x86, pti: fix boot problems from Global-bit setting

These are _very_ lightly tested. I'm throwing them out there for folks are looking for a fix.

---

From: Dave Hansen <[email protected]>

pageattr.c is not friendly when it encounters empty (zero) PTEs. The kernel linear map is exempt from these checks, but kernel text is not. This patch adds the code to also exempt kernel text from these checks. The proximate cause of these warnings was most likely an __init area that spanned a 2MB page boundary that resulted in a "zero" PMD.

Signed-off-by Dave Hansen <[email protected]>

Fixes 39114b7a7 (x86/pti: Never implicitly clear _PAGE_GLOBAL for kernel image)

Reported-by Mariusz Ceier <[email protected]>

Reported-by Aaro Koskinen <[email protected]>

Cc Andrea Arcangeli <[email protected]>

Cc Andy Lutomirski <[email protected]>

Cc Arjan van de Ven <[email protected]>

Cc Borislav Petkov <[email protected]>

Cc Dan Williams <[email protected]>

Cc David Woodhouse <[email protected]>

Cc Greg Kroah-Hartman <[email protected]>

Cc Hugh Dickins <[email protected]>

Cc Josh Poimboeuf <[email protected]>

Cc Juergen Gross <[email protected]>

Cc Kees Cook <[email protected]>

Cc Linus Torvalds <[email protected]>

Cc Nadav Amit <[email protected]>

Cc Peter Zijlstra <[email protected]>

Cc Thomas Gleixner <[email protected]>

Cc [email protected]

---

 b/arch/x86/mm/pageattr.c |   17 +++++++++++++++--
 1 file changed, 15 insertions(+), 2 deletions(-)

diff -puN arch/x86/mm/pageattr.c~pti-glb-warning-inpageattr arch/x86/mm/pageattr.c

arch/x86/mm/pageattr.c 2018-04-17 14:10:22.721395554 -0700 renamed from arch/x86/mm/pageattr.c~pti-glb-warning-inpageattr 2018-04-17 14:10:22.695395554 -0700 +15 -2

@@ -1151,6 +1151,16 @@ static int populate_pgd(struct cpa_data
        return 0;
 }
 
+bool __cpa_pfn_in_highmap(unsigned long pfn)
+{
+       /*
+        * Kernel text has an alias mapping at a high address, known
+        * here as "highmap".
+        */
+       return within_inclusive(pfn, highmap_start_pfn(),
+                       highmap_end_pfn());
+}
+
 static int __cpa_process_fault(struct cpa_data *cpa, unsigned long vaddr,
                               int primary)
 {

@@ -1183,6 +1193,10 @@ static int __cpa_process_fault(struct cp
                cpa->numpages = 1;
                cpa->pfn = __pa(vaddr) >> PAGE_SHIFT;
                return 0;
+
+       } else if (__cpa_pfn_in_highmap(cpa->pfn)) {
+               /* Faults in the highmap are OK, so do not warn: */
+               return -EFAULT;
        } else {
                WARN(1, KERN_WARNING "CPA: called for zero pte. "
                        "vaddr = %lx cpa->vaddr = %lx\n", vaddr,

@@ -1335,8 +1349,7 @@ static int cpa_process_alias(struct cpa_
         * to touch the high mapped kernel as well:
         */
        if (!within(vaddr, (unsigned long)_text, _brk_end) &&
-           within_inclusive(cpa->pfn, highmap_start_pfn(),
-                            highmap_end_pfn())) {
+           __cpa_pfn_in_highmap(cpa->pfn)) {
                unsigned long temp_cpa_vaddr = (cpa->pfn << PAGE_SHIFT) +
                                               __START_KERNEL_map - phys_base;
                alias_cpa = *cpa;

_

These are _very_ lightly tested. I'm throwing them out there for folks are looking for a fix.

---

From: Dave Hansen <[email protected]>

pageattr.c is not friendly when it encounters empty (zero) PTEs. The kernel linear map is exempt from these checks, but kernel text is not. This patch adds the code to also exempt kernel text from these checks.

The proximate cause of these warnings was most likely an __init area that spanned a 2MB page boundary that resulted in a "zero" PMD.

Quoted 3 lines

pageattr.c is not friendly when it encounters empty (zero) PTEs. The kernel linear map is exempt from these checks, but kernel text is not. This patch adds the code to also exempt kernel text from these checks.

Bah. Changelogs should tell the WHY and not the WHAT

Quoted 2 lines

The proximate cause of these warnings was most likely an __init area that spanned a 2MB page boundary that resulted in a "zero" PMD.

This doesn't make any sense at all.