On Tue, 16 Jan 2018, Joerg Roedel wrote: > On Tue, Jan 16, 2018 at 10:03:09AM -0800, Dave Hansen wrote: > > On 01/16/2018 08:36 AM, Joerg Roedel wrote: > > > + return (((ptr & ~PAGE_MASK) / sizeof(pgd_t)) < KERNEL_PGD_BOUNDARY); > > > +} > > > > One of the reasons to implement it the other way: > > > > - return (ptr & ~PAGE_MASK) < (PAGE_SIZE / 2); > > > > is that the compiler can do this all quickly. KERNEL_PGD_BOUNDARY > > depends on PAGE_OFFSET which depends on a variable. IOW, the compiler > > can't do it. > > > > How much worse is the code that this generates? > > I havn't looked at the actual code this generates, but the > (PAGE_SIZE / 2) comparison doesn't work on 32 bit where the address > space is not always evenly split. I'll look into a better way to check > this. It should be trivial enough to do return (ptr & ~PAGE_MASK) < PGD_SPLIT_SIZE); and define it PAGE_SIZE/2 for 64bit and for PAE make it depend on the configured address space split. Thanks, tglx -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxx. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"dont@xxxxxxxxx"> email@xxxxxxxxx </a>