No, also happens on ffs. If the root FS is on an affected device, it looks
like this:
root on sd0a (9e7002f8647a2403.a) swap on sd0b dump on sd0b
clock: unknown CMOS layout
panic: kernel diagnostic assertion "xs->datalen <= MINPHYS" failed: file
"../../../../dev/pci/vioblk.c", line 405
Stopped at Debugger+0x9: leave
RUN AT LEAST 'trace' AND 'ps' AND INCLUDE OUTPUT WHEN REPORTING THIS
PANIC!
IF RUNNING SMP, USE 'mach ddbcpu <#>' AND 'trace' ON OTHER PROCESSORS,
TOO.
DO NOT EVEN BOTHER REPORTING THIS WITHOUT INCLUDING THAT INFORMATION!
ddb{0}> trace
Debugger() at Debugger+0x9
panic() at panic+0xfe
__assert() at __assert+0x25
vioblk_scsi_cmd() at vioblk_scsi_cmd+0x565
scsi_xs_exec() at scsi_xs_exec+0x35
sdstart() at sdstart+0x16e
scsi_iopool_run() at scsi_iopool_run+0x5d
scsi_xsh_runqueue() at scsi_xsh_runqueue+0x13d
scsi_xsh_add() at scsi_xsh_add+0x98
sdstrategy() at sdstrategy+0x102
spec_strategy() at spec_strategy+0x53
ufs_strategy() at ufs_strategy+0x82
VOP_STRATEGY() at VOP_STRATEGY+0x3b
bread_cluster() at bread_cluster+0x2a4
ffs_read() at ffs_read+0x250
VOP_READ() at VOP_READ+0x3f
uvn_io() at uvn_io+0x2e2
uvn_get() at uvn_get+0x1f7
uvm_fault() at uvm_fault+0xdf5
trap() at trap+0x62f
--- trap (number 6) ---
end of kernel
end trace frame: 0x7f7ffffdd0c0, count: -20
acpi_pdirpa+0x43ae00:
Reproducer patch is below. I am sure it can be reproduced with any
other disk driver, but with ahci a reproducer would be a bit more
complicated because of the atascsi layer.
Maybe VOP_STRATEGY() should do a similar minphys+loop thing as physio()
does?
--- a/sys/dev/pci/vioblk.c
+++ b/sys/dev/pci/vioblk.c
@@ -63,6 +63,7 @@
#include <scsi/scsi_disk.h>
#include <scsi/scsiconf.h>
+#define MINPHYS (20*1024)
#define VIOBLK_DONE -1
struct virtio_feature_name vioblk_feature_names[] = {
@@ -157,6 +158,8 @@ vioblk_minphys(struct buf *bp, struct scsi_link *sl)
struct vioblk_softc *sc = sl->adapter_softc;
if (bp->b_bcount > sc->sc_xfer_max)
bp->b_bcount = sc->sc_xfer_max;
+ if (bp->b_bcount > MINPHYS)
+ bp->b_bcount = MINPHYS;
}
void
@@ -399,6 +402,8 @@ vioblk_scsi_cmd(struct scsi_xfer *xs)
int timeout;
int slot, ret, nsegs;
+ KASSERT(xs->datalen <= MINPHYS);
+
s = splbio();
ret = virtio_enqueue_prep(vq, &slot);
if (ret) {
On Fri, 29 Aug 2014, David Gwynne wrote:
> is it something msdos isnt doing that ffs does?
>
> On 29 Aug 2014, at 15:55, Stefan Fritsch <[email protected]> wrote:
>
> > On Friday 29 August 2014 10:34:20, David Gwynne wrote:
> >> are you expecting minphys to appear in that backtrace?
> >
> >
> > No. In that trace minphys should have limited the transfer size and I
> > had put an KASSERT() into vioblk_scsi_cmd() which verified that xs-
> >> datalen is not larger than that limit. A call to bread() causes the
> > KASSERT to trigger.
>
>
