On Mon, Aug 14, 2017 at 12:59:59PM +0100, Catalin Marinas wrote: > On Mon, Aug 07, 2017 at 07:35:56PM +0100, Mark Rutland wrote: > > Currently we define THREAD_SIZE and THREAD_SIZE order separately, with > > s/THREAD_SIZE order/THREAD_SIZE_ORDER/
Whoops; I'll fix that up now. > > +#define THREAD_SHIFT 14 > > + > > +#if THREAD_SHIFT >= PAGE_SHIFT > > +#define THREAD_SIZE_ORDER (THREAD_SHIFT - PAGE_SHIFT) > > +#endif > > + > > +#define THREAD_SIZE (UL(1) << THREAD_SHIFT) > > I haven't tried the series to this patch but it seems to me that > THREAD_SIZE_ORDER is undefined for a PAGE_SHIFT of 16. That is already the case without these patches, as we have: #ifdef CONFIG_ARM64_4K_PAGES #define THREAD_SIZE_ORDER 2 #elif defined(CONFIG_ARM64_16K_PAGES) #define THREAD_SIZE_ORDER 0 #endif ... this is also deliberate, as we'd need an order of -2 with 16K stacks and 64K pages, and this doesn't make sense. THREAD_SIZE_ORDER only matters if THREAD_SIZE >= PAGE_SIZE (or if we're using a VMAP'd stack). For 64K pages && VMAP_STACK, we'll use a 64K stack, avoiding the problem. Thanks, Mark.
