xref: /openbsd-src/sys/dev/pci/drm/i915/gt/intel_sseu.h (revision 9e88da3ec8f1a5b67c0b8f380d0056e20caf42d2)
15ca02815Sjsg /* SPDX-License-Identifier: MIT */
2c349dbc7Sjsg /*
3c349dbc7Sjsg  * Copyright © 2019 Intel Corporation
4c349dbc7Sjsg  */
5c349dbc7Sjsg 
6c349dbc7Sjsg #ifndef __INTEL_SSEU_H__
7c349dbc7Sjsg #define __INTEL_SSEU_H__
8c349dbc7Sjsg 
9c349dbc7Sjsg #include <linux/types.h>
10c349dbc7Sjsg #include <linux/kernel.h>
11c349dbc7Sjsg 
12c349dbc7Sjsg #include "i915_gem.h"
13c349dbc7Sjsg 
14c349dbc7Sjsg struct drm_i915_private;
15ad8b1aafSjsg struct intel_gt;
16ad8b1aafSjsg struct drm_printer;
17c349dbc7Sjsg 
181bb76ff1Sjsg /*
191bb76ff1Sjsg  * Maximum number of slices on older platforms.  Slices no longer exist
201bb76ff1Sjsg  * starting on Xe_HP ("gslices," "cslices," etc. are a different concept and
211bb76ff1Sjsg  * are not expressed through fusing).
221bb76ff1Sjsg  */
231bb76ff1Sjsg #define GEN_MAX_HSW_SLICES		3
241bb76ff1Sjsg 
251bb76ff1Sjsg /*
261bb76ff1Sjsg  * Maximum number of subslices that can exist within a HSW-style slice.  This
271bb76ff1Sjsg  * is only relevant to pre-Xe_HP platforms (Xe_HP and beyond use the
281bb76ff1Sjsg  * I915_MAX_SS_FUSE_BITS value below).
291bb76ff1Sjsg  */
30*9e88da3eSjsg #define GEN_MAX_SS_PER_HSW_SLICE	8
311bb76ff1Sjsg 
321bb76ff1Sjsg /*
331bb76ff1Sjsg  * Maximum number of 32-bit registers used by hardware to express the
341bb76ff1Sjsg  * enabled/disabled subslices.
351bb76ff1Sjsg  */
361bb76ff1Sjsg #define I915_MAX_SS_FUSE_REGS	2
371bb76ff1Sjsg #define I915_MAX_SS_FUSE_BITS	(I915_MAX_SS_FUSE_REGS * 32)
381bb76ff1Sjsg 
391bb76ff1Sjsg /* Maximum number of EUs that can exist within a subslice or DSS. */
401bb76ff1Sjsg #define GEN_MAX_EUS_PER_SS		16
411bb76ff1Sjsg 
421bb76ff1Sjsg #define SSEU_MAX(a, b)			((a) > (b) ? (a) : (b))
431bb76ff1Sjsg 
441bb76ff1Sjsg /* The maximum number of bits needed to express each subslice/DSS independently */
451bb76ff1Sjsg #define GEN_SS_MASK_SIZE		SSEU_MAX(I915_MAX_SS_FUSE_BITS, \
461bb76ff1Sjsg 						 GEN_MAX_HSW_SLICES * GEN_MAX_SS_PER_HSW_SLICE)
471bb76ff1Sjsg 
48c349dbc7Sjsg #define GEN_SSEU_STRIDE(max_entries)	DIV_ROUND_UP(max_entries, BITS_PER_BYTE)
491bb76ff1Sjsg #define GEN_MAX_SUBSLICE_STRIDE		GEN_SSEU_STRIDE(GEN_SS_MASK_SIZE)
501bb76ff1Sjsg #define GEN_MAX_EU_STRIDE		GEN_SSEU_STRIDE(GEN_MAX_EUS_PER_SS)
51c349dbc7Sjsg 
525ca02815Sjsg #define GEN_DSS_PER_GSLICE	4
535ca02815Sjsg #define GEN_DSS_PER_CSLICE	8
545ca02815Sjsg #define GEN_DSS_PER_MSLICE	8
555ca02815Sjsg 
561bb76ff1Sjsg #define GEN_MAX_GSLICES		(I915_MAX_SS_FUSE_BITS / GEN_DSS_PER_GSLICE)
571bb76ff1Sjsg #define GEN_MAX_CSLICES		(I915_MAX_SS_FUSE_BITS / GEN_DSS_PER_CSLICE)
581bb76ff1Sjsg 
591bb76ff1Sjsg typedef union {
601bb76ff1Sjsg 	u8 hsw[GEN_MAX_HSW_SLICES];
611bb76ff1Sjsg 
621bb76ff1Sjsg 	/* Bitmap compatible with linux/bitmap.h; may exceed size of u64 */
631bb76ff1Sjsg 	unsigned long xehp[BITS_TO_LONGS(I915_MAX_SS_FUSE_BITS)];
641bb76ff1Sjsg } intel_sseu_ss_mask_t;
651bb76ff1Sjsg 
661bb76ff1Sjsg #define XEHP_BITMAP_BITS(mask)	((int)BITS_PER_TYPE(typeof(mask.xehp)))
671bb76ff1Sjsg 
68c349dbc7Sjsg struct sseu_dev_info {
69c349dbc7Sjsg 	u8 slice_mask;
701bb76ff1Sjsg 	intel_sseu_ss_mask_t subslice_mask;
711bb76ff1Sjsg 	intel_sseu_ss_mask_t geometry_subslice_mask;
721bb76ff1Sjsg 	intel_sseu_ss_mask_t compute_subslice_mask;
731bb76ff1Sjsg 	union {
741bb76ff1Sjsg 		u16 hsw[GEN_MAX_HSW_SLICES][GEN_MAX_SS_PER_HSW_SLICE];
751bb76ff1Sjsg 		u16 xehp[I915_MAX_SS_FUSE_BITS];
761bb76ff1Sjsg 	} eu_mask;
771bb76ff1Sjsg 
78c349dbc7Sjsg 	u16 eu_total;
79c349dbc7Sjsg 	u8 eu_per_subslice;
80c349dbc7Sjsg 	u8 min_eu_in_pool;
81c349dbc7Sjsg 	/* For each slice, which subslice(s) has(have) 7 EUs (bitfield)? */
82c349dbc7Sjsg 	u8 subslice_7eu[3];
83c349dbc7Sjsg 	u8 has_slice_pg:1;
84c349dbc7Sjsg 	u8 has_subslice_pg:1;
85c349dbc7Sjsg 	u8 has_eu_pg:1;
861bb76ff1Sjsg 	/*
871bb76ff1Sjsg 	 * For Xe_HP and beyond, the hardware no longer has traditional slices
881bb76ff1Sjsg 	 * so we just report the entire DSS pool under a fake "slice 0."
891bb76ff1Sjsg 	 */
901bb76ff1Sjsg 	u8 has_xehp_dss:1;
91c349dbc7Sjsg 
92c349dbc7Sjsg 	/* Topology fields */
93c349dbc7Sjsg 	u8 max_slices;
94c349dbc7Sjsg 	u8 max_subslices;
95c349dbc7Sjsg 	u8 max_eus_per_subslice;
96c349dbc7Sjsg };
97c349dbc7Sjsg 
98c349dbc7Sjsg /*
99c349dbc7Sjsg  * Powergating configuration for a particular (context,engine).
100c349dbc7Sjsg  */
101c349dbc7Sjsg struct intel_sseu {
102c349dbc7Sjsg 	u8 slice_mask;
103c349dbc7Sjsg 	u8 subslice_mask;
104c349dbc7Sjsg 	u8 min_eus_per_subslice;
105c349dbc7Sjsg 	u8 max_eus_per_subslice;
106c349dbc7Sjsg };
107c349dbc7Sjsg 
108c349dbc7Sjsg static inline struct intel_sseu
intel_sseu_from_device_info(const struct sseu_dev_info * sseu)109c349dbc7Sjsg intel_sseu_from_device_info(const struct sseu_dev_info *sseu)
110c349dbc7Sjsg {
111c349dbc7Sjsg 	struct intel_sseu value = {
112c349dbc7Sjsg 		.slice_mask = sseu->slice_mask,
1131bb76ff1Sjsg 		.subslice_mask = sseu->subslice_mask.hsw[0],
114c349dbc7Sjsg 		.min_eus_per_subslice = sseu->max_eus_per_subslice,
115c349dbc7Sjsg 		.max_eus_per_subslice = sseu->max_eus_per_subslice,
116c349dbc7Sjsg 	};
117c349dbc7Sjsg 
118c349dbc7Sjsg 	return value;
119c349dbc7Sjsg }
120c349dbc7Sjsg 
121c349dbc7Sjsg static inline bool
intel_sseu_has_subslice(const struct sseu_dev_info * sseu,int slice,int subslice)122c349dbc7Sjsg intel_sseu_has_subslice(const struct sseu_dev_info *sseu, int slice,
123c349dbc7Sjsg 			int subslice)
124c349dbc7Sjsg {
1251bb76ff1Sjsg 	if (slice >= sseu->max_slices ||
1261bb76ff1Sjsg 	    subslice >= sseu->max_subslices)
1271bb76ff1Sjsg 		return false;
128c349dbc7Sjsg 
1291bb76ff1Sjsg 	if (sseu->has_xehp_dss)
1301bb76ff1Sjsg 		return test_bit(subslice, sseu->subslice_mask.xehp);
1311bb76ff1Sjsg 	else
1321bb76ff1Sjsg 		return sseu->subslice_mask.hsw[slice] & BIT(subslice);
1331bb76ff1Sjsg }
134c349dbc7Sjsg 
1351bb76ff1Sjsg /*
1361bb76ff1Sjsg  * Used to obtain the index of the first DSS.  Can start searching from the
1371bb76ff1Sjsg  * beginning of a specific dss group (e.g., gslice, cslice, etc.) if
1381bb76ff1Sjsg  * groupsize and groupnum are non-zero.
1391bb76ff1Sjsg  */
1401bb76ff1Sjsg static inline unsigned int
intel_sseu_find_first_xehp_dss(const struct sseu_dev_info * sseu,int groupsize,int groupnum)1411bb76ff1Sjsg intel_sseu_find_first_xehp_dss(const struct sseu_dev_info *sseu, int groupsize,
1421bb76ff1Sjsg 			       int groupnum)
1431bb76ff1Sjsg {
1441bb76ff1Sjsg 	return find_next_bit(sseu->subslice_mask.xehp,
1451bb76ff1Sjsg 			     XEHP_BITMAP_BITS(sseu->subslice_mask),
1461bb76ff1Sjsg 			     groupnum * groupsize);
147c349dbc7Sjsg }
148c349dbc7Sjsg 
149c349dbc7Sjsg void intel_sseu_set_info(struct sseu_dev_info *sseu, u8 max_slices,
150c349dbc7Sjsg 			 u8 max_subslices, u8 max_eus_per_subslice);
151c349dbc7Sjsg 
152c349dbc7Sjsg unsigned int
153c349dbc7Sjsg intel_sseu_subslice_total(const struct sseu_dev_info *sseu);
154c349dbc7Sjsg 
155c349dbc7Sjsg unsigned int
1561bb76ff1Sjsg intel_sseu_get_hsw_subslices(const struct sseu_dev_info *sseu, u8 slice);
157c349dbc7Sjsg 
1581bb76ff1Sjsg intel_sseu_ss_mask_t
1591bb76ff1Sjsg intel_sseu_get_compute_subslices(const struct sseu_dev_info *sseu);
160c349dbc7Sjsg 
161ad8b1aafSjsg void intel_sseu_info_init(struct intel_gt *gt);
162ad8b1aafSjsg 
163ad8b1aafSjsg u32 intel_sseu_make_rpcs(struct intel_gt *gt,
164c349dbc7Sjsg 			 const struct intel_sseu *req_sseu);
165c349dbc7Sjsg 
166ad8b1aafSjsg void intel_sseu_dump(const struct sseu_dev_info *sseu, struct drm_printer *p);
1671bb76ff1Sjsg void intel_sseu_print_topology(struct drm_i915_private *i915,
1681bb76ff1Sjsg 			       const struct sseu_dev_info *sseu,
169ad8b1aafSjsg 			       struct drm_printer *p);
170ad8b1aafSjsg 
1711bb76ff1Sjsg u16 intel_slicemask_from_xehp_dssmask(intel_sseu_ss_mask_t dss_mask, int dss_per_slice);
1721bb76ff1Sjsg 
1731bb76ff1Sjsg int intel_sseu_copy_eumask_to_user(void __user *to,
1741bb76ff1Sjsg 				   const struct sseu_dev_info *sseu);
1751bb76ff1Sjsg int intel_sseu_copy_ssmask_to_user(void __user *to,
1761bb76ff1Sjsg 				   const struct sseu_dev_info *sseu);
1771bb76ff1Sjsg 
1781bb76ff1Sjsg void intel_sseu_print_ss_info(const char *type,
1791bb76ff1Sjsg 			      const struct sseu_dev_info *sseu,
1801bb76ff1Sjsg 			      struct seq_file *m);
1815ca02815Sjsg 
182c349dbc7Sjsg #endif /* __INTEL_SSEU_H__ */
183