diff mbox

[RFCv2,2/6] pseries: Move hash page table allocation to reset time

Message ID 1454049744-19131-3-git-send-email-david@gibson.dropbear.id.au (mailing list archive)
State New, archived
Headers show

Commit Message

David Gibson Jan. 29, 2016, 6:42 a.m. UTC
At the moment the size of the hash page table (HPT) is fixed based on the
maximum memory allowed to the guest.  As such, we allocate the table during
machine construction, and just clear it at reset.

However, we're planning to implement a PAPR extension allowing the hash
page table to be resized at runtime.  This will mean that on reset we want
to revert it to the default size.  It also means that when migrating, we
need to make sure the destination allocates an HPT of size matching the
host, since the guest could have changed it before the migration.

This patch replaces the spapr_alloc_htab() and spapr_reset_htab() functions
with a new spapr_reallocate_hpt() function.  This is called at reset and
inbound migration only, not during machine init any more.

In addition, we add a new helper to compute the recommended hash table size
for a given RAM size.  We export this as well as spapr_reallocate_hpt(),
since we'll be needing them elsewhere in future.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
---
 hw/ppc/spapr.c         | 151 ++++++++++++++++++++++---------------------------
 include/hw/ppc/spapr.h |   3 +
 2 files changed, 71 insertions(+), 83 deletions(-)
diff mbox

Patch

diff --git a/hw/ppc/spapr.c b/hw/ppc/spapr.c
index 52b0e49..bb701e3 100644
--- a/hw/ppc/spapr.c
+++ b/hw/ppc/spapr.c
@@ -1049,81 +1049,67 @@  static void close_htab_fd(sPAPRMachineState *spapr)
     spapr->htab_fd = -1;
 }
 
-static void spapr_alloc_htab(sPAPRMachineState *spapr)
-{
-    long shift;
-    int index;
-
-    /* allocate hash page table.  For now we always make this 16mb,
-     * later we should probably make it scale to the size of guest
-     * RAM */
-
-    shift = kvmppc_reset_htab(spapr->htab_shift);
-    if (shift < 0) {
-        /*
-         * For HV KVM, host kernel will return -ENOMEM when requested
-         * HTAB size can't be allocated.
-         */
-        error_setg(&error_abort, "Failed to allocate HTAB of requested size, try with smaller maxmem");
-    } else if (shift > 0) {
-        /*
-         * Kernel handles htab, we don't need to allocate one
-         *
-         * Older kernels can fall back to lower HTAB shift values,
-         * but we don't allow booting of such guests.
-         */
-        if (shift != spapr->htab_shift) {
-            error_setg(&error_abort, "Failed to allocate HTAB of requested size, try with smaller maxmem");
+int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
+{
+    int shift;
+
+    /* We aim for a hash table of size 1/128 the size of RAM (rounded
+     * up).  The PAPR recommendation is actually 1/64 of RAM size, but
+     * that's much more than is needed for Linux guests */
+    shift = ctz64(pow2ceil(ramsize)) - 7;
+    shift = MAX(shift, 18); /* Minimum architected size */
+    shift = MIN(shift, 46); /* Maximum architected size */
+    return shift;
+}
+
+void spapr_reallocate_hpt(sPAPRMachineState *spapr, int shift, Error **errp)
+{
+    long rc;
+
+    /* Clean up any HPT info from a previous boot */
+    g_free(spapr->htab);
+    spapr->htab = NULL;
+    spapr->htab_shift = 0;
+    close_htab_fd(spapr);
+
+    rc = kvmppc_reset_htab(shift);
+    if (rc < 0) {
+        /* kernel-side HPT needed, but couldn't allocate one */
+        error_setg_errno(errp, errno,
+                         "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
+                         shift);
+        /* This is almost certainly fatal, but if the caller really
+         * wants to carry on with shift == 0, it's welcome to try */
+    } else if (rc > 0) {
+        /* kernel-side HPT allocated */
+        if (rc != shift) {
+            error_setg(errp,
+                       "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
+                       shift, rc);
         }
 
         spapr->htab_shift = shift;
         kvmppc_kern_htab = true;
     } else {
-        /* Allocate htab */
-        spapr->htab = qemu_memalign(HTAB_SIZE(spapr), HTAB_SIZE(spapr));
-
-        /* And clear it */
-        memset(spapr->htab, 0, HTAB_SIZE(spapr));
-
-        for (index = 0; index < HTAB_SIZE(spapr) / HASH_PTE_SIZE_64; index++) {
-            DIRTY_HPTE(HPTE(spapr->htab, index));
-        }
-    }
-}
-
-/*
- * Clear HTAB entries during reset.
- *
- * If host kernel has allocated HTAB, KVM_PPC_ALLOCATE_HTAB ioctl is
- * used to clear HTAB. Otherwise QEMU-allocated HTAB is cleared manually.
- */
-static void spapr_reset_htab(sPAPRMachineState *spapr)
-{
-    long shift;
-    int index;
+        /* kernel-side HPT not needed, allocate in userspace instead */
+        size_t size = 1ULL << shift;
+        int i;
 
-    shift = kvmppc_reset_htab(spapr->htab_shift);
-    if (shift < 0) {
-        error_setg(&error_abort, "Failed to reset HTAB");
-    } else if (shift > 0) {
-        if (shift != spapr->htab_shift) {
-            error_setg(&error_abort, "Requested HTAB allocation failed during reset");
+        spapr->htab = qemu_memalign(size, size);
+        if (!spapr->htab) {
+            error_setg_errno(errp, errno,
+                             "Could not allocate HPT of order %d", shift);
+            return;
         }
 
-        close_htab_fd(spapr);
-    } else {
-        memset(spapr->htab, 0, HTAB_SIZE(spapr));
+        memset(spapr->htab, 0, size);
+        spapr->htab_shift = shift;
+        kvmppc_kern_htab = false;
 
-        for (index = 0; index < HTAB_SIZE(spapr) / HASH_PTE_SIZE_64; index++) {
-            DIRTY_HPTE(HPTE(spapr->htab, index));
+        for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
+            DIRTY_HPTE(HPTE(spapr->htab, i));
         }
     }
-
-    /* Update the RMA size if necessary */
-    if (spapr->vrma_adjust) {
-        spapr->rma_size = kvmppc_rma_size(spapr_node0_size(),
-                                          spapr->htab_shift);
-    }
 }
 
 static int find_unknown_sysbus_device(SysBusDevice *sbdev, void *opaque)
@@ -1145,15 +1131,24 @@  static int find_unknown_sysbus_device(SysBusDevice *sbdev, void *opaque)
 
 static void ppc_spapr_reset(void)
 {
-    sPAPRMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
+    MachineState *machine = MACHINE(qdev_get_machine());
+    sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
     PowerPCCPU *first_ppc_cpu;
     uint32_t rtas_limit;
 
     /* Check for unknown sysbus devices */
     foreach_dynamic_sysbus_device(find_unknown_sysbus_device, NULL);
 
-    /* Reset the hash table & recalc the RMA */
-    spapr_reset_htab(spapr);
+    /* Allocate and/or reset the hash page table */
+    spapr_reallocate_hpt(spapr,
+                         spapr_hpt_shift_for_ramsize(machine->maxram_size),
+                         &error_fatal);
+
+    /* Update the RMA size if necessary */
+    if (spapr->vrma_adjust) {
+        spapr->rma_size = kvmppc_rma_size(spapr_node0_size(),
+                                          spapr->htab_shift);
+    }
 
     qemu_devices_reset();
 
@@ -1533,10 +1528,12 @@  static int htab_load(QEMUFile *f, void *opaque, int version_id)
     section_hdr = qemu_get_be32(f);
 
     if (section_hdr) {
-        /* First section, just the hash shift */
-        if (spapr->htab_shift != section_hdr) {
-            error_report("htab_shift mismatch: source %d target %d",
-                         section_hdr, spapr->htab_shift);
+        Error *local_err;
+
+        /* First section gives the htab size */
+        spapr_reallocate_hpt(spapr, section_hdr, &local_err);
+        if (local_err) {
+            error_report_err(local_err);
             return -EINVAL;
         }
         return 0;
@@ -1789,18 +1786,6 @@  static void ppc_spapr_init(MachineState *machine)
     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
 
-    /* We aim for a hash table of size 1/128 the size of RAM.  The
-     * normal rule of thumb is 1/64 the size of RAM, but that's much
-     * more than needed for the Linux guests we support. */
-    spapr->htab_shift = 18; /* Minimum architected size */
-    while (spapr->htab_shift <= 46) {
-        if ((1ULL << (spapr->htab_shift + 7)) >= machine->maxram_size) {
-            break;
-        }
-        spapr->htab_shift++;
-    }
-    spapr_alloc_htab(spapr);
-
     /* Set up Interrupt Controller before we create the VCPUs */
     spapr->icp = xics_system_init(machine,
                                   DIV_ROUND_UP(max_cpus * kvmppc_smt_threads(),
diff --git a/include/hw/ppc/spapr.h b/include/hw/ppc/spapr.h
index 098d85d..4e3dce5 100644
--- a/include/hw/ppc/spapr.h
+++ b/include/hw/ppc/spapr.h
@@ -373,6 +373,9 @@  target_ulong spapr_hypercall(PowerPCCPU *cpu, target_ulong opcode,
 int spapr_allocate_irq(int hint, bool lsi);
 int spapr_allocate_irq_block(int num, bool lsi, bool msi);
 
+int spapr_hpt_shift_for_ramsize(uint64_t ramsize);
+void spapr_reallocate_hpt(sPAPRMachineState *spapr, int shift, Error **errp);
+
 /* ibm,set-eeh-option */
 #define RTAS_EEH_DISABLE                 0
 #define RTAS_EEH_ENABLE                  1