Uninitialized variable warning in UARTSerial at -O3 #6611

pauluap · 2018-04-11T22:15:08Z

Description

Appears when complied with -O3 optimization level

Compile: UARTSerial.cpp
../drivers/UARTSerial.cpp: In member function 'void mbed::UARTSerial::tx_irq()':
../drivers/UARTSerial.cpp:314:31: warning: 'data' may be used uninitialized in this function [-Wmaybe-uninitialized]
         SerialBase::_base_putc(data);

Pull request type

[X] Fix
[ ] Refactor
[ ] New target
[ ] Feature
[ ] Breaking change

TeroJaasko · 2018-04-16T09:45:48Z

drivers/UARTSerial.cpp

@@ -305,11 +305,11 @@ void UARTSerial::rx_irq(void)
 void UARTSerial::tx_irq(void)
 {
    bool was_full = _txbuf.full();
+    char data = 0;


Isn't this now just masking the problem, which the compiler correctly warned for: if "CricularBuffer::pop()" returns false, the data -variable should not be forwarded to SerialBase::_base_putc()?

Unless there's threading issues involved here, I don't think that pop() will return false because of the check to empty() The compiler is not able to make that rationalization, so I'm not sure if testing the return value of pop() would help.

I suppose that it's writable as

char data; /* Write to the peripheral if there is something to write * and if the peripheral is available to write. */ while (SerialBase::writeable() && _txbuf.pop(data)) { SerialBase::_base_putc(data); }

Huh, it seems that the compiler was able to analyze the check to pop() after all, with that change I don't get a warning.

I switched the conditional so that it short-circuits the call to pop() if the object isn't writable.

I would go either way, within an IRQ I think that it's safe to call pop after !empty I chose the path of minimal change but could switch.

pop calls empty internally, so the end result is the same.

I think this version in the comment without the empty check is better. I think it's only the way it is because it's was imitating the read case.

Appears when complied with -O3 optimization level Compile: UARTSerial.cpp ../drivers/UARTSerial.cpp: In member function 'void mbed::UARTSerial::tx_irq()': ../drivers/UARTSerial.cpp:314:31: warning: 'data' may be used uninitialized in this function [-Wmaybe-uninitialized] SerialBase::_base_putc(data);

cmonr · 2018-04-17T04:36:32Z

Starting CI since it looks like most recent commit addresses @kjbracey-arm' feedback.

/morph build

mbed-ci · 2018-04-17T05:02:00Z

Build : SUCCESS

Build number : 1765
Build artifacts/logs : http://mbed-os.s3-website-eu-west-1.amazonaws.com/?prefix=builds/6611/

Triggering tests

/morph test
/morph uvisor-test
/morph export-build
/morph mbed2-build

kjbracey

Looks good to me. Pondered performance a bit - this does mean one more serial_writable call per IRQ if we're not saturating the serial port, but does significantly decrease the number of circular buffer calls (which are a bit clunky with IRQ enable/disable). Should be a net performance win when highly loaded, I hope.