[PATCH] parisc: Purge existing TLB entries in set_pte_at and ptep_set_wrprotect - take 2

[PATCH] parisc: Purge existing TLB entries in set_pte_at and ptep_set_wrprotect - take 2

[John David Anglin]

[-- Attachment #1: Type: text/plain, Size: 206 bytes --]

Attached is new version of patch which no longer saves and restores  
%sr1.  Lightly tested with a GCC build
and a build of Debian gtk+3.0 package.

Signed-off-by: John David Anglin <dave.anglin@bell.net>


[-- Attachment #2: purge_tlb.d.1.txt --]
[-- Type: text/plain, Size: 2286 bytes --]

diff --git a/arch/parisc/include/asm/pgtable.h b/arch/parisc/include/asm/pgtable.h
index ee99f23..2c6dedb 100644
--- a/arch/parisc/include/asm/pgtable.h
+++ b/arch/parisc/include/asm/pgtable.h
@@ -12,11 +12,10 @@
 
 #include <linux/bitops.h>
 #include <linux/spinlock.h>
+#include <linux/mm_types.h>
 #include <asm/processor.h>
 #include <asm/cache.h>
 
-struct vm_area_struct;
-
 /*
  * kern_addr_valid(ADDR) tests if ADDR is pointing to valid kernel
  * memory.  For the return value to be meaningful, ADDR must be >=
@@ -40,7 +39,14 @@ struct vm_area_struct;
         do{                                                     \
                 *(pteptr) = (pteval);                           \
         } while(0)
-#define set_pte_at(mm,addr,ptep,pteval) set_pte(ptep,pteval)
+
+extern void purge_tlb_entries(struct mm_struct *, unsigned long);
+
+#define set_pte_at(mm,addr,ptep, pteval)                        \
+        do{                                                     \
+                set_pte(ptep,pteval);                           \
+                purge_tlb_entries(mm,addr);                     \
+        } while(0)
 
 #endif /* !__ASSEMBLY__ */
 
@@ -466,6 +472,7 @@ static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long addr,
 		old = pte_val(*ptep);
 		new = pte_val(pte_wrprotect(__pte (old)));
 	} while (cmpxchg((unsigned long *) ptep, old, new) != old);
+	purge_tlb_entries(mm, addr);
 #else
 	pte_t old_pte = *ptep;
 	set_pte_at(mm, addr, ptep, pte_wrprotect(old_pte));
diff --git a/arch/parisc/kernel/cache.c b/arch/parisc/kernel/cache.c
index 48e16dc..1524fd8 100644
--- a/arch/parisc/kernel/cache.c
+++ b/arch/parisc/kernel/cache.c
@@ -419,6 +419,24 @@ void kunmap_parisc(void *addr)
 EXPORT_SYMBOL(kunmap_parisc);
 #endif
 
+void purge_tlb_entries(struct mm_struct *mm, unsigned long addr)
+{
+	unsigned long flags;
+
+	/* Note: purge_tlb_entries can be called at startup with
+	   no context.  */
+
+	/* Disable preemption while we play with %sr1.  */
+	preempt_disable();
+	mtsp(mm->context,1);
+	purge_tlb_start(flags);
+	pdtlb(addr);
+	pitlb(addr);
+	purge_tlb_end(flags);
+	preempt_enable();
+}
+EXPORT_SYMBOL(purge_tlb_entries);
+
 void __flush_tlb_range(unsigned long sid, unsigned long start,
 		       unsigned long end)
 {

[-- Attachment #3: Type: text/plain, Size: 51 bytes --]



Dave
--
John David Anglin	dave.anglin@bell.net

Re: [PATCH] parisc: Purge existing TLB entries in set_pte_at and ptep_set_wrprotect - take 2

[James Bottomley]

On Mon, 2013-01-14 at 19:45 -0500, John David Anglin wrote:
> Attached is new version of patch which no longer saves and restores  
> %sr1.  Lightly tested with a GCC build
> and a build of Debian gtk+3.0 package.

I can merge it with your previous changelog, but for ease of git users,
the accepted format is

<patch description for git>

---

<Any other details you don't want in git>

git ignores everything after the three dashes on a single line.

So your preamble should be (I also added an explanation of why we think
the problem is occurring):

This patch goes a long way toward fixing the minifail bug, and
it  significantly improves the stability of SMP machines such as
the rp3440.  When write  protecting a page for COW, we need to
purge the existing translation.  Otherwise, the COW break
doesn't occur as expected because the TLB may still have a stale entry
which allows writes.

Signed-off-by: John David Anglin <dave.anglin@bell.net>

---

v2: Attached is new version of patch which no longer saves and
restores  %sr1.  Lightly tested with a GCC build and a build of
Debian gtk+3.0 package.

James

Re: [PATCH] parisc: Purge existing TLB entries in set_pte_at and ptep_set_wrprotect - take 2

[Rolf Eike Beer]

[-- Attachment #1: Type: text/plain, Size: 557 bytes --]

James Bottomley wrote:
> On Mon, 2013-01-14 at 19:45 -0500, John David Anglin wrote:
> > Attached is new version of patch which no longer saves and restores
> > %sr1.  Lightly tested with a GCC build
> > and a build of Debian gtk+3.0 package.
> 
> I can merge it with your previous changelog, but for ease of git users,
> the accepted format is

> v2: Attached is new version of patch which no longer saves and
> restores  %sr1.  Lightly tested with a GCC build and a build of
> Debian gtk+3.0 package.

Do we want that to go back into stable kernels?

Eike

[-- Attachment #2: This is a digitally signed message part. --]
[-- Type: application/pgp-signature, Size: 198 bytes --]

Re: [PATCH] parisc: Purge existing TLB entries in set_pte_at and ptep_set_wrprotect - take 2

[Helge Deller]

Hi Rolf,

On 01/16/2013 10:44 PM, Rolf Eike Beer wrote:
> James Bottomley wrote:
>> On Mon, 2013-01-14 at 19:45 -0500, John David Anglin wrote:
>>> Attached is new version of patch which no longer saves and restores
>>> %sr1.  Lightly tested with a GCC build
>>> and a build of Debian gtk+3.0 package.
>>
>> I can merge it with your previous changelog, but for ease of git users,
>> the accepted format is
> 
>> v2: Attached is new version of patch which no longer saves and
>> restores  %sr1.  Lightly tested with a GCC build and a build of
>> Debian gtk+3.0 package.
> 
> Do we want that to go back into stable kernels?

Short answer: Yes.

Long answer:
First we need those patches upstream - which is what did not happened yet.
Then I wanted to take a look what we can send to stable.

Helge