]> xenbits.xensource.com Git - qemu-xen-unstable.git/commitdiff
hw/scsi/vmw_pvscsi: Remove assertion for kick after reset
authorElazar Leibovich <elazar.leibovich@oracle.com>
Sun, 15 Mar 2020 13:26:34 +0000 (15:26 +0200)
committerPaolo Bonzini <pbonzini@redhat.com>
Wed, 1 Apr 2020 17:00:16 +0000 (19:00 +0200)
When running Ubuntu 3.13.0-65-generic guest, QEMU sometimes crashes
during guest ACPI reset. It crashes on assert(s->rings_info_valid)
in pvscsi_process_io().

Analyzing the crash revealed that it happens when userspace issues
a sync during a reboot syscall.

Below are backtraces we gathered from the guests.

Guest backtrace when issuing PVSCSI_CMD_ADAPTER_RESET:
    pci_device_shutdown
    device_shutdown
    init_pid_ns
    init_pid_ns
    kernel_power_off
    SYSC_reboot

Guest backtrace when issuing PVSCSI_REG_OFFSET_KICK_RW_IO:
    scsi_done
    scsi_dispatch_cmd
    blk_add_timer
    scsi_request_fn
    elv_rb_add
    __blk_run_queue
    queue_unplugged
    blk_flush_plug_list
    blk_finish_plug
    ext4_writepages
    set_next_entity
    do_writepages
    __filemap_fdatawrite_range
    filemap_write_and_wait_range
    ext4_sync_file
    ext4_sync_file
    do_fsync
    sys_fsync

Since QEMU pvscsi should imitate VMware pvscsi device emulation,
we decided to imitate VMware's behavior in this case.

To check VMware behavior, we wrote a kernel module that issues
a reset to the pvscsi device and then issues a kick. We ran it on
VMware ESXi 6.5 and it seems that it simply ignores the kick.
Hence, we decided to ignore the kick as well.

Signed-off-by: Elazar Leibovich <elazar.leibovich@oracle.com>
Signed-off-by: Liran Alon <liran.alon@oracle.com>
Message-Id: <20200315132634.113632-1-liran.alon@oracle.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
hw/scsi/vmw_pvscsi.c

index c91352cf46deff47a644d30a0756be35bddf1185..ec5bf9ea34af57c8c6b9f53fd8801c7cea6ecafc 100644 (file)
@@ -719,7 +719,10 @@ pvscsi_process_io(PVSCSIState *s)
     PVSCSIRingReqDesc descr;
     hwaddr next_descr_pa;
 
-    assert(s->rings_info_valid);
+    if (!s->rings_info_valid) {
+        return;
+    }
+
     while ((next_descr_pa = pvscsi_ring_pop_req_descr(&s->rings)) != 0) {
 
         /* Only read after production index verification */