xstate.rst

  1Using XSTATE features in user space applications
  2================================================
  3
  4The x86 architecture supports floating-point extensions which are
  5enumerated via CPUID. Applications consult CPUID and use XGETBV to
  6evaluate which features have been enabled by the kernel XCR0.
  7
  8Up to AVX-512 and PKRU states, these features are automatically enabled by
  9the kernel if available. Features like AMX TILE_DATA (XSTATE component 18)
 10are enabled by XCR0 as well, but the first use of related instruction is
 11trapped by the kernel because by default the required large XSTATE buffers
 12are not allocated automatically.
 13
 14The purpose for dynamic features
 15--------------------------------
 16
 17Legacy userspace libraries often have hard-coded, static sizes for
 18alternate signal stacks, often using MINSIGSTKSZ which is typically 2KB.
 19That stack must be able to store at *least* the signal frame that the
 20kernel sets up before jumping into the signal handler. That signal frame
 21must include an XSAVE buffer defined by the CPU.
 22
 23However, that means that the size of signal stacks is dynamic, not static,
 24because different CPUs have differently-sized XSAVE buffers. A compiled-in
 25size of 2KB with existing applications is too small for new CPU features
 26like AMX. Instead of universally requiring larger stack, with the dynamic
 27enabling, the kernel can enforce userspace applications to have
 28properly-sized altstacks.
 29
 30Using dynamically enabled XSTATE features in user space applications
 31--------------------------------------------------------------------
 32
 33The kernel provides an arch_prctl(2) based mechanism for applications to
 34request the usage of such features. The arch_prctl(2) options related to
 35this are:
 36
 37-ARCH_GET_XCOMP_SUPP
 38
 39 arch_prctl(ARCH_GET_XCOMP_SUPP, &features);
 40
 41 ARCH_GET_XCOMP_SUPP stores the supported features in userspace storage of
 42 type uint64_t. The second argument is a pointer to that storage.
 43
 44-ARCH_GET_XCOMP_PERM
 45
 46 arch_prctl(ARCH_GET_XCOMP_PERM, &features);
 47
 48 ARCH_GET_XCOMP_PERM stores the features for which the userspace process
 49 has permission in userspace storage of type uint64_t. The second argument
 50 is a pointer to that storage.
 51
 52-ARCH_REQ_XCOMP_PERM
 53
 54 arch_prctl(ARCH_REQ_XCOMP_PERM, feature_nr);
 55
 56 ARCH_REQ_XCOMP_PERM allows to request permission for a dynamically enabled
 57 feature or a feature set. A feature set can be mapped to a facility, e.g.
 58 AMX, and can require one or more XSTATE components to be enabled.
 59
 60 The feature argument is the number of the highest XSTATE component which
 61 is required for a facility to work.
 62
 63When requesting permission for a feature, the kernel checks the
 64availability. The kernel ensures that sigaltstacks in the process's tasks
 65are large enough to accommodate the resulting large signal frame. It
 66enforces this both during ARCH_REQ_XCOMP_SUPP and during any subsequent
 67sigaltstack(2) calls. If an installed sigaltstack is smaller than the
 68resulting sigframe size, ARCH_REQ_XCOMP_SUPP results in -ENOSUPP. Also,
 69sigaltstack(2) results in -ENOMEM if the requested altstack is too small
 70for the permitted features.
 71
 72Permission, when granted, is valid per process. Permissions are inherited
 73on fork(2) and cleared on exec(3).
 74
 75The first use of an instruction related to a dynamically enabled feature is
 76trapped by the kernel. The trap handler checks whether the process has
 77permission to use the feature. If the process has no permission then the
 78kernel sends SIGILL to the application. If the process has permission then
 79the handler allocates a larger xstate buffer for the task so the large
 80state can be context switched. In the unlikely cases that the allocation
 81fails, the kernel sends SIGSEGV.
 82
 83AMX TILE_DATA enabling example
 84^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 85
 86Below is the example of how userspace applications enable
 87TILE_DATA dynamically:
 88
 89  1. The application first needs to query the kernel for AMX
 90     support::
 91
 92        #include <asm/prctl.h>
 93        #include <sys/syscall.h>
 94        #include <stdio.h>
 95        #include <unistd.h>
 96
 97        #ifndef ARCH_GET_XCOMP_SUPP
 98        #define ARCH_GET_XCOMP_SUPP  0x1021
 99        #endif
100
101        #ifndef ARCH_XCOMP_TILECFG
102        #define ARCH_XCOMP_TILECFG   17
103        #endif
104
105        #ifndef ARCH_XCOMP_TILEDATA
106        #define ARCH_XCOMP_TILEDATA  18
107        #endif
108
109        #define MASK_XCOMP_TILE      ((1 << ARCH_XCOMP_TILECFG) | \
110                                      (1 << ARCH_XCOMP_TILEDATA))
111
112        unsigned long features;
113        long rc;
114
115        ...
116
117        rc = syscall(SYS_arch_prctl, ARCH_GET_XCOMP_SUPP, &features);
118
119        if (!rc && (features & MASK_XCOMP_TILE) == MASK_XCOMP_TILE)
120            printf("AMX is available.\n");
121
122  2. After that, determining support for AMX, an application must
123     explicitly ask permission to use it::
124
125        #ifndef ARCH_REQ_XCOMP_PERM
126        #define ARCH_REQ_XCOMP_PERM  0x1023
127        #endif
128
129        ...
130
131        rc = syscall(SYS_arch_prctl, ARCH_REQ_XCOMP_PERM, ARCH_XCOMP_TILEDATA);
132
133        if (!rc)
134            printf("AMX is ready for use.\n");
135
136Note this example does not include the sigaltstack preparation.
137
138Dynamic features in signal frames
139---------------------------------
140
141Dynamically enabled features are not written to the signal frame upon signal
142entry if the feature is in its initial configuration.  This differs from
143non-dynamic features which are always written regardless of their
144configuration.  Signal handlers can examine the XSAVE buffer's XSTATE_BV
145field to determine if a features was written.
146
147Dynamic features for virtual machines
148-------------------------------------
149
150The permission for the guest state component needs to be managed separately
151from the host, as they are exclusive to each other. A coupled of options
152are extended to control the guest permission:
153
154-ARCH_GET_XCOMP_GUEST_PERM
155
156 arch_prctl(ARCH_GET_XCOMP_GUEST_PERM, &features);
157
158 ARCH_GET_XCOMP_GUEST_PERM is a variant of ARCH_GET_XCOMP_PERM. So it
159 provides the same semantics and functionality but for the guest
160 components.
161
162-ARCH_REQ_XCOMP_GUEST_PERM
163
164 arch_prctl(ARCH_REQ_XCOMP_GUEST_PERM, feature_nr);
165
166 ARCH_REQ_XCOMP_GUEST_PERM is a variant of ARCH_REQ_XCOMP_PERM. It has the
167 same semantics for the guest permission. While providing a similar
168 functionality, this comes with a constraint. Permission is frozen when the
169 first VCPU is created. Any attempt to change permission after that point
170 is going to be rejected. So, the permission has to be requested before the
171 first VCPU creation.
172
173Note that some VMMs may have already established a set of supported state
174components. These options are not presumed to support any particular VMM.