Re: [qemu-s390x] [PATCH 3/3] vfio-ccw: add handling for asnyc channel instructions

[Cornelia Huck]

On Mon, 17 Dec 2018 16:54:31 -0500
Eric Farman <address@hidden> wrote:

> On 11/22/2018 11:54 AM, Cornelia Huck wrote:
> > Add a region to the vfio-ccw device that can be used to submit
> > asynchronous I/O instructions. ssch continues to be handled by the
> > existing I/O region; the new region handles hsch and csch.
> > 
> > Interrupt status continues to be reported through the same channels
> > as for ssch.
> > 
> > Signed-off-by: Cornelia Huck <address@hidden>
> > ---
> >   drivers/s390/cio/Makefile           |   3 +-
> >   drivers/s390/cio/vfio_ccw_async.c   |  88 ++++++++++++++++
> >   drivers/s390/cio/vfio_ccw_drv.c     |  48 ++++++---
> >   drivers/s390/cio/vfio_ccw_fsm.c     | 158 +++++++++++++++++++++++++++-
> >   drivers/s390/cio/vfio_ccw_ops.c     |  13 ++-
> >   drivers/s390/cio/vfio_ccw_private.h |   6 ++
> >   include/uapi/linux/vfio.h           |   4 +
> >   include/uapi/linux/vfio_ccw.h       |  12 +++
> >   8 files changed, 313 insertions(+), 19 deletions(-)
> >   create mode 100644 drivers/s390/cio/vfio_ccw_async.c

> > diff --git a/drivers/s390/cio/vfio_ccw_async.c 
> > b/drivers/s390/cio/vfio_ccw_async.c
> > new file mode 100644
> > index 000000000000..8c7f51d17d70
> > --- /dev/null
> > +++ b/drivers/s390/cio/vfio_ccw_async.c
> > @@ -0,0 +1,88 @@
> > +// SPDX-License-Identifier: GPL-2.0
> > +/*
> > + * Async I/O region for vfio_ccw
> > + *
> > + * Copyright Red Hat, Inc. 2018
> > + *
> > + * Author(s): Cornelia Huck <address@hidden>
> > + */
> > +
> > +#include <linux/vfio.h>
> > +#include <linux/mdev.h>
> > +
> > +#include "vfio_ccw_private.h"
> > +
> > +static size_t vfio_ccw_async_region_read(struct vfio_ccw_private *private, 
> >  
> 
> I think this should return ssize_t ?  (same for _write, below)

Yes, ssize_t makes more sense. Changed.

(vfio_pci_regops also has size_t; should probably be changed as well.)

> 
> > +                                    char __user *buf, size_t count,
> > +                                    loff_t *ppos)
> > +{
> > +   unsigned int i = VFIO_CCW_OFFSET_TO_INDEX(*ppos) - VFIO_CCW_NUM_REGIONS;
> > +   loff_t pos = *ppos & VFIO_CCW_OFFSET_MASK;
> > +   struct ccw_cmd_region *region;
> > +
> > +   if (pos + count > sizeof(*region))
> > +           return -EINVAL;
> > +
> > +   region = private->region[i].data;
> > +   if (copy_to_user(buf, (void *)region + pos, count))
> > +           return -EFAULT;
> > +
> > +   return count;
> > +
> > +}
> > +
> > +static size_t vfio_ccw_async_region_write(struct vfio_ccw_private *private,
> > +                                     const char __user *buf, size_t count,
> > +                                     loff_t *ppos)
> > +{
> > +   unsigned int i = VFIO_CCW_OFFSET_TO_INDEX(*ppos) - VFIO_CCW_NUM_REGIONS;
> > +   loff_t pos = *ppos & VFIO_CCW_OFFSET_MASK;
> > +   struct ccw_cmd_region *region;
> > +
> > +   if (pos + count > sizeof(*region))
> > +           return -EINVAL;
> > +
> > +   if (private->state == VFIO_CCW_STATE_NOT_OPER ||
> > +       private->state == VFIO_CCW_STATE_STANDBY)
> > +           return -EACCES;
> > +
> > +   region = private->region[i].data;
> > +   if (copy_from_user((void *)region + pos, buf, count))
> > +           return -EFAULT;
> > +
> > +   switch (region->command) {
> > +   case VFIO_CCW_ASYNC_CMD_HSCH:
> > +           vfio_ccw_fsm_event(private, VFIO_CCW_EVENT_HALT_REQ);
> > +           break;
> > +   case VFIO_CCW_ASYNC_CMD_CSCH:
> > +           vfio_ccw_fsm_event(private, VFIO_CCW_EVENT_CLEAR_REQ);
> > +           break;  
> 
> I find myself wondering why we add separate VFIO_CCW_EVENT_x_REQ entries 
> for HALT and CLEAR, rather than a single VFIO_CCW_EVENT_ASYNC_REQ and a 
> switch on cmd_region->command within it to go to fsm_do_halt, 
> fsm_do_clear, or whatever.

In the end, it probably does not matter much where we do the switch.
When I started writing this, I thought I would want to allow clear in
more states than halt; but that does not make much sense (best to let
the hardware sort it out; see also the other discussions around
concurrency.)

> 
> > +   default:
> > +           return -EINVAL;
> > +   }
> > +
> > +   return region->ret_code ? region->ret_code : count;
> > +}

(...)

> > diff --git a/drivers/s390/cio/vfio_ccw_fsm.c 
> > b/drivers/s390/cio/vfio_ccw_fsm.c
> > index f94aa01f9c36..0caf77e8f377 100644
> > --- a/drivers/s390/cio/vfio_ccw_fsm.c
> > +++ b/drivers/s390/cio/vfio_ccw_fsm.c

> > @@ -102,6 +179,20 @@ static void fsm_io_busy(struct vfio_ccw_private 
> > *private,
> >     private->io_region->ret_code = -EBUSY;
> >   }
> >   
> > +static void fsm_async_error(struct vfio_ccw_private *private,
> > +                       enum vfio_ccw_event event)
> > +{
> > +   pr_err("vfio-ccw: FSM: halt/clear request from state:%d\n",
> > +          private->state);  
> 
> Worth stating whether it's a Halt or Clear here, rather than leaving it 
> ambiguous?

Not sure. Also not sure if we want to fold the events, as you suggested
above :)

This also reminds me that I need to rebase this: some details in the
handling will need to be different without the BOXED state.