PEL: Notifier support for 'host full'

The host firmware may have a limited size for its staging area
before it passes the PELs through to the OS, and this area may
fill up with PELs it can't send if there are too many or the OS
isn't up yet.

In this case, it will send down an 'Ack PEL' PLDM command with a
special response indicating this host full condition.  The PLDM
daemon will then call a method on this daemon to let it know.

This command handles the host full condition on the HostNotifier
class.  When this is set:

* The PEL that hit this condition will be put back on the queue
  to be sent again.
* No new PELs will be sent up, except as noted below
* A 60s timer will be started at the end of which another attempt
  will be made to send a PEL, in the hopes the condition went away.
  - If it didn't go away, this process will repeat.
  - If it did go away, a successful ack will be received and things
    will proceed as normal.

Signed-off-by: Matt Spinler <spinler@us.ibm.com>
Change-Id: Iaeb38f43f7acc595bcff234ba50cedf8188b3d9b

2 files changed, 115 insertions, 0 deletions

diff --git a/test/openpower-pels/host_notifier_test.cpp b/test/openpower-pels/host_notifier_test.cpp
index 62c79b2..cd30e8d 100644
--- a/test/openpower-pels/host_notifier_test.cpp
+++ b/test/openpower-pels/host_notifier_test.cpp
@@ -643,3 +643,106 @@ TEST_F(HostNotifierTest, TestPowerCycleAndAcks)
 
     EXPECT_EQ(notifier.queueSize(), 0);
 }
+
+// Test the host full condition
+TEST_F(HostNotifierTest, TestHostFull)
+{
+    // The full interaction with the host is:
+    // BMC:  new PEL available
+    // Host: ReadPELFile  (not modeled here)
+    // Host: Ack(id) (if not full), or HostFull(id)
+    // BMC: if full and any new PELs come in, don't sent them
+    // Start a timer and try again
+    // Host responds with either Ack or full
+    // and repeat
+
+    Repository repo{repoPath};
+    MockDataInterface dataIface;
+
+    sdeventplus::Event sdEvent{event};
+
+    std::unique_ptr<HostInterface> hostIface =
+        std::make_unique<MockHostInterface>(event, dataIface);
+
+    MockHostInterface& mockHostIface =
+        reinterpret_cast<MockHostInterface&>(*hostIface);
+
+    HostNotifier notifier{repo, dataIface, std::move(hostIface)};
+
+    auto send = [&mockHostIface](uint32_t id, uint32_t size) {
+        return mockHostIface.send(0);
+    };
+
+    EXPECT_CALL(mockHostIface, sendNewLogCmd(_, _))
+        .WillRepeatedly(Invoke(send));
+
+    dataIface.changeHostState(true);
+
+    // Add and dispatch/send one PEL
+    auto pel = makePEL();
+    auto id = pel->id();
+    repo.add(pel);
+    runEvents(sdEvent, 2);
+
+    EXPECT_EQ(mockHostIface.numCmdsProcessed(), 1);
+    EXPECT_EQ(notifier.queueSize(), 0);
+
+    // Host is full
+    notifier.setHostFull(id);
+
+    // It goes back on the queue
+    EXPECT_EQ(notifier.queueSize(), 1);
+
+    // The transmission state goes back to new
+    Repository::LogID i{Repository::LogID::Pel{id}};
+    auto data = repo.getPELData(i);
+    PEL pelFromRepo{*data};
+    EXPECT_EQ(pelFromRepo.hostTransmissionState(), TransmissionState::newPEL);
+
+    // Clock it, nothing should be sent still.
+    runEvents(sdEvent, 1);
+
+    EXPECT_EQ(mockHostIface.numCmdsProcessed(), 1);
+    EXPECT_EQ(notifier.queueSize(), 1);
+
+    // Add another PEL and clock it, still nothing sent
+    pel = makePEL();
+    repo.add(pel);
+    runEvents(sdEvent, 2);
+    EXPECT_EQ(mockHostIface.numCmdsProcessed(), 1);
+    EXPECT_EQ(notifier.queueSize(), 2);
+
+    // Let the host full timer expire to trigger a retry.
+    // Add some extra event passes just to be sure nothing new is sent.
+    runEvents(sdEvent, 5, mockHostIface.getHostFullRetryDelay());
+
+    // The timer expiration will send just the 1, not both
+    EXPECT_EQ(mockHostIface.numCmdsProcessed(), 2);
+    EXPECT_EQ(notifier.queueSize(), 1);
+
+    // Host still full
+    notifier.setHostFull(id);
+
+    // Let the host full timer attempt again
+    runEvents(sdEvent, 2, mockHostIface.getHostFullRetryDelay());
+    EXPECT_EQ(mockHostIface.numCmdsProcessed(), 3);
+
+    // Add yet another PEL with the retry timer expired.
+    // It shouldn't get sent out.
+    pel = makePEL();
+    repo.add(pel);
+    runEvents(sdEvent, 2);
+    EXPECT_EQ(mockHostIface.numCmdsProcessed(), 3);
+
+    // The last 2 PELs still on the queue
+    EXPECT_EQ(notifier.queueSize(), 2);
+
+    // Host no longer full, it finally acks the first PEL
+    notifier.ackPEL(id);
+
+    // Now the remaining 2 PELs will be dispatched
+    runEvents(sdEvent, 3);
+
+    EXPECT_EQ(mockHostIface.numCmdsProcessed(), 5);
+    EXPECT_EQ(notifier.queueSize(), 0);
+}
diff --git a/test/openpower-pels/mocks.hpp b/test/openpower-pels/mocks.hpp
index dcce142..5bec579 100644
--- a/test/openpower-pels/mocks.hpp
+++ b/test/openpower-pels/mocks.hpp
@@ -100,6 +100,18 @@ class MockHostInterface : public HostInterface
     }
 
     /**
+     * @brief Returns the amount of time to wait before retrying if the
+     *        host firmware's PEL storage was full and it can't store
+     *        any more logs until it is freed up somehow.
+     *
+     * @return milliseconds - The amount of time to wait
+     */
+    virtual std::chrono::milliseconds getHostFullRetryDelay() const override
+    {
+        return std::chrono::milliseconds(20);
+    }
+
+    /**
      * @brief Returns the number of commands processed
      */
     size_t numCmdsProcessed() const