linux_dsm_epyc7002/arch/x86/include/asm/pmem.h
Dan Williams 96601adb74 x86, pmem: clarify that ARCH_HAS_PMEM_API implies PMEM mapped WB
Given that a write-back (WB) mapping plus non-temporal stores is
expected to be the most efficient way to access PMEM, update the
definition of ARCH_HAS_PMEM_API to imply arch support for
WB-mapped-PMEM.  This is needed as a pre-requisite for adding PMEM to
the direct map and mapping it with struct page.

The above clarification for X86_64 means that memcpy_to_pmem() is
permitted to use the non-temporal arch_memcpy_to_pmem() rather than
needlessly fall back to default_memcpy_to_pmem() when the pcommit
instruction is not available.  When arch_memcpy_to_pmem() is not
guaranteed to flush writes out of cache, i.e. on older X86_32
implementations where non-temporal stores may just dirty cache,
ARCH_HAS_PMEM_API is simply disabled.

The default fall back for persistent memory handling remains.  Namely,
map it with the WT (write-through) cache-type and hope for the best.

arch_has_pmem_api() is updated to only indicate whether the arch
provides the proper helpers to meet the minimum "writes are visible
outside the cache hierarchy after memcpy_to_pmem() + wmb_pmem()".  Code
that cares whether wmb_pmem() actually flushes writes to pmem must now
call arch_has_wmb_pmem() directly.

Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Reviewed-by: Ross Zwisler <ross.zwisler@linux.intel.com>
[hch: set ARCH_HAS_PMEM_API=n on x86_32]
Reviewed-by: Christoph Hellwig <hch@lst.de>
[toshi: x86_32 compile fixes]
Signed-off-by: Toshi Kani <toshi.kani@hp.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2015-08-27 19:40:59 -04:00

154 lines
4.6 KiB
C

/*
* Copyright(c) 2015 Intel Corporation. All rights reserved.
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of version 2 of the GNU General Public License as
* published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful, but
* WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* General Public License for more details.
*/
#ifndef __ASM_X86_PMEM_H__
#define __ASM_X86_PMEM_H__
#include <linux/uaccess.h>
#include <asm/cacheflush.h>
#include <asm/cpufeature.h>
#include <asm/special_insns.h>
#ifdef CONFIG_ARCH_HAS_PMEM_API
/**
* arch_memcpy_to_pmem - copy data to persistent memory
* @dst: destination buffer for the copy
* @src: source buffer for the copy
* @n: length of the copy in bytes
*
* Copy data to persistent memory media via non-temporal stores so that
* a subsequent arch_wmb_pmem() can flush cpu and memory controller
* write buffers to guarantee durability.
*/
static inline void arch_memcpy_to_pmem(void __pmem *dst, const void *src,
size_t n)
{
int unwritten;
/*
* We are copying between two kernel buffers, if
* __copy_from_user_inatomic_nocache() returns an error (page
* fault) we would have already reported a general protection fault
* before the WARN+BUG.
*/
unwritten = __copy_from_user_inatomic_nocache((void __force *) dst,
(void __user *) src, n);
if (WARN(unwritten, "%s: fault copying %p <- %p unwritten: %d\n",
__func__, dst, src, unwritten))
BUG();
}
/**
* arch_wmb_pmem - synchronize writes to persistent memory
*
* After a series of arch_memcpy_to_pmem() operations this drains data
* from cpu write buffers and any platform (memory controller) buffers
* to ensure that written data is durable on persistent memory media.
*/
static inline void arch_wmb_pmem(void)
{
/*
* wmb() to 'sfence' all previous writes such that they are
* architecturally visible to 'pcommit'. Note, that we've
* already arranged for pmem writes to avoid the cache via
* arch_memcpy_to_pmem().
*/
wmb();
pcommit_sfence();
}
/**
* __arch_wb_cache_pmem - write back a cache range with CLWB
* @vaddr: virtual start address
* @size: number of bytes to write back
*
* Write back a cache range using the CLWB (cache line write back)
* instruction. This function requires explicit ordering with an
* arch_wmb_pmem() call. This API is internal to the x86 PMEM implementation.
*/
static inline void __arch_wb_cache_pmem(void *vaddr, size_t size)
{
u16 x86_clflush_size = boot_cpu_data.x86_clflush_size;
unsigned long clflush_mask = x86_clflush_size - 1;
void *vend = vaddr + size;
void *p;
for (p = (void *)((unsigned long)vaddr & ~clflush_mask);
p < vend; p += x86_clflush_size)
clwb(p);
}
/*
* copy_from_iter_nocache() on x86 only uses non-temporal stores for iovec
* iterators, so for other types (bvec & kvec) we must do a cache write-back.
*/
static inline bool __iter_needs_pmem_wb(struct iov_iter *i)
{
return iter_is_iovec(i) == false;
}
/**
* arch_copy_from_iter_pmem - copy data from an iterator to PMEM
* @addr: PMEM destination address
* @bytes: number of bytes to copy
* @i: iterator with source data
*
* Copy data from the iterator 'i' to the PMEM buffer starting at 'addr'.
* This function requires explicit ordering with an arch_wmb_pmem() call.
*/
static inline size_t arch_copy_from_iter_pmem(void __pmem *addr, size_t bytes,
struct iov_iter *i)
{
void *vaddr = (void __force *)addr;
size_t len;
/* TODO: skip the write-back by always using non-temporal stores */
len = copy_from_iter_nocache(vaddr, bytes, i);
if (__iter_needs_pmem_wb(i))
__arch_wb_cache_pmem(vaddr, bytes);
return len;
}
/**
* arch_clear_pmem - zero a PMEM memory range
* @addr: virtual start address
* @size: number of bytes to zero
*
* Write zeros into the memory range starting at 'addr' for 'size' bytes.
* This function requires explicit ordering with an arch_wmb_pmem() call.
*/
static inline void arch_clear_pmem(void __pmem *addr, size_t size)
{
void *vaddr = (void __force *)addr;
/* TODO: implement the zeroing via non-temporal writes */
if (size == PAGE_SIZE && ((unsigned long)vaddr & ~PAGE_MASK) == 0)
clear_page(vaddr);
else
memset(vaddr, 0, size);
__arch_wb_cache_pmem(vaddr, size);
}
static inline bool __arch_has_wmb_pmem(void)
{
/*
* We require that wmb() be an 'sfence', that is only guaranteed on
* 64-bit builds
*/
return static_cpu_has(X86_FEATURE_PCOMMIT);
}
#endif /* CONFIG_ARCH_HAS_PMEM_API */
#endif /* __ASM_X86_PMEM_H__ */