Andrea Arcangeli: 2.4.10pre2aa1
Aug 30, 2001, 15:03 (0 Talkback[s])
WEBINAR: On-demand Event
Replace Oracle with the NoSQL Engagement Database: Why and how leading companies are making the switch REGISTER >
Date: Thu, 30 Aug 2001 16:11:53 +0200
From: Andrea Arcangeli
Only in 2.4.10pre2aa1: 00_clear_pagetable-smp-1
Clear the pagetables only under the page_table_lock
Only in 2.4.10pre2aa1: 00_copy_strings-str-1
Skip a not needed check (if str is null strnlen_user will
take care of returning 0).
Only in 2.4.10pre2aa1: 00_delayed-writes-fix-1
Fix delayed writes for numa systems (with numa emulation enabled it
takes 10 minutes to delete a kernel tree in cache without this fix
Only in 2.4.10pre2aa1: 00_ext2-lookup-rotate-1
Heuristic to optimize stat/unlink etc... on ext2. (from 2.4.8ac8)
Only in 2.4.10pre2aa1: 00_km_user-1
Avoid wasting PAGE_SIZE*NR_CPUS bytes of kernel virtual address space.
Only in 2.4.10pre2aa1: 00_raid1-prealloc-1
Avoid losing preallocated bh for raid1. (from Neil Brown)
Only in 2.4.9aa3: 00_rwsem-17
Only in 2.4.10pre2aa1: 00_rwsem-18
Rediffed due rejects.
Only in 2.4.10pre2aa1: 00_smp_call_function-1
Fix a few races, really the two mb() are probably not needed in
practice because while writes can pass reads, the other way around
shouldn't happen on x86. The wmb() seems needed instead for the
compiler (a barrier() would be enough as well but wmb() is nicer
Only in 2.4.9aa3: 00_softirq-fixes-6_wait-network-fixes-1
Only in 2.4.10pre2aa1: 10_softirq-wait-network-fixes-2
Only in 2.4.10pre2aa1: 00_softirq-irq-1
Russell King raised the issue that the risc archs cannot easily
implement a lightweight __cpu_raise_softirq atomic with respect
to irqs. Since we are just running almost all the __cpu_raise_softirq
and cpu_raise_softirq with irq disabled right now, __cpu_raise_softirq
is been reimplemented in C and it must be run with irq disabled (same
applies to cpu_raise_softirq). Only raise_softirq (unused at the
moment) can be called with irq enabled.
Only in 2.4.10pre2aa1: 00_strnlen_user-x86-ret1-1
Andreas Schwab noticed that the x86 strnlen_user isn't returning
a value > N if N = 0. This leads to execve to return -EFAULT
instead of -E2BIG when bprm->p is zero.
Only in 2.4.9aa3: 10_prefetch-2
Only in 2.4.10pre2aa1: 10_prefetch-3
Really use prefetchw (from -ac)
Only in 2.4.9aa3: 40_blkdev-pagecache-14
Only in 2.4.10pre2aa1: 40_blkdev-pagecache-15
Only in 2.4.10pre2aa1: 50_uml-patch-2.4.9-2.bz2
Only in 2.4.9aa3: 50_uml-patch-2.4.9.bz2
Picked last update from sourceforge.
Only in 2.4.9aa3: 60_atomic-alloc-3
Only in 2.4.10pre2aa1: 60_atomic-alloc-4
Only in 2.4.9aa3: 70_mmap-rb-5
Only in 2.4.10pre2aa1: 70_mmap-rb-6
Drop all the init_mmap and INIT_MMAP.
Only in 2.4.10pre2aa1: 71_mmap-rb-6_other-archs-1
Same as above but for the other archs.