Re: [Crash-utility] ps: improve performance

Hello Qiao, I am currently on vacation and will not be able to fully review this patch until I return in September. A patch that changes such fundamental workings of the task handling functionality is going to take a significant amount of review and testing. And a quick cursory review of the patch shows several issues that are problematic. For example, the change to task_to_pid() is just wrong. That function is exported in defs.h to any other function or extension module, so it makes no sense to call hq_entry_exists(), because the hash queue is only usable during the period of time that a single command runs and has opened the queue. So perhaps Pan does not understand that the hash queue facility can only be used on a per-command basis? Also, changing the arguments to exported functions such as get_task_mem_usage() should not be done because it would break any extension module that uses it. And related to that, the ugly changes to vm_area_dump() are just unnacceptable. Does he expect every user of get_task_mem_usage() to have to go through the same hoops instead of just making a single call? Another potential problem is the use of malloc() for anything that is not permanent. If the command is interrupted for whatever reason, the memory will be leaked. GETBUF()/FREEBUF() should be used for memory allocated for the requirements of a single command. Thanks, Dave