On 08/29/2014 07:52 AM, Ard Biesheuvel wrote:
> On 29 August 2014 13:02, Edward Nevill wrote:
>> We have a need on OpenJDK to flush the entire icache. Basically this
>> need occurs when we do a garbage collection which may include garbage
>> collection and patching of the dynamically generated cod
Yes... but the JIT heap is 128Mb, and the way it flushes the cache is to
do an IC IVAU on every virtual cache line address from start .. end.
That could be up to 2E6 cache flush operations.
We had hoped that IC IALLUIS would be faster.
Maybe we end up doing this. It seems absolutely daft to me I
On 29 August 2014 13:02, Edward Nevill wrote:
> We have a need on OpenJDK to flush the entire icache. Basically this
> need occurs when we do a garbage collection which may include garbage
> collection and patching of the dynamically generated code.
>
Surely, you could restrict the icache flush t
