| Message ID | 20150605211906.20751.59875.stgit@dwillia2-desk3.amr.corp.intel.com (mailing list archive) |
|---|---|
| State | Superseded |
| Headers | show |
On Fri, Jun 5, 2015 at 2:19 PM, Dan Williams <dan.j.williams@intel.com> wrote: > +enum { > +#if BITS_PER_LONG == 64 > + PFN_SHIFT = 3, > + /* device-pfn not covered by memmap */ > + PFN_DEV = (1UL << 2), > +#else > + PFN_SHIFT = 2, > +#endif > + PFN_MASK = (1UL << PFN_SHIFT) - 1, > + PFN_SG_CHAIN = (1UL << 0), > + PFN_SG_LAST = (1UL << 1), > +}; Ugh. Just make PFN_SHIFT unconditional. Make it 2, unconditionally. Or, if you want to have more bits, make it three unconditionally, and make 'struct page' just be at least 8-byte aligned even on 32-bit. Even on 32-bit architectures, there's plenty of bits. There's no reason to "pack" this optimally. Remember: it's a page frame number, so there's the page size shifting going on in physical memory, and even if you shift the PFN by 3 - or four of five - bits unconditionally (rather than try to shift it by some minimal number), you're covering a *lot* of physical memory. Say you're a 32-bit architecture with a 4k page size, and you lose three bits to "type" bits. You still have 32+12-3=41 bits of physical address space. Which is way more than realistic for a 32-bit architecture anyway, even with PAE (or PXE or whatever ARM calls it). Not that I see persistent memory being all that relevant on 32-bit hardware anyway. So I think if you actually do want that third bit, you're better off just marking "struct page" as being __aligned__((8)) and getting the three bits unconditionally. Just make the rule be that mem_map[] has to be 8-byte aligned. Even 16-byte alignment would probably be fine. No? Linus
On Fri, Jun 5, 2015 at 2:37 PM, Linus Torvalds <torvalds@linux-foundation.org> wrote: > On Fri, Jun 5, 2015 at 2:19 PM, Dan Williams <dan.j.williams@intel.com> wrote: >> +enum { >> +#if BITS_PER_LONG == 64 >> + PFN_SHIFT = 3, >> + /* device-pfn not covered by memmap */ >> + PFN_DEV = (1UL << 2), >> +#else >> + PFN_SHIFT = 2, >> +#endif >> + PFN_MASK = (1UL << PFN_SHIFT) - 1, >> + PFN_SG_CHAIN = (1UL << 0), >> + PFN_SG_LAST = (1UL << 1), >> +}; > > Ugh. Just make PFN_SHIFT unconditional. Make it 2, unconditionally. > Or, if you want to have more bits, make it three unconditionally, and > make 'struct page' just be at least 8-byte aligned even on 32-bit. > > Even on 32-bit architectures, there's plenty of bits. There's no > reason to "pack" this optimally. Remember: it's a page frame number, > so there's the page size shifting going on in physical memory, and > even if you shift the PFN by 3 - or four of five - bits > unconditionally (rather than try to shift it by some minimal number), > you're covering a *lot* of physical memory. It is a page frame number, but page_to_pfn_t() just stores the value of the struct page pointer directly, so we really only have the pointer alignment bits. I do this so that kmap_atomic_pfn_t() can optionally call kmap_atomic() if the pfn is mapped. > > Say you're a 32-bit architecture with a 4k page size, and you lose > three bits to "type" bits. You still have 32+12-3=41 bits of physical > address space. Which is way more than realistic for a 32-bit > architecture anyway, even with PAE (or PXE or whatever ARM calls it). > Not that I see persistent memory being all that relevant on 32-bit > hardware anyway. > > So I think if you actually do want that third bit, you're better off > just marking "struct page" as being __aligned__((8)) and getting the > three bits unconditionally. Just make the rule be that mem_map[] has > to be 8-byte aligned. > > Even 16-byte alignment would probably be fine. No? > Ooh, that's great, I was already lamenting the fact that I had run out of bits. One of the reasons to go to 16-byte alignment is to have another bit to further qualify the pfn as persistent memory not just un-mapped memory. The rationale would be to generate, and verify proper usage of, __pmem annotated pointers. ...but I'm still waiting for someone to tell me I'm needlessly complicating things with a __pmem annotation [1]. [1]: https://lists.01.org/pipermail/linux-nvdimm/2015-June/001087.html
diff --git a/include/asm-generic/memory_model.h b/include/asm-generic/memory_model.h index 14909b0b9cae..1b0ae21fd8ff 100644 --- a/include/asm-generic/memory_model.h +++ b/include/asm-generic/memory_model.h @@ -70,7 +70,6 @@ #endif /* CONFIG_FLATMEM/DISCONTIGMEM/SPARSEMEM */ #define page_to_pfn __page_to_pfn -#define pfn_to_page __pfn_to_page #endif /* __ASSEMBLY__ */ diff --git a/include/asm-generic/pfn.h b/include/asm-generic/pfn.h new file mode 100644 index 000000000000..2f4ae40dc6a7 --- /dev/null +++ b/include/asm-generic/pfn.h @@ -0,0 +1,86 @@ +#ifndef __ASM_PFN_H +#define __ASM_PFN_H + +/* + * Default pfn to physical address conversion, like most arch + * page_to_phys() implementations this resolves to a dma_addr_t as it + * should be the size needed for a device to reference this address. + */ +#ifndef __pfn_to_phys +#define __pfn_to_phys(pfn) ((dma_addr_t)(pfn) << PAGE_SHIFT) +#endif + +static inline struct page *pfn_to_page(unsigned long pfn) +{ + return __pfn_to_page(pfn); +} + +/* + * __pfn_t: encapsulates a page-frame number that is optionally backed + * by memmap (struct page). This type will be used in place of a + * 'struct page *' instance in contexts where unmapped memory (usually + * persistent memory) is being referenced (scatterlists for drivers, + * biovecs for the block layer, etc). Whether a __pfn_t has a struct + * page backing is indicated by flags in the low bits of @data. + */ +typedef struct { + union { + unsigned long data; + struct page *page; + }; +} __pfn_t; + +enum { +#if BITS_PER_LONG == 64 + PFN_SHIFT = 3, + /* device-pfn not covered by memmap */ + PFN_DEV = (1UL << 2), +#else + PFN_SHIFT = 2, +#endif + PFN_MASK = (1UL << PFN_SHIFT) - 1, + PFN_SG_CHAIN = (1UL << 0), + PFN_SG_LAST = (1UL << 1), +}; + +#ifdef CONFIG_DEV_PFN +static inline bool __pfn_t_has_page(__pfn_t pfn) +{ + return (pfn.data & PFN_MASK) == 0; +} + +#else +static inline bool __pfn_t_has_page(__pfn_t pfn) +{ + return true; +} +#endif + +static inline struct page *__pfn_t_to_page(__pfn_t pfn) +{ + if (!__pfn_t_has_page(pfn)) + return NULL; + return pfn.page; +} + +static inline unsigned long __pfn_t_to_pfn(__pfn_t pfn) +{ + if (__pfn_t_has_page(pfn)) + return page_to_pfn(pfn.page); + return pfn.data >> PFN_SHIFT; +} + +static inline dma_addr_t __pfn_t_to_phys(__pfn_t pfn) +{ + if (!__pfn_t_has_page(pfn)) + return __pfn_to_phys(__pfn_t_to_pfn(pfn)); + return __pfn_to_phys(page_to_pfn(pfn.page)); +} + +static inline __pfn_t page_to_pfn_t(struct page *page) +{ + __pfn_t pfn = { .page = page }; + + return pfn; +} +#endif /* __ASM_PFN_H */ diff --git a/include/linux/mm.h b/include/linux/mm.h index 4024543b4203..ae6f9965f3dd 100644 --- a/include/linux/mm.h +++ b/include/linux/mm.h @@ -53,6 +53,7 @@ extern int sysctl_legacy_va_layout; #include <asm/page.h> #include <asm/pgtable.h> #include <asm/processor.h> +#include <asm-generic/pfn.h> #ifndef __pa_symbol #define __pa_symbol(x) __pa(RELOC_HIDE((unsigned long)(x), 0)) diff --git a/init/Kconfig b/init/Kconfig index d4f763332f9f..907ab91a5557 100644 --- a/init/Kconfig +++ b/init/Kconfig @@ -1769,6 +1769,19 @@ config PROFILING Say Y here to enable the extended profiling support mechanisms used by profilers such as OProfile. +config DEV_PFN + depends on 64BIT + bool "Support for device provided (pmem, graphics, etc) memory" if EXPERT + help + Say Y here to enable I/O to/from device provided memory, + i.e. reference memory that is not mapped. This is usually + the case if you have large quantities of persistent memory + relative to DRAM. Enabling this option may increase the + kernel size by a few kilobytes as it instructs the kernel + that a __pfn_t may reference unmapped memory. Disabling + this option instructs the kernel that a __pfn_t always + references mapped platform memory. + # # Place an empty function call at each tracepoint site. Can be # dynamically changed for a probe function.
Introduce a type that encapsulates a page-frame-number that can also be used to encode other information. This other information is the traditional "page_link" encoding in a scatterlist, but can also denote "device memory". Where "device memory" is a set of pfns that are not part of the kernel's linear mapping, but are accessed via the same memory controller as ram. The motivation for this conversion is large capacity persistent memory that does not enjoy struct page coverage, entries in memmap, by default. This type will be used in replace usage of 'struct page *' in cases where only a pfn is required, i.e. scatterlists for drivers, dma mapping api, and later biovecs for the block layer. The operations in those i/o paths that formerly required a 'struct page *' are converted to use __pfn_t aware equivalent helpers. The goal is that existing use cases of data structures referencing struct page are binary identical to the __pfn_t converted type. Only new use cases for pmem will require the new __pfn_t helper routines, i.e. __pfn_t is a struct page alias in the absence of pmem. It turns out that while 'struct page' references are used broadly in the kernel I/O stacks the usage of 'struct page' based capabilities is very shallow for block-i/o. It is only used for populating bio_vecs and scatterlists for the retrieval of dma addresses, and for temporary kernel mappings (kmap). Aside from kmap, these usages can be trivially converted to operate on a pfn. Indeed, kmap_atomic() is more problematic as it uses mm infrastructure, via struct page, to setup and track temporary kernel mappings. It would be unfortunate if the kmap infrastructure escaped its 32-bit/HIGHMEM bonds and leaked into 64-bit code. Thankfully, it seems all that is needed here is to convert kmap_atomic() callers, that want to opt-in to supporting persistent memory, to use a new kmap_atomic_pfn_t(). Where kmap_atomic_pfn_t() is enabled to re-use the existing ioremap() mapping established by the driver for persistent memory. Note, that as far as conceptually understanding __pfn_t is concerned, 'persistent memory' is really any address range in host memory not covered by memmap. Contrast this with pure iomem that is on an mmio mapped bus like PCI and cannot be converted to a dma_addr_t by "pfn << PAGE_SHIFT". Cc: H. Peter Anvin <hpa@zytor.com> Cc: Jens Axboe <axboe@kernel.dk> Cc: Tejun Heo <tj@kernel.org> Cc: Ingo Molnar <mingo@kernel.org> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Dan Williams <dan.j.williams@intel.com> --- include/asm-generic/memory_model.h | 1 include/asm-generic/pfn.h | 86 ++++++++++++++++++++++++++++++++++++ include/linux/mm.h | 1 init/Kconfig | 13 +++++ 4 files changed, 100 insertions(+), 1 deletion(-) create mode 100644 include/asm-generic/pfn.h