3.10-rc ppc64 corrupts usermem when swapping

3.10-rc ppc64 corrupts usermem when swapping

[Hugh Dickins]

Running my favourite swapping load (repeated make -j20 kernel builds
in tmpfs in parallel with repeated make -j20 kernel builds in ext4 on
loop on tmpfs file, all limited by mem=700M and swap 1.5G) on 3.10-rc
on PowerMac G5, the test dies with corrupted usermem after a few hours.

Variously, segmentation fault or Binutils assertion fail or gcc Internal
error in either or both builds: usually signs of swapping or TLB flushing
gone wrong.  Sometimes the tmpfs build breaks first, sometimes the ext4 on
loop on tmpfs, so at least it looks unrelated to loop.  No problem on x86.

This is 64-bit kernel but 4k pages and old SuSE 11.1 32-bit userspace.

I've just finished a manual bisection on arch/powerpc/mm (which might
have been a wrong guess, but has paid off): the first bad commit is
7e74c3921ad9610c0b49f28b8fc69f7480505841
"powerpc: Fix hpte_decode to use the correct decoding for page sizes".

I don't know if it's actually swapping to swap that's triggering the
problem, or a more general page reclaim or TLB flush problem.  I hit
it originally when trying to test Mel Gorman's pagevec series on top
of 3.10-rc; and though I then reproduced it without that series, it
did seem to take much longer: so I have been applying Mel's series to
speed up each step of the bisection.  But if I went back again, might
find it was just chance that I hit it sooner with Mel's series than
without.  So, you're probably safe to ignore that detail, but I
mention it just in case it turns out to have some relevance.

Something else peculiar that I've been doing in these runs, may or may
not be relevant: I've been running swapon and swapoff repeatedly in the
background, so that we're doing swapoff even while busy building.

I probably can't go into much more detail on the test (it's hard
to get the balance right, to be swapping rather than OOMing or just
running without reclaim), but can test any patches you'd like me to
try (though it may take 24 hours for me to report back usefully).

Thanks,
Hugh

Re: 3.10-rc ppc64 corrupts usermem when swapping

[Benjamin Herrenschmidt]

On Wed, 2013-05-29 at 22:47 -0700, Hugh Dickins wrote:
> Running my favourite swapping load (repeated make -j20 kernel builds
> in tmpfs in parallel with repeated make -j20 kernel builds in ext4 on
> loop on tmpfs file, all limited by mem=700M and swap 1.5G) on 3.10-rc
> on PowerMac G5, the test dies with corrupted usermem after a few hours.
> 
> Variously, segmentation fault or Binutils assertion fail or gcc Internal
> error in either or both builds: usually signs of swapping or TLB flushing
> gone wrong.  Sometimes the tmpfs build breaks first, sometimes the ext4 on
> loop on tmpfs, so at least it looks unrelated to loop.  No problem on x86.
> 
> This is 64-bit kernel but 4k pages and old SuSE 11.1 32-bit userspace.
> 
> I've just finished a manual bisection on arch/powerpc/mm (which might
> have been a wrong guess, but has paid off): the first bad commit is
> 7e74c3921ad9610c0b49f28b8fc69f7480505841
> "powerpc: Fix hpte_decode to use the correct decoding for page sizes".

Ok, I have other reasons to think is wrong. I debugged a case last week
where after kexec we still had stale TLB entries, due to the TLB cleanup
not working.

Thanks for doing that bisection ! I'll investigate ASAP (though it will
probably have to wait for tomorrow unless Paul beats me to it)

> I don't know if it's actually swapping to swap that's triggering the
> problem, or a more general page reclaim or TLB flush problem.  I hit
> it originally when trying to test Mel Gorman's pagevec series on top
> of 3.10-rc; and though I then reproduced it without that series, it
> did seem to take much longer: so I have been applying Mel's series to
> speed up each step of the bisection.  But if I went back again, might
> find it was just chance that I hit it sooner with Mel's series than
> without.  So, you're probably safe to ignore that detail, but I
> mention it just in case it turns out to have some relevance.
> 
> Something else peculiar that I've been doing in these runs, may or may
> not be relevant: I've been running swapon and swapoff repeatedly in the
> background, so that we're doing swapoff even while busy building.
> 
> I probably can't go into much more detail on the test (it's hard
> to get the balance right, to be swapping rather than OOMing or just
> running without reclaim), but can test any patches you'd like me to
> try (though it may take 24 hours for me to report back usefully).

I think it's just failing to invalidate the TLB properly. At least one
bug I can spot just looking at it:

static void native_hpte_invalidate(unsigned long slot, unsigned long vpn,
				   int psize, int ssize, int local)

   .../...

	native_lock_hpte(hptep);
	hpte_v = hptep->v;

	actual_psize = hpte_actual_psize(hptep, psize);
	if (actual_psize < 0) {
		native_unlock_hpte(hptep);
		local_irq_restore(flags);
		return;
	}

That's wrong. We must still perform the TLB invalidation even if the
hash PTE is empty.

In fact, Aneesh, this is a problem with MPSS for your THP work, I just
thought about it.

The reason is that if a hash bucket gets full, we "evict" a more/less
random entry from it. When we do that we don't invalidate the TLB
(hpte_remove) because we assume the old translation is still technically
"valid".

However that means that an hpte_invalidate *must* invalidate the TLB
later on even if it's not hitting the right entry in the hash.

However, I can see why that cannot work with THP/MPSS since you have no
way to know the page size from the PTE anymore....

So my question is, apart from hpte_decode used by kexec, which I will
fix by just blowing the whole TLB when not running phyp, why do you need
the "actual" size in invalidate and updatepp ? You really can't rely on
the size passed by the upper layers ?

Cheers,
Ben.

Re: 3.10-rc ppc64 corrupts usermem when swapping

[Aneesh Kumar K.V]

Benjamin Herrenschmidt <benh@au1.ibm.com> writes:

> On Wed, 2013-05-29 at 22:47 -0700, Hugh Dickins wrote:
>> Running my favourite swapping load (repeated make -j20 kernel builds
>> in tmpfs in parallel with repeated make -j20 kernel builds in ext4 on
>> loop on tmpfs file, all limited by mem=700M and swap 1.5G) on 3.10-rc
>> on PowerMac G5, the test dies with corrupted usermem after a few hours.
>> 
>> Variously, segmentation fault or Binutils assertion fail or gcc Internal
>> error in either or both builds: usually signs of swapping or TLB flushing
>> gone wrong.  Sometimes the tmpfs build breaks first, sometimes the ext4 on
>> loop on tmpfs, so at least it looks unrelated to loop.  No problem on x86.
>> 
>> This is 64-bit kernel but 4k pages and old SuSE 11.1 32-bit userspace.
>> 
>> I've just finished a manual bisection on arch/powerpc/mm (which might
>> have been a wrong guess, but has paid off): the first bad commit is
>> 7e74c3921ad9610c0b49f28b8fc69f7480505841
>> "powerpc: Fix hpte_decode to use the correct decoding for page sizes".
>
> Ok, I have other reasons to think is wrong. I debugged a case last week
> where after kexec we still had stale TLB entries, due to the TLB cleanup
> not working.
>
> Thanks for doing that bisection ! I'll investigate ASAP (though it will
> probably have to wait for tomorrow unless Paul beats me to it)
>
>> I don't know if it's actually swapping to swap that's triggering the
>> problem, or a more general page reclaim or TLB flush problem.  I hit
>> it originally when trying to test Mel Gorman's pagevec series on top
>> of 3.10-rc; and though I then reproduced it without that series, it
>> did seem to take much longer: so I have been applying Mel's series to
>> speed up each step of the bisection.  But if I went back again, might
>> find it was just chance that I hit it sooner with Mel's series than
>> without.  So, you're probably safe to ignore that detail, but I
>> mention it just in case it turns out to have some relevance.
>> 
>> Something else peculiar that I've been doing in these runs, may or may
>> not be relevant: I've been running swapon and swapoff repeatedly in the
>> background, so that we're doing swapoff even while busy building.
>> 
>> I probably can't go into much more detail on the test (it's hard
>> to get the balance right, to be swapping rather than OOMing or just
>> running without reclaim), but can test any patches you'd like me to
>> try (though it may take 24 hours for me to report back usefully).
>
> I think it's just failing to invalidate the TLB properly. At least one
> bug I can spot just looking at it:
>
> static void native_hpte_invalidate(unsigned long slot, unsigned long vpn,
> 				   int psize, int ssize, int local)
>
>    .../...
>
> 	native_lock_hpte(hptep);
> 	hpte_v = hptep->v;
>
> 	actual_psize = hpte_actual_psize(hptep, psize);
> 	if (actual_psize < 0) {
> 		native_unlock_hpte(hptep);
> 		local_irq_restore(flags);
> 		return;
> 	}
>
> That's wrong. We must still perform the TLB invalidation even if the
> hash PTE is empty.
>
> In fact, Aneesh, this is a problem with MPSS for your THP work, I just
> thought about it.
>
> The reason is that if a hash bucket gets full, we "evict" a more/less
> random entry from it. When we do that we don't invalidate the TLB
> (hpte_remove) because we assume the old translation is still technically
> "valid".
>

Hmm that is correct, I missed that. But to do a tlb invalidate we need
both base and actual page size. One of the reason i didn't update the
hpte_invalidate callback to take both the page sizes was because, PAPR
didn't need that for invalidate (H_REMOVE). hpte_remove did result in a
tlb invalidate there. 


> However that means that an hpte_invalidate *must* invalidate the TLB
> later on even if it's not hitting the right entry in the hash.
>
> However, I can see why that cannot work with THP/MPSS since you have no
> way to know the page size from the PTE anymore....
>
> So my question is, apart from hpte_decode used by kexec, which I will
> fix by just blowing the whole TLB when not running phyp, why do you need
> the "actual" size in invalidate and updatepp ? You really can't rely on
> the size passed by the upper layers ?

So for upstream I have below which should address the
above. Meanwhile I will see what the impact would be to do a tlb
invalidate in hpte_remove, so that we can keep both lpar and native
changes similar.


diff --git a/arch/powerpc/mm/hash_native_64.c b/arch/powerpc/mm/hash_native_64.c
index 6a2aead..6d1bd81 100644
--- a/arch/powerpc/mm/hash_native_64.c
+++ b/arch/powerpc/mm/hash_native_64.c
@@ -336,11 +336,19 @@ static long native_hpte_updatepp(unsigned long slot, unsigned long newpp,
 
 	hpte_v = hptep->v;
 	actual_psize = hpte_actual_psize(hptep, psize);
+	/*
+	 * We need to invalidate the TLB always because hpte_remove doesn't do
+	 * a tlb invalidate. If a hash bucket gets full, we "evict" a more/less
+	 * random entry from it. When we do that we don't invalidate the TLB
+	 * (hpte_remove) because we assume the old translation is still technically
+	 * "valid".
+	 */
 	if (actual_psize < 0) {
-		native_unlock_hpte(hptep);
-		return -1;
+		/* FIXME!!, will fail with when we enable hugepage support */
+		actual_psize = psize;
+		ret = -1;
+		goto err_out;
 	}
-	/* Even if we miss, we need to invalidate the TLB */
 	if (!HPTE_V_COMPARE(hpte_v, want_v)) {
 		DBG_LOW(" -> miss\n");
 		ret = -1;
@@ -350,6 +358,7 @@ static long native_hpte_updatepp(unsigned long slot, unsigned long newpp,
 		hptep->r = (hptep->r & ~(HPTE_R_PP | HPTE_R_N)) |
 			(newpp & (HPTE_R_PP | HPTE_R_N | HPTE_R_C));
 	}
+err_out:
 	native_unlock_hpte(hptep);
 
 	/* Ensure it is out of the tlb too. */
@@ -408,8 +417,9 @@ static void native_hpte_updateboltedpp(unsigned long newpp, unsigned long ea,
 		panic("could not find page to bolt\n");
 	hptep = htab_address + slot;
 	actual_psize = hpte_actual_psize(hptep, psize);
+	/* FIXME!! can this happen for bolted entry ? */
 	if (actual_psize < 0)
-		return;
+		actual_psize = psize;
 
 	/* Update the HPTE */
 	hptep->r = (hptep->r & ~(HPTE_R_PP | HPTE_R_N)) |
@@ -437,21 +447,28 @@ static void native_hpte_invalidate(unsigned long slot, unsigned long vpn,
 	hpte_v = hptep->v;
 
 	actual_psize = hpte_actual_psize(hptep, psize);
+	/*
+	 * We need to invalidate the TLB always because hpte_remove doesn't do
+	 * a tlb invalidate. If a hash bucket gets full, we "evict" a more/less
+	 * random entry from it. When we do that we don't invalidate the TLB
+	 * (hpte_remove) because we assume the old translation is still technically
+	 * "valid".
+	 */
 	if (actual_psize < 0) {
+		/* FIXME!!, will fail with when we enable hugepage support */
+		actual_psize = psize;
 		native_unlock_hpte(hptep);
-		local_irq_restore(flags);
-		return;
+		goto err_out;
 	}
-	/* Even if we miss, we need to invalidate the TLB */
 	if (!HPTE_V_COMPARE(hpte_v, want_v))
 		native_unlock_hpte(hptep);
 	else
 		/* Invalidate the hpte. NOTE: this also unlocks it */
 		hptep->v = 0;
 
+err_out:
 	/* Invalidate the TLB */
 	tlbie(vpn, psize, actual_psize, ssize, local);
-
 	local_irq_restore(flags);
 }
 

Re: 3.10-rc ppc64 corrupts usermem when swapping

[Benjamin Herrenschmidt]

On Thu, 2013-05-30 at 13:57 +0530, Aneesh Kumar K.V wrote:
> +               /* FIXME!!, will fail with when we enable hugepage
> support */

Just fix that to say "Transparent huge pages" as normal huge pages
should work fine unless I'm missing something.

Hugh, any chance you can give that a spin ? 

Cheers,
Ben.

Re: 3.10-rc ppc64 corrupts usermem when swapping

[Hugh Dickins]

On Thu, 30 May 2013, Benjamin Herrenschmidt wrote:
> On Thu, 2013-05-30 at 13:57 +0530, Aneesh Kumar K.V wrote:
> > +               /* FIXME!!, will fail with when we enable hugepage
> > support */
> 
> Just fix that to say "Transparent huge pages" as normal huge pages
> should work fine unless I'm missing something.
> 
> Hugh, any chance you can give that a spin ? 

Sure, it's now under way.  If all goes well, I'll give you a
progress report in about 15 hours time; but given the variance in
how long it took to hit, I won't feel fully confident until this
time tomorrow, when I'll update you again.

Thank you both for the great response!

Hugh

Re: 3.10-rc ppc64 corrupts usermem when swapping

[Hugh Dickins]

On Thu, 30 May 2013, Hugh Dickins wrote:
> On Thu, 30 May 2013, Benjamin Herrenschmidt wrote:
> > On Thu, 2013-05-30 at 13:57 +0530, Aneesh Kumar K.V wrote:
> > > +               /* FIXME!!, will fail with when we enable hugepage
> > > support */
> > 
> > Just fix that to say "Transparent huge pages" as normal huge pages
> > should work fine unless I'm missing something.
> > 
> > Hugh, any chance you can give that a spin ? 
> 
> Sure, it's now under way.  If all goes well, I'll give you a
> progress report in about 15 hours time; but given the variance in
> how long it took to hit, I won't feel fully confident until this
> time tomorrow, when I'll update you again.

Still running fine.  I'll leave it running a few more hours to make
sure, and then try switching to the patch Aneesh sent afterwards -
or say if you'd prefer me to switch over to that one immediately.

Hugh

Re: 3.10-rc ppc64 corrupts usermem when swapping

[Benjamin Herrenschmidt]

On Thu, 2013-05-30 at 22:05 -0700, Hugh Dickins wrote:
> > Sure, it's now under way.  If all goes well, I'll give you a
> > progress report in about 15 hours time; but given the variance in
> > how long it took to hit, I won't feel fully confident until this
> > time tomorrow, when I'll update you again.
> 
> Still running fine.  I'll leave it running a few more hours to make
> sure, and then try switching to the patch Aneesh sent afterwards -
> or say if you'd prefer me to switch over to that one immediately.

The patch you are running on is what I'll send to Linus for 3.10 (+/-
cosmetics). Aneesh second patch is a much larger rework which will be
needed for THP but that will wait for 3.11. I'm happy for you to test it
but I first want to make sure it's solid with the 3.10 fix :-)

Thanks !

Cheers,
Ben.

Re: 3.10-rc ppc64 corrupts usermem when swapping

[Hugh Dickins]

On Fri, 31 May 2013, Benjamin Herrenschmidt wrote:
> On Thu, 2013-05-30 at 22:05 -0700, Hugh Dickins wrote:
> > > Sure, it's now under way.  If all goes well, I'll give you a
> > > progress report in about 15 hours time; but given the variance in
> > > how long it took to hit, I won't feel fully confident until this
> > > time tomorrow, when I'll update you again.
> > 
> > Still running fine.  I'll leave it running a few more hours to make
> > sure, and then try switching to the patch Aneesh sent afterwards -
> > or say if you'd prefer me to switch over to that one immediately.
> 
> The patch you are running on is what I'll send to Linus for 3.10 (+/-
> cosmetics). Aneesh second patch is a much larger rework which will be
> needed for THP but that will wait for 3.11. I'm happy for you to test it
> but I first want to make sure it's solid with the 3.10 fix :-)

That makes sense.
The first patch ran fine without incident for 25 hours, I say it's good.
I've now stopped that run and started another with the second patch in.

Hugh

Re: 3.10-rc ppc64 corrupts usermem when swapping

[Benjamin Herrenschmidt]

On Fri, 2013-05-31 at 14:45 +0530, Aneesh Kumar K.V wrote:

> > The patch you are running on is what I'll send to Linus for 3.10 (+/-
> > cosmetics). Aneesh second patch is a much larger rework which will be
> > needed for THP but that will wait for 3.11. I'm happy for you to test it
> > but I first want to make sure it's solid with the 3.10 fix :-)

BTW. One concern I still have is that Hugh identified the bad commit
to be:

7e74c3921ad9610c0b49f28b8fc69f7480505841
"powerpc: Fix hpte_decode to use the correct decoding for page sizes".

However, you introduce the return on HPTE not found earlier, in

b1022fbd293564de91596b8775340cf41ad5214c
"powerpc: Decode the pte-lp-encoding bits correctly."

So while I'm still happy with the current band-aid for 3.10 and am
about to send it to Linus, the above *does* seem to indicate that
there is also something wrong with the "Fix hpte_decode..." commit,
which might not actually get the page size right...

Can you investigate ?

Cheers,
Ben.

Re: 3.10-rc ppc64 corrupts usermem when swapping

[Aneesh Kumar K.V]

Benjamin Herrenschmidt <benh@au1.ibm.com> writes:

> On Fri, 2013-05-31 at 14:45 +0530, Aneesh Kumar K.V wrote:
>
>> > The patch you are running on is what I'll send to Linus for 3.10 (+/-
>> > cosmetics). Aneesh second patch is a much larger rework which will be
>> > needed for THP but that will wait for 3.11. I'm happy for you to test it
>> > but I first want to make sure it's solid with the 3.10 fix :-)
>
> BTW. One concern I still have is that Hugh identified the bad commit
> to be:
>
> 7e74c3921ad9610c0b49f28b8fc69f7480505841
> "powerpc: Fix hpte_decode to use the correct decoding for page sizes".
>
> However, you introduce the return on HPTE not found earlier, in
>
> b1022fbd293564de91596b8775340cf41ad5214c
> "powerpc: Decode the pte-lp-encoding bits correctly."
>
> So while I'm still happy with the current band-aid for 3.10 and am
> about to send it to Linus, the above *does* seem to indicate that
> there is also something wrong with the "Fix hpte_decode..." commit,
> which might not actually get the page size right...
>
> Can you investigate ?

7e74c3921ad9610c0b49f28b8fc69f7480505841 
"powerpc: Fix hpte_decode to use the correct decoding for page sizes"
changes should only impact hpte_decode. We don't change the details
of hpte_actual_psize at all in this patch. That means we should see a
difference only with kexec right ?.

Hugh,

Will you be able to double check whether
7e74c3921ad9610c0b49f28b8fc69f7480505841 is the bad commit. The one
before that is what we changed in the patch that fixed your problem.

-aneesh

Re: 3.10-rc ppc64 corrupts usermem when swapping

[Hugh Dickins]

On Sun, 2 Jun 2013, Aneesh Kumar K.V wrote:
> Benjamin Herrenschmidt <benh@au1.ibm.com> writes:
> > On Fri, 2013-05-31 at 14:45 +0530, Aneesh Kumar K.V wrote:
> >
> >> > The patch you are running on is what I'll send to Linus for 3.10 (+/-
> >> > cosmetics). Aneesh second patch is a much larger rework which will be
> >> > needed for THP but that will wait for 3.11. I'm happy for you to test it
> >> > but I first want to make sure it's solid with the 3.10 fix :-)
> >
> > BTW. One concern I still have is that Hugh identified the bad commit
> > to be:
> >
> > 7e74c3921ad9610c0b49f28b8fc69f7480505841
> > "powerpc: Fix hpte_decode to use the correct decoding for page sizes".
> >
> > However, you introduce the return on HPTE not found earlier, in
> >
> > b1022fbd293564de91596b8775340cf41ad5214c
> > "powerpc: Decode the pte-lp-encoding bits correctly."
> >
> > So while I'm still happy with the current band-aid for 3.10 and am
> > about to send it to Linus, the above *does* seem to indicate that
> > there is also something wrong with the "Fix hpte_decode..." commit,
> > which might not actually get the page size right...
> >
> > Can you investigate ?
> 
> 7e74c3921ad9610c0b49f28b8fc69f7480505841 
> "powerpc: Fix hpte_decode to use the correct decoding for page sizes"
> changes should only impact hpte_decode. We don't change the details
> of hpte_actual_psize at all in this patch. That means we should see a
> difference only with kexec right ?.
> 
> Hugh,
> 
> Will you be able to double check whether
> 7e74c3921ad9610c0b49f28b8fc69f7480505841 is the bad commit. The one
> before that is what we changed in the patch that fixed your problem.

You are absolutely right.  I just set b1022fbd29 going, expecting
to answer you tomorrow: but got a Segmentation fault in 20 minutes
(quicker than ever seen before).  It looks as if I was running some
other kernel for the last stage of my bisection: I can't see how that
came about, but it's not very interesting now - you got it right.

Prior to trying that, I had been running your second patch, 9f70fd8cfe,
and that tested out successfully for 50 hours before I stopped it.

Hugh

Re: 3.10-rc ppc64 corrupts usermem when swapping

[Aneesh Kumar K.V]

"Aneesh Kumar K.V" <aneesh.kumar@linux.vnet.ibm.com> writes:

> Benjamin Herrenschmidt <benh@au1.ibm.com> writes:
>
>> On Wed, 2013-05-29 at 22:47 -0700, Hugh Dickins wrote:
>>> Running my favourite swapping load (repeated make -j20 kernel builds
>>> in tmpfs in parallel with repeated make -j20 kernel builds in ext4 on
>>> loop on tmpfs file, all limited by mem=700M and swap 1.5G) on 3.10-rc
>>> on PowerMac G5, the test dies with corrupted usermem after a few hours.
>>> 
>>> Variously, segmentation fault or Binutils assertion fail or gcc Internal
>>> error in either or both builds: usually signs of swapping or TLB flushing
>>> gone wrong.  Sometimes the tmpfs build breaks first, sometimes the ext4 on
>>> loop on tmpfs, so at least it looks unrelated to loop.  No problem on x86.
>>> 
>>> This is 64-bit kernel but 4k pages and old SuSE 11.1 32-bit userspace.
>>> 
>>> I've just finished a manual bisection on arch/powerpc/mm (which might
>>> have been a wrong guess, but has paid off): the first bad commit is
>>> 7e74c3921ad9610c0b49f28b8fc69f7480505841
>>> "powerpc: Fix hpte_decode to use the correct decoding for page sizes".
>>
>> Ok, I have other reasons to think is wrong. I debugged a case last week
>> where after kexec we still had stale TLB entries, due to the TLB cleanup
>> not working.
>>
>> Thanks for doing that bisection ! I'll investigate ASAP (though it will
>> probably have to wait for tomorrow unless Paul beats me to it)
>>
>>> I don't know if it's actually swapping to swap that's triggering the
>>> problem, or a more general page reclaim or TLB flush problem.  I hit
>>> it originally when trying to test Mel Gorman's pagevec series on top
>>> of 3.10-rc; and though I then reproduced it without that series, it
>>> did seem to take much longer: so I have been applying Mel's series to
>>> speed up each step of the bisection.  But if I went back again, might
>>> find it was just chance that I hit it sooner with Mel's series than
>>> without.  So, you're probably safe to ignore that detail, but I
>>> mention it just in case it turns out to have some relevance.
>>> 
>>> Something else peculiar that I've been doing in these runs, may or may
>>> not be relevant: I've been running swapon and swapoff repeatedly in the
>>> background, so that we're doing swapoff even while busy building.
>>> 
>>> I probably can't go into much more detail on the test (it's hard
>>> to get the balance right, to be swapping rather than OOMing or just
>>> running without reclaim), but can test any patches you'd like me to
>>> try (though it may take 24 hours for me to report back usefully).
>>
>> I think it's just failing to invalidate the TLB properly. At least one
>> bug I can spot just looking at it:
>>
>> static void native_hpte_invalidate(unsigned long slot, unsigned long vpn,
>> 				   int psize, int ssize, int local)
>>
>>    .../...
>>
>> 	native_lock_hpte(hptep);
>> 	hpte_v = hptep->v;
>>
>> 	actual_psize = hpte_actual_psize(hptep, psize);
>> 	if (actual_psize < 0) {
>> 		native_unlock_hpte(hptep);
>> 		local_irq_restore(flags);
>> 		return;
>> 	}
>>
>> That's wrong. We must still perform the TLB invalidation even if the
>> hash PTE is empty.
>>
>> In fact, Aneesh, this is a problem with MPSS for your THP work, I just
>> thought about it.
>>
>> The reason is that if a hash bucket gets full, we "evict" a more/less
>> random entry from it. When we do that we don't invalidate the TLB
>> (hpte_remove) because we assume the old translation is still technically
>> "valid".
>>
>
> Hmm that is correct, I missed that. But to do a tlb invalidate we need
> both base and actual page size. One of the reason i didn't update the
> hpte_invalidate callback to take both the page sizes was because, PAPR
> didn't need that for invalidate (H_REMOVE). hpte_remove did result in a
> tlb invalidate there. 
>
>
>> However that means that an hpte_invalidate *must* invalidate the TLB
>> later on even if it's not hitting the right entry in the hash.
>>
>> However, I can see why that cannot work with THP/MPSS since you have no
>> way to know the page size from the PTE anymore....
>>
>> So my question is, apart from hpte_decode used by kexec, which I will
>> fix by just blowing the whole TLB when not running phyp, why do you need
>> the "actual" size in invalidate and updatepp ? You really can't rely on
>> the size passed by the upper layers ?
>
> So for upstream I have below which should address the
> above. Meanwhile I will see what the impact would be to do a tlb
> invalidate in hpte_remove, so that we can keep both lpar and native
> changes similar.
>

How about the below ?

commit 9f70fd8cfeb7fca33ef8453197b76df46c860b52
Author: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Date:   Thu May 30 20:09:58 2013 +0530

    powerpc/mm: Always invalidate tlb on hpte invalidate and update
    
    If a hash bucket gets full, we "evict" a more/less random entry from it.
    When we do that we don't invalidate the TLB (hpte_remove) because we assume
    the old translation is still technically "valid". This implies that when
    we are invalidating or updating pte, even if HPTE entry is not valid
    we should do a tlb invalidate.
    
    Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>

diff --git a/arch/powerpc/include/asm/machdep.h b/arch/powerpc/include/asm/machdep.h
index 92386fc..801e3c6 100644
--- a/arch/powerpc/include/asm/machdep.h
+++ b/arch/powerpc/include/asm/machdep.h
@@ -36,13 +36,13 @@ struct machdep_calls {
 #ifdef CONFIG_PPC64
 	void            (*hpte_invalidate)(unsigned long slot,
 					   unsigned long vpn,
-					   int psize, int ssize,
-					   int local);
+					   int bpsize, int apsize,
+					   int ssize, int local);
 	long		(*hpte_updatepp)(unsigned long slot, 
 					 unsigned long newpp, 
 					 unsigned long vpn,
-					 int psize, int ssize,
-					 int local);
+					 int bpsize, int apsize,
+					 int ssize, int local);
 	void            (*hpte_updateboltedpp)(unsigned long newpp, 
 					       unsigned long ea,
 					       int psize, int ssize);
diff --git a/arch/powerpc/kvm/book3s_64_mmu_host.c b/arch/powerpc/kvm/book3s_64_mmu_host.c
index 3a9a1ac..176d3fd 100644
--- a/arch/powerpc/kvm/book3s_64_mmu_host.c
+++ b/arch/powerpc/kvm/book3s_64_mmu_host.c
@@ -34,7 +34,7 @@
 void kvmppc_mmu_invalidate_pte(struct kvm_vcpu *vcpu, struct hpte_cache *pte)
 {
 	ppc_md.hpte_invalidate(pte->slot, pte->host_vpn,
-			       MMU_PAGE_4K, MMU_SEGSIZE_256M,
+			       MMU_PAGE_4K, MMU_PAGE_4K, MMU_SEGSIZE_256M,
 			       false);
 }
 
diff --git a/arch/powerpc/mm/hash_low_64.S b/arch/powerpc/mm/hash_low_64.S
index 0e980ac..d3cbda6 100644
--- a/arch/powerpc/mm/hash_low_64.S
+++ b/arch/powerpc/mm/hash_low_64.S
@@ -289,9 +289,10 @@ htab_modify_pte:
 
 	/* Call ppc_md.hpte_updatepp */
 	mr	r5,r29			/* vpn */
-	li	r6,MMU_PAGE_4K		/* page size */
-	ld	r7,STK_PARAM(R9)(r1)	/* segment size */
-	ld	r8,STK_PARAM(R8)(r1)	/* get "local" param */
+	li	r6,MMU_PAGE_4K		/* base page size */
+	li	r7,MMU_PAGE_4K		/* actual page size */
+	ld	r8,STK_PARAM(R9)(r1)	/* segment size */
+	ld	r9,STK_PARAM(R8)(r1)	/* get "local" param */
 _GLOBAL(htab_call_hpte_updatepp)
 	bl	.			/* Patched by htab_finish_init() */
 
@@ -649,9 +650,10 @@ htab_modify_pte:
 
 	/* Call ppc_md.hpte_updatepp */
 	mr	r5,r29			/* vpn */
-	li	r6,MMU_PAGE_4K		/* page size */
-	ld	r7,STK_PARAM(R9)(r1)	/* segment size */
-	ld	r8,STK_PARAM(R8)(r1)	/* get "local" param */
+	li	r6,MMU_PAGE_4K		/* base page size */
+	li	r7,MMU_PAGE_4K		/* actual page size */
+	ld	r8,STK_PARAM(R9)(r1)	/* segment size */
+	ld	r9,STK_PARAM(R8)(r1)	/* get "local" param */
 _GLOBAL(htab_call_hpte_updatepp)
 	bl	.			/* patched by htab_finish_init() */
 
@@ -937,9 +939,10 @@ ht64_modify_pte:
 
 	/* Call ppc_md.hpte_updatepp */
 	mr	r5,r29			/* vpn */
-	li	r6,MMU_PAGE_64K
-	ld	r7,STK_PARAM(R9)(r1)	/* segment size */
-	ld	r8,STK_PARAM(R8)(r1)	/* get "local" param */
+	li	r6,MMU_PAGE_64K		/* base page size */
+	li	r7,MMU_PAGE_64K		/* actual page size */
+	ld	r8,STK_PARAM(R9)(r1)	/* segment size */
+	ld	r9,STK_PARAM(R8)(r1)	/* get "local" param */
 _GLOBAL(ht64_call_hpte_updatepp)
 	bl	.			/* patched by htab_finish_init() */
 
diff --git a/arch/powerpc/mm/hash_native_64.c b/arch/powerpc/mm/hash_native_64.c
index 6a2aead..121c8ef 100644
--- a/arch/powerpc/mm/hash_native_64.c
+++ b/arch/powerpc/mm/hash_native_64.c
@@ -273,61 +273,15 @@ static long native_hpte_remove(unsigned long hpte_group)
 	return i;
 }
 
-static inline int __hpte_actual_psize(unsigned int lp, int psize)
-{
-	int i, shift;
-	unsigned int mask;
-
-	/* start from 1 ignoring MMU_PAGE_4K */
-	for (i = 1; i < MMU_PAGE_COUNT; i++) {
-
-		/* invalid penc */
-		if (mmu_psize_defs[psize].penc[i] == -1)
-			continue;
-		/*
-		 * encoding bits per actual page size
-		 *        PTE LP     actual page size
-		 *    rrrr rrrz		>=8KB
-		 *    rrrr rrzz		>=16KB
-		 *    rrrr rzzz		>=32KB
-		 *    rrrr zzzz		>=64KB
-		 * .......
-		 */
-		shift = mmu_psize_defs[i].shift - LP_SHIFT;
-		if (shift > LP_BITS)
-			shift = LP_BITS;
-		mask = (1 << shift) - 1;
-		if ((lp & mask) == mmu_psize_defs[psize].penc[i])
-			return i;
-	}
-	return -1;
-}
-
-static inline int hpte_actual_psize(struct hash_pte *hptep, int psize)
-{
-	/* Look at the 8 bit LP value */
-	unsigned int lp = (hptep->r >> LP_SHIFT) & ((1 << LP_BITS) - 1);
-
-	if (!(hptep->v & HPTE_V_VALID))
-		return -1;
-
-	/* First check if it is large page */
-	if (!(hptep->v & HPTE_V_LARGE))
-		return MMU_PAGE_4K;
-
-	return __hpte_actual_psize(lp, psize);
-}
-
 static long native_hpte_updatepp(unsigned long slot, unsigned long newpp,
-				 unsigned long vpn, int psize, int ssize,
-				 int local)
+				 unsigned long vpn, int bpsize,
+				 int apsize, int ssize, int local)
 {
 	struct hash_pte *hptep = htab_address + slot;
 	unsigned long hpte_v, want_v;
 	int ret = 0;
-	int actual_psize;
 
-	want_v = hpte_encode_avpn(vpn, psize, ssize);
+	want_v = hpte_encode_avpn(vpn, bpsize, ssize);
 
 	DBG_LOW("    update(vpn=%016lx, avpnv=%016lx, group=%lx, newpp=%lx)",
 		vpn, want_v & HPTE_V_AVPN, slot, newpp);
@@ -335,13 +289,14 @@ static long native_hpte_updatepp(unsigned long slot, unsigned long newpp,
 	native_lock_hpte(hptep);
 
 	hpte_v = hptep->v;
-	actual_psize = hpte_actual_psize(hptep, psize);
-	if (actual_psize < 0) {
-		native_unlock_hpte(hptep);
-		return -1;
-	}
-	/* Even if we miss, we need to invalidate the TLB */
-	if (!HPTE_V_COMPARE(hpte_v, want_v)) {
+	/*
+	 * We need to invalidate the TLB always because hpte_remove doesn't do
+	 * a tlb invalidate. If a hash bucket gets full, we "evict" a more/less
+	 * random entry from it. When we do that we don't invalidate the TLB
+	 * (hpte_remove) because we assume the old translation is still technically
+	 * "valid".
+	 */
+	if (!HPTE_V_COMPARE(hpte_v, want_v) || !(hpte_v & HPTE_V_VALID)) {
 		DBG_LOW(" -> miss\n");
 		ret = -1;
 	} else {
@@ -353,7 +308,7 @@ static long native_hpte_updatepp(unsigned long slot, unsigned long newpp,
 	native_unlock_hpte(hptep);
 
 	/* Ensure it is out of the tlb too. */
-	tlbie(vpn, psize, actual_psize, ssize, local);
+	tlbie(vpn, bpsize, apsize, ssize, local);
 
 	return ret;
 }
@@ -394,7 +349,6 @@ static long native_hpte_find(unsigned long vpn, int psize, int ssize)
 static void native_hpte_updateboltedpp(unsigned long newpp, unsigned long ea,
 				       int psize, int ssize)
 {
-	int actual_psize;
 	unsigned long vpn;
 	unsigned long vsid;
 	long slot;
@@ -407,54 +361,82 @@ static void native_hpte_updateboltedpp(unsigned long newpp, unsigned long ea,
 	if (slot == -1)
 		panic("could not find page to bolt\n");
 	hptep = htab_address + slot;
-	actual_psize = hpte_actual_psize(hptep, psize);
-	if (actual_psize < 0)
-		return;
 
 	/* Update the HPTE */
 	hptep->r = (hptep->r & ~(HPTE_R_PP | HPTE_R_N)) |
 		(newpp & (HPTE_R_PP | HPTE_R_N));
-
-	/* Ensure it is out of the tlb too. */
-	tlbie(vpn, psize, actual_psize, ssize, 0);
+	/*
+	 * Ensure it is out of the tlb too. Bolted entries base and
+	 * actual page size will be same.
+	 */
+	tlbie(vpn, psize, psize, ssize, 0);
 }
 
 static void native_hpte_invalidate(unsigned long slot, unsigned long vpn,
-				   int psize, int ssize, int local)
+				   int bpsize, int apsize, int ssize, int local)
 {
 	struct hash_pte *hptep = htab_address + slot;
 	unsigned long hpte_v;
 	unsigned long want_v;
 	unsigned long flags;
-	int actual_psize;
 
 	local_irq_save(flags);
 
 	DBG_LOW("    invalidate(vpn=%016lx, hash: %lx)\n", vpn, slot);
 
-	want_v = hpte_encode_avpn(vpn, psize, ssize);
+	want_v = hpte_encode_avpn(vpn, bpsize, ssize);
 	native_lock_hpte(hptep);
 	hpte_v = hptep->v;
 
-	actual_psize = hpte_actual_psize(hptep, psize);
-	if (actual_psize < 0) {
-		native_unlock_hpte(hptep);
-		local_irq_restore(flags);
-		return;
-	}
-	/* Even if we miss, we need to invalidate the TLB */
-	if (!HPTE_V_COMPARE(hpte_v, want_v))
+	/*
+	 * We need to invalidate the TLB always because hpte_remove doesn't do
+	 * a tlb invalidate. If a hash bucket gets full, we "evict" a more/less
+	 * random entry from it. When we do that we don't invalidate the TLB
+	 * (hpte_remove) because we assume the old translation is still technically
+	 * "valid".
+	 */
+	if (!HPTE_V_COMPARE(hpte_v, want_v) || !(hpte_v & HPTE_V_VALID))
 		native_unlock_hpte(hptep);
 	else
 		/* Invalidate the hpte. NOTE: this also unlocks it */
 		hptep->v = 0;
 
 	/* Invalidate the TLB */
-	tlbie(vpn, psize, actual_psize, ssize, local);
+	tlbie(vpn, bpsize, apsize, ssize, local);
 
 	local_irq_restore(flags);
 }
 
+static inline int __hpte_actual_psize(unsigned int lp, int psize)
+{
+	int i, shift;
+	unsigned int mask;
+
+	/* start from 1 ignoring MMU_PAGE_4K */
+	for (i = 1; i < MMU_PAGE_COUNT; i++) {
+
+		/* invalid penc */
+		if (mmu_psize_defs[psize].penc[i] == -1)
+			continue;
+		/*
+		 * encoding bits per actual page size
+		 *        PTE LP     actual page size
+		 *    rrrr rrrz		>=8KB
+		 *    rrrr rrzz		>=16KB
+		 *    rrrr rzzz		>=32KB
+		 *    rrrr zzzz		>=64KB
+		 * .......
+		 */
+		shift = mmu_psize_defs[i].shift - LP_SHIFT;
+		if (shift > LP_BITS)
+			shift = LP_BITS;
+		mask = (1 << shift) - 1;
+		if ((lp & mask) == mmu_psize_defs[psize].penc[i])
+			return i;
+	}
+	return -1;
+}
+
 static void hpte_decode(struct hash_pte *hpte, unsigned long slot,
 			int *psize, int *apsize, int *ssize, unsigned long *vpn)
 {
diff --git a/arch/powerpc/mm/hash_utils_64.c b/arch/powerpc/mm/hash_utils_64.c
index e303a6d..2f47080 100644
--- a/arch/powerpc/mm/hash_utils_64.c
+++ b/arch/powerpc/mm/hash_utils_64.c
@@ -1232,7 +1232,11 @@ void flush_hash_page(unsigned long vpn, real_pte_t pte, int psize, int ssize,
 		slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
 		slot += hidx & _PTEIDX_GROUP_IX;
 		DBG_LOW(" sub %ld: hash=%lx, hidx=%lx\n", index, slot, hidx);
-		ppc_md.hpte_invalidate(slot, vpn, psize, ssize, local);
+		/*
+		 * We use same base page size and actual psize, because we don't
+		 * use these functions for hugepage
+		 */
+		ppc_md.hpte_invalidate(slot, vpn, psize, psize, ssize, local);
 	} pte_iterate_hashed_end();
 
 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
@@ -1365,7 +1369,8 @@ static void kernel_unmap_linear_page(unsigned long vaddr, unsigned long lmi)
 		hash = ~hash;
 	slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
 	slot += hidx & _PTEIDX_GROUP_IX;
-	ppc_md.hpte_invalidate(slot, vpn, mmu_linear_psize, mmu_kernel_ssize, 0);
+	ppc_md.hpte_invalidate(slot, vpn, mmu_linear_psize, mmu_linear_psize,
+			       mmu_kernel_ssize, 0);
 }
 
 void kernel_map_pages(struct page *page, int numpages, int enable)
diff --git a/arch/powerpc/mm/hugetlbpage-hash64.c b/arch/powerpc/mm/hugetlbpage-hash64.c
index 0f1d94a..0b7fb67 100644
--- a/arch/powerpc/mm/hugetlbpage-hash64.c
+++ b/arch/powerpc/mm/hugetlbpage-hash64.c
@@ -81,7 +81,7 @@ int __hash_page_huge(unsigned long ea, unsigned long access, unsigned long vsid,
 		slot += (old_pte & _PAGE_F_GIX) >> 12;
 
 		if (ppc_md.hpte_updatepp(slot, rflags, vpn, mmu_psize,
-					 ssize, local) == -1)
+					 mmu_psize, ssize, local) == -1)
 			old_pte &= ~_PAGE_HPTEFLAGS;
 	}
 
diff --git a/arch/powerpc/platforms/cell/beat_htab.c b/arch/powerpc/platforms/cell/beat_htab.c
index 246e1d8..c34ee4e 100644
--- a/arch/powerpc/platforms/cell/beat_htab.c
+++ b/arch/powerpc/platforms/cell/beat_htab.c
@@ -185,7 +185,8 @@ static void beat_lpar_hptab_clear(void)
 static long beat_lpar_hpte_updatepp(unsigned long slot,
 				    unsigned long newpp,
 				    unsigned long vpn,
-				    int psize, int ssize, int local)
+				    int psize, int apsize,
+				    int ssize, int local)
 {
 	unsigned long lpar_rc;
 	u64 dummy0, dummy1;
@@ -274,7 +275,8 @@ static void beat_lpar_hpte_updateboltedpp(unsigned long newpp,
 }
 
 static void beat_lpar_hpte_invalidate(unsigned long slot, unsigned long vpn,
-					 int psize, int ssize, int local)
+				      int psize, int apsize,
+				      int ssize, int local)
 {
 	unsigned long want_v;
 	unsigned long lpar_rc;
@@ -364,9 +366,10 @@ static long beat_lpar_hpte_insert_v3(unsigned long hpte_group,
  * already zero.  For now I am paranoid.
  */
 static long beat_lpar_hpte_updatepp_v3(unsigned long slot,
-				    unsigned long newpp,
-				    unsigned long vpn,
-				    int psize, int ssize, int local)
+				       unsigned long newpp,
+				       unsigned long vpn,
+				       int psize, int apsize,
+				       int ssize, int local)
 {
 	unsigned long lpar_rc;
 	unsigned long want_v;
@@ -394,7 +397,8 @@ static long beat_lpar_hpte_updatepp_v3(unsigned long slot,
 }
 
 static void beat_lpar_hpte_invalidate_v3(unsigned long slot, unsigned long vpn,
-					 int psize, int ssize, int local)
+					 int psize, int apsize,
+					 int ssize, int local)
 {
 	unsigned long want_v;
 	unsigned long lpar_rc;
diff --git a/arch/powerpc/platforms/ps3/htab.c b/arch/powerpc/platforms/ps3/htab.c
index 177a2f7..3e270e3 100644
--- a/arch/powerpc/platforms/ps3/htab.c
+++ b/arch/powerpc/platforms/ps3/htab.c
@@ -109,7 +109,8 @@ static long ps3_hpte_remove(unsigned long hpte_group)
 }
 
 static long ps3_hpte_updatepp(unsigned long slot, unsigned long newpp,
-	unsigned long vpn, int psize, int ssize, int local)
+			      unsigned long vpn, int psize, int apsize,
+			      int ssize, int local)
 {
 	int result;
 	u64 hpte_v, want_v, hpte_rs;
@@ -162,7 +163,7 @@ static void ps3_hpte_updateboltedpp(unsigned long newpp, unsigned long ea,
 }
 
 static void ps3_hpte_invalidate(unsigned long slot, unsigned long vpn,
-	int psize, int ssize, int local)
+				int psize, int apsize, int ssize, int local)
 {
 	unsigned long flags;
 	int result;
diff --git a/arch/powerpc/platforms/pseries/lpar.c b/arch/powerpc/platforms/pseries/lpar.c
index 6d62072..ca45c8f 100644
--- a/arch/powerpc/platforms/pseries/lpar.c
+++ b/arch/powerpc/platforms/pseries/lpar.c
@@ -240,7 +240,8 @@ static void pSeries_lpar_hptab_clear(void)
 static long pSeries_lpar_hpte_updatepp(unsigned long slot,
 				       unsigned long newpp,
 				       unsigned long vpn,
-				       int psize, int ssize, int local)
+				       int psize, int apsize,
+				       int ssize, int local)
 {
 	unsigned long lpar_rc;
 	unsigned long flags = (newpp & 7) | H_AVPN;
@@ -328,7 +329,8 @@ static void pSeries_lpar_hpte_updateboltedpp(unsigned long newpp,
 }
 
 static void pSeries_lpar_hpte_invalidate(unsigned long slot, unsigned long vpn,
-					 int psize, int ssize, int local)
+					 int psize, int apsize,
+					 int ssize, int local)
 {
 	unsigned long want_v;
 	unsigned long lpar_rc;
@@ -356,8 +358,10 @@ static void pSeries_lpar_hpte_removebolted(unsigned long ea,
 
 	slot = pSeries_lpar_hpte_find(vpn, psize, ssize);
 	BUG_ON(slot == -1);
-
-	pSeries_lpar_hpte_invalidate(slot, vpn, psize, ssize, 0);
+	/*
+	 * lpar doesn't use the passed actual page size
+	 */
+	pSeries_lpar_hpte_invalidate(slot, vpn, psize, 0, ssize, 0);
 }
 
 /* Flag bits for H_BULK_REMOVE */
@@ -400,8 +404,11 @@ static void pSeries_lpar_flush_hash_range(unsigned long number, int local)
 			slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
 			slot += hidx & _PTEIDX_GROUP_IX;
 			if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
+				/*
+				 * lpar doesn't use the passed actual page size
+				 */
 				pSeries_lpar_hpte_invalidate(slot, vpn, psize,
-							     ssize, local);
+							     0, ssize, local);
 			} else {
 				param[pix] = HBR_REQUEST | HBR_AVPN | slot;
 				param[pix+1] = hpte_encode_avpn(vpn, psize,