Rajagopal Ananthanarayanan wrote:
> Heads up on what I'm about to change:
>
> 1. Properly initialize invalid bits when page is allocated.
> Note that SetAllInvalid is broken ... it sets all 32-bits
> to 1. A subsequent SetRangeValid will only clear, say 8 bits,
> corresponding to the 8 valid blocks in a page. Thus, a later
> check for page validity, such as PageBlockAllValid() which
> checks for ALL 32 bits to be zero, will not find it so. Chaos.
I'm fixing that rright now.
Basically inverting the bits has show some serious problems
with the way the bits are being used.
>
>
> 2. (1) will enable the fix to delalloc case
>
> 3. (2) will allow delwri to be removed.
>
> All changes to be verified with tests as Jim indicated.
>
> --
> --------------------------------------------------------------------------
> Rajagopal Ananthanarayanan ("ananth")
> Member Technical Staff, SGI.
> --------------------------------------------------------------------------
|