Make sure we're muxed to the SP before setting the mux to the host (#2457)

I noticed that the SP hangs if you call `set_mux(HostCPU)` twice in a
row. This is because the first call switches the mux from the SP to the
host; the second call tries to read from flash while it's muxed to the
host (to find the APOB, ABL0 version, etc). When the flash is muxed to
the host, the FPGA never loads bytes into the SP's FIFO, so this hangs
the `hf` task forever.

Author: mkeeter

drv/cosmo-hf/src/hf.rs

@@ -606,6 +606,11 @@ impl idl::InOrderHostFlashImpl for ServerImpl {
         // - We will only load an APOB that's pinned to that version
         // - After booting, we'll use that version when persisting a new APOB
         if state == HfMuxState::HostCPU {
+            // We can only swap the mux state to the host if it's currently
+            // muxed to the SP, because we must read from flash (to find ABL0
+            // version, APOB state, etc) before swapping the mux.
+            self.drv.check_flash_mux_state()?;
+
             self.abl0_version = match self.find_abl0_version() {
                 Ok(v) => {
                     // If the previous ABL0 version has not been consumed by