Messages in this thread |  | | Date | Wed, 6 Sep 2000 21:52:07 +0200 (CEST) | From | Andrea Arcangeli <> | Subject | 2.2.18pre2aa2 and patches for 2.2.18pre3 |
| |
The main features of 2.2.18pre2aa2 are:
o Support for 4Gigabyte of RAM on IA32 (me and Gerhard Wichert) o Support for 2T of RAM on alpha (me) o Improved VM for high end machines with enough ram and doing heavy I/O under high memory pressure plus fixes for the MAP_SHARED oom by killing kpiod. (me) o RAW-IO (doable with bigmem enabled too). (Stephen C. Tweedie) o SMP scheduler improvements. (me and partly from 2.3.x contributed by Ingo Molnar) o LFS (>2G file on 32bit architectures) o misc fixes
Detailed description of 2.2.18pre2aa2 follows.
--------------------------------------------------------------------------- 00_4_min_percent-1
Increase the min percent of the buffer cache and page cache to 4%. (it wouldn't matter if the VM algorithms were better). (me)
00_IO-wait-2.gz
Avoid suprious unplug of the I/O queue. (me)
00_SMP-scheduler-2.2.14-G.gz
Better SMP reschedule_idle. (partly backported from 2.3.x, 2.3.x version was contributed by Ingo Molnar)
00_account-failed-buffer-tries-1
Account also the failed buffer tries during shrink_mmap. (me)
00_alpha-epoch-1
Put in sync the RTC parsing done by the kernel at boot while setting the system time, with the parsing done by the RTC driver during initialization. (Jan-Benedict Glaw <jbglaw@lug-owl.de> fixed it in 2.4.x)
00_alpha-read-unlock-SMP-race-1
SMP race fix in the read_unlock implementation of alpha. (me)
00_buf-run_task_queue.gz
Avoid suprious unplug of the I/O queue. (me)
00_delack-timer-5.gz
Additional paranoid stuff in TCP delayed acks code (if somebody can still reproduce lockups under heavy TCP load, the delack-timer-5 patch can be the solution). In 2.2.15 only what it was obviously wrong is been fixed, but at the time I did delack-timer-5 I was convinced the stuff in 2.2.15 _might_ not be enough, so if you have still lockups under 2.2.15 let us know. delack-timer-5 is reported to definitely fix the TCP lockup. (me)
00_elv-simple-latency-1
It simplify the elevator algorithm to remove the careful code that was accounting also the place in the queue where the request was inserted while setting its initial latency. This changed makes the systems much less interactive, but it fixes the tiotest numbers of course still avoiding indefinite starvation as it was happeneing before the elevator rewrite. (me, tested also by Jens Axboe)
00_inode-cleanup-3.gz
Minor inode stuff cleanedup. (me)
00_java-proc.gz
Revertd the semantic change that make differences between /proc/00000$$ and /proc/$$, this allows backwards compatibilty of a misfeature and it _won't_ hurt security. There's no downside in reverting the 2.2.13 semantic change.
00_kupdate-sigstop-2.2.11-1.gz
Allows kupdate to be stopped via SIGSTOP (currently it must be stopped by setting interval to zero via sysctl). This returns useful in the airplane where you do killall -STOP kupdate... STOP kupdate at your own risk of course ;) (me)
00_mremap-waste-virtual-stack-space-1.gz
Avoids mremap to grow in virtual addresses (and so in turn it avoids to waste virtual address space).
00_nanosleep-4.gz
Provide nanosleep usec resolution so that a signal flood doesn't hang glibc folks that correctly trust the rem field to resume the nanosleep after a syscall interruption. (without the patch nanosleep resolution is instead 10msec on IA32 and around 1msec on alpha) (me)
00_overcommit-1.gz
Make sure to not understimate the available memory (the cache and buffers may be under the min percent) (me)
00_set_rtc_mss-SMP-race-1.gz
Fix set_rtc_mss SMP race between timer irq and rtc irq. (me)
00_silent-stack-overflow-4.gz
Avoid stack to silenty overflow on the heap (make life easier to track down userspace issues). Perfect approch is not possible even using special LDT for the stack segment. The approch in the patch will work 99% of the time and it enforces a GAP between heap and stack of 1 page as default (configurable via sysctl, if turned to zero the feature is disabled completly). (me)
00_slow-gtod-SMP-race-1.gz
Fixes an SMP race in slow gettimeofday. (me)
00_stod-lost_ticks-1.gz
Check lost_ticks in settimeofday to be more precise.
00_tsc-calibration-non-compile-time-1.gz
TSC calibration must be dynamic and not a compile time thing because gettimeofday is dynamic and it depends on the TSCs to be in sync. (me)
00_version
Change to the Makefile EXTRAVERSION.
10_bigmem-2.2.18pre2-18.bz2
BIGMEM production code (IA32 and alpha). This one fixes a memory leak during swap (almost unreproducibe, found it while reading the sources only) and some alpha-bigmem bug. It's now possible to generate core dumps larger than 2G (the previous limitation was due a few ext2 bugs and a binfmt_elf.c bug). I just includes the large_shm patch so it should be possible to allocate more than 2g in one shm segment. (me and Gerhard Wichert)
13_bigmem-rawio-2.2.17pre19aa1-3.gz
SCT's rawio bigmem capable. (Stephen C. Tweedie)
14_bigmem-rawio-lvm-2.2.15pre13aa1-6.gz
LVM compatible with the 2.3.x one. (Heinz Mauelshagen)
30_rtclight-2.2.15pre13aa1-1.gz
Needed by alpha SMP. (won't be needed anymore in 2.4.x)
40_lfs-2.2.18pre2aa2-15.bz2
LFS: >2G file support for 32bit archs. This also includes the new getdents64 and fcntl64 syscalls in sync with the 2.4.0-test8-pre5 user<-> kernel API. It also fixes some bug present in the previous 2.2.x LFS patches.
50_robust-initrd-2
Relocate the initrd image ASAP (btw, on alpha ASAP is sooner than on IA32), to avoid corrupting it while allocating the boot memory (like the mem_map_t array). This is good for robusteness and it's mainly interesting when bigmem is enabled.
60_VM-global-2.2.18pre2aa1-6.gz
Kills kpiod removing the OOM while paging out MAP_SHARED segments during a low memory condition. Fixes mmap00[123] on 8mbyte of RAM, and I bet I fixes as well Alan's DBMS OOM troubles. The way I could remove kpiod is to add a per-process fs_locks information and to check that information instead of deadlocking into the semaphore or into the superblock lock.
Fixes in a alternate (more efficient) way the GFP race that is currently handled by the free_before_allocate logic. The alternate way is to have a per-process freelist and to refill the global queue only once the tasks allocate something for itself.
Accounts protected buffers as unfreeable (fixes OOM with big ramdisk in memory).
Rewrite the sync_page_buffers logic. Now it's a per-bh bitflag information that tell us if a dirty buffer is hanging around for a too long time w/o becoming freeable (so we block on such buffer only once necessary and never when there's no high pressure). If there's no high pressure and __GFP_IO is set we try to write with WRITEA that doesn't block but that allows us to fill the I/O queue with the less interesting buffers in the VM. Never waits on the buffer to become unlocked (that's too aggressive and that was necessary only to workaround the pagingout of MAP_SHARED segments that wasn't able to do proper write throttling due the always async kpiod).
Rewrite the async buffer flushing wakeup points ala 2.4.x way.
Avoids swain/swapout while writing heavily to disk (no-swapout).
Adds lowlatency control checks in shrink_mmap and swap_out.
The same VM-global-patch backported to clean 2.2.18pre2 is here (the only difference is that it includes the above account-buffer-tries and it doesn't remove the suprious tq_disk run during readahead that isn't a VM related change):
ftp://ftp.us.kernel.org/pub/linux/kernel/people/andrea/patches/v2.2/2.2.18pre2/VM-global-2.2.18pre2-6.bz2
70_cond-sched-1
Implements conditional_schedule(). (Ingo Molnar)
71_copy-user-reschedule-2
Add low latency schedule entry points into the copy_user(), this fixes read(2)/write(2) hangs. (Originally implemented by Ingo Molnar in the lowlatency patch, but this patch is not equivalent, it's a very subset and it's different the reason I put the conditional schedule _after_ the copy is because we want the higher timeslice possible once we return to userspace to avoid live locks during swap) ---------------------------------------------------------------------------
The 2.2.18pre2aa1 patch is here:
ftp://ftp.us.kernel.org/pub/linux/kernel/people/andrea/kernels/v2.2/2.2.18pre2aa2.bz2
As usual all the separate patches are in:
ftp://ftp.us.kernel.org/pub/linux/kernel/people/andrea/kernels/v2.2/2.2.18pre2aa2/
and they can be applyed in `ls` order to generate the 2.2.18pre2aa2 kernel.
All the patches that I propose for inclusion into 2.2.18pre3 are here (the directory is called 2.2.18pre2 because the patches are against 2.2.18pre2).
ftp://ftp.us.kernel.org/pub/linux/kernel/people/andrea/proposed/v2.2/2.2.18pre2/
They are all orthogonal and they can be applied in any order regardless of each other.
Andrea
PS. If you'll run with more than 4giga of RAM on alpha, you'll need also to fix procps if you want to read the correct MM values with `free`, `top` and `vmstat`. procps takes all the memory information with byte granularity in `int` words (and not long words). I fixed procps this way:
ftp://ftp.suse.com/pub/people/andrea/procps/procps-2.0.2-bigmem-1.gz
PPS. 2.2.18pre2 doesn't included riskious stuff compared to 2.2.17, thus don't worry applying that pre-patch. In case you need to be based on 2.2.17 for whatever reason the 2.2.18pre2aa2.bz2 patch will just apply cleanly to 2.2.17 too.
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org Please read the FAQ at http://www.tux.org/lkml/
|  |