Tvrtko was looking through the execbuffer-ioctl and noticed that the uABI was tightly coupled to our internal engine identifiers. Close inspection also revealed that we leak those internal engine identifiers through the busy-ioctl, and those internal identifiers already do not match the user identifiers. Fortuitiously, there is only one user of the set of busy rings from the busy-ioctl, and they only wish to choose between the RENDER and the BLT engines. Let's fix the userspace ABI while we still can. v2: Update the uAPI documentation to explain the identifiers.
Showing the changes to include/uapi (1 of 5 files). Show all
Unified · Split
| @@ -812,10 +812,35 @@ struct drm_i915_gem_busy { · 811 unmodified lines | ||
| 812 | 812 | /** Handle of the buffer to check for busy */ |
| 813 | 813 | __u32 handle; |
| 814 | 814 | |
| 815 | − /** Return busy status (1 if busy, 0 if idle). | |
| 816 | − * The high word is used to indicate on which rings the object | |
| 817 | − * currently resides: | |
| 818 | − * 16:31 - busy (r or r/w) rings (16 render, 17 bsd, 18 blt, etc) | |
| 815 | + /** Return busy status | |
| 816 | + * | |
| 817 | + * A return of 0 implies that the object is idle (after | |
| 818 | + * having flushed any pending activity), and a non-zero return that | |
| 819 | + * the object is still in-flight on the GPU. (The GPU has not yet | |
| 820 | + * signaled completion for all pending requests that reference the | |
| 821 | + * object.) | |
| 822 | + * | |
| 823 | + * The returned dword is split into two fields to indicate both | |
| 824 | + * the engines on which the object is being read, and the | |
| 825 | + * engine on which it is currently being written (if any). | |
| 826 | + * | |
| 827 | + * The low word (bits 0:15) indicate if the object is being written | |
| 828 | + * to by any engine (there can only be one, as the GEM implicit | |
| 829 | + * synchronisation rules force writes to be serialised). Only the | |
| 830 | + * engine for the last write is reported. | |
| 831 | + * | |
| 832 | + * The high word (bits 16:31) are a bitmask of which engines are | |
| 833 | + * currently reading from the object. Multiple engines may be | |
| 834 | + * reading from the object simultaneously. | |
| 835 | + * | |
| 836 | + * The value of each engine is the same as specified in the | |
| 837 | + * EXECBUFFER2 ioctl, i.e. I915_EXEC_RENDER, I915_EXEC_BSD etc. | |
| 838 | + * Note I915_EXEC_DEFAULT is a symbolic value and is mapped to | |
| 839 | + * the I915_EXEC_RENDER engine for execution, and so it is never | |
| 840 | + * reported as active itself. Some hardware may have parallel | |
| 841 | + * execution engines, e.g. multiple media engines, which are | |
| 842 | + * mapped to the same identifier in the EXECBUFFER2 ioctl and | |
| 843 | + * so are not separately reported for busyness. | |
| 819 | 844 | */ |
| 820 | 845 | __u32 busy; |
| 821 | 846 | }; |