Re: [patch] tls-2.5.30-A1

From: Luca Barbieri (
Date: Wed Aug 07 2002 - 18:21:48 EST

> your patch looks good to me - as long as we want to keep those 2 TLS
> entries and nothing more. (which i believe we want.) If even more TLS
> entries are to be made possible then a cleaner TLS enumeration interface
> has to be used like Christoph mentioned - although i dont think we really
> want that, 3 or more entries would be a stretch i think.
I think that 2 are enough.
Flat 32-bit programs set ds=es=ss=__USER_DS and cs=__USER_CS so they
only have fs and gs left.
16-bit programs and other odd ones can use the LDT support.

As for the interface I would suggest replacing the current one with a
single interface for LDT and GDT modifications that would provide the
following parameters:

unsigned table
- GDTAVAIL: GDT starting from first TLS
- GDTABS: GDT starting from 0
- AUTO: starts with the 2 TLS entries and proceeds with LDT

unsigned operation
- set: copy to kernel space (enlarge table if necessary). If root, don't
check validity for speed, otherwise check to ensure the user is not e.g.
putting call gates to CPL 0 code.
- set1: like set, but passes a single entry directly in the num and ptr
- get: copy from kernel space
- free: free memory and lower limits. If entry = 0 and num = ~0,
completely frees table.
- map: only for LDT and for root, allows to directly point to a user
memory range
- movekern: when support for per-task GDT is implemented, this would
allow to change the entries used for kernel entries. This would be
implemented with per-CPU IDTs and maybe dynamically generated code.
Useful for virtualization programs.

unsigned entry
- first entry affected. ~0 for first unused entry.

unsigned num
- number of entries affected

void* ptr
- pointer to read/write entries from

(table and operations may be merged)

Return value: first entry changed

e.g. libpthread would use table = AUTO, operation = set1, entry = ~0.

For the LDT things would be implemented as usual. For the GDT the
initial implementation would just modify TLS entries.
In future, support for dynamically allocated per-task GDTs could be

I would implement this by adding ops to sys_modify_ldt.

BTW, tls_desc1/tls_desc2 would IMHO be better as gdt_desc[2].

I don't plan to implement this myself.

To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to
More majordomo info at
Please read the FAQ at

This archive was generated by hypermail 2b29 : Wed Aug 07 2002 - 22:00:38 EST