blob: 3d68ef6d2266cb66b3d07c578191b80c5348e0e2 [file] [log] [blame]
venkatesh.pallipadi@intel.com2e5d9c82008-03-18 17:00:14 -07001/*
2 * Handle caching attributes in page tables (PAT)
3 *
4 * Authors: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
5 * Suresh B Siddha <suresh.b.siddha@intel.com>
6 *
7 * Loosely based on earlier PAT patchset from Eric Biederman and Andi Kleen.
8 */
9
Ingo Molnarad2cde12008-09-30 13:20:45 +020010#include <linux/seq_file.h>
venkatesh.pallipadi@intel.come7f260a2008-03-18 17:00:21 -070011#include <linux/bootmem.h>
venkatesh.pallipadi@intel.comfec09622008-07-18 16:08:14 -070012#include <linux/debugfs.h>
Ingo Molnarad2cde12008-09-30 13:20:45 +020013#include <linux/kernel.h>
Ingo Molnar92b9af92009-02-28 14:09:27 +010014#include <linux/module.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090015#include <linux/slab.h>
Ingo Molnarad2cde12008-09-30 13:20:45 +020016#include <linux/mm.h>
17#include <linux/fs.h>
Venkatesh Pallipadi335ef892009-07-10 09:57:36 -070018#include <linux/rbtree.h>
venkatesh.pallipadi@intel.com2e5d9c82008-03-18 17:00:14 -070019
venkatesh.pallipadi@intel.com2e5d9c82008-03-18 17:00:14 -070020#include <asm/cacheflush.h>
Ingo Molnarad2cde12008-09-30 13:20:45 +020021#include <asm/processor.h>
22#include <asm/tlbflush.h>
Jack Steinerfd12a0d2009-11-19 14:23:41 -060023#include <asm/x86_init.h>
Ingo Molnarad2cde12008-09-30 13:20:45 +020024#include <asm/pgtable.h>
venkatesh.pallipadi@intel.com2e5d9c82008-03-18 17:00:14 -070025#include <asm/fcntl.h>
Ingo Molnarad2cde12008-09-30 13:20:45 +020026#include <asm/e820.h>
venkatesh.pallipadi@intel.com2e5d9c82008-03-18 17:00:14 -070027#include <asm/mtrr.h>
Ingo Molnarad2cde12008-09-30 13:20:45 +020028#include <asm/page.h>
29#include <asm/msr.h>
30#include <asm/pat.h>
venkatesh.pallipadi@intel.come7f260a2008-03-18 17:00:21 -070031#include <asm/io.h>
venkatesh.pallipadi@intel.com2e5d9c82008-03-18 17:00:14 -070032
venkatesh.pallipadi@intel.combe5a0c12010-02-10 11:57:06 -080033#include "pat_internal.h"
34
Thomas Gleixner8d4a4302008-05-08 09:18:43 +020035#ifdef CONFIG_X86_PAT
Andreas Herrmann499f8f82008-06-10 16:06:21 +020036int __read_mostly pat_enabled = 1;
venkatesh.pallipadi@intel.com2e5d9c82008-03-18 17:00:14 -070037
Marcin Slusarz1ee4bd92009-04-10 22:47:17 +020038static inline void pat_disable(const char *reason)
Thomas Gleixner8d4a4302008-05-08 09:18:43 +020039{
Andreas Herrmann499f8f82008-06-10 16:06:21 +020040 pat_enabled = 0;
Thomas Gleixner8d4a4302008-05-08 09:18:43 +020041 printk(KERN_INFO "%s\n", reason);
42}
venkatesh.pallipadi@intel.com2e5d9c82008-03-18 17:00:14 -070043
Andrew Mortonbe524fb2008-05-29 00:01:28 -070044static int __init nopat(char *str)
venkatesh.pallipadi@intel.com2e5d9c82008-03-18 17:00:14 -070045{
Thomas Gleixner8d4a4302008-05-08 09:18:43 +020046 pat_disable("PAT support disabled.");
venkatesh.pallipadi@intel.com2e5d9c82008-03-18 17:00:14 -070047 return 0;
48}
49early_param("nopat", nopat);
H. Peter Anvin75a04812009-01-22 16:17:05 -080050#else
51static inline void pat_disable(const char *reason)
52{
53 (void)reason;
54}
Thomas Gleixner8d4a4302008-05-08 09:18:43 +020055#endif
venkatesh.pallipadi@intel.com2e5d9c82008-03-18 17:00:14 -070056
Venki Pallipadi77b52b42008-05-05 19:09:10 -070057
venkatesh.pallipadi@intel.combe5a0c12010-02-10 11:57:06 -080058int pat_debug_enable;
Ingo Molnarad2cde12008-09-30 13:20:45 +020059
Venki Pallipadi77b52b42008-05-05 19:09:10 -070060static int __init pat_debug_setup(char *str)
61{
venkatesh.pallipadi@intel.combe5a0c12010-02-10 11:57:06 -080062 pat_debug_enable = 1;
Venki Pallipadi77b52b42008-05-05 19:09:10 -070063 return 0;
64}
65__setup("debugpat", pat_debug_setup);
66
Thomas Gleixner8d4a4302008-05-08 09:18:43 +020067static u64 __read_mostly boot_pat_state;
venkatesh.pallipadi@intel.com2e5d9c82008-03-18 17:00:14 -070068
69enum {
70 PAT_UC = 0, /* uncached */
71 PAT_WC = 1, /* Write combining */
72 PAT_WT = 4, /* Write Through */
73 PAT_WP = 5, /* Write Protected */
74 PAT_WB = 6, /* Write Back (default) */
75 PAT_UC_MINUS = 7, /* UC, but can be overriden by MTRR */
76};
77
Andreas Herrmanncd7a4e92008-06-10 16:05:39 +020078#define PAT(x, y) ((u64)PAT_ ## y << ((x)*8))
venkatesh.pallipadi@intel.com2e5d9c82008-03-18 17:00:14 -070079
80void pat_init(void)
81{
82 u64 pat;
Roland Dreiere23a8b62009-09-23 15:35:35 -070083 bool boot_cpu = !boot_pat_state;
venkatesh.pallipadi@intel.com2e5d9c82008-03-18 17:00:14 -070084
Andreas Herrmann499f8f82008-06-10 16:06:21 +020085 if (!pat_enabled)
venkatesh.pallipadi@intel.com2e5d9c82008-03-18 17:00:14 -070086 return;
87
H. Peter Anvin75a04812009-01-22 16:17:05 -080088 if (!cpu_has_pat) {
89 if (!boot_pat_state) {
90 pat_disable("PAT not supported by CPU.");
91 return;
92 } else {
93 /*
94 * If this happens we are on a secondary CPU, but
95 * switched to PAT on the boot CPU. We have no way to
96 * undo PAT.
97 */
98 printk(KERN_ERR "PAT enabled, "
99 "but not supported by secondary CPU\n");
100 BUG();
101 }
Thomas Gleixner8d4a4302008-05-08 09:18:43 +0200102 }
venkatesh.pallipadi@intel.com2e5d9c82008-03-18 17:00:14 -0700103
104 /* Set PWT to Write-Combining. All other bits stay the same */
105 /*
106 * PTE encoding used in Linux:
107 * PAT
108 * |PCD
109 * ||PWT
110 * |||
111 * 000 WB _PAGE_CACHE_WB
112 * 001 WC _PAGE_CACHE_WC
113 * 010 UC- _PAGE_CACHE_UC_MINUS
114 * 011 UC _PAGE_CACHE_UC
115 * PAT bit unused
116 */
Andreas Herrmanncd7a4e92008-06-10 16:05:39 +0200117 pat = PAT(0, WB) | PAT(1, WC) | PAT(2, UC_MINUS) | PAT(3, UC) |
118 PAT(4, WB) | PAT(5, WC) | PAT(6, UC_MINUS) | PAT(7, UC);
venkatesh.pallipadi@intel.com2e5d9c82008-03-18 17:00:14 -0700119
120 /* Boot CPU check */
Thomas Gleixner8d4a4302008-05-08 09:18:43 +0200121 if (!boot_pat_state)
venkatesh.pallipadi@intel.com2e5d9c82008-03-18 17:00:14 -0700122 rdmsrl(MSR_IA32_CR_PAT, boot_pat_state);
venkatesh.pallipadi@intel.com2e5d9c82008-03-18 17:00:14 -0700123
124 wrmsrl(MSR_IA32_CR_PAT, pat);
Roland Dreiere23a8b62009-09-23 15:35:35 -0700125
126 if (boot_cpu)
127 printk(KERN_INFO "x86 PAT enabled: cpu %d, old 0x%Lx, new 0x%Lx\n",
128 smp_processor_id(), boot_pat_state, pat);
venkatesh.pallipadi@intel.com2e5d9c82008-03-18 17:00:14 -0700129}
130
131#undef PAT
132
Pallipadi, Venkatesh9e41a492010-02-10 15:26:07 -0800133static DEFINE_SPINLOCK(memtype_lock); /* protects memtype accesses */
Venkatesh Pallipadi335ef892009-07-10 09:57:36 -0700134
venkatesh.pallipadi@intel.com2e5d9c82008-03-18 17:00:14 -0700135/*
136 * Does intersection of PAT memory type and MTRR memory type and returns
137 * the resulting memory type as PAT understands it.
138 * (Type in pat and mtrr will not have same value)
139 * The intersection is based on "Effective Memory Type" tables in IA-32
140 * SDM vol 3a
141 */
Hugh Dickins6cf514f2008-06-16 18:42:43 +0100142static unsigned long pat_x_mtrr_type(u64 start, u64 end, unsigned long req_type)
venkatesh.pallipadi@intel.com2e5d9c82008-03-18 17:00:14 -0700143{
Venki Pallipadic26421d2008-05-29 12:01:44 -0700144 /*
145 * Look for MTRR hint to get the effective type in case where PAT
146 * request is for WB.
147 */
Andreas Herrmanndd0c7c42008-06-18 15:38:57 +0200148 if (req_type == _PAGE_CACHE_WB) {
149 u8 mtrr_type;
150
151 mtrr_type = mtrr_type_lookup(start, end);
Suresh Siddhab6ff32d2009-04-09 14:26:51 -0700152 if (mtrr_type != MTRR_TYPE_WRBACK)
153 return _PAGE_CACHE_UC_MINUS;
154
155 return _PAGE_CACHE_WB;
Andreas Herrmanndd0c7c42008-06-18 15:38:57 +0200156 }
157
158 return req_type;
venkatesh.pallipadi@intel.com2e5d9c82008-03-18 17:00:14 -0700159}
160
John Dykstrafa835232012-05-25 16:12:46 -0500161struct pagerange_state {
162 unsigned long cur_pfn;
163 int ram;
164 int not_ram;
165};
166
167static int
168pagerange_is_ram_callback(unsigned long initial_pfn, unsigned long total_nr_pages, void *arg)
169{
170 struct pagerange_state *state = arg;
171
172 state->not_ram |= initial_pfn > state->cur_pfn;
173 state->ram |= total_nr_pages > 0;
174 state->cur_pfn = initial_pfn + total_nr_pages;
175
176 return state->ram && state->not_ram;
177}
178
Yasuaki Ishimatsu3709c852010-07-22 14:57:35 +0900179static int pat_pagerange_is_ram(resource_size_t start, resource_size_t end)
Suresh Siddhabe03d9e2009-02-11 11:20:23 -0800180{
John Dykstrafa835232012-05-25 16:12:46 -0500181 int ret = 0;
182 unsigned long start_pfn = start >> PAGE_SHIFT;
183 unsigned long end_pfn = (end + PAGE_SIZE - 1) >> PAGE_SHIFT;
184 struct pagerange_state state = {start_pfn, 0, 0};
Suresh Siddhabe03d9e2009-02-11 11:20:23 -0800185
John Dykstrafa835232012-05-25 16:12:46 -0500186 /*
187 * For legacy reasons, physical address range in the legacy ISA
188 * region is tracked as non-RAM. This will allow users of
189 * /dev/mem to map portions of legacy ISA region, even when
190 * some of those portions are listed(or not even listed) with
191 * different e820 types(RAM/reserved/..)
192 */
193 if (start_pfn < ISA_END_ADDRESS >> PAGE_SHIFT)
194 start_pfn = ISA_END_ADDRESS >> PAGE_SHIFT;
Suresh Siddhabe03d9e2009-02-11 11:20:23 -0800195
John Dykstrafa835232012-05-25 16:12:46 -0500196 if (start_pfn < end_pfn) {
197 ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn,
198 &state, pagerange_is_ram_callback);
Suresh Siddhabe03d9e2009-02-11 11:20:23 -0800199 }
200
John Dykstrafa835232012-05-25 16:12:46 -0500201 return (ret > 0) ? -1 : (state.ram ? 1 : 0);
Suresh Siddhabe03d9e2009-02-11 11:20:23 -0800202}
203
venkatesh.pallipadi@intel.come7f260a2008-03-18 17:00:21 -0700204/*
Venkatesh Pallipadif5841742009-07-10 09:57:38 -0700205 * For RAM pages, we use page flags to mark the pages with appropriate type.
206 * Here we do two pass:
207 * - Find the memtype of all the pages in the range, look for any conflicts
208 * - In case of no conflicts, set the new memtype for pages in the range
Suresh Siddha9542ada2008-09-24 08:53:33 -0700209 */
210static int reserve_ram_pages_type(u64 start, u64 end, unsigned long req_type,
Ingo Molnarad2cde12008-09-30 13:20:45 +0200211 unsigned long *new_type)
Suresh Siddha9542ada2008-09-24 08:53:33 -0700212{
213 struct page *page;
Venkatesh Pallipadif5841742009-07-10 09:57:38 -0700214 u64 pfn;
215
216 if (req_type == _PAGE_CACHE_UC) {
217 /* We do not support strong UC */
218 WARN_ON_ONCE(1);
219 req_type = _PAGE_CACHE_UC_MINUS;
220 }
221
222 for (pfn = (start >> PAGE_SHIFT); pfn < (end >> PAGE_SHIFT); ++pfn) {
223 unsigned long type;
224
225 page = pfn_to_page(pfn);
226 type = get_page_memtype(page);
227 if (type != -1) {
Bjorn Helgaas365811d2012-05-29 15:06:29 -0700228 printk(KERN_INFO "reserve_ram_pages_type failed [mem %#010Lx-%#010Lx], track 0x%lx, req 0x%lx\n",
229 start, end - 1, type, req_type);
Venkatesh Pallipadif5841742009-07-10 09:57:38 -0700230 if (new_type)
231 *new_type = type;
232
233 return -EBUSY;
234 }
235 }
236
237 if (new_type)
238 *new_type = req_type;
Suresh Siddha9542ada2008-09-24 08:53:33 -0700239
240 for (pfn = (start >> PAGE_SHIFT); pfn < (end >> PAGE_SHIFT); ++pfn) {
241 page = pfn_to_page(pfn);
Venkatesh Pallipadif5841742009-07-10 09:57:38 -0700242 set_page_memtype(page, req_type);
Suresh Siddha9542ada2008-09-24 08:53:33 -0700243 }
244 return 0;
Suresh Siddha9542ada2008-09-24 08:53:33 -0700245}
246
247static int free_ram_pages_type(u64 start, u64 end)
248{
249 struct page *page;
Venkatesh Pallipadif5841742009-07-10 09:57:38 -0700250 u64 pfn;
Suresh Siddha9542ada2008-09-24 08:53:33 -0700251
252 for (pfn = (start >> PAGE_SHIFT); pfn < (end >> PAGE_SHIFT); ++pfn) {
253 page = pfn_to_page(pfn);
Venkatesh Pallipadif5841742009-07-10 09:57:38 -0700254 set_page_memtype(page, -1);
Suresh Siddha9542ada2008-09-24 08:53:33 -0700255 }
256 return 0;
Suresh Siddha9542ada2008-09-24 08:53:33 -0700257}
258
259/*
venkatesh.pallipadi@intel.come7f260a2008-03-18 17:00:21 -0700260 * req_type typically has one of the:
261 * - _PAGE_CACHE_WB
262 * - _PAGE_CACHE_WC
263 * - _PAGE_CACHE_UC_MINUS
264 * - _PAGE_CACHE_UC
265 *
Andreas Herrmannac979912008-06-20 22:01:49 +0200266 * If new_type is NULL, function will return an error if it cannot reserve the
267 * region with req_type. If new_type is non-NULL, function will return
268 * available type in new_type in case of no error. In case of any error
venkatesh.pallipadi@intel.come7f260a2008-03-18 17:00:21 -0700269 * it will return a negative return value.
270 */
venkatesh.pallipadi@intel.com2e5d9c82008-03-18 17:00:14 -0700271int reserve_memtype(u64 start, u64 end, unsigned long req_type,
Ingo Molnarad2cde12008-09-30 13:20:45 +0200272 unsigned long *new_type)
venkatesh.pallipadi@intel.com2e5d9c82008-03-18 17:00:14 -0700273{
venkatesh.pallipadi@intel.combe5a0c12010-02-10 11:57:06 -0800274 struct memtype *new;
venkatesh.pallipadi@intel.com2e5d9c82008-03-18 17:00:14 -0700275 unsigned long actual_type;
Suresh Siddha9542ada2008-09-24 08:53:33 -0700276 int is_range_ram;
Ingo Molnarad2cde12008-09-30 13:20:45 +0200277 int err = 0;
venkatesh.pallipadi@intel.com2e5d9c82008-03-18 17:00:14 -0700278
Ingo Molnarad2cde12008-09-30 13:20:45 +0200279 BUG_ON(start >= end); /* end is exclusive */
Andreas Herrmann69e26be2008-06-20 22:03:06 +0200280
Andreas Herrmann499f8f82008-06-10 16:06:21 +0200281 if (!pat_enabled) {
venkatesh.pallipadi@intel.come7f260a2008-03-18 17:00:21 -0700282 /* This is identical to page table setting without PAT */
Andreas Herrmannac979912008-06-20 22:01:49 +0200283 if (new_type) {
Xiaotian Feng83ea05e2009-11-10 17:23:07 +0800284 if (req_type == _PAGE_CACHE_WC)
Venkatesh Pallipadi5fc51742009-07-10 09:57:32 -0700285 *new_type = _PAGE_CACHE_UC_MINUS;
Andreas Herrmannac979912008-06-20 22:01:49 +0200286 else
287 *new_type = req_type & _PAGE_CACHE_MASK;
venkatesh.pallipadi@intel.come7f260a2008-03-18 17:00:21 -0700288 }
venkatesh.pallipadi@intel.com2e5d9c82008-03-18 17:00:14 -0700289 return 0;
290 }
291
292 /* Low ISA region is always mapped WB in page table. No need to track */
H. Peter Anvin8a271382009-11-23 14:49:20 -0800293 if (x86_platform.is_untracked_pat_range(start, end)) {
Andreas Herrmannac979912008-06-20 22:01:49 +0200294 if (new_type)
295 *new_type = _PAGE_CACHE_WB;
venkatesh.pallipadi@intel.com2e5d9c82008-03-18 17:00:14 -0700296 return 0;
297 }
298
Suresh Siddhab6ff32d2009-04-09 14:26:51 -0700299 /*
300 * Call mtrr_lookup to get the type hint. This is an
301 * optimization for /dev/mem mmap'ers into WB memory (BIOS
302 * tools and ACPI tools). Use WB request for WB memory and use
303 * UC_MINUS otherwise.
304 */
305 actual_type = pat_x_mtrr_type(start, end, req_type & _PAGE_CACHE_MASK);
venkatesh.pallipadi@intel.com2e5d9c82008-03-18 17:00:14 -0700306
Suresh Siddha95971342009-01-13 10:21:30 -0800307 if (new_type)
308 *new_type = actual_type;
309
Suresh Siddhabe03d9e2009-02-11 11:20:23 -0800310 is_range_ram = pat_pagerange_is_ram(start, end);
Venkatesh Pallipadif5841742009-07-10 09:57:38 -0700311 if (is_range_ram == 1) {
312
Venkatesh Pallipadif5841742009-07-10 09:57:38 -0700313 err = reserve_ram_pages_type(start, end, req_type, new_type);
Venkatesh Pallipadif5841742009-07-10 09:57:38 -0700314
315 return err;
316 } else if (is_range_ram < 0) {
Suresh Siddha9542ada2008-09-24 08:53:33 -0700317 return -EINVAL;
Venkatesh Pallipadif5841742009-07-10 09:57:38 -0700318 }
Suresh Siddha9542ada2008-09-24 08:53:33 -0700319
Venkatesh Pallipadi6a4f3b52010-06-10 17:45:01 -0700320 new = kzalloc(sizeof(struct memtype), GFP_KERNEL);
Andreas Herrmannac979912008-06-20 22:01:49 +0200321 if (!new)
venkatesh.pallipadi@intel.com2e5d9c82008-03-18 17:00:14 -0700322 return -ENOMEM;
323
Ingo Molnarad2cde12008-09-30 13:20:45 +0200324 new->start = start;
325 new->end = end;
326 new->type = actual_type;
venkatesh.pallipadi@intel.com2e5d9c82008-03-18 17:00:14 -0700327
venkatesh.pallipadi@intel.com2e5d9c82008-03-18 17:00:14 -0700328 spin_lock(&memtype_lock);
329
Pallipadi, Venkatesh9e41a492010-02-10 15:26:07 -0800330 err = rbt_memtype_check_insert(new, new_type);
venkatesh.pallipadi@intel.com2e5d9c82008-03-18 17:00:14 -0700331 if (err) {
Bjorn Helgaas365811d2012-05-29 15:06:29 -0700332 printk(KERN_INFO "reserve_memtype failed [mem %#010Lx-%#010Lx], track %s, req %s\n",
333 start, end - 1,
334 cattr_name(new->type), cattr_name(req_type));
Andreas Herrmannac979912008-06-20 22:01:49 +0200335 kfree(new);
venkatesh.pallipadi@intel.com2e5d9c82008-03-18 17:00:14 -0700336 spin_unlock(&memtype_lock);
Ingo Molnarad2cde12008-09-30 13:20:45 +0200337
venkatesh.pallipadi@intel.com2e5d9c82008-03-18 17:00:14 -0700338 return err;
339 }
340
venkatesh.pallipadi@intel.com2e5d9c82008-03-18 17:00:14 -0700341 spin_unlock(&memtype_lock);
Andreas Herrmann3e9c83b2008-06-20 22:04:02 +0200342
Bjorn Helgaas365811d2012-05-29 15:06:29 -0700343 dprintk("reserve_memtype added [mem %#010Lx-%#010Lx], track %s, req %s, ret %s\n",
344 start, end - 1, cattr_name(new->type), cattr_name(req_type),
Andreas Herrmann3e9c83b2008-06-20 22:04:02 +0200345 new_type ? cattr_name(*new_type) : "-");
346
venkatesh.pallipadi@intel.com2e5d9c82008-03-18 17:00:14 -0700347 return err;
348}
349
350int free_memtype(u64 start, u64 end)
351{
venkatesh.pallipadi@intel.com2e5d9c82008-03-18 17:00:14 -0700352 int err = -EINVAL;
Suresh Siddha9542ada2008-09-24 08:53:33 -0700353 int is_range_ram;
Xiaotian Feng20413f22010-05-26 09:51:10 +0800354 struct memtype *entry;
venkatesh.pallipadi@intel.com2e5d9c82008-03-18 17:00:14 -0700355
Andreas Herrmann69e26be2008-06-20 22:03:06 +0200356 if (!pat_enabled)
venkatesh.pallipadi@intel.com2e5d9c82008-03-18 17:00:14 -0700357 return 0;
venkatesh.pallipadi@intel.com2e5d9c82008-03-18 17:00:14 -0700358
359 /* Low ISA region is always mapped WB. No need to track */
H. Peter Anvin8a271382009-11-23 14:49:20 -0800360 if (x86_platform.is_untracked_pat_range(start, end))
venkatesh.pallipadi@intel.com2e5d9c82008-03-18 17:00:14 -0700361 return 0;
venkatesh.pallipadi@intel.com2e5d9c82008-03-18 17:00:14 -0700362
Suresh Siddhabe03d9e2009-02-11 11:20:23 -0800363 is_range_ram = pat_pagerange_is_ram(start, end);
Venkatesh Pallipadif5841742009-07-10 09:57:38 -0700364 if (is_range_ram == 1) {
365
Venkatesh Pallipadif5841742009-07-10 09:57:38 -0700366 err = free_ram_pages_type(start, end);
Venkatesh Pallipadif5841742009-07-10 09:57:38 -0700367
368 return err;
369 } else if (is_range_ram < 0) {
Suresh Siddha9542ada2008-09-24 08:53:33 -0700370 return -EINVAL;
Venkatesh Pallipadif5841742009-07-10 09:57:38 -0700371 }
Suresh Siddha9542ada2008-09-24 08:53:33 -0700372
venkatesh.pallipadi@intel.com2e5d9c82008-03-18 17:00:14 -0700373 spin_lock(&memtype_lock);
Xiaotian Feng20413f22010-05-26 09:51:10 +0800374 entry = rbt_memtype_erase(start, end);
venkatesh.pallipadi@intel.com2e5d9c82008-03-18 17:00:14 -0700375 spin_unlock(&memtype_lock);
376
Xiaotian Feng20413f22010-05-26 09:51:10 +0800377 if (!entry) {
Bjorn Helgaas365811d2012-05-29 15:06:29 -0700378 printk(KERN_INFO "%s:%d freeing invalid memtype [mem %#010Lx-%#010Lx]\n",
379 current->comm, current->pid, start, end - 1);
Xiaotian Feng20413f22010-05-26 09:51:10 +0800380 return -EINVAL;
venkatesh.pallipadi@intel.com2e5d9c82008-03-18 17:00:14 -0700381 }
venkatesh.pallipadi@intel.com6997ab42008-03-18 17:00:25 -0700382
Xiaotian Feng20413f22010-05-26 09:51:10 +0800383 kfree(entry);
384
Bjorn Helgaas365811d2012-05-29 15:06:29 -0700385 dprintk("free_memtype request [mem %#010Lx-%#010Lx]\n", start, end - 1);
Ingo Molnarad2cde12008-09-30 13:20:45 +0200386
Xiaotian Feng20413f22010-05-26 09:51:10 +0800387 return 0;
venkatesh.pallipadi@intel.com2e5d9c82008-03-18 17:00:14 -0700388}
389
venkatesh.pallipadi@intel.comf0970c12008-03-18 17:00:20 -0700390
Venkatesh Pallipadi9fd126bc2009-07-10 09:57:34 -0700391/**
Venkatesh Pallipadi637b86e2009-07-10 09:57:39 -0700392 * lookup_memtype - Looksup the memory type for a physical address
393 * @paddr: physical address of which memory type needs to be looked up
394 *
395 * Only to be called when PAT is enabled
396 *
397 * Returns _PAGE_CACHE_WB, _PAGE_CACHE_WC, _PAGE_CACHE_UC_MINUS or
398 * _PAGE_CACHE_UC
399 */
400static unsigned long lookup_memtype(u64 paddr)
401{
402 int rettype = _PAGE_CACHE_WB;
403 struct memtype *entry;
404
H. Peter Anvin8a271382009-11-23 14:49:20 -0800405 if (x86_platform.is_untracked_pat_range(paddr, paddr + PAGE_SIZE))
Venkatesh Pallipadi637b86e2009-07-10 09:57:39 -0700406 return rettype;
407
408 if (pat_pagerange_is_ram(paddr, paddr + PAGE_SIZE)) {
409 struct page *page;
Venkatesh Pallipadi637b86e2009-07-10 09:57:39 -0700410 page = pfn_to_page(paddr >> PAGE_SHIFT);
411 rettype = get_page_memtype(page);
Venkatesh Pallipadi637b86e2009-07-10 09:57:39 -0700412 /*
413 * -1 from get_page_memtype() implies RAM page is in its
414 * default state and not reserved, and hence of type WB
415 */
416 if (rettype == -1)
417 rettype = _PAGE_CACHE_WB;
418
419 return rettype;
420 }
421
422 spin_lock(&memtype_lock);
423
Pallipadi, Venkatesh9e41a492010-02-10 15:26:07 -0800424 entry = rbt_memtype_lookup(paddr);
Venkatesh Pallipadi637b86e2009-07-10 09:57:39 -0700425 if (entry != NULL)
426 rettype = entry->type;
427 else
428 rettype = _PAGE_CACHE_UC_MINUS;
429
430 spin_unlock(&memtype_lock);
431 return rettype;
432}
433
434/**
Venkatesh Pallipadi9fd126bc2009-07-10 09:57:34 -0700435 * io_reserve_memtype - Request a memory type mapping for a region of memory
436 * @start: start (physical address) of the region
437 * @end: end (physical address) of the region
438 * @type: A pointer to memtype, with requested type. On success, requested
439 * or any other compatible type that was available for the region is returned
440 *
441 * On success, returns 0
442 * On failure, returns non-zero
443 */
444int io_reserve_memtype(resource_size_t start, resource_size_t end,
445 unsigned long *type)
446{
H. Peter Anvinb8551922009-08-26 17:17:51 -0700447 resource_size_t size = end - start;
Venkatesh Pallipadi9fd126bc2009-07-10 09:57:34 -0700448 unsigned long req_type = *type;
449 unsigned long new_type;
450 int ret;
451
H. Peter Anvinb8551922009-08-26 17:17:51 -0700452 WARN_ON_ONCE(iomem_map_sanity_check(start, size));
Venkatesh Pallipadi9fd126bc2009-07-10 09:57:34 -0700453
454 ret = reserve_memtype(start, end, req_type, &new_type);
455 if (ret)
456 goto out_err;
457
H. Peter Anvinb8551922009-08-26 17:17:51 -0700458 if (!is_new_memtype_allowed(start, size, req_type, new_type))
Venkatesh Pallipadi9fd126bc2009-07-10 09:57:34 -0700459 goto out_free;
460
H. Peter Anvinb8551922009-08-26 17:17:51 -0700461 if (kernel_map_sync_memtype(start, size, new_type) < 0)
Venkatesh Pallipadi9fd126bc2009-07-10 09:57:34 -0700462 goto out_free;
463
464 *type = new_type;
465 return 0;
466
467out_free:
468 free_memtype(start, end);
469 ret = -EBUSY;
470out_err:
471 return ret;
472}
473
474/**
475 * io_free_memtype - Release a memory type mapping for a region of memory
476 * @start: start (physical address) of the region
477 * @end: end (physical address) of the region
478 */
479void io_free_memtype(resource_size_t start, resource_size_t end)
480{
481 free_memtype(start, end);
482}
483
venkatesh.pallipadi@intel.comf0970c12008-03-18 17:00:20 -0700484pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
485 unsigned long size, pgprot_t vma_prot)
486{
487 return vma_prot;
488}
489
Ingo Molnard0926332008-07-18 00:26:59 +0200490#ifdef CONFIG_STRICT_DEVMEM
491/* This check is done in drivers/char/mem.c in case of STRICT_DEVMEM*/
Venki Pallipadi0124cec2008-04-26 11:32:12 -0700492static inline int range_is_allowed(unsigned long pfn, unsigned long size)
493{
494 return 1;
495}
496#else
Ravikiran G Thirumalai9e41bff2008-10-30 13:59:21 -0700497/* This check is needed to avoid cache aliasing when PAT is enabled */
Venki Pallipadi0124cec2008-04-26 11:32:12 -0700498static inline int range_is_allowed(unsigned long pfn, unsigned long size)
499{
500 u64 from = ((u64)pfn) << PAGE_SHIFT;
501 u64 to = from + size;
502 u64 cursor = from;
503
Ravikiran G Thirumalai9e41bff2008-10-30 13:59:21 -0700504 if (!pat_enabled)
505 return 1;
506
Venki Pallipadi0124cec2008-04-26 11:32:12 -0700507 while (cursor < to) {
508 if (!devmem_is_allowed(pfn)) {
Bjorn Helgaas365811d2012-05-29 15:06:29 -0700509 printk(KERN_INFO "Program %s tried to access /dev/mem between [mem %#010Lx-%#010Lx]\n",
510 current->comm, from, to - 1);
Venki Pallipadi0124cec2008-04-26 11:32:12 -0700511 return 0;
512 }
513 cursor += PAGE_SIZE;
514 pfn++;
515 }
516 return 1;
517}
Ingo Molnard0926332008-07-18 00:26:59 +0200518#endif /* CONFIG_STRICT_DEVMEM */
Venki Pallipadi0124cec2008-04-26 11:32:12 -0700519
venkatesh.pallipadi@intel.comf0970c12008-03-18 17:00:20 -0700520int phys_mem_access_prot_allowed(struct file *file, unsigned long pfn,
521 unsigned long size, pgprot_t *vma_prot)
522{
Suresh Siddha0c3c8a12009-04-09 14:26:52 -0700523 unsigned long flags = _PAGE_CACHE_WB;
venkatesh.pallipadi@intel.comf0970c12008-03-18 17:00:20 -0700524
Venki Pallipadi0124cec2008-04-26 11:32:12 -0700525 if (!range_is_allowed(pfn, size))
526 return 0;
527
Christoph Hellwig6b2f3d12009-10-27 11:05:28 +0100528 if (file->f_flags & O_DSYNC)
venkatesh.pallipadi@intel.com28df82e2008-08-20 16:45:52 -0700529 flags = _PAGE_CACHE_UC_MINUS;
venkatesh.pallipadi@intel.comf0970c12008-03-18 17:00:20 -0700530
531#ifdef CONFIG_X86_32
532 /*
533 * On the PPro and successors, the MTRRs are used to set
534 * memory types for physical addresses outside main memory,
535 * so blindly setting UC or PWT on those pages is wrong.
536 * For Pentiums and earlier, the surround logic should disable
537 * caching for the high addresses through the KEN pin, but
538 * we maintain the tradition of paranoia in this code.
539 */
Andreas Herrmann499f8f82008-06-10 16:06:21 +0200540 if (!pat_enabled &&
Andreas Herrmanncd7a4e92008-06-10 16:05:39 +0200541 !(boot_cpu_has(X86_FEATURE_MTRR) ||
542 boot_cpu_has(X86_FEATURE_K6_MTRR) ||
543 boot_cpu_has(X86_FEATURE_CYRIX_ARR) ||
544 boot_cpu_has(X86_FEATURE_CENTAUR_MCR)) &&
545 (pfn << PAGE_SHIFT) >= __pa(high_memory)) {
venkatesh.pallipadi@intel.come7f260a2008-03-18 17:00:21 -0700546 flags = _PAGE_CACHE_UC;
venkatesh.pallipadi@intel.comf0970c12008-03-18 17:00:20 -0700547 }
548#endif
549
venkatesh.pallipadi@intel.come7f260a2008-03-18 17:00:21 -0700550 *vma_prot = __pgprot((pgprot_val(*vma_prot) & ~_PAGE_CACHE_MASK) |
551 flags);
venkatesh.pallipadi@intel.comf0970c12008-03-18 17:00:20 -0700552 return 1;
553}
venkatesh.pallipadi@intel.come7f260a2008-03-18 17:00:21 -0700554
venkatesh.pallipadi@intel.com58993292008-12-18 11:41:30 -0800555/*
Venkatesh Pallipadi7880f742009-02-24 17:35:13 -0800556 * Change the memory type for the physial address range in kernel identity
557 * mapping space if that range is a part of identity map.
558 */
559int kernel_map_sync_memtype(u64 base, unsigned long size, unsigned long flags)
560{
561 unsigned long id_sz;
562
Venkatesh Pallipadi5fc51742009-07-10 09:57:32 -0700563 if (base >= __pa(high_memory))
Venkatesh Pallipadi7880f742009-02-24 17:35:13 -0800564 return 0;
565
566 id_sz = (__pa(high_memory) < base + size) ?
567 __pa(high_memory) - base :
568 size;
569
570 if (ioremap_change_attr((unsigned long)__va(base), id_sz, flags) < 0) {
Bjorn Helgaas365811d2012-05-29 15:06:29 -0700571 printk(KERN_INFO "%s:%d ioremap_change_attr failed %s "
572 "for [mem %#010Lx-%#010Lx]\n",
Venkatesh Pallipadi7880f742009-02-24 17:35:13 -0800573 current->comm, current->pid,
574 cattr_name(flags),
Bjorn Helgaas365811d2012-05-29 15:06:29 -0700575 base, (unsigned long long)(base + size-1));
Venkatesh Pallipadi7880f742009-02-24 17:35:13 -0800576 return -EINVAL;
577 }
578 return 0;
579}
580
581/*
venkatesh.pallipadi@intel.com58993292008-12-18 11:41:30 -0800582 * Internal interface to reserve a range of physical memory with prot.
583 * Reserved non RAM regions only and after successful reserve_memtype,
584 * this func also keeps identity mapping (if any) in sync with this new prot.
585 */
venkatesh.pallipadi@intel.comcdecff62009-01-09 16:13:12 -0800586static int reserve_pfn_range(u64 paddr, unsigned long size, pgprot_t *vma_prot,
587 int strict_prot)
venkatesh.pallipadi@intel.com58993292008-12-18 11:41:30 -0800588{
589 int is_ram = 0;
Venkatesh Pallipadi7880f742009-02-24 17:35:13 -0800590 int ret;
venkatesh.pallipadi@intel.comcdecff62009-01-09 16:13:12 -0800591 unsigned long want_flags = (pgprot_val(*vma_prot) & _PAGE_CACHE_MASK);
Suresh Siddha0c3c8a12009-04-09 14:26:52 -0700592 unsigned long flags = want_flags;
venkatesh.pallipadi@intel.com58993292008-12-18 11:41:30 -0800593
Suresh Siddhabe03d9e2009-02-11 11:20:23 -0800594 is_ram = pat_pagerange_is_ram(paddr, paddr + size);
venkatesh.pallipadi@intel.com58993292008-12-18 11:41:30 -0800595
Suresh Siddhabe03d9e2009-02-11 11:20:23 -0800596 /*
Venkatesh Pallipadid886c732009-07-10 09:57:41 -0700597 * reserve_pfn_range() for RAM pages. We do not refcount to keep
598 * track of number of mappings of RAM pages. We can assert that
599 * the type requested matches the type of first page in the range.
Suresh Siddhabe03d9e2009-02-11 11:20:23 -0800600 */
Venkatesh Pallipadid886c732009-07-10 09:57:41 -0700601 if (is_ram) {
602 if (!pat_enabled)
603 return 0;
604
605 flags = lookup_memtype(paddr);
606 if (want_flags != flags) {
Bjorn Helgaas365811d2012-05-29 15:06:29 -0700607 printk(KERN_WARNING "%s:%d map pfn RAM range req %s for [mem %#010Lx-%#010Lx], got %s\n",
Venkatesh Pallipadid886c732009-07-10 09:57:41 -0700608 current->comm, current->pid,
609 cattr_name(want_flags),
610 (unsigned long long)paddr,
Bjorn Helgaas365811d2012-05-29 15:06:29 -0700611 (unsigned long long)(paddr + size - 1),
Venkatesh Pallipadid886c732009-07-10 09:57:41 -0700612 cattr_name(flags));
613 *vma_prot = __pgprot((pgprot_val(*vma_prot) &
614 (~_PAGE_CACHE_MASK)) |
615 flags);
616 }
Pallipadi, Venkatesh4bb9c5c2009-03-12 17:45:27 -0700617 return 0;
Venkatesh Pallipadid886c732009-07-10 09:57:41 -0700618 }
venkatesh.pallipadi@intel.com58993292008-12-18 11:41:30 -0800619
620 ret = reserve_memtype(paddr, paddr + size, want_flags, &flags);
621 if (ret)
622 return ret;
623
624 if (flags != want_flags) {
Suresh Siddha1adcaaf2009-08-17 13:23:50 -0700625 if (strict_prot ||
626 !is_new_memtype_allowed(paddr, size, want_flags, flags)) {
venkatesh.pallipadi@intel.comcdecff62009-01-09 16:13:12 -0800627 free_memtype(paddr, paddr + size);
628 printk(KERN_ERR "%s:%d map pfn expected mapping type %s"
Bjorn Helgaas365811d2012-05-29 15:06:29 -0700629 " for [mem %#010Lx-%#010Lx], got %s\n",
venkatesh.pallipadi@intel.comcdecff62009-01-09 16:13:12 -0800630 current->comm, current->pid,
631 cattr_name(want_flags),
632 (unsigned long long)paddr,
Bjorn Helgaas365811d2012-05-29 15:06:29 -0700633 (unsigned long long)(paddr + size - 1),
venkatesh.pallipadi@intel.comcdecff62009-01-09 16:13:12 -0800634 cattr_name(flags));
635 return -EINVAL;
636 }
637 /*
638 * We allow returning different type than the one requested in
639 * non strict case.
640 */
641 *vma_prot = __pgprot((pgprot_val(*vma_prot) &
642 (~_PAGE_CACHE_MASK)) |
643 flags);
venkatesh.pallipadi@intel.com58993292008-12-18 11:41:30 -0800644 }
645
Venkatesh Pallipadi7880f742009-02-24 17:35:13 -0800646 if (kernel_map_sync_memtype(paddr, size, flags) < 0) {
venkatesh.pallipadi@intel.com58993292008-12-18 11:41:30 -0800647 free_memtype(paddr, paddr + size);
venkatesh.pallipadi@intel.com58993292008-12-18 11:41:30 -0800648 return -EINVAL;
649 }
650 return 0;
651}
652
653/*
654 * Internal interface to free a range of physical memory.
655 * Frees non RAM regions only.
656 */
657static void free_pfn_range(u64 paddr, unsigned long size)
658{
659 int is_ram;
660
Suresh Siddhabe03d9e2009-02-11 11:20:23 -0800661 is_ram = pat_pagerange_is_ram(paddr, paddr + size);
venkatesh.pallipadi@intel.com58993292008-12-18 11:41:30 -0800662 if (is_ram == 0)
663 free_memtype(paddr, paddr + size);
664}
665
666/*
667 * track_pfn_vma_copy is called when vma that is covering the pfnmap gets
668 * copied through copy_page_range().
669 *
670 * If the vma has a linear pfn mapping for the entire range, we get the prot
671 * from pte and reserve the entire vma range with single reserve_pfn_range call.
venkatesh.pallipadi@intel.com58993292008-12-18 11:41:30 -0800672 */
673int track_pfn_vma_copy(struct vm_area_struct *vma)
674{
H. Peter Anvinc1c15b62008-12-23 10:10:40 -0800675 resource_size_t paddr;
venkatesh.pallipadi@intel.com982d7892008-12-19 13:47:28 -0800676 unsigned long prot;
Pallipadi, Venkatesh4b065042009-04-08 15:37:16 -0700677 unsigned long vma_size = vma->vm_end - vma->vm_start;
venkatesh.pallipadi@intel.comcdecff62009-01-09 16:13:12 -0800678 pgprot_t pgprot;
venkatesh.pallipadi@intel.com58993292008-12-18 11:41:30 -0800679
venkatesh.pallipadi@intel.com58993292008-12-18 11:41:30 -0800680 if (is_linear_pfn_mapping(vma)) {
681 /*
venkatesh.pallipadi@intel.com982d7892008-12-19 13:47:28 -0800682 * reserve the whole chunk covered by vma. We need the
683 * starting address and protection from pte.
venkatesh.pallipadi@intel.com58993292008-12-18 11:41:30 -0800684 */
Pallipadi, Venkatesh4b065042009-04-08 15:37:16 -0700685 if (follow_phys(vma, vma->vm_start, 0, &prot, &paddr)) {
venkatesh.pallipadi@intel.com58993292008-12-18 11:41:30 -0800686 WARN_ON_ONCE(1);
venkatesh.pallipadi@intel.com982d7892008-12-19 13:47:28 -0800687 return -EINVAL;
venkatesh.pallipadi@intel.com58993292008-12-18 11:41:30 -0800688 }
venkatesh.pallipadi@intel.comcdecff62009-01-09 16:13:12 -0800689 pgprot = __pgprot(prot);
690 return reserve_pfn_range(paddr, vma_size, &pgprot, 1);
venkatesh.pallipadi@intel.com58993292008-12-18 11:41:30 -0800691 }
692
venkatesh.pallipadi@intel.com58993292008-12-18 11:41:30 -0800693 return 0;
venkatesh.pallipadi@intel.com58993292008-12-18 11:41:30 -0800694}
695
696/*
697 * track_pfn_vma_new is called when a _new_ pfn mapping is being established
698 * for physical range indicated by pfn and size.
699 *
700 * prot is passed in as a parameter for the new mapping. If the vma has a
701 * linear pfn mapping for the entire range reserve the entire vma range with
702 * single reserve_pfn_range call.
venkatesh.pallipadi@intel.com58993292008-12-18 11:41:30 -0800703 */
venkatesh.pallipadi@intel.come4b866e2009-01-09 16:13:11 -0800704int track_pfn_vma_new(struct vm_area_struct *vma, pgprot_t *prot,
venkatesh.pallipadi@intel.com58993292008-12-18 11:41:30 -0800705 unsigned long pfn, unsigned long size)
706{
Venkatesh Pallipadi10876372009-07-10 09:57:40 -0700707 unsigned long flags;
H. Peter Anvinc1c15b62008-12-23 10:10:40 -0800708 resource_size_t paddr;
Pallipadi, Venkatesh4b065042009-04-08 15:37:16 -0700709 unsigned long vma_size = vma->vm_end - vma->vm_start;
venkatesh.pallipadi@intel.com58993292008-12-18 11:41:30 -0800710
venkatesh.pallipadi@intel.com58993292008-12-18 11:41:30 -0800711 if (is_linear_pfn_mapping(vma)) {
712 /* reserve the whole chunk starting from vm_pgoff */
H. Peter Anvinc1c15b62008-12-23 10:10:40 -0800713 paddr = (resource_size_t)vma->vm_pgoff << PAGE_SHIFT;
venkatesh.pallipadi@intel.comcdecff62009-01-09 16:13:12 -0800714 return reserve_pfn_range(paddr, vma_size, prot, 0);
venkatesh.pallipadi@intel.com58993292008-12-18 11:41:30 -0800715 }
716
Venkatesh Pallipadi10876372009-07-10 09:57:40 -0700717 if (!pat_enabled)
718 return 0;
719
720 /* for vm_insert_pfn and friends, we set prot based on lookup */
721 flags = lookup_memtype(pfn << PAGE_SHIFT);
722 *prot = __pgprot((pgprot_val(vma->vm_page_prot) & (~_PAGE_CACHE_MASK)) |
723 flags);
724
venkatesh.pallipadi@intel.com58993292008-12-18 11:41:30 -0800725 return 0;
venkatesh.pallipadi@intel.com58993292008-12-18 11:41:30 -0800726}
727
728/*
729 * untrack_pfn_vma is called while unmapping a pfnmap for a region.
730 * untrack can be called for a specific region indicated by pfn and size or
731 * can be for the entire vma (in which case size can be zero).
732 */
733void untrack_pfn_vma(struct vm_area_struct *vma, unsigned long pfn,
734 unsigned long size)
735{
H. Peter Anvinc1c15b62008-12-23 10:10:40 -0800736 resource_size_t paddr;
Pallipadi, Venkatesh4b065042009-04-08 15:37:16 -0700737 unsigned long vma_size = vma->vm_end - vma->vm_start;
venkatesh.pallipadi@intel.com58993292008-12-18 11:41:30 -0800738
venkatesh.pallipadi@intel.com58993292008-12-18 11:41:30 -0800739 if (is_linear_pfn_mapping(vma)) {
740 /* free the whole chunk starting from vm_pgoff */
H. Peter Anvinc1c15b62008-12-23 10:10:40 -0800741 paddr = (resource_size_t)vma->vm_pgoff << PAGE_SHIFT;
venkatesh.pallipadi@intel.com58993292008-12-18 11:41:30 -0800742 free_pfn_range(paddr, vma_size);
743 return;
744 }
venkatesh.pallipadi@intel.com58993292008-12-18 11:41:30 -0800745}
746
venkatesh.pallipadi@intel.com2520bd32008-12-18 11:41:32 -0800747pgprot_t pgprot_writecombine(pgprot_t prot)
748{
749 if (pat_enabled)
750 return __pgprot(pgprot_val(prot) | _PAGE_CACHE_WC);
751 else
752 return pgprot_noncached(prot);
753}
Ingo Molnar92b9af92009-02-28 14:09:27 +0100754EXPORT_SYMBOL_GPL(pgprot_writecombine);
venkatesh.pallipadi@intel.com2520bd32008-12-18 11:41:32 -0800755
Andreas Herrmann012f09e2008-08-06 16:23:08 +0200756#if defined(CONFIG_DEBUG_FS) && defined(CONFIG_X86_PAT)
venkatesh.pallipadi@intel.comfec09622008-07-18 16:08:14 -0700757
venkatesh.pallipadi@intel.comfec09622008-07-18 16:08:14 -0700758static struct memtype *memtype_get_idx(loff_t pos)
759{
venkatesh.pallipadi@intel.combe5a0c12010-02-10 11:57:06 -0800760 struct memtype *print_entry;
761 int ret;
venkatesh.pallipadi@intel.comfec09622008-07-18 16:08:14 -0700762
venkatesh.pallipadi@intel.combe5a0c12010-02-10 11:57:06 -0800763 print_entry = kzalloc(sizeof(struct memtype), GFP_KERNEL);
venkatesh.pallipadi@intel.comfec09622008-07-18 16:08:14 -0700764 if (!print_entry)
765 return NULL;
766
767 spin_lock(&memtype_lock);
Pallipadi, Venkatesh9e41a492010-02-10 15:26:07 -0800768 ret = rbt_memtype_copy_nth_element(print_entry, pos);
venkatesh.pallipadi@intel.comfec09622008-07-18 16:08:14 -0700769 spin_unlock(&memtype_lock);
Ingo Molnarad2cde12008-09-30 13:20:45 +0200770
venkatesh.pallipadi@intel.combe5a0c12010-02-10 11:57:06 -0800771 if (!ret) {
772 return print_entry;
773 } else {
774 kfree(print_entry);
775 return NULL;
776 }
venkatesh.pallipadi@intel.comfec09622008-07-18 16:08:14 -0700777}
778
779static void *memtype_seq_start(struct seq_file *seq, loff_t *pos)
780{
781 if (*pos == 0) {
782 ++*pos;
783 seq_printf(seq, "PAT memtype list:\n");
784 }
785
786 return memtype_get_idx(*pos);
787}
788
789static void *memtype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
790{
791 ++*pos;
792 return memtype_get_idx(*pos);
793}
794
795static void memtype_seq_stop(struct seq_file *seq, void *v)
796{
797}
798
799static int memtype_seq_show(struct seq_file *seq, void *v)
800{
801 struct memtype *print_entry = (struct memtype *)v;
802
803 seq_printf(seq, "%s @ 0x%Lx-0x%Lx\n", cattr_name(print_entry->type),
804 print_entry->start, print_entry->end);
805 kfree(print_entry);
Ingo Molnarad2cde12008-09-30 13:20:45 +0200806
venkatesh.pallipadi@intel.comfec09622008-07-18 16:08:14 -0700807 return 0;
808}
809
Tobias Klauserd535e432009-09-04 15:53:09 +0200810static const struct seq_operations memtype_seq_ops = {
venkatesh.pallipadi@intel.comfec09622008-07-18 16:08:14 -0700811 .start = memtype_seq_start,
812 .next = memtype_seq_next,
813 .stop = memtype_seq_stop,
814 .show = memtype_seq_show,
815};
816
817static int memtype_seq_open(struct inode *inode, struct file *file)
818{
819 return seq_open(file, &memtype_seq_ops);
820}
821
822static const struct file_operations memtype_fops = {
823 .open = memtype_seq_open,
824 .read = seq_read,
825 .llseek = seq_lseek,
826 .release = seq_release,
827};
828
829static int __init pat_memtype_list_init(void)
830{
Xiaotian Fengdd4377b2009-11-26 19:53:48 +0800831 if (pat_enabled) {
832 debugfs_create_file("pat_memtype_list", S_IRUSR,
833 arch_debugfs_dir, NULL, &memtype_fops);
834 }
venkatesh.pallipadi@intel.comfec09622008-07-18 16:08:14 -0700835 return 0;
836}
837
838late_initcall(pat_memtype_list_init);
839
Andreas Herrmann012f09e2008-08-06 16:23:08 +0200840#endif /* CONFIG_DEBUG_FS && CONFIG_X86_PAT */