Tue Nov 13 22:41:37 2007 +0000
122 Fix buffer overflow if server sends corrupt response to small
124 [CIFS] Fix buffer overflow if server sends corrupt response to small
127 In SendReceive() function in transport.c - it memcpy's
128 message payload into a buffer passed via out_buf param. The function
129 assumes that all buffers are of size (CIFSMaxBufSize +
130 MAX_CIFS_HDR_SIZE) , unfortunately it is also called with smaller
131 (MAX_CIFS_SMALL_BUFFER_SIZE) buffers. There are eight callers
132 (SMB worker functions) which are primarily affected by this change:
134 TreeDisconnect, uLogoff, Close, findClose, SetFileSize, SetFileTimes,
137 CC: Dave Kleikamp <shaggy@austin.ibm.com>
138 CC: Przemyslaw Wegrzyn <czajnik@czajsoft.pl>
139 Acked-by: Jeff Layton <jlayton@redhat.com>
140 Signed-off-by: Steve French <sfrench@us.ibm.com>
141 Acked-by: Jeff Mahoney <jeffm@suse.com>
142 Signed-off-by: Oliver Pinter <oliver.pntr@gmail.com>
144 commit 54ddb8eee828607fb70a553b9473fe671fc2c87e
145 Author: Oliver Pinter <oliver.pntr@gmail.com>
146 Date: Tue Feb 26 17:06:12 2008 +0100
148 linux v2.6.22.20-op1-rc1
150 Signed-off-by: Oliver Pinter <oliver.pntr@gmail.com>
152 commit dbd68c840f69fe6721fb609d0eb8a5d548e76a99
153 Author: Kurt Garloff <garloff@suse.de>
154 Date: Mon Feb 18 21:16:27 2008 +0100
156 make (low) swappiness safer to use
159 make swappiness safer to use
160 has been added to the -mm tree. Its filename is
161 make-swappiness-safer-to-use.patch
163 See http://www.zip.com.au/~akpm/linux/patches/stuff/added-to-mm.txt to find
164 out what to do about this
166 #Subject: make swappiness safer to use
167 #From: Andrea Arcangeli <andrea@suse.de>
169 Swappiness isn't a safe sysctl. Setting it to 0 for example can hang a
170 system. That's a corner case but even setting it to 10 or lower can waste
171 enormous amounts of cpu without making much progress. We've customers who
172 wants to use swappiness but they can't because of the current
173 implementation (if you change it so the system stops swapping it really
174 stops swapping and nothing works sane anymore if you really had to swap
175 something to make progress).
177 This patch from Kurt Garloff makes swappiness safer to use (no more huge
178 cpu usage or hangs with low swappiness values).
180 I think the prev_priority can also be nuked since it wastes 4 bytes per
181 zone (that would be an incremental patch but I wait the nr_scan_[in]active
182 to be nuked first for similar reasons). Clearly somebody at some point
183 noticed how broken that thing was and they had to add min(priority,
184 prev_priority) to give it some reliability, but they didn't go the last
185 mile to nuke prev_priority too. Calculating distress only in function of
186 not-racy priority is correct and sure more than enough without having to
187 add randomness into the equation.
189 Patch is tested on older kernels but it compiles and it's quite simple
192 Overall I'm not very satisified by the swappiness tweak, since it doesn't
193 rally do anything with the dirty pagecache that may be inactive. We need
194 another kind of tweak that controls the inactive scan and tunes the
195 can_writepage feature (not yet in mainline despite having submitted it a
196 few times), not only the active one. That new tweak will tell the kernel
197 how hard to scan the inactive list for pure clean pagecache (something the
198 mainline kernel isn't capable of yet). We already have that feature
199 working in all our enterprise kernels with the default reasonable tune, or
200 they can't even run a readonly backup with tar without triggering huge
201 write I/O. I think it should be available also in mainline later.
203 Cc: Nick Piggin <npiggin@suse.de>
204 Signed-off-by: Kurt Garloff <garloff@suse.de>
205 Signed-off-by: Andrea Arcangeli <andrea@suse.de>
206 Signed-off-by: Fengguang Wu <wfg@mail.ustc.edu.cn>
207 Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
208 Signed-off-by: Oliver Pinter <oliver.pntr@gmail.com>
210 commit eef55466b00265cfe843ae58edd608a29ad0fc5e
211 Author: David Howells <dhowells@redhat.com>
212 Date: Thu Feb 21 16:12:45 2008 +0000
214 MM: Fix macro argument substitution in PageHead() and PageTail()
216 Fix macro argument substitution in PageHead() and PageTail() - 'page' should
217 have brackets surrounding it (commit 6d7779538f765963ced45a3fa4bed7ba8d2c277d).
219 Signed-off-by: David Howells <dhowells@redhat.com>
220 Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
221 Signed-off-by: Oliver Pinter <oliver.pntr@gmail.com>
223 commit 2cd082a9a5860afa64b032fc6df230e25d913289
224 Author: Thomas Gleixner <tglx@linutronix.de>
225 Date: Wed Feb 20 00:29:02 2008 +0100
227 genirq: do not leave interupts enabled on free_irq
229 commit 89d694b9dbe769ca1004e01db0ca43964806a611
231 The default_disable() function was changed in commit:
233 76d2160147f43f982dfe881404cfde9fd0a9da21
234 genirq: do not mask interrupts by default
236 It removed the mask function in favour of the default delayed
237 interrupt disabling. Unfortunately this also broke the shutdown in
238 free_irq() when the last handler is removed from the interrupt for
239 those architectures which rely on the default implementations. Now we
240 can end up with a enabled interrupt line after the last handler was
241 removed, which can result in spurious interrupts.
243 Fix this by adding a default_shutdown function, which is only
244 installed, when the irqchip implementation does provide neither a
245 shutdown nor a disable function.
247 Pointed-out-by: Michael Hennerich <Michael.Hennerich@analog.com>
248 Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
249 Acked-by: Ingo Molnar <mingo@elte.hu>
250 Tested-by: Michael Hennerich <Michael.Hennerich@analog.com>
251 Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
252 Signed-off-by: Oliver Pinter <oliver.pntr@gmail.com>
254 commit c3bec7bc563415560734d190f33083716bb1cb68
255 Author: Ingo Molnar <mingo@elte.hu>
256 Date: Fri Feb 15 20:58:22 2008 +0100
258 x86_64: CPA, fix cache attribute inconsistency bug, v2.6.22 backport
260 fix CPA cache attribute bug in v2.6.2[234]. When phys_base is nonzero
261 (when CONFIG_RELOCATABLE=y) then change_page_attr_addr() miscalculates
262 the secondary alias address by -14 MB (depending on the configured
265 The default 64-bit kernels of Fedora and Ubuntu are affected:
267 $ grep RELOCA /boot/config-2.6.23.9-85.fc8
270 $ grep RELOC /boot/config-2.6.22-14-generic
273 and probably on many other distros as well.
275 the bug affects all pages in the first 40 MB of physical RAM that
276 are allocated by some subsystem that does ioremap_nocache() on them:
278 if (__pa(address) < KERNEL_TEXT_SIZE) {
280 Hence we might leave page table entries with inconsistent cache
281 attributes around (pages mapped at both UnCacheable and Write-Back),
282 and we can also set the wrong kernel text pages to UnCacheable.
284 the effects of this bug can be random slowdowns and other misbehavior.
285 If for example AGP allocates its aperture pages into the first 40 MB
286 of physical RAM, then the -14 MB bug might mark random kernel texto
287 pages as uncacheable, slowing down a random portion of the 64-bit
288 kernel until the AGP driver is unloaded.
290 Signed-off-by: Ingo Molnar <mingo@elte.hu>
291 Acked-by: Thomas Gleixner <tglx@linutronix.de>
292 Signed-off-by: Oliver Pinter <oliver.pntr@gmail.com>