pseries: Use smaller default hash page tables when guest can resize

We've now implemented a PAPR extension allowing PAPR guest to resize
their hash page table (HPT) during runtime.

This patch makes use of that facility to allocate smaller HPTs by default.
Specifically when a guest is aware of the HPT resize facility, qemu sizes
the HPT to the initial memory size, rather than the maximum memory size on
the assumption that the guest will resize its HPT if necessary for hot
plugged memory.

When the initial memory size is much smaller than the maximum memory size
(a common configuration with e.g. oVirt / RHEV) then this can save
significant memory on the HPT.

If the guest does *not* advertise HPT resize awareness when it makes the
ibm,client-architecture-support call, qemu resizes the HPT for maxmimum
memory size (unless it's been configured not to allow such guests at all).

For now we make that reallocation assuming the guest has not yet used the
HPT at all.  That's true in practice, but not, strictly, an architectural
or PAPR requirement.  If we need to in future we can fix this by having
the client-architecture-support call reboot the guest with the revised
HPT size (the client-architecture-support call is explicitly permitted to
trigger a reboot in this way).

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Suraj Jitindar Singh <sjitindarsingh@gmail.com>

1 files changed, 29 insertions, 0 deletions

diff --git a/hw/ppc/spapr_hcall.c b/hw/ppc/spapr_hcall.c
index f69ce4f..436f5e2 100644
--- a/hw/ppc/spapr_hcall.c
+++ b/hw/ppc/spapr_hcall.c
@@ -1473,6 +1473,35 @@ static target_ulong h_client_architecture_support(PowerPCCPU *cpu,
     guest_radix = spapr_ovec_test(ov5_guest, OV5_MMU_RADIX_300);
     spapr_ovec_clear(ov5_guest, OV5_MMU_RADIX_300);
 
+    /*
+     * HPT resizing is a bit of a special case, because when enabled
+     * we assume an HPT guest will support it until it says it
+     * doesn't, instead of assuming it won't support it until it says
+     * it does.  Strictly speaking that approach could break for
+     * guests which don't make a CAS call, but those are so old we
+     * don't care about them.  Without that assumption we'd have to
+     * make at least a temporary allocation of an HPT sized for max
+     * memory, which could be impossibly difficult under KVM HV if
+     * maxram is large.
+     */
+    if (!guest_radix && !spapr_ovec_test(ov5_guest, OV5_HPT_RESIZE)) {
+        int maxshift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
+
+        if (spapr->resize_hpt == SPAPR_RESIZE_HPT_REQUIRED) {
+            error_report(
+                "h_client_architecture_support: Guest doesn't support HPT resizing, but resize-hpt=required");
+            exit(1);
+        }
+
+        if (spapr->htab_shift < maxshift) {
+            /* Guest doesn't know about HPT resizing, so we
+             * pre-emptively resize for the maximum permitted RAM.  At
+             * the point this is called, nothing should have been
+             * entered into the existing HPT */
+            spapr_reallocate_hpt(spapr, maxshift, &error_fatal);
+        }
+    }
+
     /* NOTE: there are actually a number of ov5 bits where input from the
      * guest is always zero, and the platform/QEMU enables them independently
      * of guest input. To model these properly we'd want some sort of mask,