[PATCH] x86_64-always-use-cpuid-80000008-to-figure-out-mtrr fix
[linux-2.6/kvm.git] / arch / i386 / kernel / cpu / mtrr / generic.c
blob9f7a7ea6388dbf84f28be95b73484d63c38e243a
1 /* This only handles 32bit MTRR on 32bit hosts. This is strictly wrong
2 because MTRRs can span upto 40 bits (36bits on most modern x86) */
3 #include <linux/init.h>
4 #include <linux/slab.h>
5 #include <linux/mm.h>
6 #include <asm/io.h>
7 #include <asm/mtrr.h>
8 #include <asm/msr.h>
9 #include <asm/system.h>
10 #include <asm/cpufeature.h>
11 #include <asm/tlbflush.h>
12 #include "mtrr.h"
14 struct mtrr_state {
15 struct mtrr_var_range *var_ranges;
16 mtrr_type fixed_ranges[NUM_FIXED_RANGES];
17 unsigned char enabled;
18 mtrr_type def_type;
21 static unsigned long smp_changes_mask;
22 static struct mtrr_state mtrr_state = {};
24 /* Get the MSR pair relating to a var range */
25 static void __init
26 get_mtrr_var_range(unsigned int index, struct mtrr_var_range *vr)
28 rdmsr(MTRRphysBase_MSR(index), vr->base_lo, vr->base_hi);
29 rdmsr(MTRRphysMask_MSR(index), vr->mask_lo, vr->mask_hi);
32 static void __init
33 get_fixed_ranges(mtrr_type * frs)
35 unsigned int *p = (unsigned int *) frs;
36 int i;
38 rdmsr(MTRRfix64K_00000_MSR, p[0], p[1]);
40 for (i = 0; i < 2; i++)
41 rdmsr(MTRRfix16K_80000_MSR + i, p[2 + i * 2], p[3 + i * 2]);
42 for (i = 0; i < 8; i++)
43 rdmsr(MTRRfix4K_C0000_MSR + i, p[6 + i * 2], p[7 + i * 2]);
46 /* Grab all of the MTRR state for this CPU into *state */
47 void __init get_mtrr_state(void)
49 unsigned int i;
50 struct mtrr_var_range *vrs;
51 unsigned lo, dummy;
53 if (!mtrr_state.var_ranges) {
54 mtrr_state.var_ranges = kmalloc(num_var_ranges * sizeof (struct mtrr_var_range),
55 GFP_KERNEL);
56 if (!mtrr_state.var_ranges)
57 return;
59 vrs = mtrr_state.var_ranges;
61 for (i = 0; i < num_var_ranges; i++)
62 get_mtrr_var_range(i, &vrs[i]);
63 get_fixed_ranges(mtrr_state.fixed_ranges);
65 rdmsr(MTRRdefType_MSR, lo, dummy);
66 mtrr_state.def_type = (lo & 0xff);
67 mtrr_state.enabled = (lo & 0xc00) >> 10;
70 /* Free resources associated with a struct mtrr_state */
71 void __init finalize_mtrr_state(void)
73 if (mtrr_state.var_ranges)
74 kfree(mtrr_state.var_ranges);
75 mtrr_state.var_ranges = NULL;
78 /* Some BIOS's are fucked and don't set all MTRRs the same! */
79 void __init mtrr_state_warn(void)
81 unsigned long mask = smp_changes_mask;
83 if (!mask)
84 return;
85 if (mask & MTRR_CHANGE_MASK_FIXED)
86 printk(KERN_WARNING "mtrr: your CPUs had inconsistent fixed MTRR settings\n");
87 if (mask & MTRR_CHANGE_MASK_VARIABLE)
88 printk(KERN_WARNING "mtrr: your CPUs had inconsistent variable MTRR settings\n");
89 if (mask & MTRR_CHANGE_MASK_DEFTYPE)
90 printk(KERN_WARNING "mtrr: your CPUs had inconsistent MTRRdefType settings\n");
91 printk(KERN_INFO "mtrr: probably your BIOS does not setup all CPUs.\n");
92 printk(KERN_INFO "mtrr: corrected configuration.\n");
95 /* Doesn't attempt to pass an error out to MTRR users
96 because it's quite complicated in some cases and probably not
97 worth it because the best error handling is to ignore it. */
98 void mtrr_wrmsr(unsigned msr, unsigned a, unsigned b)
100 if (wrmsr_safe(msr, a, b) < 0)
101 printk(KERN_ERR
102 "MTRR: CPU %u: Writing MSR %x to %x:%x failed\n",
103 smp_processor_id(), msr, a, b);
106 int generic_get_free_region(unsigned long base, unsigned long size)
107 /* [SUMMARY] Get a free MTRR.
108 <base> The starting (base) address of the region.
109 <size> The size (in bytes) of the region.
110 [RETURNS] The index of the region on success, else -1 on error.
113 int i, max;
114 mtrr_type ltype;
115 unsigned long lbase;
116 unsigned lsize;
118 max = num_var_ranges;
119 for (i = 0; i < max; ++i) {
120 mtrr_if->get(i, &lbase, &lsize, &ltype);
121 if (lsize == 0)
122 return i;
124 return -ENOSPC;
127 void generic_get_mtrr(unsigned int reg, unsigned long *base,
128 unsigned int *size, mtrr_type * type)
130 unsigned int mask_lo, mask_hi, base_lo, base_hi;
132 rdmsr(MTRRphysMask_MSR(reg), mask_lo, mask_hi);
133 if ((mask_lo & 0x800) == 0) {
134 /* Invalid (i.e. free) range */
135 *base = 0;
136 *size = 0;
137 *type = 0;
138 return;
141 rdmsr(MTRRphysBase_MSR(reg), base_lo, base_hi);
143 /* Work out the shifted address mask. */
144 mask_lo = size_or_mask | mask_hi << (32 - PAGE_SHIFT)
145 | mask_lo >> PAGE_SHIFT;
147 /* This works correctly if size is a power of two, i.e. a
148 contiguous range. */
149 *size = -mask_lo;
150 *base = base_hi << (32 - PAGE_SHIFT) | base_lo >> PAGE_SHIFT;
151 *type = base_lo & 0xff;
154 static int set_fixed_ranges(mtrr_type * frs)
156 unsigned int *p = (unsigned int *) frs;
157 int changed = FALSE;
158 int i;
159 unsigned int lo, hi;
161 rdmsr(MTRRfix64K_00000_MSR, lo, hi);
162 if (p[0] != lo || p[1] != hi) {
163 mtrr_wrmsr(MTRRfix64K_00000_MSR, p[0], p[1]);
164 changed = TRUE;
167 for (i = 0; i < 2; i++) {
168 rdmsr(MTRRfix16K_80000_MSR + i, lo, hi);
169 if (p[2 + i * 2] != lo || p[3 + i * 2] != hi) {
170 mtrr_wrmsr(MTRRfix16K_80000_MSR + i, p[2 + i * 2],
171 p[3 + i * 2]);
172 changed = TRUE;
176 for (i = 0; i < 8; i++) {
177 rdmsr(MTRRfix4K_C0000_MSR + i, lo, hi);
178 if (p[6 + i * 2] != lo || p[7 + i * 2] != hi) {
179 mtrr_wrmsr(MTRRfix4K_C0000_MSR + i, p[6 + i * 2],
180 p[7 + i * 2]);
181 changed = TRUE;
184 return changed;
187 /* Set the MSR pair relating to a var range. Returns TRUE if
188 changes are made */
189 static int set_mtrr_var_ranges(unsigned int index, struct mtrr_var_range *vr)
191 unsigned int lo, hi;
192 int changed = FALSE;
194 rdmsr(MTRRphysBase_MSR(index), lo, hi);
195 if ((vr->base_lo & 0xfffff0ffUL) != (lo & 0xfffff0ffUL)
196 || (vr->base_hi & (size_and_mask >> (32 - PAGE_SHIFT))) !=
197 (hi & (size_and_mask >> (32 - PAGE_SHIFT)))) {
198 mtrr_wrmsr(MTRRphysBase_MSR(index), vr->base_lo, vr->base_hi);
199 changed = TRUE;
202 rdmsr(MTRRphysMask_MSR(index), lo, hi);
204 if ((vr->mask_lo & 0xfffff800UL) != (lo & 0xfffff800UL)
205 || (vr->mask_hi & (size_and_mask >> (32 - PAGE_SHIFT))) !=
206 (hi & (size_and_mask >> (32 - PAGE_SHIFT)))) {
207 mtrr_wrmsr(MTRRphysMask_MSR(index), vr->mask_lo, vr->mask_hi);
208 changed = TRUE;
210 return changed;
213 static unsigned long set_mtrr_state(u32 deftype_lo, u32 deftype_hi)
214 /* [SUMMARY] Set the MTRR state for this CPU.
215 <state> The MTRR state information to read.
216 <ctxt> Some relevant CPU context.
217 [NOTE] The CPU must already be in a safe state for MTRR changes.
218 [RETURNS] 0 if no changes made, else a mask indication what was changed.
221 unsigned int i;
222 unsigned long change_mask = 0;
224 for (i = 0; i < num_var_ranges; i++)
225 if (set_mtrr_var_ranges(i, &mtrr_state.var_ranges[i]))
226 change_mask |= MTRR_CHANGE_MASK_VARIABLE;
228 if (set_fixed_ranges(mtrr_state.fixed_ranges))
229 change_mask |= MTRR_CHANGE_MASK_FIXED;
231 /* Set_mtrr_restore restores the old value of MTRRdefType,
232 so to set it we fiddle with the saved value */
233 if ((deftype_lo & 0xff) != mtrr_state.def_type
234 || ((deftype_lo & 0xc00) >> 10) != mtrr_state.enabled) {
235 deftype_lo |= (mtrr_state.def_type | mtrr_state.enabled << 10);
236 change_mask |= MTRR_CHANGE_MASK_DEFTYPE;
239 return change_mask;
243 static unsigned long cr4 = 0;
244 static u32 deftype_lo, deftype_hi;
245 static DEFINE_SPINLOCK(set_atomicity_lock);
248 * Since we are disabling the cache don't allow any interrupts - they
249 * would run extremely slow and would only increase the pain. The caller must
250 * ensure that local interrupts are disabled and are reenabled after post_set()
251 * has been called.
254 static void prepare_set(void)
256 unsigned long cr0;
258 /* Note that this is not ideal, since the cache is only flushed/disabled
259 for this CPU while the MTRRs are changed, but changing this requires
260 more invasive changes to the way the kernel boots */
262 spin_lock(&set_atomicity_lock);
264 /* Enter the no-fill (CD=1, NW=0) cache mode and flush caches. */
265 cr0 = read_cr0() | 0x40000000; /* set CD flag */
266 write_cr0(cr0);
267 wbinvd();
269 /* Save value of CR4 and clear Page Global Enable (bit 7) */
270 if ( cpu_has_pge ) {
271 cr4 = read_cr4();
272 write_cr4(cr4 & ~X86_CR4_PGE);
275 /* Flush all TLBs via a mov %cr3, %reg; mov %reg, %cr3 */
276 __flush_tlb();
278 /* Save MTRR state */
279 rdmsr(MTRRdefType_MSR, deftype_lo, deftype_hi);
281 /* Disable MTRRs, and set the default type to uncached */
282 mtrr_wrmsr(MTRRdefType_MSR, deftype_lo & 0xf300UL, deftype_hi);
285 static void post_set(void)
287 /* Flush TLBs (no need to flush caches - they are disabled) */
288 __flush_tlb();
290 /* Intel (P6) standard MTRRs */
291 mtrr_wrmsr(MTRRdefType_MSR, deftype_lo, deftype_hi);
293 /* Enable caches */
294 write_cr0(read_cr0() & 0xbfffffff);
296 /* Restore value of CR4 */
297 if ( cpu_has_pge )
298 write_cr4(cr4);
299 spin_unlock(&set_atomicity_lock);
302 static void generic_set_all(void)
304 unsigned long mask, count;
305 unsigned long flags;
307 local_irq_save(flags);
308 prepare_set();
310 /* Actually set the state */
311 mask = set_mtrr_state(deftype_lo,deftype_hi);
313 post_set();
314 local_irq_restore(flags);
316 /* Use the atomic bitops to update the global mask */
317 for (count = 0; count < sizeof mask * 8; ++count) {
318 if (mask & 0x01)
319 set_bit(count, &smp_changes_mask);
320 mask >>= 1;
325 static void generic_set_mtrr(unsigned int reg, unsigned long base,
326 unsigned long size, mtrr_type type)
327 /* [SUMMARY] Set variable MTRR register on the local CPU.
328 <reg> The register to set.
329 <base> The base address of the region.
330 <size> The size of the region. If this is 0 the region is disabled.
331 <type> The type of the region.
332 <do_safe> If TRUE, do the change safely. If FALSE, safety measures should
333 be done externally.
334 [RETURNS] Nothing.
337 unsigned long flags;
339 local_irq_save(flags);
340 prepare_set();
342 if (size == 0) {
343 /* The invalid bit is kept in the mask, so we simply clear the
344 relevant mask register to disable a range. */
345 mtrr_wrmsr(MTRRphysMask_MSR(reg), 0, 0);
346 } else {
347 mtrr_wrmsr(MTRRphysBase_MSR(reg), base << PAGE_SHIFT | type,
348 (base & size_and_mask) >> (32 - PAGE_SHIFT));
349 mtrr_wrmsr(MTRRphysMask_MSR(reg), -size << PAGE_SHIFT | 0x800,
350 (-size & size_and_mask) >> (32 - PAGE_SHIFT));
353 post_set();
354 local_irq_restore(flags);
357 int generic_validate_add_page(unsigned long base, unsigned long size, unsigned int type)
359 unsigned long lbase, last;
361 /* For Intel PPro stepping <= 7, must be 4 MiB aligned
362 and not touch 0x70000000->0x7003FFFF */
363 if (is_cpu(INTEL) && boot_cpu_data.x86 == 6 &&
364 boot_cpu_data.x86_model == 1 &&
365 boot_cpu_data.x86_mask <= 7) {
366 if (base & ((1 << (22 - PAGE_SHIFT)) - 1)) {
367 printk(KERN_WARNING "mtrr: base(0x%lx000) is not 4 MiB aligned\n", base);
368 return -EINVAL;
370 if (!(base + size < 0x70000000 || base > 0x7003FFFF) &&
371 (type == MTRR_TYPE_WRCOMB
372 || type == MTRR_TYPE_WRBACK)) {
373 printk(KERN_WARNING "mtrr: writable mtrr between 0x70000000 and 0x7003FFFF may hang the CPU.\n");
374 return -EINVAL;
378 if (base + size < 0x100) {
379 printk(KERN_WARNING "mtrr: cannot set region below 1 MiB (0x%lx000,0x%lx000)\n",
380 base, size);
381 return -EINVAL;
383 /* Check upper bits of base and last are equal and lower bits are 0
384 for base and 1 for last */
385 last = base + size - 1;
386 for (lbase = base; !(lbase & 1) && (last & 1);
387 lbase = lbase >> 1, last = last >> 1) ;
388 if (lbase != last) {
389 printk(KERN_WARNING "mtrr: base(0x%lx000) is not aligned on a size(0x%lx000) boundary\n",
390 base, size);
391 return -EINVAL;
393 return 0;
397 static int generic_have_wrcomb(void)
399 unsigned long config, dummy;
400 rdmsr(MTRRcap_MSR, config, dummy);
401 return (config & (1 << 10));
404 int positive_have_wrcomb(void)
406 return 1;
409 /* generic structure...
411 struct mtrr_ops generic_mtrr_ops = {
412 .use_intel_if = 1,
413 .set_all = generic_set_all,
414 .get = generic_get_mtrr,
415 .get_free_region = generic_get_free_region,
416 .set = generic_set_mtrr,
417 .validate_add_page = generic_validate_add_page,
418 .have_wrcomb = generic_have_wrcomb,