Dear Srinivas... > If the threads share the same(single) PGD then AFAIK they must be > sharing the descriptor task_struct . No. each thread has its own unique task structure, but they can share certain resource. In this case, since we are talking about threads which are forked with CLONE_VM, they shares same same address space (mm_struct) >because the threads extract the > the PGD from the mm_struct in the descriptor. Correct. But remember, threads (which is forked with CLONE_VM) shares same mm_struct, so eventually they share same PGD. >Now how do we account > for the thread local storage ...if everything is shared ?? iam > confused about this . Shared here refers that each thread might walk through the same VMA, same page tables etc etc. However, with the help of segmentation (LDT/GDT), this sharing problem can be solved. TLS of each thread is defined as a segment in GDT/LDT entry and gs/fs register is used as segment selector which point to this segment (it is the work of gcc). The segment descriptor is renewed to reflect the current condition of TLS of a thread after context switch. Since gcc converts access toward the TLS as offset from certain based address , by combining with the "base" value taken from the segment descriptor pointed by gs or fs register and combine it with the "offset", you will access the TLS of the related thread Also, segment descriptor contains "segment limit", so there is no way a thread could access other thread TLS (assuming there is no bug on TLS creation which cause overlapping TLS area). If you force it access a memory address beyond the segment limit, you will get a fault. Hope it helps.... like always, CMIIW people regards Mulyadi -- Kernelnewbies: Help each other learn about the Linux kernel. Archive: http://mail.nl.linux.org/kernelnewbies/ FAQ: http://kernelnewbies.org/faq/
