xref: /linux/Documentation/virt/kvm/s390/s390-diag.rst (revision 34f7c6e7d4396090692a09789db231e12cb4762b)
1.. SPDX-License-Identifier: GPL-2.0
2
3=============================
4The s390 DIAGNOSE call on KVM
5=============================
6
7KVM on s390 supports the DIAGNOSE call for making hypercalls, both for
8native hypercalls and for selected hypercalls found on other s390
9hypervisors.
10
11Note that bits are numbered as by the usual s390 convention (most significant
12bit on the left).
13
14
15General remarks
16---------------
17
18DIAGNOSE calls by the guest cause a mandatory intercept. This implies
19all supported DIAGNOSE calls need to be handled by either KVM or its
20userspace.
21
22All DIAGNOSE calls supported by KVM use the RS-a format::
23
24  --------------------------------------
25  |  '83'  | R1 | R3 | B2 |     D2     |
26  --------------------------------------
27  0        8    12   16   20           31
28
29The second-operand address (obtained by the base/displacement calculation)
30is not used to address data. Instead, bits 48-63 of this address specify
31the function code, and bits 0-47 are ignored.
32
33The supported DIAGNOSE function codes vary by the userspace used. For
34DIAGNOSE function codes not specific to KVM, please refer to the
35documentation for the s390 hypervisors defining them.
36
37
38DIAGNOSE function code 'X'500' - KVM virtio functions
39-----------------------------------------------------
40
41If the function code specifies 0x500, various virtio-related functions
42are performed.
43
44General register 1 contains the virtio subfunction code. Supported
45virtio subfunctions depend on KVM's userspace. Generally, userspace
46provides either s390-virtio (subcodes 0-2) or virtio-ccw (subcode 3).
47
48Upon completion of the DIAGNOSE instruction, general register 2 contains
49the function's return code, which is either a return code or a subcode
50specific value.
51
52Subcode 0 - s390-virtio notification and early console printk
53    Handled by userspace.
54
55Subcode 1 - s390-virtio reset
56    Handled by userspace.
57
58Subcode 2 - s390-virtio set status
59    Handled by userspace.
60
61Subcode 3 - virtio-ccw notification
62    Handled by either userspace or KVM (ioeventfd case).
63
64    General register 2 contains a subchannel-identification word denoting
65    the subchannel of the virtio-ccw proxy device to be notified.
66
67    General register 3 contains the number of the virtqueue to be notified.
68
69    General register 4 contains a 64bit identifier for KVM usage (the
70    kvm_io_bus cookie). If general register 4 does not contain a valid
71    identifier, it is ignored.
72
73    After completion of the DIAGNOSE call, general register 2 may contain
74    a 64bit identifier (in the kvm_io_bus cookie case), or a negative
75    error value, if an internal error occurred.
76
77    See also the virtio standard for a discussion of this hypercall.
78
79
80DIAGNOSE function code 'X'501 - KVM breakpoint
81----------------------------------------------
82
83If the function code specifies 0x501, breakpoint functions may be performed.
84This function code is handled by userspace.
85
86This diagnose function code has no subfunctions and uses no parameters.
87
88
89DIAGNOSE function code 'X'9C - Voluntary Time Slice Yield
90---------------------------------------------------------
91
92General register 1 contains the target CPU address.
93
94In a guest of a hypervisor like LPAR, KVM or z/VM using shared host CPUs,
95DIAGNOSE with function code 0x9c may improve system performance by
96yielding the host CPU on which the guest CPU is running to be assigned
97to another guest CPU, preferably the logical CPU containing the specified
98target CPU.
99
100
101DIAG 'X'9C forwarding
102+++++++++++++++++++++
103
104The guest may send a DIAGNOSE 0x9c in order to yield to a certain
105other vcpu. An example is a Linux guest that tries to yield to the vcpu
106that is currently holding a spinlock, but not running.
107
108However, on the host the real cpu backing the vcpu may itself not be
109running.
110Forwarding the DIAGNOSE 0x9c initially sent by the guest to yield to
111the backing cpu will hopefully cause that cpu, and thus subsequently
112the guest's vcpu, to be scheduled.
113
114
115diag9c_forwarding_hz
116    KVM kernel parameter allowing to specify the maximum number of DIAGNOSE
117    0x9c forwarding per second in the purpose of avoiding a DIAGNOSE 0x9c
118    forwarding storm.
119    A value of 0 turns the forwarding off.
120