]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blobdiff - arch/x86/mm/pat.c
x86: Enable PAT to use cache mode translation tables
[mirror_ubuntu-bionic-kernel.git] / arch / x86 / mm / pat.c
index 657438858e8358745484c828158634fba41a5b1c..4c601276a5563e38bc8b7310cdb15684f6e61a2a 100644 (file)
@@ -31,6 +31,7 @@
 #include <asm/io.h>
 
 #include "pat_internal.h"
+#include "mm_internal.h"
 
 #ifdef CONFIG_X86_PAT
 int __read_mostly pat_enabled = 1;
@@ -75,6 +76,52 @@ enum {
        PAT_UC_MINUS = 7,       /* UC, but can be overriden by MTRR */
 };
 
+#define CM(c) (_PAGE_CACHE_MODE_ ## c)
+
+static enum page_cache_mode pat_get_cache_mode(unsigned pat_val, char *msg)
+{
+       enum page_cache_mode cache;
+       char *cache_mode;
+
+       switch (pat_val) {
+       case PAT_UC:       cache = CM(UC);       cache_mode = "UC  "; break;
+       case PAT_WC:       cache = CM(WC);       cache_mode = "WC  "; break;
+       case PAT_WT:       cache = CM(WT);       cache_mode = "WT  "; break;
+       case PAT_WP:       cache = CM(WP);       cache_mode = "WP  "; break;
+       case PAT_WB:       cache = CM(WB);       cache_mode = "WB  "; break;
+       case PAT_UC_MINUS: cache = CM(UC_MINUS); cache_mode = "UC- "; break;
+       default:           cache = CM(WB);       cache_mode = "WB  "; break;
+       }
+
+       memcpy(msg, cache_mode, 4);
+
+       return cache;
+}
+
+#undef CM
+
+/*
+ * Update the cache mode to pgprot translation tables according to PAT
+ * configuration.
+ * Using lower indices is preferred, so we start with highest index.
+ */
+void pat_init_cache_modes(void)
+{
+       int i;
+       enum page_cache_mode cache;
+       char pat_msg[33];
+       u64 pat;
+
+       rdmsrl(MSR_IA32_CR_PAT, pat);
+       pat_msg[32] = 0;
+       for (i = 7; i >= 0; i--) {
+               cache = pat_get_cache_mode((pat >> (i * 8)) & 7,
+                                          pat_msg + 4 * i);
+               update_cache_mode_entry(i, cache);
+       }
+       pr_info("PAT configuration [0-7]: %s\n", pat_msg);
+}
+
 #define PAT(x, y)      ((u64)PAT_ ## y << ((x)*8))
 
 void pat_init(void)
@@ -124,8 +171,7 @@ void pat_init(void)
        wrmsrl(MSR_IA32_CR_PAT, pat);
 
        if (boot_cpu)
-               printk(KERN_INFO "x86 PAT enabled: cpu %d, old 0x%Lx, new 0x%Lx\n",
-                      smp_processor_id(), boot_pat_state, pat);
+               pat_init_cache_modes();
 }
 
 #undef PAT
@@ -139,20 +185,21 @@ static DEFINE_SPINLOCK(memtype_lock);     /* protects memtype accesses */
  * The intersection is based on "Effective Memory Type" tables in IA-32
  * SDM vol 3a
  */
-static unsigned long pat_x_mtrr_type(u64 start, u64 end, unsigned long req_type)
+static unsigned long pat_x_mtrr_type(u64 start, u64 end,
+                                    enum page_cache_mode req_type)
 {
        /*
         * Look for MTRR hint to get the effective type in case where PAT
         * request is for WB.
         */
-       if (req_type == _PAGE_CACHE_WB) {
+       if (req_type == _PAGE_CACHE_MODE_WB) {
                u8 mtrr_type;
 
                mtrr_type = mtrr_type_lookup(start, end);
                if (mtrr_type != MTRR_TYPE_WRBACK)
-                       return _PAGE_CACHE_UC_MINUS;
+                       return _PAGE_CACHE_MODE_UC_MINUS;
 
-               return _PAGE_CACHE_WB;
+               return _PAGE_CACHE_MODE_WB;
        }
 
        return req_type;
@@ -207,25 +254,26 @@ static int pat_pagerange_is_ram(resource_size_t start, resource_size_t end)
  * - Find the memtype of all the pages in the range, look for any conflicts
  * - In case of no conflicts, set the new memtype for pages in the range
  */
-static int reserve_ram_pages_type(u64 start, u64 end, unsigned long req_type,
-                                 unsigned long *new_type)
+static int reserve_ram_pages_type(u64 start, u64 end,
+                                 enum page_cache_mode req_type,
+                                 enum page_cache_mode *new_type)
 {
        struct page *page;
        u64 pfn;
 
-       if (req_type == _PAGE_CACHE_UC) {
+       if (req_type == _PAGE_CACHE_MODE_UC) {
                /* We do not support strong UC */
                WARN_ON_ONCE(1);
-               req_type = _PAGE_CACHE_UC_MINUS;
+               req_type = _PAGE_CACHE_MODE_UC_MINUS;
        }
 
        for (pfn = (start >> PAGE_SHIFT); pfn < (end >> PAGE_SHIFT); ++pfn) {
-               unsigned long type;
+               enum page_cache_mode type;
 
                page = pfn_to_page(pfn);
                type = get_page_memtype(page);
                if (type != -1) {
-                       printk(KERN_INFO "reserve_ram_pages_type failed [mem %#010Lx-%#010Lx], track 0x%lx, req 0x%lx\n",
+                       pr_info("reserve_ram_pages_type failed [mem %#010Lx-%#010Lx], track 0x%x, req 0x%x\n",
                                start, end - 1, type, req_type);
                        if (new_type)
                                *new_type = type;
@@ -258,21 +306,21 @@ static int free_ram_pages_type(u64 start, u64 end)
 
 /*
  * req_type typically has one of the:
- * - _PAGE_CACHE_WB
- * - _PAGE_CACHE_WC
- * - _PAGE_CACHE_UC_MINUS
- * - _PAGE_CACHE_UC
+ * - _PAGE_CACHE_MODE_WB
+ * - _PAGE_CACHE_MODE_WC
+ * - _PAGE_CACHE_MODE_UC_MINUS
+ * - _PAGE_CACHE_MODE_UC
  *
  * If new_type is NULL, function will return an error if it cannot reserve the
  * region with req_type. If new_type is non-NULL, function will return
  * available type in new_type in case of no error. In case of any error
  * it will return a negative return value.
  */
-int reserve_memtype(u64 start, u64 end, unsigned long req_type,
-                   unsigned long *new_type)
+int reserve_memtype(u64 start, u64 end, enum page_cache_mode req_type,
+                   enum page_cache_mode *new_type)
 {
        struct memtype *new;
-       unsigned long actual_type;
+       enum page_cache_mode actual_type;
        int is_range_ram;
        int err = 0;
 
@@ -281,10 +329,10 @@ int reserve_memtype(u64 start, u64 end, unsigned long req_type,
        if (!pat_enabled) {
                /* This is identical to page table setting without PAT */
                if (new_type) {
-                       if (req_type == _PAGE_CACHE_WC)
-                               *new_type = _PAGE_CACHE_UC_MINUS;
+                       if (req_type == _PAGE_CACHE_MODE_WC)
+                               *new_type = _PAGE_CACHE_MODE_UC_MINUS;
                        else
-                               *new_type = req_type & _PAGE_CACHE_MASK;
+                               *new_type = req_type;
                }
                return 0;
        }
@@ -292,7 +340,7 @@ int reserve_memtype(u64 start, u64 end, unsigned long req_type,
        /* Low ISA region is always mapped WB in page table. No need to track */
        if (x86_platform.is_untracked_pat_range(start, end)) {
                if (new_type)
-                       *new_type = _PAGE_CACHE_WB;
+                       *new_type = _PAGE_CACHE_MODE_WB;
                return 0;
        }
 
@@ -302,7 +350,7 @@ int reserve_memtype(u64 start, u64 end, unsigned long req_type,
         * tools and ACPI tools). Use WB request for WB memory and use
         * UC_MINUS otherwise.
         */
-       actual_type = pat_x_mtrr_type(start, end, req_type & _PAGE_CACHE_MASK);
+       actual_type = pat_x_mtrr_type(start, end, req_type);
 
        if (new_type)
                *new_type = actual_type;
@@ -394,12 +442,12 @@ int free_memtype(u64 start, u64 end)
  *
  * Only to be called when PAT is enabled
  *
- * Returns _PAGE_CACHE_WB, _PAGE_CACHE_WC, _PAGE_CACHE_UC_MINUS or
- * _PAGE_CACHE_UC
+ * Returns _PAGE_CACHE_MODE_WB, _PAGE_CACHE_MODE_WC, _PAGE_CACHE_MODE_UC_MINUS
+ * or _PAGE_CACHE_MODE_UC
  */
-static unsigned long lookup_memtype(u64 paddr)
+static enum page_cache_mode lookup_memtype(u64 paddr)
 {
-       int rettype = _PAGE_CACHE_WB;
+       enum page_cache_mode rettype = _PAGE_CACHE_MODE_WB;
        struct memtype *entry;
 
        if (x86_platform.is_untracked_pat_range(paddr, paddr + PAGE_SIZE))
@@ -414,7 +462,7 @@ static unsigned long lookup_memtype(u64 paddr)
                 * default state and not reserved, and hence of type WB
                 */
                if (rettype == -1)
-                       rettype = _PAGE_CACHE_WB;
+                       rettype = _PAGE_CACHE_MODE_WB;
 
                return rettype;
        }
@@ -425,7 +473,7 @@ static unsigned long lookup_memtype(u64 paddr)
        if (entry != NULL)
                rettype = entry->type;
        else
-               rettype = _PAGE_CACHE_UC_MINUS;
+               rettype = _PAGE_CACHE_MODE_UC_MINUS;
 
        spin_unlock(&memtype_lock);
        return rettype;
@@ -442,11 +490,11 @@ static unsigned long lookup_memtype(u64 paddr)
  * On failure, returns non-zero
  */
 int io_reserve_memtype(resource_size_t start, resource_size_t end,
-                       unsigned long *type)
+                       enum page_cache_mode *type)
 {
        resource_size_t size = end - start;
-       unsigned long req_type = *type;
-       unsigned long new_type;
+       enum page_cache_mode req_type = *type;
+       enum page_cache_mode new_type;
        int ret;
 
        WARN_ON_ONCE(iomem_map_sanity_check(start, size));
@@ -520,13 +568,13 @@ static inline int range_is_allowed(unsigned long pfn, unsigned long size)
 int phys_mem_access_prot_allowed(struct file *file, unsigned long pfn,
                                unsigned long size, pgprot_t *vma_prot)
 {
-       unsigned long flags = _PAGE_CACHE_WB;
+       enum page_cache_mode pcm = _PAGE_CACHE_MODE_WB;
 
        if (!range_is_allowed(pfn, size))
                return 0;
 
        if (file->f_flags & O_DSYNC)
-               flags = _PAGE_CACHE_UC_MINUS;
+               pcm = _PAGE_CACHE_MODE_UC_MINUS;
 
 #ifdef CONFIG_X86_32
        /*
@@ -543,12 +591,12 @@ int phys_mem_access_prot_allowed(struct file *file, unsigned long pfn,
              boot_cpu_has(X86_FEATURE_CYRIX_ARR) ||
              boot_cpu_has(X86_FEATURE_CENTAUR_MCR)) &&
            (pfn << PAGE_SHIFT) >= __pa(high_memory)) {
-               flags = _PAGE_CACHE_UC;
+               pcm = _PAGE_CACHE_MODE_UC;
        }
 #endif
 
        *vma_prot = __pgprot((pgprot_val(*vma_prot) & ~_PAGE_CACHE_MASK) |
-                            flags);
+                            cachemode2protval(pcm));
        return 1;
 }
 
@@ -556,7 +604,8 @@ int phys_mem_access_prot_allowed(struct file *file, unsigned long pfn,
  * Change the memory type for the physial address range in kernel identity
  * mapping space if that range is a part of identity map.
  */
-int kernel_map_sync_memtype(u64 base, unsigned long size, unsigned long flags)
+int kernel_map_sync_memtype(u64 base, unsigned long size,
+                           enum page_cache_mode pcm)
 {
        unsigned long id_sz;
 
@@ -574,11 +623,11 @@ int kernel_map_sync_memtype(u64 base, unsigned long size, unsigned long flags)
                                __pa(high_memory) - base :
                                size;
 
-       if (ioremap_change_attr((unsigned long)__va(base), id_sz, flags) < 0) {
+       if (ioremap_change_attr((unsigned long)__va(base), id_sz, pcm) < 0) {
                printk(KERN_INFO "%s:%d ioremap_change_attr failed %s "
                        "for [mem %#010Lx-%#010Lx]\n",
                        current->comm, current->pid,
-                       cattr_name(flags),
+                       cattr_name(pcm),
                        base, (unsigned long long)(base + size-1));
                return -EINVAL;
        }
@@ -595,8 +644,8 @@ static int reserve_pfn_range(u64 paddr, unsigned long size, pgprot_t *vma_prot,
 {
        int is_ram = 0;
        int ret;
-       unsigned long want_flags = (pgprot_val(*vma_prot) & _PAGE_CACHE_MASK);
-       unsigned long flags = want_flags;
+       enum page_cache_mode want_pcm = pgprot2cachemode(*vma_prot);
+       enum page_cache_mode pcm = want_pcm;
 
        is_ram = pat_pagerange_is_ram(paddr, paddr + size);
 
@@ -609,36 +658,36 @@ static int reserve_pfn_range(u64 paddr, unsigned long size, pgprot_t *vma_prot,
                if (!pat_enabled)
                        return 0;
 
-               flags = lookup_memtype(paddr);
-               if (want_flags != flags) {
+               pcm = lookup_memtype(paddr);
+               if (want_pcm != pcm) {
                        printk(KERN_WARNING "%s:%d map pfn RAM range req %s for [mem %#010Lx-%#010Lx], got %s\n",
                                current->comm, current->pid,
-                               cattr_name(want_flags),
+                               cattr_name(want_pcm),
                                (unsigned long long)paddr,
                                (unsigned long long)(paddr + size - 1),
-                               cattr_name(flags));
+                               cattr_name(pcm));
                        *vma_prot = __pgprot((pgprot_val(*vma_prot) &
-                                             (~_PAGE_CACHE_MASK)) |
-                                            flags);
+                                            (~_PAGE_CACHE_MASK)) |
+                                            cachemode2protval(pcm));
                }
                return 0;
        }
 
-       ret = reserve_memtype(paddr, paddr + size, want_flags, &flags);
+       ret = reserve_memtype(paddr, paddr + size, want_pcm, &pcm);
        if (ret)
                return ret;
 
-       if (flags != want_flags) {
+       if (pcm != want_pcm) {
                if (strict_prot ||
-                   !is_new_memtype_allowed(paddr, size, want_flags, flags)) {
+                   !is_new_memtype_allowed(paddr, size, want_pcm, pcm)) {
                        free_memtype(paddr, paddr + size);
                        printk(KERN_ERR "%s:%d map pfn expected mapping type %s"
                                " for [mem %#010Lx-%#010Lx], got %s\n",
                                current->comm, current->pid,
-                               cattr_name(want_flags),
+                               cattr_name(want_pcm),
                                (unsigned long long)paddr,
                                (unsigned long long)(paddr + size - 1),
-                               cattr_name(flags));
+                               cattr_name(pcm));
                        return -EINVAL;
                }
                /*
@@ -647,10 +696,10 @@ static int reserve_pfn_range(u64 paddr, unsigned long size, pgprot_t *vma_prot,
                 */
                *vma_prot = __pgprot((pgprot_val(*vma_prot) &
                                      (~_PAGE_CACHE_MASK)) |
-                                    flags);
+                                    cachemode2protval(pcm));
        }
 
-       if (kernel_map_sync_memtype(paddr, size, flags) < 0) {
+       if (kernel_map_sync_memtype(paddr, size, pcm) < 0) {
                free_memtype(paddr, paddr + size);
                return -EINVAL;
        }
@@ -709,7 +758,7 @@ int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
                    unsigned long pfn, unsigned long addr, unsigned long size)
 {
        resource_size_t paddr = (resource_size_t)pfn << PAGE_SHIFT;
-       unsigned long flags;
+       enum page_cache_mode pcm;
 
        /* reserve the whole chunk starting from paddr */
        if (addr == vma->vm_start && size == (vma->vm_end - vma->vm_start)) {
@@ -728,18 +777,18 @@ int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
         * For anything smaller than the vma size we set prot based on the
         * lookup.
         */
-       flags = lookup_memtype(paddr);
+       pcm = lookup_memtype(paddr);
 
        /* Check memtype for the remaining pages */
        while (size > PAGE_SIZE) {
                size -= PAGE_SIZE;
                paddr += PAGE_SIZE;
-               if (flags != lookup_memtype(paddr))
+               if (pcm != lookup_memtype(paddr))
                        return -EINVAL;
        }
 
        *prot = __pgprot((pgprot_val(vma->vm_page_prot) & (~_PAGE_CACHE_MASK)) |
-                        flags);
+                        cachemode2protval(pcm));
 
        return 0;
 }
@@ -747,15 +796,15 @@ int track_pfn_remap(struct vm_area_struct *vma, pgprot_t *prot,
 int track_pfn_insert(struct vm_area_struct *vma, pgprot_t *prot,
                     unsigned long pfn)
 {
-       unsigned long flags;
+       enum page_cache_mode pcm;
 
        if (!pat_enabled)
                return 0;
 
        /* Set prot based on lookup */
-       flags = lookup_memtype((resource_size_t)pfn << PAGE_SHIFT);
+       pcm = lookup_memtype((resource_size_t)pfn << PAGE_SHIFT);
        *prot = __pgprot((pgprot_val(vma->vm_page_prot) & (~_PAGE_CACHE_MASK)) |
-                        flags);
+                        cachemode2protval(pcm));
 
        return 0;
 }
@@ -791,7 +840,8 @@ void untrack_pfn(struct vm_area_struct *vma, unsigned long pfn,
 pgprot_t pgprot_writecombine(pgprot_t prot)
 {
        if (pat_enabled)
-               return __pgprot(pgprot_val(prot) | _PAGE_CACHE_WC);
+               return __pgprot(pgprot_val(prot) |
+                               cachemode2protval(_PAGE_CACHE_MODE_WC));
        else
                return pgprot_noncached(prot);
 }