Re: [PATCH 3/3] Allow removal of current Slab

["Durrant, Paul" <xadimgnik@xxxxxxxxx>]

On 07/06/2023 12:53, Owen Smith wrote:
> Then it looks like a minor reordering is needed. The check for 
> CacheMaskCount(Slab) == CacheMaskSize(Slab) should be made before the 
> attempt to call CacheGetObjectFromSlab(Slab), as if Slab is full, 
> CacheGetObjectFromSlab will fail and return NULL, triggering the ASSERT
> i.e.
> if (Cache->Cursor != &Cache->SlabList) {
>    PLIST_ENTRY ListEntry = Cache->Cursor;
>    PXENBUS_CACHE_SLAB Slab;
>    Slab = CONTAINING_RECORD(ListEntry, XENBUS_CACHE_SLAB, ListEntry);
>    if (CacheMaskCount(Slab->Allocated) == CacheMaskSize(Slab->Allocated)) {
>      // this slab is full, try next slab
>      Cache->Cursor = Slab->ListEntry.Flink;
>      goto again;
>    }
>    // there are free objects in this slab, this should always work
>    Object = CacheGetObjectFromSlab(Slab);
>    ASSERT(Object != NULL);
> }
[snip]

The question is why can we end up with Cache->Cursor != 
&Cache->SlabList, but yet the slab is full? This is not a valid state 
for the cache to be in.
  Paul