I'm using an ATtiny1634 for my project and I wanted to make sure the external 8MHz crystal was soldered correctly before I set the fuse to select it at boot. So here's my code from what I found in the data sheet:
void setup()
{
// Select external 8MHz crystal
CCP = 0xD8; // ATtiny1634 Signature
CLKSR = 0b1101 | _BV(CSTR);
loop_until_bit_is_set(CLKSR, OSCRDY);
// Set prescaler to unity
CCP = 0xD8; // ATtiny1634 Signature
CLKPR = 0;
loop_until_bit_is_set(CLKSR, OSCRDY);
}
This is just the setup function that I'm interested in here. In a loop function, I toggle the state of an output that is connected to a buzzer (with no internal oscillator) so I can hear the resulting tone. Of course I introduced some delay to make the sound audible.
There are 2 things I noticed. First the prescaler doesn't seem to change — I should hear a sound about 8 times higher but it stays the same. Next the micro-controller sometimes seems to stall, then I hear a screetching noise coming from the buffer and the frequency slowly ramps up to the expected tone. The tone is stable after, say, 8-10 seconds.
Is this normal? Or have I missed something?
Strangely enough the buzzer makes its screetchy noise whenever I touch the XTAL pins pads (which the crystal is soldered on) with my scope probe. That suggests the crystal is actually working… right? That [slow clock rise] doesn't happen if I comment out the clock selection. And it doesn't matter if I use the 10x range probe setting or not. I also used two 15pF with the crystal, as advised in the data sheet (between 12 and 22pF).
EDIT: It turns out the first issue was a PEBCAK. For one test I had compiled the program and set F_CPU to 1MHz on the command line (with the default prescaler value, aka 8). For the second test, prescaler programmatically set to 1, I had used the default F_CPU value from my makefile, that is 8MHz. It's therefore absolutely normal that didn't hear a difference in the generated tone.
I'd just forgotten that so the remaining issue is that of the external crystal. I still cannot rule out a bad solder joint, which is why I wanted to select the clock dynamically.
EDIT: Below is a view of the crystal and tracks.
I placed the crystal and caps as close to the MCU as I could. I've already used such a design for an ATmega64M1 without an issue.
EDIT: And below is complete the assembly code (for testing purposes) — now with proper register names.
main: format de fichier elf32-avr
Déassemblage de la section .text :
00000000 <__vectors>:
0: 37 c0 rjmp .+110 ; 0x70 <__ctors_end>
4: 49 c0 rjmp .+146 ; 0x98 <__bad_interrupt>
8: 47 c0 rjmp .+142 ; 0x98 <__bad_interrupt>
c: 45 c0 rjmp .+138 ; 0x98 <__bad_interrupt>
10: 43 c0 rjmp .+134 ; 0x98 <__bad_interrupt>
14: 41 c0 rjmp .+130 ; 0x98 <__bad_interrupt>
18: 3f c0 rjmp .+126 ; 0x98 <__bad_interrupt>
1c: 3d c0 rjmp .+122 ; 0x98 <__bad_interrupt>
20: 3b c0 rjmp .+118 ; 0x98 <__bad_interrupt>
24: 39 c0 rjmp .+114 ; 0x98 <__bad_interrupt>
28: 37 c0 rjmp .+110 ; 0x98 <__bad_interrupt>
2c: 35 c0 rjmp .+106 ; 0x98 <__bad_interrupt>
30: 33 c0 rjmp .+102 ; 0x98 <__bad_interrupt>
34: 31 c0 rjmp .+98 ; 0x98 <__bad_interrupt>
38: 2f c0 rjmp .+94 ; 0x98 <__bad_interrupt>
3c: 2d c0 rjmp .+90 ; 0x98 <__bad_interrupt>
40: 2b c0 rjmp .+86 ; 0x98 <__bad_interrupt>
44: 29 c0 rjmp .+82 ; 0x98 <__bad_interrupt>
48: 27 c0 rjmp .+78 ; 0x98 <__bad_interrupt>
4c: 25 c0 rjmp .+74 ; 0x98 <__bad_interrupt>
50: 23 c0 rjmp .+70 ; 0x98 <__bad_interrupt>
54: 21 c0 rjmp .+66 ; 0x98 <__bad_interrupt>
58: 1f c0 rjmp .+62 ; 0x98 <__bad_interrupt>
5c: 1d c0 rjmp .+58 ; 0x98 <__bad_interrupt>
60: 1b c0 rjmp .+54 ; 0x98 <__bad_interrupt>
64: 19 c0 rjmp .+50 ; 0x98 <__bad_interrupt>
68: 17 c0 rjmp .+46 ; 0x98 <__bad_interrupt>
6c: 15 c0 rjmp .+42 ; 0x98 <__bad_interrupt>
...
00000070 <__ctors_end>:
70: 11 24 eor r1, r1
72: 1f be out SREG, r1 ; 63
74: cf ef ldi r28, 0xFF ; 255
76: d4 e0 ldi r29, 0x04 ; 4
78: de bf out SPH, r29 ; 62
7a: cd bf out SPL, r28 ; 61
0000007c <setup>:
#define SELECT C,2
#define BUZZER A,6
void setup()
{
// Wait for clock to be stable (probably useless here but...)
7c: 02 b6 in r0, CLKSR ; 50
7e: 07 fe sbrs r0, 7
80: fd cf rjmp .-6 ; 0x7c <setup>
CCP = CCPSIG;
CLKSR = 0b1101 | _BV(CSTR); // Select 8MHz external crystal
loop_until_bit_is_set(CLKSR, OSCRDY);
#endif
// Change clock prescaler to unity
82: 88 ed ldi r24, 0xD8 ; 216
84: 8f bd out CCP, r24 ; 47
clock_prescale_set(clock_div_1);
86: 13 be out CLKPR, r1 ; 51
//~ CCP = CCPSIG;
88: 02 b6 in r0, CLKSR ; 50
8a: 07 fe sbrs r0, 7
8c: fd cf rjmp .-6 ; 0x88 <setup+0xc>
//~ CLKPR = 0;
loop_until_bit_is_set(CLKSR, OSCRDY);
8e: 42 9a sbi DDRC, 2 ; 8
// Setup pins
90: 86 9a sbi DDRA, 6 ; 16
set_output(SELECT);
92: 4a 9a sbi PORTC, 2 ; 9
set_output(BUZZER);
94: 0b d0 rcall .+22 ; 0xac <main>
96: 0c c0 rjmp .+24 ; 0xb0 <_exit>
00000098 <__bad_interrupt>:
98: b3 cf rjmp .-154 ; 0x0 <__vectors>
0000009a <loop>:
set_pin(SELECT);
}
void loop()
9a: 91 b3 in r25, PORTA ; 17
9c: 80 e4 ldi r24, 0x40 ; 64
9e: 89 27 eor r24, r25
a0: 81 bb out PORTA, r24 ; 17
milliseconds can be achieved.
*/
void
_delay_loop_2(uint16_t __count)
{
__asm__ volatile (
a2: 84 ef ldi r24, 0xF4 ; 244
a4: 91 e0 ldi r25, 0x01 ; 1
a6: 01 97 sbiw r24, 0x01 ; 1
a8: f1 f7 brne .-4 ; 0xa6 <loop+0xc>
aa: 08 95 ret
000000ac <main>:
{
toggle_pin(BUZZER);
_delay_us(250);
}
ac: f6 df rcall .-20 ; 0x9a <loop>
ae: fe cf rjmp .-4 ; 0xac <main>
000000b0 <_exit>:
b0: f8 94 cli
000000b2 <__stop_program>:
b2: ff cf rjmp .-2 ; 0xb2 <__stop_program>
Best Answer
I confirm this is indeed how to change the MCU clock programmatically, so there's nothing wrong in the code. The issue about the slow start was, like I suspected, a bad solder joint under the crystal pads.
With a heat gun I heated up the board (with the crystal) to 180°C-220°C (after a pre-heat stage around 150°C), re-did the solder joint with fresh solder and enough flux and... voilà! the "slow start" issue is gone and I can now see the 8MHz sine on both pins of the ATtiny1634 [EDIT: without disturbing the signal, this time].