Provide fix for issue #5835 - Tickers update should be atomic

[mprse]

Description

Perform update of the present time in critical section.

Status

READY

Migrations

NO

Steps to test or reproduce

While working on new lp ticker driver for NRF51_DK board (#5629), some problems with lp ticker has been detected.

Performed test:

volatile int cnt = 0;
void isr()
{
    cnt++;
}

void test()
{
    {
        Timer timer;
        LowPowerTicker ticker;
        timer.start();

        ticker.attach_us(isr, 1000);

        int val = 0;
        int buf = 0;

        while(val < 1) {
            val = timer.read();

            if(val != buf) {
                printf("--> val: %d \n", val);
                buf = val;
            }

        }
        ticker.detach();
        printf("---> cnt: %d \n", cnt);
    }
}

We expect that isr() function will be executed about 1000 times and exit from while loop is after one second:

val: 1
cnt: ¬1000

Test results without this fix NRF51_DK/GCC_ARM:

[1516363480.03][CONN][RXD] --> val: 3
[1516363480.05][CONN][RXD] ---> cnt: 3237

[1516363445.63][CONN][RXD] --> val: 3
[1516363445.65][CONN][RXD] ---> cnt: 3228

[1516362979.66][CONN][RXD] --> val: 3
[1516362979.67][CONN][RXD] ---> cnt: 3219

Test results with this fix NRF51_DK/GCC_ARM:

[1516363621.83][CONN][RXD] --> val: 1
[1516363621.86][CONN][RXD] ---> cnt: 1016

[1516363576.84][CONN][RXD] --> val: 1
[1516363576.86][CONN][RXD] ---> cnt: 1016

[1516362931.19][CONN][RXD] --> val: 1
[1516362931.20][CONN][RXD] ---> cnt: 1016

[0xc0170]

Thanks for the testing before/after 👍

[0xc0170]

/morph build

[mbed-ci]

Build : SUCCESS

Build number : 898
Build artifacts/logs : http://mbed-os.s3-website-eu-west-1.amazonaws.com/?prefix=builds/5889/

Triggering tests

/morph test
/morph uvisor-test
/morph export-build

[pan-]

If ever the time spent in critical section is too long, I suppose this can be rewritten in way where all the computation happen outside of the critical section.

The only parts that needs protection are data reads and data writes.

[mbed-ci]

Exporter Build : SUCCESS

Build number : 573
Build artifacts/logs : http://mbed-os.s3-website-eu-west-1.amazonaws.com/?prefix=builds/exporter/5889/

[cmonr]

/morph test

[mbed-ci]

Test : FAILURE

Build number : 742
Test logs :http://mbed-os-logs.s3-website-us-west-1.amazonaws.com/?prefix=logs/5889/742

[c1728p9]

One alternative you may want to consider is synchronizing only (and all) the public functions. Remaining function which could benefit from synchronization:

• ticker_set_handler
• ticker_irq_handler
• ticker_read_us
• ticker_read (implicitly synchronized if ticker_read_us is)

[mprse]

Provided fix has influence on tests-mbed_drivers-ticker. This test now fails on NRF51_DK/GCC_ARM.
Probably time spent in critical section is too long.

[mprse]

As suggested by @c1728p9 synchronized only (and all) the public functions.

Tested (tests-mbed_drivers-ticker) on the following platforms: NRF51_DK, K64F, NUCLEO_F070RB (all tool-chains) with positive results.

Please review.

[0xc0170]

/morph build

[cmonr]

/morph build

[mbed-ci]

Build : SUCCESS

Build number : 907
Build artifacts/logs : http://mbed-os.s3-website-eu-west-1.amazonaws.com/?prefix=builds/5889/

Triggering tests

/morph test
/morph uvisor-test
/morph export-build

[mbed-ci]

Exporter Build : SUCCESS

Build number : 582
Build artifacts/logs : http://mbed-os.s3-website-eu-west-1.amazonaws.com/?prefix=builds/exporter/5889/

[mbed-ci]

Test : SUCCESS

Build number : 749
Test logs :http://mbed-os-logs.s3-website-us-west-1.amazonaws.com/?prefix=logs/5889/749