I knew the original was undefined because Ian showed me the relevant section from the standard. I'm actually surprised that it doesn't work in GCC. Using -fno-strict-overflow doesn't help either. I think at the optimizations that the kernel uses -O2 and -Os the original "works". Anyway, the old code is definitely wrong. But the new code is also undefined because we are subtracting from INT_MIN. I imagine how GCC could handle the undefined behavior in an unexpected way so the new code is probably fine. But we may as well just be pedantic. b_mask = (b_chans < 32) ? ((1U << b_chans) - 1) : 0xffffffff; regards, dan carpenter _______________________________________________ devel mailing list devel@xxxxxxxxxxxxxxxxxxxxxx http://driverdev.linuxdriverproject.org/mailman/listinfo/driverdev-devel
