RFR(S): JDK-8035952 Remove use of JVM_Open, JVM_Read and JVM_Close functions from serviceability code

[Martin Buchholz]

C infrastructure for OpenJDK has always been unsatisfactory.

JVM_Read was never documented, but did do EINTR system call restarting via
macros such as RESTARTABLE.  Now that you've stopped using JVM_Read,
doesn't mean that the need to restart system calls has gone away.  But
there's still no good place for shared openjdk C infrastructure.


On Mon, Mar 3, 2014 at 1:32 PM, Alan Bateman <Alan.Bateman at oracle.com>wrote:

> On 03/03/2014 10:30, Frederic Parain wrote:
>
>> :
>>
>> I'd prefer to keep open() and read() as they are,
>> for the sake of simplicity. In the unlikely case
>> where one of these calls gets interrupted, it would
>> simply generate a transient error in the value returned
>> by OperatingSystem.getCommittedVirtualMemorySize()
>> with a easily identifiable value of -1.
>>
> open is okay (as you say, open64 isn't interesting here). On EINTR then
> open can fail but it seems be just fifos so probably not a concern with
> /proc. However for read then I think we should handle it (I agree with
> Martin that it's generally a bug if we don't). I realize it's not a big
> issue but might be odd for
> getCommittedVirtualMemorySize be intermittently "not supported" .
>
> -Alan
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.openjdk.java.net/pipermail/serviceability-dev/attachments/20140303/652e126d/attachment.html