Vikram Rajguru
2018-09-28 18:08:49 UTC
Hi,
There is a MOESI protocol deadlock problem when dealing with a large number
of cores (larger than 32 cores), which has not been solved yet though it
has been posted years ago.
I found that the first L2 cache miss request cannot be correctly delivered
to its directory controller at the beginning of the simulation when the
core count exceeds 32. This results in a fact that L2 cache controller
keeps executing zz_recycleL1RequestQueue, and eventually it leads to the
system dump.
Can anyone give me a hint why the core number can potentially incur this
problem?
Thanks,
There is a MOESI protocol deadlock problem when dealing with a large number
of cores (larger than 32 cores), which has not been solved yet though it
has been posted years ago.
I found that the first L2 cache miss request cannot be correctly delivered
to its directory controller at the beginning of the simulation when the
core count exceeds 32. This results in a fact that L2 cache controller
keeps executing zz_recycleL1RequestQueue, and eventually it leads to the
system dump.
Can anyone give me a hint why the core number can potentially incur this
problem?
Thanks,