Google Groups no longer supports new Usenet posts or subscriptions. Historical content remains viewable.
Dismiss

Alpha and thread-local storage

1 view
Skip to first unread message

Havard Eidnes

unread,
Mar 28, 2011, 2:58:08 PM3/28/11
to
Hi,

Joerg has worked on TLS. He claims that the only thing missing
for alpha is that inside cpu_switchto, the process unique
register needs to be written.

Attached is my attempt at cargo-culting this code, but someone
who knows how PAL_wrunique should be called should take a closer
look, and preferably test the change.

Best regards,

- Håvard

diff

Jason Thorpe

unread,
Mar 28, 2011, 3:35:13 PM3/28/11
to

On Mar 28, 2011, at 11:58 AM, Havard Eidnes wrote:

> Hi,
>
> Joerg has worked on TLS. He claims that the only thing missing
> for alpha is that inside cpu_switchto, the process unique
> register needs to be written.

If it's in the hwpcb, doesn't the PAL call that does the context switch already save/restore it for you?

I have an Alpha ARM here in my office for nostalgia's sake, so I'll try to take a look while I gobble my lunch.

-- thorpej


--
Posted automagically by a mail2news gateway at muc.de e.V.
Please direct questions, flames, donations, etc. to news-...@muc.de

Jason Thorpe

unread,
Mar 28, 2011, 3:58:20 PM3/28/11
to

On Mar 28, 2011, at 11:58 AM, Havard Eidnes wrote:

Confirmed by reading Alpha ARM 2nd Edition, OSF/1 PALcode section 2.2.9 (Swap Process Context), if the process unique requester is implemented, swpctx does the unique value swap for you. rdunique and wrunique on systems that lack the internal register access it from the current HWPCB. So, there is no need to actually do this separately in the context switch routine.

>
> Best regards,
>
> - Håvard
> Index: alpha/genassym.cf
> ===================================================================
> RCS file: /cvsroot/src/sys/arch/alpha/alpha/genassym.cf,v
> retrieving revision 1.19
> diff -u -r1.19 genassym.cf
> --- alpha/genassym.cf 15 Dec 2010 01:33:37 -0000 1.19
> +++ alpha/genassym.cf 28 Mar 2011 09:10:05 -0000
> @@ -162,6 +162,7 @@
> define PCB_CONTEXT offsetof(struct pcb, pcb_context[0])
> define PCB_ONFAULT offsetof(struct pcb, pcb_onfault)
> define PCB_ACCESSADDR offsetof(struct pcb, pcb_accessaddr)
> +define PCB_APCB_UNIQUE offsetof(struct pcb, pcb_hw.apcb_unique)
>
> # Offsets into struct fpstate, for save, restore
> define FPREG_FPR_REGS offsetof(struct fpreg, fpr_regs[0])
> Index: alpha/locore.s
> ===================================================================
> RCS file: /cvsroot/src/sys/arch/alpha/alpha/locore.s,v
> retrieving revision 1.120
> diff -u -r1.120 locore.s
> --- alpha/locore.s 7 Jul 2010 01:17:49 -0000 1.120
> +++ alpha/locore.s 28 Mar 2011 09:10:05 -0000
> @@ -697,6 +697,12 @@
> ldq s0, L_PCB(s2)
>
> /*
> + * Save process unique register.
> + */
> + ldq a0, PCB_APCB_UNIQUE(s0)
> + call_pal PAL_wrunique
> +
> + /*
> * Check for restartable atomic sequences (RAS).
> */
> ldq a0, L_PROC(s2) /* first ras_lookup() arg */

Jason Thorpe

unread,
Mar 29, 2011, 12:13:37 AM3/29/11
to

On Mar 28, 2011, at 12:58 PM, Jason Thorpe wrote:

> process unique requester

Wow, can't even blame auto-complete for that one ...

"register"

0 new messages