2 PAT (Page Attribute Table)
4 x86 Page Attribute Table (PAT) allows for setting the memory attribute at the
5 page level granularity. PAT is complementary to the MTRR settings which allows
6 for setting of memory types over physical address ranges. However, PAT is
7 more flexible than MTRR due to its capability to set attributes at page level
8 and also due to the fact that there are no hardware limitations on number of
9 such attribute settings allowed. Added flexibility comes with guidelines for
10 not having memory type aliasing for the same physical memory with multiple
13 PAT allows for different types of memory attributes. The most commonly used
14 ones that will be supported at this time are Write-back, Uncached,
15 Write-combined, Write-through and Uncached Minus.
21 There are many different APIs in the kernel that allows setting of memory
22 attributes at the page level. In order to avoid aliasing, these interfaces
23 should be used thoughtfully. Below is a table of interfaces available,
24 their intended usage and their memory attribute relationships. Internally,
25 these APIs use a reserve_memtype()/free_memtype() interface on the physical
26 address range to avoid any aliasing.
29 -------------------------------------------------------------------
30 API | RAM | ACPI,... | Reserved/Holes |
31 -----------------------|----------|------------|------------------|
33 ioremap | -- | UC- | UC- |
35 ioremap_cache | -- | WB | WB |
37 ioremap_uc | -- | UC | UC |
39 ioremap_nocache | -- | UC- | UC- |
41 ioremap_wc | -- | -- | WC |
43 ioremap_wt | -- | -- | WT |
45 set_memory_uc | UC- | -- | -- |
48 set_memory_wc | WC | -- | -- |
51 pci sysfs resource | -- | -- | UC- |
53 pci sysfs resource_wc | -- | -- | WC |
54 is IORESOURCE_PREFETCH| | | |
56 pci proc | -- | -- | UC- |
57 !PCIIOC_WRITE_COMBINE | | | |
59 pci proc | -- | -- | WC |
60 PCIIOC_WRITE_COMBINE | | | |
62 /dev/mem | -- | WB/WC/UC- | WB/WC/UC- |
65 /dev/mem | -- | UC- | UC- |
66 mmap SYNC flag | | | |
68 /dev/mem | -- | WB/WC/UC- | WB/WC/UC- |
69 mmap !SYNC flag | |(from exist-| (from exist- |
70 and | | ing alias)| ing alias) |
71 any alias to this area| | | |
73 /dev/mem | -- | WB | WB |
74 mmap !SYNC flag | | | |
75 no alias to this area | | | |
79 /dev/mem | -- | -- | UC- |
80 mmap !SYNC flag | | | |
81 no alias to this area | | | |
85 -------------------------------------------------------------------
87 Advanced APIs for drivers
88 -------------------------
89 A. Exporting pages to users with remap_pfn_range, io_remap_pfn_range,
92 Drivers wanting to export some pages to userspace do it by using mmap
93 interface and a combination of
95 2) io_remap_pfn_range() or remap_pfn_range() or vm_insert_pfn()
97 With PAT support, a new API pgprot_writecombine is being added. So, drivers can
98 continue to use the above sequence, with either pgprot_noncached() or
99 pgprot_writecombine() in step 1, followed by step 2.
101 In addition, step 2 internally tracks the region as UC or WC in memtype
102 list in order to ensure no conflicting mapping.
104 Note that this set of APIs only works with IO (non RAM) regions. If driver
105 wants to export a RAM region, it has to do set_memory_uc() or set_memory_wc()
106 as step 0 above and also track the usage of those pages and use set_memory_wb()
107 before the page is freed to free pool.
109 MTRR effects on PAT / non-PAT systems
110 -------------------------------------
112 The following table provides the effects of using write-combining MTRRs when
113 using ioremap*() calls on x86 for both non-PAT and PAT systems. Ideally
114 mtrr_add() usage will be phased out in favor of arch_phys_wc_add() which will
115 be a no-op on PAT enabled systems. The region over which a arch_phys_wc_add()
116 is made, should already have been ioremapped with WC attributes or PAT entries,
117 this can be done by using ioremap_wc() / set_memory_wc(). Devices which
118 combine areas of IO memory desired to remain uncacheable with areas where
119 write-combining is desirable should consider use of ioremap_uc() followed by
120 set_memory_wc() to white-list effective write-combined areas. Such use is
121 nevertheless discouraged as the effective memory type is considered
122 implementation defined, yet this strategy can be used as last resort on devices
123 with size-constrained regions where otherwise MTRR write-combining would
124 otherwise not be effective.
126 ----------------------------------------------------------------------
127 MTRR Non-PAT PAT Linux ioremap value Effective memory type
128 ----------------------------------------------------------------------
134 WC 000 WB _PAGE_CACHE_MODE_WB WC | WC
135 WC 001 WC _PAGE_CACHE_MODE_WC WC* | WC
136 WC 010 UC- _PAGE_CACHE_MODE_UC_MINUS WC* | UC
137 WC 011 UC _PAGE_CACHE_MODE_UC UC | UC
138 ----------------------------------------------------------------------
140 (*) denotes implementation defined and is discouraged
144 -- in the above table mean "Not suggested usage for the API". Some of the --'s
145 are strictly enforced by the kernel. Some others are not really enforced
146 today, but may be enforced in future.
148 For ioremap and pci access through /sys or /proc - The actual type returned
149 can be more restrictive, in case of any existing aliasing for that address.
150 For example: If there is an existing uncached mapping, a new ioremap_wc can
151 return uncached mapping in place of write-combine requested.
153 set_memory_[uc|wc] and set_memory_wb should be used in pairs, where driver will
154 first make a region uc or wc and switch it back to wb after use.
156 Over time writes to /proc/mtrr will be deprecated in favor of using PAT based
157 interfaces. Users writing to /proc/mtrr are suggested to use above interfaces.
159 Drivers should use ioremap_[uc|wc] to access PCI BARs with [uc|wc] access
162 Drivers should use set_memory_[uc|wc] to set access type for RAM ranges.
168 With CONFIG_DEBUG_FS enabled, PAT memtype list can be examined by
170 # mount -t debugfs debugfs /sys/kernel/debug
171 # cat /sys/kernel/debug/x86/pat_memtype_list
173 uncached-minus @ 0x7fadf000-0x7fae0000
174 uncached-minus @ 0x7fb19000-0x7fb1a000
175 uncached-minus @ 0x7fb1a000-0x7fb1b000
176 uncached-minus @ 0x7fb1b000-0x7fb1c000
177 uncached-minus @ 0x7fb1c000-0x7fb1d000
178 uncached-minus @ 0x7fb1d000-0x7fb1e000
179 uncached-minus @ 0x7fb1e000-0x7fb25000
180 uncached-minus @ 0x7fb25000-0x7fb26000
181 uncached-minus @ 0x7fb26000-0x7fb27000
182 uncached-minus @ 0x7fb27000-0x7fb28000
183 uncached-minus @ 0x7fb28000-0x7fb2e000
184 uncached-minus @ 0x7fb2e000-0x7fb2f000
185 uncached-minus @ 0x7fb2f000-0x7fb30000
186 uncached-minus @ 0x7fb31000-0x7fb32000
187 uncached-minus @ 0x80000000-0x90000000
189 This list shows physical address ranges and various PAT settings used to
190 access those physical address ranges.
192 Another, more verbose way of getting PAT related debug messages is with
193 "debugpat" boot parameter. With this parameter, various debug messages are
194 printed to dmesg log.