summaryrefslogtreecommitdiff
path: root/arch/powerpc/platforms/pseries/mobility.c
diff options
context:
space:
mode:
Diffstat (limited to 'arch/powerpc/platforms/pseries/mobility.c')
-rw-r--r--arch/powerpc/platforms/pseries/mobility.c358
1 files changed, 299 insertions, 59 deletions
diff --git a/arch/powerpc/platforms/pseries/mobility.c b/arch/powerpc/platforms/pseries/mobility.c
index 2f73cb5bf12d..ea4d6a660e0d 100644
--- a/arch/powerpc/platforms/pseries/mobility.c
+++ b/arch/powerpc/platforms/pseries/mobility.c
@@ -12,9 +12,11 @@
#include <linux/cpu.h>
#include <linux/kernel.h>
#include <linux/kobject.h>
+#include <linux/nmi.h>
#include <linux/sched.h>
#include <linux/smp.h>
#include <linux/stat.h>
+#include <linux/stop_machine.h>
#include <linux/completion.h>
#include <linux/device.h>
#include <linux/delay.h>
@@ -59,18 +61,10 @@ static int mobility_rtas_call(int token, char *buf, s32 scope)
return rc;
}
-static int delete_dt_node(__be32 phandle)
+static int delete_dt_node(struct device_node *dn)
{
- struct device_node *dn;
-
- dn = of_find_node_by_phandle(be32_to_cpu(phandle));
- if (!dn)
- return -ENOENT;
-
pr_debug("removing node %pOFfp\n", dn);
-
dlpar_detach_node(dn);
- of_node_put(dn);
return 0;
}
@@ -135,10 +129,9 @@ static int update_dt_property(struct device_node *dn, struct property **prop,
return 0;
}
-static int update_dt_node(__be32 phandle, s32 scope)
+static int update_dt_node(struct device_node *dn, s32 scope)
{
struct update_props_workarea *upwa;
- struct device_node *dn;
struct property *prop = NULL;
int i, rc, rtas_rc;
char *prop_data;
@@ -155,14 +148,8 @@ static int update_dt_node(__be32 phandle, s32 scope)
if (!rtas_buf)
return -ENOMEM;
- dn = of_find_node_by_phandle(be32_to_cpu(phandle));
- if (!dn) {
- kfree(rtas_buf);
- return -ENOENT;
- }
-
upwa = (struct update_props_workarea *)&rtas_buf[0];
- upwa->phandle = phandle;
+ upwa->phandle = cpu_to_be32(dn->phandle);
do {
rtas_rc = mobility_rtas_call(update_properties_token, rtas_buf,
@@ -208,11 +195,12 @@ static int update_dt_node(__be32 phandle, s32 scope)
rc = update_dt_property(dn, &prop, prop_name,
vd, prop_data);
if (rc) {
- printk(KERN_ERR "Could not update %s"
- " property\n", prop_name);
+ pr_err("updating %s property failed: %d\n",
+ prop_name, rc);
}
prop_data += vd;
+ break;
}
cond_resched();
@@ -221,26 +209,18 @@ static int update_dt_node(__be32 phandle, s32 scope)
cond_resched();
} while (rtas_rc == 1);
- of_node_put(dn);
kfree(rtas_buf);
return 0;
}
-static int add_dt_node(__be32 parent_phandle, __be32 drc_index)
+static int add_dt_node(struct device_node *parent_dn, __be32 drc_index)
{
struct device_node *dn;
- struct device_node *parent_dn;
int rc;
- parent_dn = of_find_node_by_phandle(be32_to_cpu(parent_phandle));
- if (!parent_dn)
- return -ENOENT;
-
dn = dlpar_configure_connector(drc_index, parent_dn);
- if (!dn) {
- of_node_put(parent_dn);
+ if (!dn)
return -ENOENT;
- }
rc = dlpar_attach_node(dn, parent_dn);
if (rc)
@@ -248,7 +228,6 @@ static int add_dt_node(__be32 parent_phandle, __be32 drc_index)
pr_debug("added node %pOFfp\n", dn);
- of_node_put(parent_dn);
return rc;
}
@@ -261,7 +240,7 @@ int pseries_devicetree_update(s32 scope)
update_nodes_token = rtas_token("ibm,update-nodes");
if (update_nodes_token == RTAS_UNKNOWN_SERVICE)
- return -EINVAL;
+ return 0;
rtas_buf = kzalloc(RTAS_DATA_BUF_SIZE, GFP_KERNEL);
if (!rtas_buf)
@@ -281,22 +260,31 @@ int pseries_devicetree_update(s32 scope)
data++;
for (i = 0; i < node_count; i++) {
+ struct device_node *np;
__be32 phandle = *data++;
__be32 drc_index;
+ np = of_find_node_by_phandle(be32_to_cpu(phandle));
+ if (!np) {
+ pr_warn("Failed lookup: phandle 0x%x for action 0x%x\n",
+ be32_to_cpu(phandle), action);
+ continue;
+ }
+
switch (action) {
case DELETE_DT_NODE:
- delete_dt_node(phandle);
+ delete_dt_node(np);
break;
case UPDATE_DT_NODE:
- update_dt_node(phandle, scope);
+ update_dt_node(np, scope);
break;
case ADD_DT_NODE:
drc_index = *data++;
- add_dt_node(phandle, drc_index);
+ add_dt_node(np, drc_index);
break;
}
+ of_node_put(np);
cond_resched();
}
}
@@ -311,21 +299,8 @@ int pseries_devicetree_update(s32 scope)
void post_mobility_fixup(void)
{
int rc;
- int activate_fw_token;
-
- activate_fw_token = rtas_token("ibm,activate-firmware");
- if (activate_fw_token == RTAS_UNKNOWN_SERVICE) {
- printk(KERN_ERR "Could not make post-mobility "
- "activate-fw call.\n");
- return;
- }
-
- do {
- rc = rtas_call(activate_fw_token, 0, 1, NULL);
- } while (rtas_busy_delay(rc));
- if (rc)
- printk(KERN_ERR "Post-mobility activate-fw failed: %d\n", rc);
+ rtas_activate_firmware();
/*
* We don't want CPUs to go online/offline while the device
@@ -342,8 +317,7 @@ void post_mobility_fixup(void)
rc = pseries_devicetree_update(MIGRATION_SCOPE);
if (rc)
- printk(KERN_ERR "Post-mobility device tree update "
- "failed: %d\n", rc);
+ pr_err("device tree update failed: %d\n", rc);
cacheinfo_rebuild();
@@ -358,6 +332,279 @@ void post_mobility_fixup(void)
return;
}
+static int poll_vasi_state(u64 handle, unsigned long *res)
+{
+ unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
+ long hvrc;
+ int ret;
+
+ hvrc = plpar_hcall(H_VASI_STATE, retbuf, handle);
+ switch (hvrc) {
+ case H_SUCCESS:
+ ret = 0;
+ *res = retbuf[0];
+ break;
+ case H_PARAMETER:
+ ret = -EINVAL;
+ break;
+ case H_FUNCTION:
+ ret = -EOPNOTSUPP;
+ break;
+ case H_HARDWARE:
+ default:
+ pr_err("unexpected H_VASI_STATE result %ld\n", hvrc);
+ ret = -EIO;
+ break;
+ }
+ return ret;
+}
+
+static int wait_for_vasi_session_suspending(u64 handle)
+{
+ unsigned long state;
+ int ret;
+
+ /*
+ * Wait for transition from H_VASI_ENABLED to
+ * H_VASI_SUSPENDING. Treat anything else as an error.
+ */
+ while (true) {
+ ret = poll_vasi_state(handle, &state);
+
+ if (ret != 0 || state == H_VASI_SUSPENDING) {
+ break;
+ } else if (state == H_VASI_ENABLED) {
+ ssleep(1);
+ } else {
+ pr_err("unexpected H_VASI_STATE result %lu\n", state);
+ ret = -EIO;
+ break;
+ }
+ }
+
+ /*
+ * Proceed even if H_VASI_STATE is unavailable. If H_JOIN or
+ * ibm,suspend-me are also unimplemented, we'll recover then.
+ */
+ if (ret == -EOPNOTSUPP)
+ ret = 0;
+
+ return ret;
+}
+
+static void prod_single(unsigned int target_cpu)
+{
+ long hvrc;
+ int hwid;
+
+ hwid = get_hard_smp_processor_id(target_cpu);
+ hvrc = plpar_hcall_norets(H_PROD, hwid);
+ if (hvrc == H_SUCCESS)
+ return;
+ pr_err_ratelimited("H_PROD of CPU %u (hwid %d) error: %ld\n",
+ target_cpu, hwid, hvrc);
+}
+
+static void prod_others(void)
+{
+ unsigned int cpu;
+
+ for_each_online_cpu(cpu) {
+ if (cpu != smp_processor_id())
+ prod_single(cpu);
+ }
+}
+
+static u16 clamp_slb_size(void)
+{
+ u16 prev = mmu_slb_size;
+
+ slb_set_size(SLB_MIN_SIZE);
+
+ return prev;
+}
+
+static int do_suspend(void)
+{
+ u16 saved_slb_size;
+ int status;
+ int ret;
+
+ pr_info("calling ibm,suspend-me on CPU %i\n", smp_processor_id());
+
+ /*
+ * The destination processor model may have fewer SLB entries
+ * than the source. We reduce mmu_slb_size to a safe minimum
+ * before suspending in order to minimize the possibility of
+ * programming non-existent entries on the destination. If
+ * suspend fails, we restore it before returning. On success
+ * the OF reconfig path will update it from the new device
+ * tree after resuming on the destination.
+ */
+ saved_slb_size = clamp_slb_size();
+
+ ret = rtas_ibm_suspend_me(&status);
+ if (ret != 0) {
+ pr_err("ibm,suspend-me error: %d\n", status);
+ slb_set_size(saved_slb_size);
+ }
+
+ return ret;
+}
+
+static int do_join(void *arg)
+{
+ atomic_t *counter = arg;
+ long hvrc;
+ int ret;
+
+ /* Must ensure MSR.EE off for H_JOIN. */
+ hard_irq_disable();
+ hvrc = plpar_hcall_norets(H_JOIN);
+
+ switch (hvrc) {
+ case H_CONTINUE:
+ /*
+ * All other CPUs are offline or in H_JOIN. This CPU
+ * attempts the suspend.
+ */
+ ret = do_suspend();
+ break;
+ case H_SUCCESS:
+ /*
+ * The suspend is complete and this cpu has received a
+ * prod.
+ */
+ ret = 0;
+ break;
+ case H_BAD_MODE:
+ case H_HARDWARE:
+ default:
+ ret = -EIO;
+ pr_err_ratelimited("H_JOIN error %ld on CPU %i\n",
+ hvrc, smp_processor_id());
+ break;
+ }
+
+ if (atomic_inc_return(counter) == 1) {
+ pr_info("CPU %u waking all threads\n", smp_processor_id());
+ prod_others();
+ }
+ /*
+ * Execution may have been suspended for several seconds, so
+ * reset the watchdog.
+ */
+ touch_nmi_watchdog();
+ return ret;
+}
+
+/*
+ * Abort reason code byte 0. We use only the 'Migrating partition' value.
+ */
+enum vasi_aborting_entity {
+ ORCHESTRATOR = 1,
+ VSP_SOURCE = 2,
+ PARTITION_FIRMWARE = 3,
+ PLATFORM_FIRMWARE = 4,
+ VSP_TARGET = 5,
+ MIGRATING_PARTITION = 6,
+};
+
+static void pseries_cancel_migration(u64 handle, int err)
+{
+ u32 reason_code;
+ u32 detail;
+ u8 entity;
+ long hvrc;
+
+ entity = MIGRATING_PARTITION;
+ detail = abs(err) & 0xffffff;
+ reason_code = (entity << 24) | detail;
+
+ hvrc = plpar_hcall_norets(H_VASI_SIGNAL, handle,
+ H_VASI_SIGNAL_CANCEL, reason_code);
+ if (hvrc)
+ pr_err("H_VASI_SIGNAL error: %ld\n", hvrc);
+}
+
+static int pseries_suspend(u64 handle)
+{
+ const unsigned int max_attempts = 5;
+ unsigned int retry_interval_ms = 1;
+ unsigned int attempt = 1;
+ int ret;
+
+ while (true) {
+ atomic_t counter = ATOMIC_INIT(0);
+ unsigned long vasi_state;
+ int vasi_err;
+
+ ret = stop_machine(do_join, &counter, cpu_online_mask);
+ if (ret == 0)
+ break;
+ /*
+ * Encountered an error. If the VASI stream is still
+ * in Suspending state, it's likely a transient
+ * condition related to some device in the partition
+ * and we can retry in the hope that the cause has
+ * cleared after some delay.
+ *
+ * A better design would allow drivers etc to prepare
+ * for the suspend and avoid conditions which prevent
+ * the suspend from succeeding. For now, we have this
+ * mitigation.
+ */
+ pr_notice("Partition suspend attempt %u of %u error: %d\n",
+ attempt, max_attempts, ret);
+
+ if (attempt == max_attempts)
+ break;
+
+ vasi_err = poll_vasi_state(handle, &vasi_state);
+ if (vasi_err == 0) {
+ if (vasi_state != H_VASI_SUSPENDING) {
+ pr_notice("VASI state %lu after failed suspend\n",
+ vasi_state);
+ break;
+ }
+ } else if (vasi_err != -EOPNOTSUPP) {
+ pr_err("VASI state poll error: %d", vasi_err);
+ break;
+ }
+
+ pr_notice("Will retry partition suspend after %u ms\n",
+ retry_interval_ms);
+
+ msleep(retry_interval_ms);
+ retry_interval_ms *= 10;
+ attempt++;
+ }
+
+ return ret;
+}
+
+static int pseries_migrate_partition(u64 handle)
+{
+ int ret;
+
+ ret = wait_for_vasi_session_suspending(handle);
+ if (ret)
+ return ret;
+
+ ret = pseries_suspend(handle);
+ if (ret == 0)
+ post_mobility_fixup();
+ else
+ pseries_cancel_migration(handle, ret);
+
+ return ret;
+}
+
+int rtas_syscall_dispatch_ibm_suspend_me(u64 handle)
+{
+ return pseries_migrate_partition(handle);
+}
+
static ssize_t migration_store(struct class *class,
struct class_attribute *attr, const char *buf,
size_t count)
@@ -369,17 +616,10 @@ static ssize_t migration_store(struct class *class,
if (rc)
return rc;
- do {
- rc = rtas_ibm_suspend_me(streamid);
- if (rc == -EAGAIN)
- ssleep(1);
- } while (rc == -EAGAIN);
-
+ rc = pseries_migrate_partition(streamid);
if (rc)
return rc;
- post_mobility_fixup();
-
return count;
}
generated by cgit v1.2.3 (git 2.39.0) at 2024-09-11 12:18:59 +0300