Received: from mailserv2.iuinc.com (qmailr@mailserv2.iuinc.com [206.245.164.55]) by puffin.external.hp.com (8.8.7/8.8.7) with SMTP id PAA17454 for ; Wed, 24 Nov 1999 15:45:30 -0700 From: Stan Sieler Message-Id: <199911242246.OAA31733@opus.allegro.com> Subject: Re: itmr rollover in delay loop To: frowand@cup.hp.com (Frank Rowand) Date: Wed, 24 Nov 1999 14:46:17 -0800 (PST) Cc: parisc-linux@thepuffingroup.com In-Reply-To: <199911241936.LAA19634@xsvr4.cup.hp.com> from "Frank Rowand" at Nov 24, 1999 11:36:25 AM MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii List-ID: Re: > > >> Look at the loop. What we do is basically > >> > >> cr16 = mfctl(16); > >> while(((cr16+loops)-mfctl(16))>0); > > > >You definitely don't want to do the above! > > > >Even ignoring the possibility of an interrupt that takes us away > >for awhile, there's the simple possibility that cr16 might > >roll over during your loop. > > Careful use of unsigned variables fixes the problem. The easiest way to > explain this is by a small sample program: ... Actually, unsigned integers cause the problem, as seen below. I'd been assuming cr16 and mfctl were uint32, not int32, hence my concern about a possible bug. (I don't have easy access to the source code) NOTE: with "int32 cr16" and "int32 mfctl (int crnum);" I don't see any problems. Assumptions: cr16 is a uint32 (note: better to use than "unsigned long int") mfctl is a uint32 function loops is a reasonable value The loop ending will be *missed* if the precise value of CR16: original_cr16 + loops happens to be missed (again, using uint32). This can easily happen if CR16 is ticking faster than the clock (which is allowed by the architecture) *or* can happen if an interrupt occurs shortly before CR16 hits the final expected value. Note that the unsigned arithmetic means that the subtraction in the "while" control will always be > 0 except for the precise time when cr16 = original_cr16 + loops. Here's sample code demonstrating the problem, along with sample output. (the following should loop 5 times...but it loops far more!) Startup: cr16 = 0x1e241, loops = 5, cr16 + loops = 0x1e246 (loop, cr16 = 0x 1e242), ((cr16 + loops) - mfctl(16)) = 0x 4) (loop, cr16 = 0x 1e243), ((cr16 + loops) - mfctl(16)) = 0x 3) (loop, cr16 = 0x 1e244), ((cr16 + loops) - mfctl(16)) = 0x 2) (loop, cr16 = 0x 1e245), ((cr16 + loops) - mfctl(16)) = 0x 1) (extra tick) (loop, cr16 = 0x 1e247), ((cr16 + loops) - mfctl(16)) = 0xffffffff) (loop, cr16 = 0x 1e248), ((cr16 + loops) - mfctl(16)) = 0xfffffffe) (loop, cr16 = 0x 1e249), ((cr16 + loops) - mfctl(16)) = 0xfffffffd) (loop, cr16 = 0x 1e24a), ((cr16 + loops) - mfctl(16)) = 0xfffffffc) (loop, cr16 = 0x 1e24b), ((cr16 + loops) - mfctl(16)) = 0xfffffffb) (loop, cr16 = 0x 1e24c), ((cr16 + loops) - mfctl(16)) = 0xfffffffa) (loop, cr16 = 0x 1e24d), ((cr16 + loops) - mfctl(16)) = 0xfffffff9) (loop, cr16 = 0x 1e24e), ((cr16 + loops) - mfctl(16)) = 0xfffffff8) (loop, cr16 = 0x 1e24f), ((cr16 + loops) - mfctl(16)) = 0xfffffff7) (loop, cr16 = 0x 1e250), ((cr16 + loops) - mfctl(16)) = 0xfffffff6) (loop, cr16 = 0x 1e251), ((cr16 + loops) - mfctl(16)) = 0xfffffff5) (loop, cr16 = 0x 1e252), ((cr16 + loops) - mfctl(16)) = 0xfffffff4) (loop, cr16 = 0x 1e253), ((cr16 + loops) - mfctl(16)) = 0xfffffff3) oops: we've looped too many times! -----------------------------cut here for source code------------------------- unsigned long cr16; unsigned long simulated_cr16; unsigned long loops; /***************************************************************/ unsigned long mfctl (int crnum) /* ticks once per call ... sometimes twice :) */ { simulated_cr16++; /* To demonstrate the problem, let's always make it */ /* tick more than once at *exactly* the worst time... */ if ((cr16 + loops) == simulated_cr16) { simulated_cr16++; printf (" (extra tick)\n"); } return simulated_cr16; } /* end mfctl proc */ /***************************************************************/ main() { int actual_loops; simulated_cr16 = 123456; /* actual value irrelevant */ cr16 = mfctl (16); /* every call to mfctl(16) causes a simulated tick loops = 5; /* desired number of loops */ actual_loops = 0; /* track number we've actually done */ printf ("Startup: cr16 = 0x%x, loops = %d, cr16 + loops = 0x%x\n", cr16, loops, (cr16 + loops)); while (((cr16 + loops) - mfctl (16)) > 0) { printf (" (loop, cr16 = 0x%8x), ((cr16 + loops) - mfctl(16)) = 0x%8x)\n", simulated_cr16, ((cr16 + loops) - simulated_cr16)); if (actual_loops++ > (loops + 10)) { printf ("oops: we've looped too many times!\n"); exit (1); } } /* while loop */ printf ("\nExited normally from delay loop. At exit:\n"); printf ("cr16 = 0x%8x, ((cr16 + loops) - mfctl(16)) = 0x%8x)\n", simulated_cr16, ((cr16 + loops) - simulated_cr16)); } /* end main proc */ /***************************************************************/ -- Stan Sieler sieler@allegro.com www.allegro.com/sieler/wanted/index.html www.allegro.com/sieler