svn commit: r188350 - in head/sys: amd64/amd64 arm/arm dev/usb2/core i386/i386 ia64/ia64 sys

[Scott Long]

Marcel Moolenaar wrote:
> 
> On Feb 8, 2009, at 4:05 PM, M. Warner Losh wrote:
> 
>> In message: <9E40268E-A2E2-4CAA-AAFE-EB2491175CEE at mac.com>
>>            Marcel Moolenaar <xcllnt at mac.com> writes:
>> :
>> : On Feb 8, 2009, at 2:54 PM, Warner Losh wrote:
>> :
>> : > Author: imp
>> : > Date: Sun Feb  8 22:54:58 2009
>> : > New Revision: 188350
>> : > URL: http://svn.freebsd.org/changeset/base/188350
>> : >
>> : > Log:
>> : >  When bouncing pages, allow a new option to preserve the intra-page
>> : >  offset.  This is needed for the ehci hardware buffer rings that
>> : > assume
>> : >  this behavior.
>> :
>> : I thought we ended up with always doing that?
>> :
>> : Bounce buffers may or may not be used, so if the page offset
>> : must be zero for bounce buffers, they must be zero for the
>> : original DMA request. I see no value in re-aligning the DMA
>> : request to a page boundary when bounce buffers are used, but
>> : it is required in some cases to not realign. Hence, why not
>> : always (and unconditionally) preserve the alignment?
>>
>> I'm not entirely sure either.  However, at this stage of the game I'd
>> rather be a little conservative in what we change and do a more
>> careful and thorough analysis.
> 
> Fair enough.
> 

Busdma allows you to request bouncing for realignment.  Unconditionally 
preserving the page offset when bouncing defeats this.  Also, it has 
never preserved the offset.  I have no idea if this was intentional or
not.

Scott