scsi: esp: Defer command completion until previous interrupts have been handled

The guest OS reads RSTAT, RSEQ, and RINTR, and expects those registers
to reflect a consistent state. However, it is possible that the registers
can change after RSTAT was read, but before RINTR is read, when
esp_command_complete() is called.

Guest OS		qemu
--------		----
[handle interrupt]
Read RSTAT
			esp_command_complete()
			 RSTAT = STAT_ST
			 esp_dma_done()
			  RSTAT |= STAT_TC
			  RSEQ = 0
			  RINTR = INTR_BS

Read RSEQ
Read RINTR		RINTR = 0
			RSTAT &= ~STAT_TC
			RSEQ = SEQ_CD

The guest OS would then try to handle INTR_BS combined with an old
value of RSTAT. This sometimes resulted in lost events, spurious
interrupts, guest OS confusion, and stalled SCSI operations.
A typical guest error log (observed with various versions of Linux)
looks as follows.

scsi host1: Spurious irq, sreg=13.
...
scsi host1: Aborting command [84531f10:2a]
scsi host1: Current command [f882eea8:35]
scsi host1: Queued command [84531f10:2a]
scsi host1:  Active command [f882eea8:35]
scsi host1: Dumping command log
scsi host1: ent[15] CMD val[44] sreg[90] seqreg[00] sreg2[00] ireg[20] ss[00] event[0c]
scsi host1: ent[16] CMD val[01] sreg[90] seqreg[00] sreg2[00] ireg[20] ss[02] event[0c]
scsi host1: ent[17] CMD val[43] sreg[90] seqreg[00] sreg2[00] ireg[20] ss[02] event[0c]
scsi host1: ent[18] EVENT val[0d] sreg[92] seqreg[04] sreg2[00] ireg[18] ss[00] event[0c]
...

Defer handling command completion until previous interrupts have been
handled to fix the problem.

Signed-off-by: Guenter Roeck <linux@roeck-us.net>

1 files changed, 2 insertions, 0 deletions

diff --git a/include/hw/scsi/esp.h b/include/hw/scsi/esp.h
index 682a0d2de0..adab63d1c9 100644
--- a/include/hw/scsi/esp.h
+++ b/include/hw/scsi/esp.h
@@ -23,6 +23,8 @@ struct ESPState {
     int32_t ti_size;
     uint32_t ti_rptr, ti_wptr;
     uint32_t status;
+    uint32_t deferred_status;
+    bool deferred_complete;
     uint32_t dma;
     uint8_t ti_buf[TI_BUFSZ];
     SCSIBus bus;