On Wed, Apr 12, 2017 at 11:53 AM, Balbir Singh [off-list ref] wrote:

On Wed, 2017-04-12 at 03:42 +1000, Oliver O'Halloran wrote:

quoted

From: Rashmica Gupta <redacted>

Adds support for removing bolted (i.e kernel linear mapping) mappings on
powernv. This is needed to support memory hot unplug operations which
are required for the teardown of DAX/PMEM devices.

Cc: Rashmica Gupta <redacted>
Cc: Anton Blanchard <redacted>
Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
---
Could the original author of this add their S-o-b? I pulled it out of
Rashmica's memtrace patch, but I remember someone saying Anton wrote
it originally.
---
 arch/powerpc/mm/hash_native_64.c | 31 +++++++++++++++++++++++++++++++
 1 file changed, 31 insertions(+)

diff --git a/arch/powerpc/mm/hash_native_64.c b/arch/powerpc/mm/hash_native_64.c
index 65bb8f33b399..9ba91d4905a4 100644
--- a/arch/powerpc/mm/hash_native_64.c
+++ b/arch/powerpc/mm/hash_native_64.c

@@ -407,6 +407,36 @@ static void native_hpte_updateboltedpp(unsigned long newpp, unsigned long ea,
      tlbie(vpn, psize, psize, ssize, 0);
 }

+/*
+ * Remove a bolted kernel entry. Memory hotplug uses this.
+ *
+ * No need to lock here because we should be the only user.

As long as this is after the necessary isolation and is called from
arch_remove_memory(), I think we should be fine

quoted

+ */
+static int native_hpte_removebolted(unsigned long ea, int psize, int ssize)
+{
+     unsigned long vpn;
+     unsigned long vsid;
+     long slot;
+     struct hash_pte *hptep;
+
+     vsid = get_kernel_vsid(ea, ssize);
+     vpn = hpt_vpn(ea, vsid, ssize);
+
+     slot = native_hpte_find(vpn, psize, ssize);
+     if (slot == -1)
+             return -ENOENT;

If slot == -1, it means someone else removed the HPTE entry? Are we racing?
I suspect we should never hit this situation during hotunplug, specifically
since this is bolted.

Or the slot was never populated in the first place. I'd rather keep
the current behaviour since it aligns with the behaviour of
pSeries_lpar_hpte_removebolted and we might hit these situations in
the future if the sub-section hotplug patches are merged (big if...).

quoted

+
+     hptep = htab_address + slot;
+
+     /* Invalidate the hpte */
+     hptep->v = 0;

Under DEBUG or otherwise, I would add more checks like

1. was hpte_v & HPTE_V_VALID and BOLTED set? If not, we've already invalidated
that hpte and we can skip the tlbie. Since this was bolted you might be right
that it is always valid and bolted

A VM_WARN_ON() if the bolted bit is clear might be appropriate. We
don't need to check the valid bit since hpte_native_find() will fail
if it's cleared.

quoted

+
+     /* Invalidate the TLB */
+     tlbie(vpn, psize, psize, ssize, 0);

The API also does not clear linear_map_hash_slots[] under DEBUG_PAGEALLOC

I'm not sure what API you're referring to here. The tracking for
linear_map_hash_slots[] is agnostic of mmu_hash_ops so we shouldn't be
touching it here. It also looks like DEBUG_PAGEALLOC is a bit broken
with hotplugged memory anyway so I think that's a fix for a different
patch.

quoted

+     return 0;
+}
+
+

Balbir Singh.