Hi!
On Fri, Oct 18, 2019 at 04:53:26PM +0100, Mark Cave-Ayland wrote:
On 18/10/2019 10:52, Segher Boessenkool wrote:
On Fri, Oct 18, 2019 at 08:48:53AM +0100, Mark Cave-Ayland wrote:
Otherwise the compiler may not understand that a read from the assigned buffer is comming from memory-mapped IO space and fail to update accordingly.
I think you should not tell the compiler that whatever writes to ch writes the same value multiple times in sequence, instead?
Can you explain a bit more about what you mean? The issue seems to be related to this:
static volatile unsigned char *kbd_dev;
void kbd_init(phys_addr_t base) { kbd_dev = (unsigned char *)ofmem_map_io(base, 2 * 4); kbd_dev += 4; }
... ...
unsigned char keyboard_readdata(void) { volatile unsigned char ch;
while (!keyboard_dataready()) { } do { ch = kbd_dev[2] & 0xff; ...}
Without the volatile marking on ch it seems with optimisation enabled that ch gets optimised into a register causing strange intermittent keyboard behaviour. I guess since kbd_dev is a pointer to something of unknown size, the compiler doesn't understand that kbd_dev[2] should also be considered volatile?
No, the compiler understands that fine.
How is ch *used*?
[ It's a lot better to treat I/O as I/O, accessed via access functions, than to treat it as memory, exposed in C as some variables. I/O is not memory, and you cannot access it exactly the same way, normally. But that aside. ]
Segher