As usual I was not clear enough
You are right, but:
1. if the L2 cache is configured as write-through what does it bring you to have FB cached? I consider that FB operations are like GPU's, you mostly do writes (and if you do reads then writes, you have to flush) so if the caches are WT there's no gain (unless having the accesses cached brings some other benefit such as better write merging, but I could not find that in C-A8 TRM)
2. if an area of memory where you mostly do writes is configured as cacheable and write allocate is enabled, you lose performance because the first write to each new cache line has to wait for line fill to complete (+ the cache flush once you're done)
3. you don't know how the L2 cache is configured, WT or WB.
Anyway that's a moot point, Squidge told the obvious thing: this only concerns the boot process, Linux will map the FB its own way