On Wed, 8 May 2024 20:10:15 GMT, Serguei Spitsyn <sspit...@openjdk.org> wrote:
>> I think you are suggesting that all monitors be created up front during
>> util_intialize(). The debug agent in the past has always created them lazily
>> (and sometimes some of them end up never being created because they are not
>> needed). I don't really see a big advantage in creating them all up front.
>>
>> Alex's suggestion was a very different one, and was just a suggestion to
>> initialize all the DebugRawMonitor ranks up front rather than when the
>> jMonitorID is created, and the reason for the suggestion was to avoid having
>> to grab the dbgRawMonitor when setting the rank, but I wasn't convinced that
>> it actually allowed you to not grab the dbgRawMonitor.
>
> I just wanted to say that Alex's suggestion is in a similar direction.
> I do not see a big advantage to save just on creation of a few monitors. It
> is not a scalability problem as their number is constant. Creation of the
> monitors at initialization time is a simplification as we do not care about
> synchronization at this phase. Also, it is easier to understand because there
> is no interaction with other threads.
The init code would have to know the name of each monitor since it is no longer
being passed in. Something like the following might work:
static DebugRawMonitor dbg_monitors[] = {
{ NULL, "JDWP VM_DEATH Lock", vmDeathLockForDebugLoop_Rank, NULL, 0 },
{ NULL, "JDWP Callback Block", callbackBlock_Rank, NULL, 0 },
{ NULL, "JDWP Callback Lock", callbackLock_Rank, NULL, 0 },
...
};
And then the init() function can iterate over the array and allocate the
monitor for each entry. Note that this array needs to be kept in sync with the
DebugRawMonitorRank enum (same entries and in the same order). I can assert
during the initialization that the rank field for each entry is equal to its
array index and that the array size is NUM_DEBUG_RAW_MONITORS.
-------------
PR Review Comment: https://git.openjdk.org/jdk/pull/19044#discussion_r1594777084
