hello,
I'm working on a 9260 board running linux 2.6.36.
I try to toggle some GPIO, this works fine if I run a C program using sysfs method, but it's too slow for me.
I try to toggle these GPIOs with another C program using mmap, and it's works a little faster.
My favorite language is C++, so I try the mmap method with, and compile it with g++. And all I get is a "segmentation Fault" at the first instruction using the returning value of mmap
...
m_MapBase = ::mmap(0, MAP_SIZE, PROT_READ | PROT_WRITE, MAP_SHARED, m_gpios, AT91_SYS & ~MAP_MASK);
....
uint32_t value = *(( volatile uint32_t*)( m_MapBase + ((PMC_OFFSET + PMC_PCSR) & MAP_MASK)));
-> Segmentation Fault !

I try to recompile the C program with g++, and I get the same seg fault as in C++ program...

Is there a problem between C++ and mmap ?

any idea is welcome...
thanks

Pierre

Well consider what the pointer returned (m_MapBase) actually is, and if it is valid.

If it is not look at m_gpios, for the C version of mmap() this should be the handle from opening "/dev/mem", is that valid?

Then consider if (PMC_OFFSET + PMC_PCSR) > MAP_SIZE, or if MAP_SIZE is a power of 2.

Do some more basic debugging, provide more context.

Of course m_gpio is a handle :
m_gpios = open("/dev/mem", O_RDWR | O_SYNC)
and its value is ok :

(gdb) p m_gpios
$1 = 4

#define MAP_SIZE 4096UL
#define PMC_OFFSET 0xC00
#define PMC_PCSR 0x0018 // Peripheral Clock Status Register - Read-only
so I try to read inside the map, not outside

Here is a gdb screen :

Program received signal SIGSEGV, Segmentation fault.
0x000096f0 in cFastGPIO::Init (this=0xbea11c78) at cFastGPIO.cpp:48
48 uint32_t value = *(( volatile uint32_t*)( m_MapBase + ((PMC_OF...
(gdb) p value
$1 = 4
(gdb) p m_MapBase
$2 = (void *) 0x400df000

the thing is surprising for me , is the fact that the (quite) same code is running well in C!