Step 33 of 56

PCF85063A: a clock that tells you when not to trust it

BCD calendar registers, a programmable alarm, and one flag that answers the only question that matters after a power loss.

The typed calendar

A small fixed-address (0x51) real-time clock. The time is not a blob of BCD bytes in the API — every field is a constrained subtype, so an impossible date cannot be constructed:

subtype Year_Number   is Natural range 2000 .. 2099;
subtype Month_Number  is Natural range 1 .. 12;
subtype Day_Number    is Natural range 1 .. 31;
subtype Hour_Number   is Natural range 0 .. 23;      --  24-hour mode only
subtype Minute_Number is Natural range 0 .. 59;
subtype Second_Number is Natural range 0 .. 59;

type Weekday is (Sunday, Monday, Tuesday, Wednesday, Thursday, Friday, Saturday);
type Time is record ... end record;

Day-of-week is worth knowing about: in the chip it is just a free 0 .. 6 counter that nothing validates against the date. The Weekday naming is a convention the driver applies, not something the hardware enforces — so the chip will happily tell you a Tuesday that isn't one if something wrote it wrong.

The flag that matters

procedure Get_Time (Dev : Device; T : out Time;
                    Valid : out Boolean; Result : out Status);

Valid is the whole point of using an RTC chip. It reflects the oscillator-stop flag: False means power was lost and the time it just handed you is meaningless. A clock that returns a plausible-looking wrong time is worse than one that admits it does not know, so branch on Valid before believing T — that is the difference between resuming a schedule and corrupting a log.

Set_Time stops the clock around the write, as the datasheet requires, and clears the flag.

The rest

Alarm on second, minute, hour, day or weekday — any subset — asserting an active-low open-drain INT line. Attach it with the .Interrupts child, observing the callback rules: latch in the handler, do the I2C read in a task.
Stop_Clock halts the counters via Control_1.STOP without changing the loaded time.
Reset returns every register to its power-on default, which also stops the clock.

Reads use the auto-incrementing pointer, and the pointer survives the STOP between the two transactions — so a one-byte write followed by a separate read streams correctly without needing repeated START. Up to 400 kHz.

Step 33 of 56

PCF85063A: a clock that tells you when not to trust it

BCD calendar registers, a programmable alarm, and one flag that answers the only question that matters after a power loss.

The typed calendar

A small fixed-address (0x51) real-time clock. The time is not a blob of BCD bytes in the API — every field is a constrained subtype, so an impossible date cannot be constructed:

subtype Year_Number   is Natural range 2000 .. 2099;
subtype Month_Number  is Natural range 1 .. 12;
subtype Day_Number    is Natural range 1 .. 31;
subtype Hour_Number   is Natural range 0 .. 23;      --  24-hour mode only
subtype Minute_Number is Natural range 0 .. 59;
subtype Second_Number is Natural range 0 .. 59;

type Weekday is (Sunday, Monday, Tuesday, Wednesday, Thursday, Friday, Saturday);
type Time is record ... end record;

Day-of-week is worth knowing about: in the chip it is just a free 0 .. 6 counter that nothing validates against the date. The Weekday naming is a convention the driver applies, not something the hardware enforces — so the chip will happily tell you a Tuesday that isn't one if something wrote it wrong.

The flag that matters

procedure Get_Time (Dev : Device; T : out Time;
                    Valid : out Boolean; Result : out Status);

Valid is the whole point of using an RTC chip. It reflects the oscillator-stop flag: False means power was lost and the time it just handed you is meaningless. A clock that returns a plausible-looking wrong time is worse than one that admits it does not know, so branch on Valid before believing T — that is the difference between resuming a schedule and corrupting a log.

Set_Time stops the clock around the write, as the datasheet requires, and clears the flag.

The rest

Alarm on second, minute, hour, day or weekday — any subset — asserting an active-low open-drain INT line. Attach it with the .Interrupts child, observing the callback rules: latch in the handler, do the I2C read in a task.
Stop_Clock halts the counters via Control_1.STOP without changing the loaded time.
Reset returns every register to its power-on default, which also stops the clock.

Reads use the auto-incrementing pointer, and the pointer survives the STOP between the two transactions — so a one-byte write followed by a separate read streams correctly without needing repeated START. Up to 400 kHz.

Step 33 of 56

PCF85063A: a clock that tells you when not to trust it

BCD calendar registers, a programmable alarm, and one flag that answers the only question that matters after a power loss.

The typed calendar

A small fixed-address (0x51) real-time clock. The time is not a blob of BCD bytes in the API — every field is a constrained subtype, so an impossible date cannot be constructed:

subtype Year_Number   is Natural range 2000 .. 2099;
subtype Month_Number  is Natural range 1 .. 12;
subtype Day_Number    is Natural range 1 .. 31;
subtype Hour_Number   is Natural range 0 .. 23;      --  24-hour mode only
subtype Minute_Number is Natural range 0 .. 59;
subtype Second_Number is Natural range 0 .. 59;

type Weekday is (Sunday, Monday, Tuesday, Wednesday, Thursday, Friday, Saturday);
type Time is record ... end record;

Day-of-week is worth knowing about: in the chip it is just a free 0 .. 6 counter that nothing validates against the date. The Weekday naming is a convention the driver applies, not something the hardware enforces — so the chip will happily tell you a Tuesday that isn't one if something wrote it wrong.

The flag that matters

procedure Get_Time (Dev : Device; T : out Time;
                    Valid : out Boolean; Result : out Status);

Valid is the whole point of using an RTC chip. It reflects the oscillator-stop flag: False means power was lost and the time it just handed you is meaningless. A clock that returns a plausible-looking wrong time is worse than one that admits it does not know, so branch on Valid before believing T — that is the difference between resuming a schedule and corrupting a log.

Set_Time stops the clock around the write, as the datasheet requires, and clears the flag.

The rest

Alarm on second, minute, hour, day or weekday — any subset — asserting an active-low open-drain INT line. Attach it with the .Interrupts child, observing the callback rules: latch in the handler, do the I2C read in a task.
Stop_Clock halts the counters via Control_1.STOP without changing the loaded time.
Reset returns every register to its power-on default, which also stops the clock.

Reads use the auto-incrementing pointer, and the pointer survives the STOP between the two transactions — so a one-byte write followed by a separate read streams correctly without needing repeated START. Up to 400 kHz.

Step 33 of 56

PCF85063A: a clock that tells you when not to trust it

BCD calendar registers, a programmable alarm, and one flag that answers the only question that matters after a power loss.

The typed calendar

A small fixed-address (0x51) real-time clock. The time is not a blob of BCD bytes in the API — every field is a constrained subtype, so an impossible date cannot be constructed:

subtype Year_Number   is Natural range 2000 .. 2099;
subtype Month_Number  is Natural range 1 .. 12;
subtype Day_Number    is Natural range 1 .. 31;
subtype Hour_Number   is Natural range 0 .. 23;      --  24-hour mode only
subtype Minute_Number is Natural range 0 .. 59;
subtype Second_Number is Natural range 0 .. 59;

type Weekday is (Sunday, Monday, Tuesday, Wednesday, Thursday, Friday, Saturday);
type Time is record ... end record;

Day-of-week is worth knowing about: in the chip it is just a free 0 .. 6 counter that nothing validates against the date. The Weekday naming is a convention the driver applies, not something the hardware enforces — so the chip will happily tell you a Tuesday that isn't one if something wrote it wrong.

The flag that matters

procedure Get_Time (Dev : Device; T : out Time;
                    Valid : out Boolean; Result : out Status);

Valid is the whole point of using an RTC chip. It reflects the oscillator-stop flag: False means power was lost and the time it just handed you is meaningless. A clock that returns a plausible-looking wrong time is worse than one that admits it does not know, so branch on Valid before believing T — that is the difference between resuming a schedule and corrupting a log.

Set_Time stops the clock around the write, as the datasheet requires, and clears the flag.

The rest

Alarm on second, minute, hour, day or weekday — any subset — asserting an active-low open-drain INT line. Attach it with the .Interrupts child, observing the callback rules: latch in the handler, do the I2C read in a task.
Stop_Clock halts the counters via Control_1.STOP without changing the loaded time.
Reset returns every register to its power-on default, which also stops the clock.

Reads use the auto-incrementing pointer, and the pointer survives the STOP between the two transactions — so a one-byte write followed by a separate read streams correctly without needing repeated START. Up to 400 kHz.

Step 33 of 56

PCF85063A: a clock that tells you when not to trust it

BCD calendar registers, a programmable alarm, and one flag that answers the only question that matters after a power loss.

The typed calendar

A small fixed-address (0x51) real-time clock. The time is not a blob of BCD bytes in the API — every field is a constrained subtype, so an impossible date cannot be constructed:

subtype Year_Number   is Natural range 2000 .. 2099;
subtype Month_Number  is Natural range 1 .. 12;
subtype Day_Number    is Natural range 1 .. 31;
subtype Hour_Number   is Natural range 0 .. 23;      --  24-hour mode only
subtype Minute_Number is Natural range 0 .. 59;
subtype Second_Number is Natural range 0 .. 59;

type Weekday is (Sunday, Monday, Tuesday, Wednesday, Thursday, Friday, Saturday);
type Time is record ... end record;

Day-of-week is worth knowing about: in the chip it is just a free 0 .. 6 counter that nothing validates against the date. The Weekday naming is a convention the driver applies, not something the hardware enforces — so the chip will happily tell you a Tuesday that isn't one if something wrote it wrong.

The flag that matters

procedure Get_Time (Dev : Device; T : out Time;
                    Valid : out Boolean; Result : out Status);

Valid is the whole point of using an RTC chip. It reflects the oscillator-stop flag: False means power was lost and the time it just handed you is meaningless. A clock that returns a plausible-looking wrong time is worse than one that admits it does not know, so branch on Valid before believing T — that is the difference between resuming a schedule and corrupting a log.

Set_Time stops the clock around the write, as the datasheet requires, and clears the flag.

The rest

Alarm on second, minute, hour, day or weekday — any subset — asserting an active-low open-drain INT line. Attach it with the .Interrupts child, observing the callback rules: latch in the handler, do the I2C read in a task.
Stop_Clock halts the counters via Control_1.STOP without changing the loaded time.
Reset returns every register to its power-on default, which also stops the clock.

Reads use the auto-incrementing pointer, and the pointer survives the STOP between the two transactions — so a one-byte write followed by a separate read streams correctly without needing repeated START. Up to 400 kHz.