From: lsorense@csclub.uwaterloo.ca (Lennart Sorensen)
To: Rodolfo Giometti <giometti@enneenne.com>
Cc: Andrew Morton <akpm@linux-foundation.org>,
Alan Cox <alan@lxorguk.ukuu.org.uk>,
linux-kernel@vger.kernel.org
Subject: Re: LinuxPPS low-level IRQs timestamps & ldisc
Date: Sun, 1 Jun 2008 15:26:13 -0400 [thread overview]
Message-ID: <20080601192613.GC16164@csclub.uwaterloo.ca> (raw)
In-Reply-To: <20080601161510.GA26854@enneenne.com>
On Sun, Jun 01, 2008 at 06:15:10PM +0200, Rodolfo Giometti wrote:
> Hello,
>
> I write to both of you since my questions are related. :)
>
> My last modifications to LinuxPPS are focused on adding a PPS ldisc
> and a low-level IRQs timestamps recording.
>
> First I added a new dcd_change() ldisc method which can be used as
> follow:
>
> static void pps_tty_dcd_change(struct tty_struct *tty, unsigned int status)
> {
> int id = (int) tty->disc_data;
> struct timespec __ts;
> struct pps_ktime ts;
>
> /* First of all we get the time stamp... */
> getnstimeofday(&__ts);
>
> /* ... and translate it to PPS time data struct */
> ts.sec = __ts.tv_sec;
> ts.nsec = __ts.tv_nsec;
>
> /* Now do the PPS event report */
> pps_event(id, &ts,
> status ? PPS_CAPTUREASSERT : PPS_CAPTURECLEAR, tty);
>
> pr_debug("[STDev] PPS %s at %lu on source #%d\n",
> status ? "assert" : "clear", jiffies, id);
> }
>
> However this solution gives very low precision timestamps so that's
> why I propose to add the following code to register IRQ timestamps as
> soon as possible (here the code for x86_32 architecture):
>
> diff --git a/arch/x86/kernel/irq_32.c b/arch/x86/kernel/irq_32.c
> index d3fde94..a3b8457 100644
> --- a/arch/x86/kernel/irq_32.c
> +++ b/arch/x86/kernel/irq_32.c
> @@ -15,6 +15,7 @@
> #include <linux/notifier.h>
> #include <linux/cpu.h>
> #include <linux/delay.h>
> +#include <linux/pps.h>
>
> #include <asm/apic.h>
> #include <asm/uaccess.h>
> @@ -25,6 +26,11 @@ EXPORT_PER_CPU_SYMBOL(irq_stat);
> DEFINE_PER_CPU(struct pt_regs *, irq_regs);
> EXPORT_PER_CPU_SYMBOL(irq_regs);
>
> +#ifdef CONFIG_PPS_IRQ_EVENTS
> +struct pps_ktime pps_irq_ts[NR_IRQS];
> +EXPORT_SYMBOL(pps_irq_ts);
> +#endif
> +
> /*
> * 'what should we do if we get a hw irq event on an illegal vector'.
> * each architecture has to answer this themselves.
> @@ -76,6 +82,12 @@ fastcall unsigned int do_IRQ(struct pt_regs *regs)
> union irq_ctx *curctx, *irqctx;
> u32 *isp;
> #endif
> +#ifdef CONFIG_PPS_IRQ_EVENTS
> + struct timespec __ts;
> +
> + /* Get IRQ timestamps as soon as possible for the PPS layer */
> + getnstimeofday(&__ts);
> +#endif
How expensive is this call? Would it be cheaper to check whether this
IRQ has anyone requesting the time info before doing it? I suspect most
systems will have at most one IRQ that actually needs this done, so does
doing it for all IRQs make sense?
> if (unlikely((unsigned)irq >= NR_IRQS)) {
> printk(KERN_EMERG "%s: cannot handle IRQ %d\n",
> @@ -83,6 +95,12 @@ fastcall unsigned int do_IRQ(struct pt_regs *regs)
> BUG();
> }
>
> +#ifdef CONFIG_PPS_IRQ_EVENTS
> + /* Then, after sanity check, store the IRQ timestamp */
> + pps_irq_ts[irq].sec = __ts.tv_sec;
> + pps_irq_ts[irq].nsec = __ts.tv_nsec;
> +#endif
> +
> old_regs = set_irq_regs(regs);
> irq_enter();
> #ifdef CONFIG_DEBUG_STACKOVERFLOW
>
> This allow the following modifications to the dcd_change() method:
>
> static void pps_tty_dcd_change(struct tty_struct *tty, unsigned int irq,
> unsigned int status)
> {
> int id = (int) tty->disc_data;
>
> pps_event(id, &pps_irq_ts[irq],
> status ? PPS_CAPTUREASSERT : PPS_CAPTURECLEAR, tty);
>
> pr_debug("[IRQev] PPS %s at %lu on source #%d\n",
> status ? "assert" : "clear", jiffies, id);
> }
>
> Note that in this case I need also the "irq" parameter.
>
> This solution gives very high timestamps precision! Better that
> before! :D
>
> What do you think about this solution? More precisely:
>
> 1) The low-level IRQ patch can be acceptable?
>
> 2) The new dcd_change() method is well implemented? Can I add the
> "irq" parameter or I can find it somewhere?
>
> Thanks a lot for your suggestions,
I certainly like high precision time stamps. Very useful for ntp.
--
Len Sorensen
next prev parent reply other threads:[~2008-06-01 19:26 UTC|newest]
Thread overview: 13+ messages / expand[flat|nested] mbox.gz Atom feed top
2008-06-01 16:15 Rodolfo Giometti
2008-06-01 19:26 ` Lennart Sorensen [this message]
2008-06-02 6:51 ` Alan Cox
2008-06-02 9:25 ` Rodolfo Giometti
2008-06-02 14:02 ` Alan Cox
2008-06-02 14:56 ` Rodolfo Giometti
2008-06-02 15:48 ` Alan Cox
2008-06-02 17:22 ` Rodolfo Giometti
2008-06-02 19:35 ` Alan Cox
2008-06-02 20:21 ` Matti Aarnio
2008-06-03 13:40 ` Alan Cox
2008-06-02 17:09 ` Lennart Sorensen
2008-06-02 16:59 ` Alan Cox
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=20080601192613.GC16164@csclub.uwaterloo.ca \
--to=lsorense@csclub.uwaterloo.ca \
--cc=akpm@linux-foundation.org \
--cc=alan@lxorguk.ukuu.org.uk \
--cc=giometti@enneenne.com \
--cc=linux-kernel@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
all inboxes | Powered by JetHome®