Re: [ECOS] [Fwd: Re: [ECOS] sntp problems]

[Gary Thomas <gary at mlbassoc dot com>]

Gary Thomas wrote:
> Gary Thomas wrote:
> > Laurie Gellatly wrote:
> > > Couldn't find anyting about the timezone offset.
> > > Looks like it needs to be added in when the time is adjusted back to 
> > > 1900.
> > >
> > > My other problem with the locking up turns out to be the new 
> > > impolementation
> > > of the I2C clock chip
> > > I'm using and the set_current_time function that calls cyg_drv_dsr_lock.
> > > Trouble is that the I2C clock is interrupt driven and the 
> > > cyg_drv_dsr_lock
> > > prevents the DSR from running.
> > > I'm using a dallas 1307 variant but I don't see anything in that code 
> > > that
> > > tries to address this.
> > > Does anyone know why cyg_drv_dsr_lock is used in set_current_time?
> > > Anyone using an interrupt based I2C clock that might have addressed 
> > > this?
> >
> > Any piece of code that calls cyg_drv_dsr_lock() and does not provide a
> > way to call cyg_drv_dsr_unlock() is broken.  If this is the case, it's no
> > wonder that your system comes to a halt when a NTP update comes in.
> >
> > The 'set_current_time()' function is not part of eCos, so it's difficult
> > to analyze this much further.
>
> Sorry, I missed this in .../io/wallclock/current/src/wallclock.cxx
>
> Nonetheless, that's where I'd start looking at the problem.  As mentioned
> in the note you quoted, this function should be using a mutex for 
> protection,
> not DSR locking.

Please try the attached patch (untested) and let me know if it helps.

> > > Thanks            ...Laurie:{)
> > >
> > > -----Original Message-----
> > > From: ecos-discuss-owner@ecos.sourceware.org
> > > [mailto:ecos-discuss-owner@ecos.sourceware.org]On Behalf Of Laurie
> > > Gellatly
> > > Sent: Friday, 12 January 2007 9:47 PM
> > > To: eCos discussion
> > > Subject: RE: [ECOS] [Fwd: Re: [ECOS] sntp problems]
> > >
> > >
> > > Thanks Jonathan,
> > > That looks promising. I'll check it out.
> > >
> > >             ...Laurie:{)
> > >
> > > -----Original Message-----
> > > From: ecos-discuss-owner@ecos.sourceware.org
> > > [mailto:ecos-discuss-owner@ecos.sourceware.org]On Behalf Of Jonathan
> > > Larmour
> > > Sent: Friday, 12 January 2007 1:44 AM
> > > To: eCos discussion
> > > Subject: [ECOS] [Fwd: Re: [ECOS] sntp problems]
> > >
> > >
> > > The attached post was mis-directed. - Jifl
> > >
> > >
> > > --
> > > Before posting, please read the FAQ: http://ecos.sourceware.org/fom/ecos
> > > and search the list archive: http://ecos.sourceware.org/ml/ecos-discuss


--
------------------------------------------------------------
Gary Thomas                 |  Consulting for the
MLB Associates              |    Embedded world
------------------------------------------------------------

Index: io/wallclock/current/src/wallclock.cxx
===================================================================
RCS file: /misc/cvsfiles/ecos/packages/io/wallclock/current/src/wallclock.cxx,v
retrieving revision 1.6
diff -u -5 -p -r1.6 wallclock.cxx
--- io/wallclock/current/src/wallclock.cxx	23 May 2002 23:06:38 -0000	1.6
+++ io/wallclock/current/src/wallclock.cxx	12 Jan 2007 12:46:41 -0000
@@ -85,20 +85,25 @@ static Cyg_WallClock wallclock_instance 
 #ifndef CYGSEM_WALLCLOCK_SET_GET_MODE
 static cyg_uint32 epoch_ticks;
 static cyg_uint32 epoch_time_stamp;
 #endif
 
+static cyg_drv_mutex_t wallclock_lock;
+
 Cyg_WallClock *Cyg_WallClock::wallclock;
 
 //-----------------------------------------------------------------------------
 // Constructor
 
 Cyg_WallClock::Cyg_WallClock()
 {
     // install instance pointer
     wallclock = &wallclock_instance;
 
+    // Initialize lock used for mutually exclusive access to hardware
+    cyg_drv_mutex_init(&wallclock_lock);
+
     // Always allow low-level driver to initialize clock, even though it
     // may not be necessary for set-get mode.
     init_hw_seconds();
 }
 
@@ -109,38 +114,38 @@ Cyg_WallClock::Cyg_WallClock()
 
 cyg_uint32 Cyg_WallClock::get_current_time()
 {
     cyg_uint32 res;
 
-    cyg_drv_dsr_lock();
+    while (!cyg_drv_mutex_lock(&wallclock_lock));
 
 #ifdef CYGSEM_WALLCLOCK_SET_GET_MODE
     res = get_hw_seconds();
 #else
     res = epoch_time_stamp + get_hw_seconds() - epoch_ticks;
 #endif
 
-    cyg_drv_dsr_unlock();
+    cyg_drv_mutex_unlock(&wallclock_lock);
 
     return res;
 }
 
 //-----------------------------------------------------------------------------
 // Sets the clock. Argument is seconds elapsed since 1970-01-01 00:00:00.
 // This may involve reading or writing to the hardware, so it may take
 // anything up to a second to complete.
 void Cyg_WallClock::set_current_time( cyg_uint32 time_stamp )
 {
-    cyg_drv_dsr_lock();
+    while (!cyg_drv_mutex_lock(&wallclock_lock));
 
 #ifdef CYGSEM_WALLCLOCK_SET_GET_MODE
     set_hw_seconds(time_stamp);
 #else
     epoch_time_stamp    = time_stamp;
     epoch_ticks         = get_hw_seconds();
 #endif
 
-    cyg_drv_dsr_unlock();
+    cyg_drv_mutex_unlock(&wallclock_lock);
 }
 
 //-----------------------------------------------------------------------------
 // End of devs/wallclock/wallclock.cxx

-- 
Before posting, please read the FAQ: http://ecos.sourceware.org/fom/ecos
and search the list archive: http://ecos.sourceware.org/ml/ecos-discuss