From: Russ Anderson <rja@sgi.com>
To: linux-ia64@vger.kernel.org
Subject: Re: [patch] per cpu MCA/INIT save areas (take 2)
Date: Wed, 17 Nov 2004 17:47:51 +0000 [thread overview]
Message-ID: <200411171747.iAHHlpS3110348@ben.americas.sgi.com> (raw)
In-Reply-To: <200411122327.iACNRR5h131335@ben.americas.sgi.com>
Tony Luck wrote:
>
> >Can you remind me why k3? Are you worried about the TR mapping the
> >per-CPU region getting corrupted?
>
> This isn't for correctness, or even efficiency (since this is a
> very low frequency execution path). It is just to make the code
> easier to read and maintain. The current code in the MCA path that
> loops through the array of tlb info structures looking for the
> one that matches the current cpus cr.lid is quite ugly. Replicating
> it in the INIT code to find the per-cpu save+stack area will add
> to the ugliness. Having the phys address of the per-cpu area in
> ar.k3 would clean things up.
>
> For the INIT case it would be safe to just "tpa" a per-cpu address,
> but in the MCA case we shouldn't trust the TR mapping.
So one way to hide the uglyness is to put it in a macro, such
as in the example at the bottom.
That cleans up the current place (in mca_asm.S) where it is called to be:
// The following code purges TC and TR entries. Then reload all TC entries.
// Purge percpu data TC entries.
begin_tlb_purge_and_reload:
MCA_SAVE_AREA_PTR(r2, lbl1, err);;
addl r2=IA64_MCA_TLB_INFO,r2;;
mov r17=r2
;;
One technical concern is that the routine uses r16, r17, r18, r19, r20.
Optimizing is may squeeze out a couple but the bigger question is what
scratch registers are available? Specificly, ia64_os_mca_proc_state_dump
saves off r16+ registers, so it would be good not to trash them before they
are saved.
Other than that issue, it looks like a macro could be used instead of k3.
----------------------------------------------------------------------
/*
* MCA_SAVE_AREA_PTR() gets the physical address pointer to this cpus
* MCA/INIT save area.
* paddr (register) ends up holding the physical address.
*/
#define MCA_SAVE_AREA_PTR(paddr, tlbl, tmperr) \
mov r16=cr.lid;; \
LOAD_PHYSICAL(p0,r17,ia64_mca_tlb_list);; \
mov r19=0;; \
mov r20=NR_CPUS;; \
tlbl: cmp.eq p6,p7=r19,r20;; \
(p6) br.spnt.few tmperr;; \
ld8 r18=[r17],IA64_MCA_TLB_INFO_SIZE;; \
add r19=1,r19;; \
cmp.eq p6,p7=r18,r16;; \
(p7) br.sptk.few 1b;; \
adds r17=-IA64_MCA_TLB_INFO_SIZE,r17;; \
mov paddr = r17;;
----------------------------------------------------------------------
Thanks,
--
Russ Anderson, OS RAS/Partitioning Project Lead
SGI - Silicon Graphics Inc rja@sgi.com
next prev parent reply other threads:[~2004-11-17 17:47 UTC|newest]
Thread overview: 17+ messages / expand[flat|nested] mbox.gz Atom feed top
2004-11-12 23:27 [patch] per cpu MCA/INIT save areas (take 2) Russ Anderson
2004-11-13 7:05 ` David Mosberger
2004-11-13 23:24 ` Russ Anderson
2004-11-14 0:07 ` Jack Steiner
2004-11-14 20:07 ` Luck, Tony
2004-11-15 19:13 ` David Mosberger
2004-11-17 17:47 ` Russ Anderson [this message]
2004-11-17 18:59 ` Luck, Tony
2004-11-18 21:36 ` Russ Anderson
2004-11-18 22:18 ` Luck, Tony
2004-11-18 22:56 ` Russ Anderson
2004-11-23 23:36 ` Russ Anderson
2004-11-24 2:41 ` Keith Owens
2004-11-24 23:32 ` Luck, Tony
2004-12-03 21:33 ` Russ Anderson
2004-12-09 4:15 ` Russ Anderson
2004-12-09 4:35 ` Keith Owens
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=200411171747.iAHHlpS3110348@ben.americas.sgi.com \
--to=rja@sgi.com \
--cc=linux-ia64@vger.kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox