2.4.10-pre3

2.4.10-pre3

[Linus Torvalds]

Ok, there's been a number of small merges, and VM cleanups etc. This
should avoid kswapd spending any more CPU time than before, and fixes a
number of annoying bugs. ChangeLog appended,

		Linus

----

pre3:
 - Johannes Erdfelt, Oliver Neukum: USB printer driver race fix
 - John Byrne: fix stupid i386-SMP irq stack layout bug
 - Andreas Bombe, me: yenta IO window fix
 - Neil Brown: raid1 buffer state fix
 - David Miller, Paul Mackerras: fix up sparc and ppc respectively for kmap/kbd_rate
 - Matija Nalis: umsdos fixes, and make it possible to boot up with umsdos
 - Francois Romieu: fix bugs in dscc4 driver
 - Andy Grover: new PCI config space access functions (eventually for ACPI)
 - Albert Cranford: fix incorrect e2fsprog data from ver_linux script
 - Dave Jones: re-sync x86 setup code, fix macsonic kmalloc use
 - Johannes Erdfelt: remove obsolete plusb USB driver
 - Andries Brouwer: fix USB compact flash version info, add blksize ioctls

pre2:
 - Al Viro: block device cleanups
 - Marcelo Tosatti: make bounce buffer allocations more robust (it's ok
   for them to do IO, just not cause recursive bounce IO. So allow them)
 - Anton Altaparmakov: NTFS update (1.1.17)
 - Paul Mackerras: PPC update (big re-org)
 - Petko Manolov: USB pegasus driver fixes
 - David Miller: networking and sparc updates
 - Trond Myklebust: Export atomic_dec_and_lock
 - OGAWA Hirofumi: find and fix umsdos "filldir" users that were broken
   by the 64-bit-cleanups. Fix msdos warnings.
 - Al Viro: superblock handling cleanups and race fixes
 - Johannes Erdfelt++: USB updates

pre1:
 - Jeff Hartmann: DRM AGP/alpha cleanups
 - Ben LaHaise: highmem user pagecopy/clear optimization
 - Vojtech Pavlik: VIA IDE driver update
 - Herbert Xu: make cramfs work with HIGHMEM pages
 - David Fennell: awe32 ram size detection improvement
 - Istvan Varadi: umsdos EMD filename bug fix
 - Keith Owens: make min/max work for pointers too
 - Jan Kara: quota initialization fix
 - Brad Hards: Kaweth USB driver update (enable, and fix endianness)
 - Ralf Baechle: MIPS updates
 - David Gibson: airport driver update
 - Rogier Wolff: firestream ATM driver multi-phy support
 - Daniel Phillips: swap read page referenced set - avoid swap thrashing

Re: 2.4.10-pre3 - bug report

[Andrey Nekrasov]

Hello Linus Torvalds,

> pre3:
>  - Johannes Erdfelt, Oliver Neukum: USB printer driver race fix
>  - John Byrne: fix stupid i386-SMP irq stack layout bug
>  - Andreas Bombe, me: yenta IO window fix
>  - Neil Brown: raid1 buffer state fix
>  - David Miller, Paul Mackerras: fix up sparc and ppc respectively for kmap/kbd_rate
>  - Matija Nalis: umsdos fixes, and make it possible to boot up with umsdos
>  - Francois Romieu: fix bugs in dscc4 driver
>  - Andy Grover: new PCI config space access functions (eventually for ACPI)
>  - Albert Cranford: fix incorrect e2fsprog data from ver_linux script
>  - Dave Jones: re-sync x86 setup code, fix macsonic kmalloc use
>  - Johannes Erdfelt: remove obsolete plusb USB driver
>  - Andries Brouwer: fix USB compact flash version info, add blksize ioctls

1.  hardware configuration:
		 - Intel Server ISP1100 + Promise Ultra100 controller
		 - 1Gb RAM

2.  ...
		CONFIG_HIGHMEM4G=y
    CONFIG_HIGHMEM=y
		...
		# CONFIG_SMP is not set
		# CONFIG_X86_UP_IOAPIC is not set
	  ...

3.  Test:

		run
		test:/spylog/test/tiobench-0.3.1 # ./tiobench.pl 
		No size specified, using 1792 MB
	
    syslog level "kern.*":
	
ep  2 05:20:18 test kernel: Inspecting /boot/System.map
Sep  2 05:20:18 test kernel: Loaded 12509 symbols from /boot/System.map.
Sep  2 05:20:18 test kernel: Symbols match kernel version 2.4.10.
Sep  2 05:20:18 test kernel: No module symbols loaded - kernel modules not enabled. 
Sep  2 05:24:52 test kernel: __alloc_pages: 0-order allocation failed (gfp=0x70/1).
Sep  2 05:24:54 test last message repeated 77 times
Sep  2 05:24:54 test kernel: __alloc_pages: 0-order allocation failed (gfp=0x30/1).
Sep  2 05:24:54 test kernel: __alloc_pages: 0-order allocation failed (gfp=0x70/1).
Sep  2 05:24:59 test last message repeated 133 times
Sep  2 05:24:59 test kernel: __alloc_pages: 0-order allocation failed (gfp=0x20/0).
Sep  2 05:24:59 test kernel: __alloc_pages: 0-order allocation failed (gfp=0x70/1).
Sep  2 05:25:15 test last message repeated 505 times
Sep  2 05:25:15 test kernel: __alloc_pages: 0-order allocation failed (gfp=0x30/1).
Sep  2 05:25:15 test kernel: __alloc_pages: 0-order allocation failed (gfp=0x70/1).
Sep  2 05:25:31 test last message repeated 482 times
Sep  2 05:28:00 test kernel: __alloc_pages: 0-order allocation failed (gfp=0x70/1).
Sep  2 05:28:04 test last message repeated 140 times
Sep  2 05:28:04 test kernel: __alloc_pages: 0-order allocation failed (gfp=0x20/0).
Sep  2 05:28:04 test kernel: __alloc_pages: 0-order allocation failed (gfp=0x70/1).
Sep  2 05:28:07 test last message repeated 97 times
Sep  2 05:28:07 test kernel: __alloc_pages: 0-order allocation failed (gfp=0x20/0).
Sep  2 05:28:07 test kernel: __alloc_pages: 0-order allocation failed (gfp=0x70/1).

Need other test, that search this bug?
(MySQL server version: 3.23.40 also work wrong)

-- 
bye.
Andrey Nekrasov, SpyLOG.

Re: 2.4.10-pre3 - bug report

[Linus Torvalds]

In article <20010902053338.A26119@spylog.ru> you write:
>
>		run
>		test:/spylog/test/tiobench-0.3.1 # ./tiobench.pl 
>		No size specified, using 1792 MB
>	
>    syslog level "kern.*":
>	
>ep  2 05:20:18 test kernel: Inspecting /boot/System.map
>Sep  2 05:20:18 test kernel: Loaded 12509 symbols from /boot/System.map.
>Sep  2 05:20:18 test kernel: Symbols match kernel version 2.4.10.
>Sep  2 05:20:18 test kernel: No module symbols loaded - kernel modules not enabled. 
>Sep  2 05:24:52 test kernel: __alloc_pages: 0-order allocation failed (gfp=0x70/1).
>Sep  2 05:24:54 test last message repeated 77 times
>Sep  2 05:24:54 test kernel: __alloc_pages: 0-order allocation failed (gfp=0x30/1).

These are bounce buffer allocations - they do fail, but the failures
should be temporary and the machine should make progress. 

There are some nasty issues with HIGHMEM that will be seriously improved
during 2.5.x when we start doing IO directly from highmem for
controllers that can handle it, but that 2.4.x is not likely to really
fix.  So you should expect to see messages like the above that are about
"we couldn't allocate memory for bounce buffers", and they _will_ imply
that performance isn't going to be as good as it possibly should be, but
at the same time it shouldn't be a real problem either.

Does the machine stay up and perform reasonably well?

(Where "reasonable" doesn't necessarily mean "really well", but means
"it's not absolutely horrible").

		Linus

Re: 2.4.10-pre3 - bug report

[arjan]

In article <200109020348.f823mCs01183@penguin.transmeta.com> you wrote:
> These are bounce buffer allocations - they do fail, but the failures
> should be temporary and the machine should make progress. 

The patch I sent to linux-mm a week ago fixes this for "normal" loads.
It obviously doesn't fix the worst case scenario, but in the testing in the
Red Hat lab we've seen it 1) improve performance and 2) get rid of VM
problems caused by highmem; under normal load. Under the insane load we also
put our kernels on, it doesn't make much difference, the obvious deadlock is
still there.

Greetings,
    Arjan van de Ven