(Moving back to the gem5-dev list as this could be of broader interest.) Hi Andreas,
That's a blast from the past... it looks like there are two semi-separate uses of that flag, although the flag it self has the same semantics (indicating that the supplier has/had an exclusive copy, i.e., E/M state vs. O state): 1. The original application (http://repo.gem5.org/gem5/rev/000ab733f1eb) was to avoid unnecessary snoops in multilevel cache hierarchies. Say you have multiple L2s where each L2 has multiple L1s. If you request an exclusive block, and get supplied a block by a peer L1 attached to the same L2, then without the "supply exclusive" bit, you don't know whether or not you need to send out a "downward" invalidate to make sure the other L2s and their associated L1s get invalidated. The bit tells you that the supplying cache had the block in E or M state, so there's no need to broadcast an invalidation more widely. Obviously you can't rely on !sharedAsserted at this point since you haven't even snooped those other caches yet (and the whole point is to avoid doing so). 2. A few years later (http://repo.gem5.org/gem5/rev/d1ba390671ec) I reused this bit for writebacks to indicate that the original block was in M state and not O state. The commit message sums it up pretty well: If we write back an exclusive copy, we now mark it as such, so the cache receiving the writeback can mark its copy as exclusive. This avoids some unnecessary upgrade requests when a cache later tries to re-acquire exclusive access to the block. Hope that helps... Steve On Mon, Mar 9, 2015 at 7:33 AM, Andreas Hansson <andreas.hans...@arm.com> wrote: > Question for you concerning a change you made quite some years back: the > setSupplyExclusive/isSupplyExclusive, why were those added? For the most > part the cache seems to use “!sharedAsserted” to conclude that something is > exclusive, but for write backs you added the exclusive flags above in > addition to the shared flag. Do you remember what the logic was? > > Thanks, > > Andreas > > > _______________________________________________ gem5-dev mailing list gem5-dev@gem5.org http://m5sim.org/mailman/listinfo/gem5-dev
