summaryrefslogtreecommitdiff
path: root/arch/powerpc/kvm/book3s_hv_ras.c
diff options
context:
space:
mode:
Diffstat (limited to 'arch/powerpc/kvm/book3s_hv_ras.c')
-rw-r--r--arch/powerpc/kvm/book3s_hv_ras.c176
1 files changed, 176 insertions, 0 deletions
diff --git a/arch/powerpc/kvm/book3s_hv_ras.c b/arch/powerpc/kvm/book3s_hv_ras.c
index 93b5f5c9b445..0fa70a9618d7 100644
--- a/arch/powerpc/kvm/book3s_hv_ras.c
+++ b/arch/powerpc/kvm/book3s_hv_ras.c
@@ -13,6 +13,9 @@
#include <linux/kernel.h>
#include <asm/opal.h>
#include <asm/mce.h>
+#include <asm/machdep.h>
+#include <asm/cputhreads.h>
+#include <asm/hmi.h>
/* SRR1 bits for machine check on POWER7 */
#define SRR1_MC_LDSTERR (1ul << (63-42))
@@ -140,3 +143,176 @@ long kvmppc_realmode_machine_check(struct kvm_vcpu *vcpu)
{
return kvmppc_realmode_mc_power7(vcpu);
}
+
+/* Check if dynamic split is in force and return subcore size accordingly. */
+static inline int kvmppc_cur_subcore_size(void)
+{
+ if (local_paca->kvm_hstate.kvm_split_mode)
+ return local_paca->kvm_hstate.kvm_split_mode->subcore_size;
+
+ return threads_per_subcore;
+}
+
+void kvmppc_subcore_enter_guest(void)
+{
+ int thread_id, subcore_id;
+
+ thread_id = cpu_thread_in_core(local_paca->paca_index);
+ subcore_id = thread_id / kvmppc_cur_subcore_size();
+
+ local_paca->sibling_subcore_state->in_guest[subcore_id] = 1;
+}
+
+void kvmppc_subcore_exit_guest(void)
+{
+ int thread_id, subcore_id;
+
+ thread_id = cpu_thread_in_core(local_paca->paca_index);
+ subcore_id = thread_id / kvmppc_cur_subcore_size();
+
+ local_paca->sibling_subcore_state->in_guest[subcore_id] = 0;
+}
+
+static bool kvmppc_tb_resync_required(void)
+{
+ if (test_and_set_bit(CORE_TB_RESYNC_REQ_BIT,
+ &local_paca->sibling_subcore_state->flags))
+ return false;
+
+ return true;
+}
+
+static void kvmppc_tb_resync_done(void)
+{
+ clear_bit(CORE_TB_RESYNC_REQ_BIT,
+ &local_paca->sibling_subcore_state->flags);
+}
+
+/*
+ * kvmppc_realmode_hmi_handler() is called only by primary thread during
+ * guest exit path.
+ *
+ * There are multiple reasons why HMI could occur, one of them is
+ * Timebase (TB) error. If this HMI is due to TB error, then TB would
+ * have been in stopped state. The opal hmi handler Will fix it and
+ * restore the TB value with host timebase value. For HMI caused due
+ * to non-TB errors, opal hmi handler will not touch/restore TB register
+ * and hence there won't be any change in TB value.
+ *
+ * Since we are not sure about the cause of this HMI, we can't be sure
+ * about the content of TB register whether it holds guest or host timebase
+ * value. Hence the idea is to resync the TB on every HMI, so that we
+ * know about the exact state of the TB value. Resync TB call will
+ * restore TB to host timebase.
+ *
+ * Things to consider:
+ * - On TB error, HMI interrupt is reported on all the threads of the core
+ * that has encountered TB error irrespective of split-core mode.
+ * - The very first thread on the core that get chance to fix TB error
+ * would rsync the TB with local chipTOD value.
+ * - The resync TB is a core level action i.e. it will sync all the TBs
+ * in that core independent of split-core mode. This means if we trigger
+ * TB sync from a thread from one subcore, it would affect TB values of
+ * sibling subcores of the same core.
+ *
+ * All threads need to co-ordinate before making opal hmi handler.
+ * All threads will use sibling_subcore_state->in_guest[] (shared by all
+ * threads in the core) in paca which holds information about whether
+ * sibling subcores are in Guest mode or host mode. The in_guest[] array
+ * is of size MAX_SUBCORE_PER_CORE=4, indexed using subcore id to set/unset
+ * subcore status. Only primary threads from each subcore is responsible
+ * to set/unset its designated array element while entering/exiting the
+ * guset.
+ *
+ * After invoking opal hmi handler call, one of the thread (of entire core)
+ * will need to resync the TB. Bit 63 from subcore state bitmap flags
+ * (sibling_subcore_state->flags) will be used to co-ordinate between
+ * primary threads to decide who takes up the responsibility.
+ *
+ * This is what we do:
+ * - Primary thread from each subcore tries to set resync required bit[63]
+ * of paca->sibling_subcore_state->flags.
+ * - The first primary thread that is able to set the flag takes the
+ * responsibility of TB resync. (Let us call it as thread leader)
+ * - All other threads which are in host will call
+ * wait_for_subcore_guest_exit() and wait for in_guest[0-3] from
+ * paca->sibling_subcore_state to get cleared.
+ * - All the primary thread will clear its subcore status from subcore
+ * state in_guest[] array respectively.
+ * - Once all primary threads clear in_guest[0-3], all of them will invoke
+ * opal hmi handler.
+ * - Now all threads will wait for TB resync to complete by invoking
+ * wait_for_tb_resync() except the thread leader.
+ * - Thread leader will do a TB resync by invoking opal_resync_timebase()
+ * call and the it will clear the resync required bit.
+ * - All other threads will now come out of resync wait loop and proceed
+ * with individual execution.
+ * - On return of this function, primary thread will signal all
+ * secondary threads to proceed.
+ * - All secondary threads will eventually call opal hmi handler on
+ * their exit path.
+ */
+
+long kvmppc_realmode_hmi_handler(void)
+{
+ int ptid = local_paca->kvm_hstate.ptid;
+ bool resync_req;
+
+ /* This is only called on primary thread. */
+ BUG_ON(ptid != 0);
+ __this_cpu_inc(irq_stat.hmi_exceptions);
+
+ /*
+ * By now primary thread has already completed guest->host
+ * partition switch but haven't signaled secondaries yet.
+ * All the secondary threads on this subcore is waiting
+ * for primary thread to signal them to go ahead.
+ *
+ * For threads from subcore which isn't in guest, they all will
+ * wait until all other subcores on this core exit the guest.
+ *
+ * Now set the resync required bit. If you are the first to
+ * set this bit then kvmppc_tb_resync_required() function will
+ * return true. For rest all other subcores
+ * kvmppc_tb_resync_required() will return false.
+ *
+ * If resync_req == true, then this thread is responsible to
+ * initiate TB resync after hmi handler has completed.
+ * All other threads on this core will wait until this thread
+ * clears the resync required bit flag.
+ */
+ resync_req = kvmppc_tb_resync_required();
+
+ /* Reset the subcore status to indicate it has exited guest */
+ kvmppc_subcore_exit_guest();
+
+ /*
+ * Wait for other subcores on this core to exit the guest.
+ * All the primary threads and threads from subcore that are
+ * not in guest will wait here until all subcores are out
+ * of guest context.
+ */
+ wait_for_subcore_guest_exit();
+
+ /*
+ * At this point we are sure that primary threads from each
+ * subcore on this core have completed guest->host partition
+ * switch. Now it is safe to call HMI handler.
+ */
+ if (ppc_md.hmi_exception_early)
+ ppc_md.hmi_exception_early(NULL);
+
+ /*
+ * Check if this thread is responsible to resync TB.
+ * All other threads will wait until this thread completes the
+ * TB resync.
+ */
+ if (resync_req) {
+ opal_resync_timebase();
+ /* Reset TB resync req bit */
+ kvmppc_tb_resync_done();
+ } else {
+ wait_for_tb_resync();
+ }
+ return 0;
+}