On Wed, 2015-12-09 at 08:25 -0800, Dan Williams wrote: > On Thu, Dec 3, 2015 at 12:35 PM, Toshi Kani <toshi.kani@xxxxxxx> wrote: > > On Thu, 2015-12-03 at 11:01 -0800, Linus Torvalds wrote: > > > On Thu, Dec 3, 2015 at 10:40 AM, Borislav Petkov <bp@xxxxxxxxx> > > > wrote: > > > > On Thu, Dec 03, 2015 at 11:54:19AM -0700, Toshi Kani wrote: > > > > > Adding a new type for regular memory will require inspecting the > > > > > codes using IORESOURCE_MEM currently, and modify them to use the > > > > > new type if their target ranges are regular memory. There are > > > > > many references to this type across multiple architectures and > > > > > drivers, which make this inspection and testing challenging. > > > > > > > > What's wrong with adding a new type_flags to struct resource and > > > > not touching IORESOURCE_* at all? > > > > > > Bah. Both of these ideas are bogus. > > > > > > Just add a new flag. The bits are already modifiers that you can > > > *combine* to show what kind of resource it is, and we already have > > > things like IORESOURCE_PREFETCH etc, that are in *addition* to the > > > normal IORESOURCE_MEM bit. > > > > > > Just add another modifier: IORESOURCE_RAM. > > > > > > So it would still show up as IORESOURCE_MEM, but it would have > > > additional information specifying that it's actually RAM. > > > > > > If somebody does something like > > > > > > if (res->flags == IORESOURCE_MEM) > > > > > > then they are already completely broken and won't work *anyway*. It's > > > a bitmask, bit a set of values. > > > > Yes, if we can assign new modifiers, that will be quite simple. :-) I > > assume we can allocate new bits from the remaining free bits as > > follows. > > > > +#define IORESOURCE_SYSTEM_RAM 0x01000000 /* System RAM */ > > +#define IORESOURCE_PMEM 0x02000000 /* Persistent memory */ > > #define IORESOURCE_EXCLUSIVE 0x08000000 /* Userland may not map > > this resource */ > > > > Note, SYSTEM_RAM represents the OS memory, i.e. "System RAM", not any > > RAM ranges. > > > > With the new modifiers, region_intersect() can check these ranges. One > > caveat is that the modifiers are not very extensible for new types as > > they are bit maps. region_intersect() will no longer be capable of > > checking any regions with any given name. I think this is OK since > > this function was introduced recently, and is only used for checking > > "System RAM" and "Persistent Memory" (with this patch series). > > IORESOURCE_PMEM is not descriptive enough for the two different types > of pmem in the kernel. How about we go with just > IORESOURCE_SYSTEM_RAM for now since "is_ram()" checks are common. Let > the rest continue to be checked by strcmp(). > > For example the nvdimm-e820 driver cares about "Persistent Memory > (legacy)", while other forms of pmem may just be "reserved" and only > the driver knows that it is pmem. An IORESOURCE_PMEM would not be > reliable nor descriptive enough. Agreed. I will introduce a new type for System RAM, and leave the strcmp check for other types. Thanks, -Toshi -- 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>