2.4.19pre2aa2

2.4.19pre2aa2

[Andrea Arcangeli]

URL:

	http://www.us.kernel.org/pub/linux/kernel/people/andrea/kernels/v2.4/2.4.19pre2aa2.gz
	http://www.us.kernel.org/pub/linux/kernel/people/andrea/kernels/v2.4/2.4.19pre2aa2/

Only in 2.4.19pre2aa2: 00_F_DUPFD-fcntl-rlimit-1

	Return -EINVAL if F_DUPFD and arg is negative or greater than or equal
	to {OPEN_MAX} as required by SuS and LSB. -EMFILE has to be returned
	only if we couldn't find an empty fd slot in the range, but the range
	was valid. From Thorsten Kukuk.

Only in 2.4.19pre2aa2: 00_amd-viper-7441-guessed-1

	Let amd74xx recognize the 7441 amd chipset, it works and I needed it
	mainly to set ->highmem = 1 and to skip the bounce buffers on my
	desktop.  (Tried also mode 5 and it failed, so I #undef __CAN_MODE_5
	back)

Only in 2.4.19pre2aa1: 00_block-highmem-all-18b-5.gz
Only in 2.4.19pre2aa2: 00_block-highmem-all-18b-6.gz

	Set hwif->highmem = 1 also with via and amd chipsets.

Only in 2.4.19pre2aa2: 00_cpu-affinity-rml-1

	Include it for now, it's handy for specialized usages, but please don't
	rely on this API, never include it in any software until it hits some
	mainline branch first. This should be considered as an unofficial
	debugging/benchmarking trick only for now. The fact it's an API by
	name it allows to include it but that's all.

	Patch from Robert Love.

Only in 2.4.19pre2aa2: 00_dnotify-fl_owner-1

	Let only the fl_owner to drop the dnotify (otherwise
	a modprobe triggered by konqueror would lead konqueror
	not to notice any longer the directory updates).

	Based on alternate patch from Waldo Bastian.

Only in 2.4.19pre2aa2: 00_fdatasync-FIFO-1

	writepage in FIFO order (more likely to be increasing physically
	consecutive on disk). fdatawait is fine to be left LIFO, so
	it's also less likely to reschedule in wait_on_page.

	Patch from Andrew Morton, keep up the good work! :)

Only in 2.4.19pre2aa1: 00_ia64-dma-1
Only in 2.4.19pre2aa2: 00_ia64-dma-2
Only in 2.4.19pre2aa1: 00_nfs-rdplus-2
Only in 2.4.19pre2aa2: 00_nfs-rdplus-3
Only in 2.4.19pre2aa1: 10_rawio-vary-io-3
Only in 2.4.19pre2aa2: 10_rawio-vary-io-4
Only in 2.4.19pre2aa1: 60_pagecache-atomic-3
Only in 2.4.19pre2aa2: 60_pagecache-atomic-4
Only in 2.4.19pre2aa1: 60_tux-config-stuff-1
Only in 2.4.19pre2aa2: 60_tux-config-stuff-2
Only in 2.4.19pre2aa1: 60_tux-dprintk-2
Only in 2.4.19pre2aa2: 60_tux-dprintk-3
Only in 2.4.19pre2aa1: 60_tux-exports-1
Only in 2.4.19pre2aa2: 60_tux-exports-2
Only in 2.4.19pre2aa1: 60_tux-syscall-2
Only in 2.4.19pre2aa2: 60_tux-syscall-3
Only in 2.4.19pre2aa1: 60_tux-sysctl-2
Only in 2.4.19pre2aa2: 60_tux-sysctl-3

	Rediffed so that it applies cleanly with -F1.

Only in 2.4.19pre2aa1: 00_lvm-1.0.2-2.gz

	Dropped (thanks to Christoph Hellwig for noticing the silly backout of
	1.0.3 :).

Only in 2.4.19pre2aa1: 00_max_bytes-1
Only in 2.4.19pre2aa2: 00_max_bytes-2

	Fixed an (harmless) merging error due "patch" being too permissive w/o
	-F1.  Thanks to Chip Salzenberg for noticing it. I defaulted to always
	use patch -F1 in the future and I recommend the same to everybody.

Only in 2.4.19pre2aa2: 00_mprotect_msync-ENOMEM-1

	According to SuS and LSB mprotect and msync have to return -ENOMEM
	and not -EFAULT when somebody passes unmapped regions as working
	areas. Patch from Thorsten Kukuk.

Only in 2.4.19pre2aa2: 00_rawio-forward-blkdev-ioctl-1

	Allow ioctl on the rawdevices, forward them to the bd_op->ioctl
	of the bind blockdevice.

	Patch from Andi Kleen.

Only in 2.4.19pre2aa1: 10_lvm-snapshot-check-1
Only in 2.4.19pre2aa2: 10_lvm-snapshot-check-2
Only in 2.4.19pre2aa1: 70_xfs-4.gz
Only in 2.4.19pre2aa2: 70_xfs-5.gz

	Rediffed.

Only in 2.4.19pre2aa1: 10_vm-29
Only in 2.4.19pre2aa2: 10_vm-30

	Fix deadlock. 2.4.19pre2aa1 could deadlock due a missing UnlockPage().
	All previous -aa kernels was not affected by such bug, only 2.4.19pre2aa1
	could deadlock. Made some other vm change and cleanup, including the
	drop of wait_on_page and the exclusive flag to the lock_page waitqueue.

Only in 2.4.19pre2aa2: 71_xfs-blksize-1

	Advertise the fs blkszie as PAGE_SIZE, so that the db libs doesn't run
	wild when on top of MD. I think on-disk format shouldn't depend on the
	underlying device device, they should depend on the fs only. The stripe
	size hint should be used only to fill two bios of the right size etc...
	but I'd prefer if they wouldn't affect the ondisk format of files. Even
	PAGE_SIZE doesn't look completly correct, but at least it reduces the
	problem to moving an harddisk from an alpha/ia64 to an x86 (x86-64
	doesn't change the page_size), which is a bit less common.

	Patch from Andi Kleen.

Andrea

Re: 2.4.19pre2aa2

[Alan Cox]

> Only in 2.4.19pre2aa2: 00_amd-viper-7441-guessed-1
> 
> 	Let amd74xx recognize the 7441 amd chipset, it works and I needed it
> 	mainly to set ->highmem = 1 and to skip the bounce buffers on my
> 	desktop.  (Tried also mode 5 and it failed, so I #undef __CAN_MODE_5
> 	back)

The correct AMD 7441 fixes are in the IDE patch and have been for a few
months. They were supplied by AMD and work a treat. I don't believe there is
any reason they require the new IDE infrastructure. They are howeve 32bit
still so the 64bit IDE will be nice

Re: 2.4.19pre2aa2

[Andrea Arcangeli]

On Mon, Mar 11, 2002 at 03:34:52PM +0000, Alan Cox wrote:
> > Only in 2.4.19pre2aa2: 00_amd-viper-7441-guessed-1
> > 
> > 	Let amd74xx recognize the 7441 amd chipset, it works and I needed it
> > 	mainly to set ->highmem = 1 and to skip the bounce buffers on my
> > 	desktop.  (Tried also mode 5 and it failed, so I #undef __CAN_MODE_5
> > 	back)
> 
> The correct AMD 7441 fixes are in the IDE patch and have been for a few
> months. They were supplied by AMD and work a treat. I don't believe there is
> any reason they require the new IDE infrastructure. They are howeve 32bit
> still so the 64bit IDE will be nice

thanks for the info. I will merge the IDE patch then (with the
additional modification to enable high-IO, that is why I looked into
it). btw, while making that change, I was also wondering that it would
be simpler to enable the highio in the common ide-dma part, rather than
in the chipsets tunings, the highio is completly unrelated to the fact
we compile amd7xxx or viaxxx into the kernel or not. but I didn't made
that change because the amd7xxx driver was working fine for me and also
because of possibly broken chipsets with the 31th bit of the bus address
disconnected, just to stay on the very safe side and not to trigger
hardware (not software) bugs.

Andrea

Re: 2.4.19pre2aa2

[Jens Axboe]

On Mon, Mar 11 2002, Andrea Arcangeli wrote:
> On Mon, Mar 11, 2002 at 03:34:52PM +0000, Alan Cox wrote:
> > > Only in 2.4.19pre2aa2: 00_amd-viper-7441-guessed-1
> > > 
> > > 	Let amd74xx recognize the 7441 amd chipset, it works and I needed it
> > > 	mainly to set ->highmem = 1 and to skip the bounce buffers on my
> > > 	desktop.  (Tried also mode 5 and it failed, so I #undef __CAN_MODE_5
> > > 	back)
> > 
> > The correct AMD 7441 fixes are in the IDE patch and have been for a few
> > months. They were supplied by AMD and work a treat. I don't believe there is
> > any reason they require the new IDE infrastructure. They are howeve 32bit
> > still so the 64bit IDE will be nice
> 
> thanks for the info. I will merge the IDE patch then (with the
> additional modification to enable high-IO, that is why I looked into
> it). btw, while making that change, I was also wondering that it would
> be simpler to enable the highio in the common ide-dma part, rather than
> in the chipsets tunings, the highio is completly unrelated to the fact
> we compile amd7xxx or viaxxx into the kernel or not. but I didn't made
> that change because the amd7xxx driver was working fine for me and also
> because of possibly broken chipsets with the 31th bit of the bus address
> disconnected, just to stay on the very safe side and not to trigger
> hardware (not software) bugs.

We can probably safely just drop the highio flag in the hwif now. I just
added it way back then as a safeguard, it might be a better idea to just
have potentially buggy chipsets set their dma mask appropriately and let
ide-dma enable the right bounce address (if any).

I'll update the block-highmem for 2.4.19-pre3 now that the IDE merge is
in.

-- 
Jens Axboe

Re: 2.4.19pre2aa2

[Andrea Arcangeli]

On Tue, Mar 12, 2002 at 09:11:14AM +0100, Jens Axboe wrote:
> On Mon, Mar 11 2002, Andrea Arcangeli wrote:
> > On Mon, Mar 11, 2002 at 03:34:52PM +0000, Alan Cox wrote:
> > > > Only in 2.4.19pre2aa2: 00_amd-viper-7441-guessed-1
> > > > 
> > > > 	Let amd74xx recognize the 7441 amd chipset, it works and I needed it
> > > > 	mainly to set ->highmem = 1 and to skip the bounce buffers on my
> > > > 	desktop.  (Tried also mode 5 and it failed, so I #undef __CAN_MODE_5
> > > > 	back)
> > > 
> > > The correct AMD 7441 fixes are in the IDE patch and have been for a few
> > > months. They were supplied by AMD and work a treat. I don't believe there is
> > > any reason they require the new IDE infrastructure. They are howeve 32bit
> > > still so the 64bit IDE will be nice
> > 
> > thanks for the info. I will merge the IDE patch then (with the
> > additional modification to enable high-IO, that is why I looked into
> > it). btw, while making that change, I was also wondering that it would
> > be simpler to enable the highio in the common ide-dma part, rather than
> > in the chipsets tunings, the highio is completly unrelated to the fact
> > we compile amd7xxx or viaxxx into the kernel or not. but I didn't made
> > that change because the amd7xxx driver was working fine for me and also
> > because of possibly broken chipsets with the 31th bit of the bus address
> > disconnected, just to stay on the very safe side and not to trigger
> > hardware (not software) bugs.
> 
> We can probably safely just drop the highio flag in the hwif now. I just
> added it way back then as a safeguard, it might be a better idea to just
> have potentially buggy chipsets set their dma mask appropriately and let
> ide-dma enable the right bounce address (if any).
> 
> I'll update the block-highmem for 2.4.19-pre3 now that the IDE merge is
> in.

I will send you my latest version, you probably want to hack on top of
it I think, I synched up the VM part and fixed some silly bug (like
pages_to_phys on alpha).

Andrea