Ivan, I agree, esp. for the JITted code buffer if the architecture has no distinction for code and data TLB supports. So probably this large page support should be designed as a lower layer in memory management. A longer vision can be that, the lower layer MM (under GC and other memory functionalities of JVM) is built underneath all JVM components serving for system level MM primitives. This is not very difficult, I guess.
And I think the codes for large page are learned from GCv4.1. :-) You should have been credited in Yu-Nan's patch description. Thank you, Ivan! Thanks, xiaofeng On 1/25/07, Ivan Volosyuk <[EMAIL PROTECTED]> wrote:
Is the VM arguments for large TLB pages support are compatible with the same of GCv4.1 or different? Just interesting. One more observation, it makes sense to use large TLB pages for all VM including classloader pools and other code not just for GC. -- Ivan On 1/24/07, Xiao-Feng Li <[EMAIL PROTECTED]> wrote: > On 1/24/07, Geir Magnusson Jr. <[EMAIL PROTECTED]> wrote: > > > > On Jan 23, 2007, at 1:15 PM, yunan He wrote: > > > > > Hi, I submitted a patch in JIRA-3042 ( > > > http://issues.apache.org/jira/browse/HARMONY-3042) . > > > > > > This patch is to enable GCv5 to run with 32bit Linux. It has also > > > large page > > > support for performance improvement. It works with MSVC/ICC in > > > Windows and > > > GCC in Linux. > > > > Very cool! Didn't GCv5 run on 32-bit linux before? > > Actually it did, but with only default configuration. This patch > enables it with multiple different configurations in Linux ( such as, > generational or non-generational, static or dynamic mem mapping, small > page or large page, etc.). > > Thanks, > xiaofeng > > > geir > > > > > > > > Dear committer, please help to apply this patch at gc_gen > > > directory. Thanks.
