Regarding Jens' Zero-Bounce Highmem I/O Patch

Regarding Jens' Zero-Bounce Highmem I/O Patch

[Peter Wong]

In order to use Jens' zero-bounce highmem I/O patch against 2.4.6,
a small modification for the patch is needed. Simply replace
GFP_BUFFER by GFP_NOIO in block-highmem-all-5.gz, which can be obtained
at http://www.kernel.org/pub/linux/kernel/people/axboe/patches/2.4.6-pre1/.

However, there is another problem. For both 2.4.5 and 2.4.6 with
Jens' patches, if the kernels are built with 4GB highmem, they
boot without problems. But if the kernels are built with 64GB
highmem, the kernels hang right after uncompressing Linux. Has
anyone seen this problem?

Thanks,
Peter

Wai Yee Peter Wong
IBM Linux Technology Center, Performance Analysis
email: wpeter@us.ibm.com
Office: (512) 838-9272, T/L 678-9272; Fax: (512) 838-4663

Re: [Lse-tech] Regarding Jens' Zero-Bounce Highmem I/O Patch

[Christoph Hellwig]

On Wed, Sep 19, 2001 at 05:04:59PM -0500, Peter Wong wrote:
> In order to use Jens' zero-bounce highmem I/O patch against 2.4.6,
> a small modification for the patch is needed. Simply replace
> GFP_BUFFER by GFP_NOIO in block-highmem-all-5.gz, which can be obtained
> at http://www.kernel.org/pub/linux/kernel/people/axboe/patches/2.4.6-pre1/.
> 
> However, there is another problem. For both 2.4.5 and 2.4.6 with
> Jens' patches, if the kernels are built with 4GB highmem, they
> boot without problems. But if the kernels are built with 64GB
> highmem, the kernels hang right after uncompressing Linux. Has
> anyone seen this problem?

block-highmem-all-5 is _very_ _very_ old.
Please upgrade to a recent kernel and patch and report again.

I had a system with PAE enabled and 2.4.9ac + some hacked block-highmem
patch running very well here.  (Probably it still runns :))

	Christoph

-- 
Of course it doesn't work. We've performed a software upgrade.

Re: Regarding Jens' Zero-Bounce Highmem I/O Patch

[Jens Axboe]

On Wed, Sep 19 2001, Peter Wong wrote:
> In order to use Jens' zero-bounce highmem I/O patch against 2.4.6,
> a small modification for the patch is needed. Simply replace
> GFP_BUFFER by GFP_NOIO in block-highmem-all-5.gz, which can be obtained
> at http://www.kernel.org/pub/linux/kernel/people/axboe/patches/2.4.6-pre1/.
> 
> However, there is another problem. For both 2.4.5 and 2.4.6 with
> Jens' patches, if the kernels are built with 4GB highmem, they
> boot without problems. But if the kernels are built with 64GB
> highmem, the kernels hang right after uncompressing Linux. Has
> anyone seen this problem?

It's a "solved" problem (well sort of, I decided against including the <
4GB memory zone which was the cause of this problem) in this old
version. Use a more recent one and you'll be fine.

Or, use the old one and backout the zone changes.

-- 
Jens Axboe