]> git.proxmox.com Git - mirror_qemu.git/commitdiff
spapr: Handle VMX/VSX presence as an spapr capability flag
authorDavid Gibson <david@gibson.dropbear.id.au>
Thu, 7 Dec 2017 06:08:47 +0000 (17:08 +1100)
committerDavid Gibson <david@gibson.dropbear.id.au>
Tue, 16 Jan 2018 22:35:24 +0000 (09:35 +1100)
We currently have some conditionals in the spapr device tree code to decide
whether or not to advertise the availability of the VMX (aka Altivec) and
VSX vector extensions to the guest, based on whether the guest cpu has
those features.

This can lead to confusion and subtle failures on migration, since it makes
a guest visible change based only on host capabilities.  We now have a
better mechanism for this, in spapr capabilities flags, which explicitly
depend on user options rather than host capabilities.

Rework the advertisement of VSX and VMX based on a new VSX capability.  We
no longer bother with a conditional for VMX support, because every CPU
that's ever been supported by the pseries machine type supports VMX.

NOTE: Some userspace distributions (e.g. RHEL7.4) already rely on
availability of VSX in libc, so using cap-vsx=off may lead to a fatal
SIGILL in init.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Greg Kurz <groug@kaod.org>
hw/ppc/spapr.c
hw/ppc/spapr_caps.c
include/hw/ppc/spapr.h

index 3451d0806d92a7a3632e6a9a6d5b5a92f8aa2a69..eca9f11c913fb825bc061fb5e18b0211c8a05de8 100644 (file)
@@ -557,14 +557,16 @@ static void spapr_populate_cpu_dt(CPUState *cs, void *fdt, int offset,
                           segs, sizeof(segs))));
     }
 
-    /* Advertise VMX/VSX (vector extensions) if available
-     *   0 / no property == no vector extensions
+    /* Advertise VSX (vector extensions) if available
      *   1               == VMX / Altivec available
-     *   2               == VSX available */
-    if (env->insns_flags & PPC_ALTIVEC) {
-        uint32_t vmx = (env->insns_flags2 & PPC2_VSX) ? 2 : 1;
-
-        _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", vmx)));
+     *   2               == VSX available
+     *
+     * Only CPUs for which we create core types in spapr_cpu_core.c
+     * are possible, and all of those have VMX */
+    if (spapr_has_cap(spapr, SPAPR_CAP_VSX)) {
+        _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 2)));
+    } else {
+        _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 1)));
     }
 
     /* Advertise DFP (Decimal Floating Point) if available
@@ -3832,7 +3834,7 @@ static void spapr_machine_class_init(ObjectClass *oc, void *data)
      */
     mc->numa_mem_align_shift = 28;
 
-    smc->default_caps = spapr_caps(0);
+    smc->default_caps = spapr_caps(SPAPR_CAP_VSX);
     spapr_caps_add_properties(smc, &error_abort);
 }
 
@@ -3914,7 +3916,7 @@ static void spapr_machine_2_11_class_options(MachineClass *mc)
     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
 
     spapr_machine_2_12_class_options(mc);
-    smc->default_caps = spapr_caps(SPAPR_CAP_HTM);
+    smc->default_caps = spapr_caps(SPAPR_CAP_HTM | SPAPR_CAP_VSX);
     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_11);
 }
 
index cad40fe49af50621f6568e5572ef1f61e01ca7e0..7c855c67ad4008895339de081947a71b1e7a76a5 100644 (file)
@@ -57,6 +57,19 @@ static void cap_htm_allow(sPAPRMachineState *spapr, Error **errp)
     }
 }
 
+static void cap_vsx_allow(sPAPRMachineState *spapr, Error **errp)
+{
+    PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
+    CPUPPCState *env = &cpu->env;
+
+    /* Allowable CPUs in spapr_cpu_core.c should already have gotten
+     * rid of anything that doesn't do VMX */
+    g_assert(env->insns_flags & PPC_ALTIVEC);
+    if (!(env->insns_flags2 & PPC2_VSX)) {
+        error_setg(errp, "VSX support not available, try cap-vsx=off");
+    }
+}
+
 static sPAPRCapabilityInfo capability_table[] = {
     {
         .name = "htm",
@@ -65,6 +78,13 @@ static sPAPRCapabilityInfo capability_table[] = {
         .allow = cap_htm_allow,
         /* TODO: add cap_htm_disallow */
     },
+    {
+        .name = "vsx",
+        .description = "Allow Vector Scalar Extensions (VSX)",
+        .flag = SPAPR_CAP_VSX,
+        .allow = cap_vsx_allow,
+        /* TODO: add cap_vsx_disallow */
+    },
 };
 
 static sPAPRCapabilities default_caps_with_cpu(sPAPRMachineState *spapr,
@@ -81,6 +101,11 @@ static sPAPRCapabilities default_caps_with_cpu(sPAPRMachineState *spapr,
         caps.mask &= ~SPAPR_CAP_HTM;
     }
 
+    if (!ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_06,
+                          0, spapr->max_compat_pvr)) {
+        caps.mask &= ~SPAPR_CAP_VSX;
+    }
+
     return caps;
 }
 
index 5c85f39c3b9d520d49e92b1bc77f0049733b8f03..148a03d189c0f4550d2fcabbac0947e449901d88 100644 (file)
@@ -59,6 +59,9 @@ typedef enum {
 /* Hardware Transactional Memory */
 #define SPAPR_CAP_HTM               0x0000000000000001ULL
 
+/* Vector Scalar Extensions */
+#define SPAPR_CAP_VSX               0x0000000000000002ULL
+
 typedef struct sPAPRCapabilities sPAPRCapabilities;
 struct sPAPRCapabilities {
     uint64_t mask;