Re: [Qemu-devel] [PATCH RFC 1/2] vfio-ccw: forward halt/clear to device if supported

[Pierre Morel <pmorel@linux.ibm.com>]

On 15/05/2018 18:01, Cornelia Huck wrote:
> On Fri, 11 May 2018 11:53:52 +0200
> Pierre Morel <pmorel@linux.ibm.com> wrote:
>
>> On 09/05/2018 17:49, Cornelia Huck wrote:
>>> The initial version of vfio-ccw did not support forwarding of the
>>> halt or clear functions to the device, and we had to emulate them
>>> instead.
>>>
>>> For versions of the vfio-ccw kernel implementation that indeed do
>>> support halt/clear (by indicating them in the fctl of the scsw in
>>> the io_region), we can simply start making use of it. If the kernel
>>> does not support handling halt/clear, fall back to emulation.
>>>
>>> Signed-off-by: Cornelia Huck <cohuck@redhat.com>
>>> ---
>>>    hw/s390x/css.c         | 32 ++++++++++++++++++++++++++++----
>>>    hw/vfio/ccw.c          | 11 +++++++++--
>>>    include/hw/s390x/css.h | 10 +++++++---
>>>    3 files changed, 44 insertions(+), 9 deletions(-)
>>>
>>> diff --git a/hw/s390x/css.c b/hw/s390x/css.c
>>> index 301bf1772f..b6727d0607 100644
>>> --- a/hw/s390x/css.c
>>> +++ b/hw/s390x/css.c
>>> @@ -1180,6 +1180,16 @@ static void sch_handle_start_func_virtual(SubchDev *sch)
>>>    
>>>    }
>>>    
>>> +static IOInstEnding sch_handle_clear_func_passthrough(SubchDev *sch)
>>> +{
>>> +    return s390_ccw_cmd_request(sch);
>>> +}
>>> +
>>> +static IOInstEnding sch_handle_halt_func_passthrough(SubchDev *sch)
>>> +{
>>> +    return s390_ccw_cmd_request(sch);
>>> +}
>>> +
>>>    static IOInstEnding sch_handle_start_func_passthrough(SubchDev *sch)
>>>    {
>>>    
>>> @@ -1233,13 +1243,27 @@ IOInstEnding do_subchannel_work_virtual(SubchDev *sch)
>>>    IOInstEnding do_subchannel_work_passthrough(SubchDev *sch)
>>>    {
>>>        SCSW *s = &sch->curr_status.scsw;
>>> +    static bool no_halt_clear;
>>>    
>>> +    /* if the kernel does not support halt/clear, fall back to emulation */
>>>        if (s->ctrl & SCSW_FCTL_CLEAR_FUNC) {
>>> -        /* TODO: Clear handling */
>>> -        sch_handle_clear_func(sch);
>>> +        if (no_halt_clear) {
>>> +            sch_handle_clear_func(sch);
>>> +        } else {
>>> +            if (sch_handle_clear_func_passthrough(sch) == IOINST_OPNOTSUPP) {
>>> +                no_halt_clear = true;
>>> +                sch_handle_halt_func(sch);
>>> +            }
>>> +        }
>>>        } else if (s->ctrl & SCSW_FCTL_HALT_FUNC) {
>>> -        /* TODO: Halt handling */
>>> -        sch_handle_halt_func(sch);
>>> +        if (no_halt_clear) {
>>> +            sch_handle_halt_func(sch);
>>> +        } else {
>>> +            if (sch_handle_halt_func_passthrough(sch) == IOINST_OPNOTSUPP) {
>>> +                no_halt_clear = true;
>>> +                sch_handle_halt_func(sch);
>>> +            }
>>> +        }
>>>        } else if (s->ctrl & SCSW_FCTL_START_FUNC) {
>>>            return sch_handle_start_func_passthrough(sch);
>>>        }
>>> diff --git a/hw/vfio/ccw.c b/hw/vfio/ccw.c
>>> index e67392c5f9..247901ae41 100644
>>> --- a/hw/vfio/ccw.c
>>> +++ b/hw/vfio/ccw.c
>>> @@ -60,6 +60,7 @@ static IOInstEnding vfio_ccw_handle_request(SubchDev *sch)
>>>    
>>>        memset(region, 0, sizeof(*region));
>>>    
>>> +    /* orb is only valid for ssch, but does not hurt for other functions */
>>>        memcpy(region->orb_area, &sch->orb, sizeof(ORB));
>>>        memcpy(region->scsw_area, &sch->curr_status.scsw, sizeof(SCSW));
>>>    
>>> @@ -70,8 +71,12 @@ again:
>>>            if (errno == EAGAIN) {
>>>                goto again;
>>>            }
>>> -        error_report("vfio-ccw: wirte I/O region failed with errno=%d", errno);
>>> -        ret = -errno;
>>> +        /* handle not supported operations like halt/clear on older kernels */
>>> +        if (ret != -EOPNOTSUPP) {
>>> +            error_report("vfio-ccw: write I/O region failed with errno=%d",
>>> +                         errno);
>>> +            ret = -errno;
>>> +        }
>>>        } else {
>>>            ret = region->ret_code;
>>>        }
>>> @@ -83,6 +88,8 @@ again:
>>>        case -ENODEV:
>>>        case -EACCES:
>>>            return IOINST_CC_NOT_OPERATIONAL;
>>> +    case -EOPNOTSUPP:
>>> +        return IOINST_OPNOTSUPP;
>>>        case -EFAULT:
>>>        default:
>>>            sch_gen_unit_exception(sch);
>>> diff --git a/include/hw/s390x/css.h b/include/hw/s390x/css.h
>>> index 35facb47d2..e33f26882b 100644
>>> --- a/include/hw/s390x/css.h
>>> +++ b/include/hw/s390x/css.h
>>> @@ -100,9 +100,11 @@ typedef struct CcwDataStream {
>>>    } CcwDataStream;
>>>    
>>>    /*
>>> - * IO instructions conclude according to this. Currently we have only
>>> - * cc codes. Valid values are 0, 1, 2, 3 and the generic semantic for
>>> + * IO instructions conclude according to this. One class of values are
>>> + * cc codes: Valid values are 0, 1, 2, 3 and the generic semantic for
>>>     * IO instructions is described briefly. For more details consult the PoP.
>>> + * Additionally, other endings may occur due to internal processing errors
>>> + * like lack of support for an operation.
>>>     */
>>>    typedef enum IOInstEnding {
>>>        /* produced expected result */
>>> @@ -112,7 +114,9 @@ typedef enum IOInstEnding {
>>>        /* inst. ineffective because busy with previously initiated function */
>>>        IOINST_CC_BUSY = 2,
>>>        /* inst. ineffective because not operational */
>>> -    IOINST_CC_NOT_OPERATIONAL = 3
>>> +    IOINST_CC_NOT_OPERATIONAL = 3,
>>> +    /* internal: operation not supported */
>>> +    IOINST_OPNOTSUPP = 4
>>>    } IOInstEnding;
>>>    
>>>    typedef struct SubchDev SubchDev;
>>
>> Couldn't we introduce ABI versioning ?
> Can you elaborate what you're referring to?
>
> If you mean checking capabilities of the kernel or so: If we can avoid
> that and just try (and stop if it does not work), I'd prefer that (no
> dependencies).

VFIO_CHECK_EXTENSION is already used in different drivers
for this kind of interface extension.
We could use it to setup appropriate callbacks for scsh/csch/xsch/hsch
depending on the extension argument.

>
> The IOINST_OPNOTSUPP is a bit ugly, but I did not see a more elegant
> way to pass 'not supported' up to the caller.
>

-- 
Pierre Morel
Linux/KVM/QEMU in Böblingen - Germany