| Message ID | 20190514234248.36203-2-farman@linux.ibm.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | s390: vfio-ccw fixes | expand |
On 05/14/2019 07:42 PM, Eric Farman wrote: > Per the POPs [1], when processing an interrupt the SCSW.CPA field of an > IRB generally points to 8 bytes after the last CCW that was executed > (there are exceptions, but this is the most common behavior). > > In the case of an error, this points us to the first un-executed CCW > in the chain. But in the case of normal I/O, the address points beyond > the end of the chain. While the guest generally only cares about this > when possibly restarting a channel program after error recovery, we > should convert the address even in the good scenario so that we provide > a consistent, valid, response upon I/O completion. > > [1] Figure 16-6 in SA22-7832-11. The footnotes in that table also state > that this is true even if the resulting address is invalid or protected, > but moving to the end of the guest chain should not be a surprise. > > Signed-off-by: Eric Farman <farman@linux.ibm.com> > --- > drivers/s390/cio/vfio_ccw_cp.c | 6 +++++- > 1 file changed, 5 insertions(+), 1 deletion(-) > > diff --git a/drivers/s390/cio/vfio_ccw_cp.c b/drivers/s390/cio/vfio_ccw_cp.c > index 384b3987eeb4..41f48b8790bc 100644 > --- a/drivers/s390/cio/vfio_ccw_cp.c > +++ b/drivers/s390/cio/vfio_ccw_cp.c > @@ -870,7 +870,11 @@ void cp_update_scsw(struct channel_program *cp, union scsw *scsw) > */ > list_for_each_entry(chain, &cp->ccwchain_list, next) { > ccw_head = (u32)(u64)chain->ch_ccw; > - if (is_cpa_within_range(cpa, ccw_head, chain->ch_len)) { > + /* > + * On successful execution, cpa points just beyond the end > + * of the chain. > + */ > + if (is_cpa_within_range(cpa, ccw_head, chain->ch_len + 1)) { > /* > * (cpa - ccw_head) is the offset value of the host > * physical ccw to its chain head. > Reviewed-by: Farhan Ali <alifm@linux.ibm.com>
diff --git a/drivers/s390/cio/vfio_ccw_cp.c b/drivers/s390/cio/vfio_ccw_cp.c index 384b3987eeb4..41f48b8790bc 100644 --- a/drivers/s390/cio/vfio_ccw_cp.c +++ b/drivers/s390/cio/vfio_ccw_cp.c @@ -870,7 +870,11 @@ void cp_update_scsw(struct channel_program *cp, union scsw *scsw) */ list_for_each_entry(chain, &cp->ccwchain_list, next) { ccw_head = (u32)(u64)chain->ch_ccw; - if (is_cpa_within_range(cpa, ccw_head, chain->ch_len)) { + /* + * On successful execution, cpa points just beyond the end + * of the chain. + */ + if (is_cpa_within_range(cpa, ccw_head, chain->ch_len + 1)) { /* * (cpa - ccw_head) is the offset value of the host * physical ccw to its chain head.
Per the POPs [1], when processing an interrupt the SCSW.CPA field of an IRB generally points to 8 bytes after the last CCW that was executed (there are exceptions, but this is the most common behavior). In the case of an error, this points us to the first un-executed CCW in the chain. But in the case of normal I/O, the address points beyond the end of the chain. While the guest generally only cares about this when possibly restarting a channel program after error recovery, we should convert the address even in the good scenario so that we provide a consistent, valid, response upon I/O completion. [1] Figure 16-6 in SA22-7832-11. The footnotes in that table also state that this is true even if the resulting address is invalid or protected, but moving to the end of the guest chain should not be a surprise. Signed-off-by: Eric Farman <farman@linux.ibm.com> --- drivers/s390/cio/vfio_ccw_cp.c | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-)