Hi Paul... Looks like it's me who learn from you :) But the positive side is, I am forced to re-read about inline assembly :D > finally translated into this: > > movl %1,%0 > > or > > movl %%fs:%1,%0 > > (that macro__percpu_seg() means "" on a non-SMP machine, or > "%%fs:" on a SMP machine.) > > Let's say I have a non-SMP machine, how dose that give me the > pointer? I googled my answer, found that, every CPU has a private > data zone, it contains the pointer to current task_struct. And I still > have a question: where is that segment (pointed by which segment > register)? And where in that segment containes the pointer we want? OK, you do better job than me on the macro expansion :) For your above question, since I don't have 2.6.22.x source code in hand, I can only offer a guess. I think the address is simply fetched from per_cpu_current_task variable. Kernel usually refer to DS when retrieving data. So it's DS:<offset>. And since normally data segment starts from 0 address, it's equal to access toward <address>. Does that helps you? NB: I see an important point here. Why "current" is no longer derived by substracting kernel's %esp? Is it faster now by simply referencing to (aligned?) in-RAM pointer? regards, Mulyadi -- To unsubscribe from this list: send an email with "unsubscribe kernelnewbies" to ecartis@xxxxxxxxxxxx Please read the FAQ at http://kernelnewbies.org/FAQ
