cachepc-linux

Fork of AMDESE/linux with modifications for CachePC side-channel attack
git clone https://git.sinitax.com/sinitax/cachepc-linux
Log | Files | Refs | README | LICENSE | sfeed.txt

vmmgp100.c (16647B)


      1/*
      2 * Copyright 2017 Red Hat Inc.
      3 *
      4 * Permission is hereby granted, free of charge, to any person obtaining a
      5 * copy of this software and associated documentation files (the "Software"),
      6 * to deal in the Software without restriction, including without limitation
      7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
      8 * and/or sell copies of the Software, and to permit persons to whom the
      9 * Software is furnished to do so, subject to the following conditions:
     10 *
     11 * The above copyright notice and this permission notice shall be included in
     12 * all copies or substantial portions of the Software.
     13 *
     14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
     15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
     16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
     17 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
     18 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
     19 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
     20 * OTHER DEALINGS IN THE SOFTWARE.
     21 */
     22#include "vmm.h"
     23
     24#include <core/client.h>
     25#include <subdev/fb.h>
     26#include <subdev/ltc.h>
     27#include <subdev/timer.h>
     28#include <engine/gr.h>
     29
     30#include <nvif/ifc00d.h>
     31#include <nvif/unpack.h>
     32
     33static void
     34gp100_vmm_pfn_unmap(struct nvkm_vmm *vmm,
     35		    struct nvkm_mmu_pt *pt, u32 ptei, u32 ptes)
     36{
     37	struct device *dev = vmm->mmu->subdev.device->dev;
     38	dma_addr_t addr;
     39
     40	nvkm_kmap(pt->memory);
     41	while (ptes--) {
     42		u32 datalo = nvkm_ro32(pt->memory, pt->base + ptei * 8 + 0);
     43		u32 datahi = nvkm_ro32(pt->memory, pt->base + ptei * 8 + 4);
     44		u64 data   = (u64)datahi << 32 | datalo;
     45		if ((data & (3ULL << 1)) != 0) {
     46			addr = (data >> 8) << 12;
     47			dma_unmap_page(dev, addr, PAGE_SIZE, DMA_BIDIRECTIONAL);
     48		}
     49		ptei++;
     50	}
     51	nvkm_done(pt->memory);
     52}
     53
     54static bool
     55gp100_vmm_pfn_clear(struct nvkm_vmm *vmm,
     56		    struct nvkm_mmu_pt *pt, u32 ptei, u32 ptes)
     57{
     58	bool dma = false;
     59	nvkm_kmap(pt->memory);
     60	while (ptes--) {
     61		u32 datalo = nvkm_ro32(pt->memory, pt->base + ptei * 8 + 0);
     62		u32 datahi = nvkm_ro32(pt->memory, pt->base + ptei * 8 + 4);
     63		u64 data   = (u64)datahi << 32 | datalo;
     64		if ((data & BIT_ULL(0)) && (data & (3ULL << 1)) != 0) {
     65			VMM_WO064(pt, vmm, ptei * 8, data & ~BIT_ULL(0));
     66			dma = true;
     67		}
     68		ptei++;
     69	}
     70	nvkm_done(pt->memory);
     71	return dma;
     72}
     73
     74static void
     75gp100_vmm_pgt_pfn(struct nvkm_vmm *vmm, struct nvkm_mmu_pt *pt,
     76		  u32 ptei, u32 ptes, struct nvkm_vmm_map *map)
     77{
     78	struct device *dev = vmm->mmu->subdev.device->dev;
     79	dma_addr_t addr;
     80
     81	nvkm_kmap(pt->memory);
     82	for (; ptes; ptes--, map->pfn++) {
     83		u64 data = 0;
     84
     85		if (!(*map->pfn & NVKM_VMM_PFN_V))
     86			continue;
     87
     88		if (!(*map->pfn & NVKM_VMM_PFN_W))
     89			data |= BIT_ULL(6); /* RO. */
     90
     91		if (!(*map->pfn & NVKM_VMM_PFN_A))
     92			data |= BIT_ULL(7); /* Atomic disable. */
     93
     94		if (!(*map->pfn & NVKM_VMM_PFN_VRAM)) {
     95			addr = *map->pfn >> NVKM_VMM_PFN_ADDR_SHIFT;
     96			addr = dma_map_page(dev, pfn_to_page(addr), 0,
     97					    PAGE_SIZE, DMA_BIDIRECTIONAL);
     98			if (!WARN_ON(dma_mapping_error(dev, addr))) {
     99				data |= addr >> 4;
    100				data |= 2ULL << 1; /* SYSTEM_COHERENT_MEMORY. */
    101				data |= BIT_ULL(3); /* VOL. */
    102				data |= BIT_ULL(0); /* VALID. */
    103			}
    104		} else {
    105			data |= (*map->pfn & NVKM_VMM_PFN_ADDR) >> 4;
    106			data |= BIT_ULL(0); /* VALID. */
    107		}
    108
    109		VMM_WO064(pt, vmm, ptei++ * 8, data);
    110	}
    111	nvkm_done(pt->memory);
    112}
    113
    114static inline void
    115gp100_vmm_pgt_pte(struct nvkm_vmm *vmm, struct nvkm_mmu_pt *pt,
    116		  u32 ptei, u32 ptes, struct nvkm_vmm_map *map, u64 addr)
    117{
    118	u64 data = (addr >> 4) | map->type;
    119
    120	map->type += ptes * map->ctag;
    121
    122	while (ptes--) {
    123		VMM_WO064(pt, vmm, ptei++ * 8, data);
    124		data += map->next;
    125	}
    126}
    127
    128static void
    129gp100_vmm_pgt_sgl(struct nvkm_vmm *vmm, struct nvkm_mmu_pt *pt,
    130		  u32 ptei, u32 ptes, struct nvkm_vmm_map *map)
    131{
    132	VMM_MAP_ITER_SGL(vmm, pt, ptei, ptes, map, gp100_vmm_pgt_pte);
    133}
    134
    135static void
    136gp100_vmm_pgt_dma(struct nvkm_vmm *vmm, struct nvkm_mmu_pt *pt,
    137		  u32 ptei, u32 ptes, struct nvkm_vmm_map *map)
    138{
    139	if (map->page->shift == PAGE_SHIFT) {
    140		VMM_SPAM(vmm, "DMAA %08x %08x PTE(s)", ptei, ptes);
    141		nvkm_kmap(pt->memory);
    142		while (ptes--) {
    143			const u64 data = (*map->dma++ >> 4) | map->type;
    144			VMM_WO064(pt, vmm, ptei++ * 8, data);
    145			map->type += map->ctag;
    146		}
    147		nvkm_done(pt->memory);
    148		return;
    149	}
    150
    151	VMM_MAP_ITER_DMA(vmm, pt, ptei, ptes, map, gp100_vmm_pgt_pte);
    152}
    153
    154static void
    155gp100_vmm_pgt_mem(struct nvkm_vmm *vmm, struct nvkm_mmu_pt *pt,
    156		  u32 ptei, u32 ptes, struct nvkm_vmm_map *map)
    157{
    158	VMM_MAP_ITER_MEM(vmm, pt, ptei, ptes, map, gp100_vmm_pgt_pte);
    159}
    160
    161static void
    162gp100_vmm_pgt_sparse(struct nvkm_vmm *vmm,
    163		     struct nvkm_mmu_pt *pt, u32 ptei, u32 ptes)
    164{
    165	/* VALID_FALSE + VOL tells the MMU to treat the PTE as sparse. */
    166	VMM_FO064(pt, vmm, ptei * 8, BIT_ULL(3) /* VOL. */, ptes);
    167}
    168
    169static const struct nvkm_vmm_desc_func
    170gp100_vmm_desc_spt = {
    171	.unmap = gf100_vmm_pgt_unmap,
    172	.sparse = gp100_vmm_pgt_sparse,
    173	.mem = gp100_vmm_pgt_mem,
    174	.dma = gp100_vmm_pgt_dma,
    175	.sgl = gp100_vmm_pgt_sgl,
    176	.pfn = gp100_vmm_pgt_pfn,
    177	.pfn_clear = gp100_vmm_pfn_clear,
    178	.pfn_unmap = gp100_vmm_pfn_unmap,
    179};
    180
    181static void
    182gp100_vmm_lpt_invalid(struct nvkm_vmm *vmm,
    183		      struct nvkm_mmu_pt *pt, u32 ptei, u32 ptes)
    184{
    185	/* VALID_FALSE + PRIV tells the MMU to ignore corresponding SPTEs. */
    186	VMM_FO064(pt, vmm, ptei * 8, BIT_ULL(5) /* PRIV. */, ptes);
    187}
    188
    189static const struct nvkm_vmm_desc_func
    190gp100_vmm_desc_lpt = {
    191	.invalid = gp100_vmm_lpt_invalid,
    192	.unmap = gf100_vmm_pgt_unmap,
    193	.sparse = gp100_vmm_pgt_sparse,
    194	.mem = gp100_vmm_pgt_mem,
    195};
    196
    197static inline void
    198gp100_vmm_pd0_pte(struct nvkm_vmm *vmm, struct nvkm_mmu_pt *pt,
    199		  u32 ptei, u32 ptes, struct nvkm_vmm_map *map, u64 addr)
    200{
    201	u64 data = (addr >> 4) | map->type;
    202
    203	map->type += ptes * map->ctag;
    204
    205	while (ptes--) {
    206		VMM_WO128(pt, vmm, ptei++ * 0x10, data, 0ULL);
    207		data += map->next;
    208	}
    209}
    210
    211static void
    212gp100_vmm_pd0_mem(struct nvkm_vmm *vmm, struct nvkm_mmu_pt *pt,
    213		  u32 ptei, u32 ptes, struct nvkm_vmm_map *map)
    214{
    215	VMM_MAP_ITER_MEM(vmm, pt, ptei, ptes, map, gp100_vmm_pd0_pte);
    216}
    217
    218static inline bool
    219gp100_vmm_pde(struct nvkm_mmu_pt *pt, u64 *data)
    220{
    221	switch (nvkm_memory_target(pt->memory)) {
    222	case NVKM_MEM_TARGET_VRAM: *data |= 1ULL << 1; break;
    223	case NVKM_MEM_TARGET_HOST: *data |= 2ULL << 1;
    224		*data |= BIT_ULL(3); /* VOL. */
    225		break;
    226	case NVKM_MEM_TARGET_NCOH: *data |= 3ULL << 1; break;
    227	default:
    228		WARN_ON(1);
    229		return false;
    230	}
    231	*data |= pt->addr >> 4;
    232	return true;
    233}
    234
    235static void
    236gp100_vmm_pd0_pde(struct nvkm_vmm *vmm, struct nvkm_vmm_pt *pgd, u32 pdei)
    237{
    238	struct nvkm_vmm_pt *pgt = pgd->pde[pdei];
    239	struct nvkm_mmu_pt *pd = pgd->pt[0];
    240	u64 data[2] = {};
    241
    242	if (pgt->pt[0] && !gp100_vmm_pde(pgt->pt[0], &data[0]))
    243		return;
    244	if (pgt->pt[1] && !gp100_vmm_pde(pgt->pt[1], &data[1]))
    245		return;
    246
    247	nvkm_kmap(pd->memory);
    248	VMM_WO128(pd, vmm, pdei * 0x10, data[0], data[1]);
    249	nvkm_done(pd->memory);
    250}
    251
    252static void
    253gp100_vmm_pd0_sparse(struct nvkm_vmm *vmm,
    254		     struct nvkm_mmu_pt *pt, u32 pdei, u32 pdes)
    255{
    256	/* VALID_FALSE + VOL_BIG tells the MMU to treat the PDE as sparse. */
    257	VMM_FO128(pt, vmm, pdei * 0x10, BIT_ULL(3) /* VOL_BIG. */, 0ULL, pdes);
    258}
    259
    260static void
    261gp100_vmm_pd0_unmap(struct nvkm_vmm *vmm,
    262		    struct nvkm_mmu_pt *pt, u32 pdei, u32 pdes)
    263{
    264	VMM_FO128(pt, vmm, pdei * 0x10, 0ULL, 0ULL, pdes);
    265}
    266
    267static void
    268gp100_vmm_pd0_pfn_unmap(struct nvkm_vmm *vmm,
    269			struct nvkm_mmu_pt *pt, u32 ptei, u32 ptes)
    270{
    271	struct device *dev = vmm->mmu->subdev.device->dev;
    272	dma_addr_t addr;
    273
    274	nvkm_kmap(pt->memory);
    275	while (ptes--) {
    276		u32 datalo = nvkm_ro32(pt->memory, pt->base + ptei * 16 + 0);
    277		u32 datahi = nvkm_ro32(pt->memory, pt->base + ptei * 16 + 4);
    278		u64 data   = (u64)datahi << 32 | datalo;
    279
    280		if ((data & (3ULL << 1)) != 0) {
    281			addr = (data >> 8) << 12;
    282			dma_unmap_page(dev, addr, 1UL << 21, DMA_BIDIRECTIONAL);
    283		}
    284		ptei++;
    285	}
    286	nvkm_done(pt->memory);
    287}
    288
    289static bool
    290gp100_vmm_pd0_pfn_clear(struct nvkm_vmm *vmm,
    291			struct nvkm_mmu_pt *pt, u32 ptei, u32 ptes)
    292{
    293	bool dma = false;
    294
    295	nvkm_kmap(pt->memory);
    296	while (ptes--) {
    297		u32 datalo = nvkm_ro32(pt->memory, pt->base + ptei * 16 + 0);
    298		u32 datahi = nvkm_ro32(pt->memory, pt->base + ptei * 16 + 4);
    299		u64 data   = (u64)datahi << 32 | datalo;
    300
    301		if ((data & BIT_ULL(0)) && (data & (3ULL << 1)) != 0) {
    302			VMM_WO064(pt, vmm, ptei * 16, data & ~BIT_ULL(0));
    303			dma = true;
    304		}
    305		ptei++;
    306	}
    307	nvkm_done(pt->memory);
    308	return dma;
    309}
    310
    311static void
    312gp100_vmm_pd0_pfn(struct nvkm_vmm *vmm, struct nvkm_mmu_pt *pt,
    313		  u32 ptei, u32 ptes, struct nvkm_vmm_map *map)
    314{
    315	struct device *dev = vmm->mmu->subdev.device->dev;
    316	dma_addr_t addr;
    317
    318	nvkm_kmap(pt->memory);
    319	for (; ptes; ptes--, map->pfn++) {
    320		u64 data = 0;
    321
    322		if (!(*map->pfn & NVKM_VMM_PFN_V))
    323			continue;
    324
    325		if (!(*map->pfn & NVKM_VMM_PFN_W))
    326			data |= BIT_ULL(6); /* RO. */
    327
    328		if (!(*map->pfn & NVKM_VMM_PFN_A))
    329			data |= BIT_ULL(7); /* Atomic disable. */
    330
    331		if (!(*map->pfn & NVKM_VMM_PFN_VRAM)) {
    332			addr = *map->pfn >> NVKM_VMM_PFN_ADDR_SHIFT;
    333			addr = dma_map_page(dev, pfn_to_page(addr), 0,
    334					    1UL << 21, DMA_BIDIRECTIONAL);
    335			if (!WARN_ON(dma_mapping_error(dev, addr))) {
    336				data |= addr >> 4;
    337				data |= 2ULL << 1; /* SYSTEM_COHERENT_MEMORY. */
    338				data |= BIT_ULL(3); /* VOL. */
    339				data |= BIT_ULL(0); /* VALID. */
    340			}
    341		} else {
    342			data |= (*map->pfn & NVKM_VMM_PFN_ADDR) >> 4;
    343			data |= BIT_ULL(0); /* VALID. */
    344		}
    345
    346		VMM_WO064(pt, vmm, ptei++ * 16, data);
    347	}
    348	nvkm_done(pt->memory);
    349}
    350
    351static const struct nvkm_vmm_desc_func
    352gp100_vmm_desc_pd0 = {
    353	.unmap = gp100_vmm_pd0_unmap,
    354	.sparse = gp100_vmm_pd0_sparse,
    355	.pde = gp100_vmm_pd0_pde,
    356	.mem = gp100_vmm_pd0_mem,
    357	.pfn = gp100_vmm_pd0_pfn,
    358	.pfn_clear = gp100_vmm_pd0_pfn_clear,
    359	.pfn_unmap = gp100_vmm_pd0_pfn_unmap,
    360};
    361
    362static void
    363gp100_vmm_pd1_pde(struct nvkm_vmm *vmm, struct nvkm_vmm_pt *pgd, u32 pdei)
    364{
    365	struct nvkm_vmm_pt *pgt = pgd->pde[pdei];
    366	struct nvkm_mmu_pt *pd = pgd->pt[0];
    367	u64 data = 0;
    368
    369	if (!gp100_vmm_pde(pgt->pt[0], &data))
    370		return;
    371
    372	nvkm_kmap(pd->memory);
    373	VMM_WO064(pd, vmm, pdei * 8, data);
    374	nvkm_done(pd->memory);
    375}
    376
    377static const struct nvkm_vmm_desc_func
    378gp100_vmm_desc_pd1 = {
    379	.unmap = gf100_vmm_pgt_unmap,
    380	.sparse = gp100_vmm_pgt_sparse,
    381	.pde = gp100_vmm_pd1_pde,
    382};
    383
    384const struct nvkm_vmm_desc
    385gp100_vmm_desc_16[] = {
    386	{ LPT, 5,  8, 0x0100, &gp100_vmm_desc_lpt },
    387	{ PGD, 8, 16, 0x1000, &gp100_vmm_desc_pd0 },
    388	{ PGD, 9,  8, 0x1000, &gp100_vmm_desc_pd1 },
    389	{ PGD, 9,  8, 0x1000, &gp100_vmm_desc_pd1 },
    390	{ PGD, 2,  8, 0x1000, &gp100_vmm_desc_pd1 },
    391	{}
    392};
    393
    394const struct nvkm_vmm_desc
    395gp100_vmm_desc_12[] = {
    396	{ SPT, 9,  8, 0x1000, &gp100_vmm_desc_spt },
    397	{ PGD, 8, 16, 0x1000, &gp100_vmm_desc_pd0 },
    398	{ PGD, 9,  8, 0x1000, &gp100_vmm_desc_pd1 },
    399	{ PGD, 9,  8, 0x1000, &gp100_vmm_desc_pd1 },
    400	{ PGD, 2,  8, 0x1000, &gp100_vmm_desc_pd1 },
    401	{}
    402};
    403
    404int
    405gp100_vmm_valid(struct nvkm_vmm *vmm, void *argv, u32 argc,
    406		struct nvkm_vmm_map *map)
    407{
    408	const enum nvkm_memory_target target = nvkm_memory_target(map->memory);
    409	const struct nvkm_vmm_page *page = map->page;
    410	union {
    411		struct gp100_vmm_map_vn vn;
    412		struct gp100_vmm_map_v0 v0;
    413	} *args = argv;
    414	struct nvkm_device *device = vmm->mmu->subdev.device;
    415	struct nvkm_memory *memory = map->memory;
    416	u8  kind, kind_inv, priv, ro, vol;
    417	int kindn, aper, ret = -ENOSYS;
    418	const u8 *kindm;
    419
    420	map->next = (1ULL << page->shift) >> 4;
    421	map->type = 0;
    422
    423	if (!(ret = nvif_unpack(ret, &argv, &argc, args->v0, 0, 0, false))) {
    424		vol  = !!args->v0.vol;
    425		ro   = !!args->v0.ro;
    426		priv = !!args->v0.priv;
    427		kind =   args->v0.kind;
    428	} else
    429	if (!(ret = nvif_unvers(ret, &argv, &argc, args->vn))) {
    430		vol  = target == NVKM_MEM_TARGET_HOST;
    431		ro   = 0;
    432		priv = 0;
    433		kind = 0x00;
    434	} else {
    435		VMM_DEBUG(vmm, "args");
    436		return ret;
    437	}
    438
    439	aper = vmm->func->aper(target);
    440	if (WARN_ON(aper < 0))
    441		return aper;
    442
    443	kindm = vmm->mmu->func->kind(vmm->mmu, &kindn, &kind_inv);
    444	if (kind >= kindn || kindm[kind] == kind_inv) {
    445		VMM_DEBUG(vmm, "kind %02x", kind);
    446		return -EINVAL;
    447	}
    448
    449	if (kindm[kind] != kind) {
    450		u64 tags = nvkm_memory_size(memory) >> 16;
    451		if (aper != 0 || !(page->type & NVKM_VMM_PAGE_COMP)) {
    452			VMM_DEBUG(vmm, "comp %d %02x", aper, page->type);
    453			return -EINVAL;
    454		}
    455
    456		ret = nvkm_memory_tags_get(memory, device, tags,
    457					   nvkm_ltc_tags_clear,
    458					   &map->tags);
    459		if (ret) {
    460			VMM_DEBUG(vmm, "comp %d", ret);
    461			return ret;
    462		}
    463
    464		if (map->tags->mn) {
    465			tags = map->tags->mn->offset + (map->offset >> 16);
    466			map->ctag |= ((1ULL << page->shift) >> 16) << 36;
    467			map->type |= tags << 36;
    468			map->next |= map->ctag;
    469		} else {
    470			kind = kindm[kind];
    471		}
    472	}
    473
    474	map->type |= BIT(0);
    475	map->type |= (u64)aper << 1;
    476	map->type |= (u64) vol << 3;
    477	map->type |= (u64)priv << 5;
    478	map->type |= (u64)  ro << 6;
    479	map->type |= (u64)kind << 56;
    480	return 0;
    481}
    482
    483static int
    484gp100_vmm_fault_cancel(struct nvkm_vmm *vmm, void *argv, u32 argc)
    485{
    486	struct nvkm_device *device = vmm->mmu->subdev.device;
    487	union {
    488		struct gp100_vmm_fault_cancel_v0 v0;
    489	} *args = argv;
    490	int ret = -ENOSYS;
    491	u32 aper;
    492
    493	if ((ret = nvif_unpack(ret, &argv, &argc, args->v0, 0, 0, false)))
    494		return ret;
    495
    496	/* Translate MaxwellFaultBufferA instance pointer to the same
    497	 * format as the NV_GR_FECS_CURRENT_CTX register.
    498	 */
    499	aper = (args->v0.inst >> 8) & 3;
    500	args->v0.inst >>= 12;
    501	args->v0.inst |= aper << 28;
    502	args->v0.inst |= 0x80000000;
    503
    504	if (!WARN_ON(nvkm_gr_ctxsw_pause(device))) {
    505		if (nvkm_gr_ctxsw_inst(device) == args->v0.inst) {
    506			gf100_vmm_invalidate(vmm, 0x0000001b
    507					     /* CANCEL_TARGETED. */ |
    508					     (args->v0.hub    << 20) |
    509					     (args->v0.gpc    << 15) |
    510					     (args->v0.client << 9));
    511		}
    512		WARN_ON(nvkm_gr_ctxsw_resume(device));
    513	}
    514
    515	return 0;
    516}
    517
    518static int
    519gp100_vmm_fault_replay(struct nvkm_vmm *vmm, void *argv, u32 argc)
    520{
    521	union {
    522		struct gp100_vmm_fault_replay_vn vn;
    523	} *args = argv;
    524	int ret = -ENOSYS;
    525
    526	if (!(ret = nvif_unvers(ret, &argv, &argc, args->vn))) {
    527		gf100_vmm_invalidate(vmm, 0x0000000b); /* REPLAY_GLOBAL. */
    528	}
    529
    530	return ret;
    531}
    532
    533int
    534gp100_vmm_mthd(struct nvkm_vmm *vmm,
    535	       struct nvkm_client *client, u32 mthd, void *argv, u32 argc)
    536{
    537	switch (mthd) {
    538	case GP100_VMM_VN_FAULT_REPLAY:
    539		return gp100_vmm_fault_replay(vmm, argv, argc);
    540	case GP100_VMM_VN_FAULT_CANCEL:
    541		return gp100_vmm_fault_cancel(vmm, argv, argc);
    542	default:
    543		break;
    544	}
    545	return -EINVAL;
    546}
    547
    548void
    549gp100_vmm_invalidate_pdb(struct nvkm_vmm *vmm, u64 addr)
    550{
    551	struct nvkm_device *device = vmm->mmu->subdev.device;
    552	nvkm_wr32(device, 0x100cb8, lower_32_bits(addr));
    553	nvkm_wr32(device, 0x100cec, upper_32_bits(addr));
    554}
    555
    556void
    557gp100_vmm_flush(struct nvkm_vmm *vmm, int depth)
    558{
    559	u32 type = (5 /* CACHE_LEVEL_UP_TO_PDE3 */ - depth) << 24;
    560	if (atomic_read(&vmm->engref[NVKM_SUBDEV_BAR]))
    561		type |= 0x00000004; /* HUB_ONLY */
    562	type |= 0x00000001; /* PAGE_ALL */
    563	gf100_vmm_invalidate(vmm, type);
    564}
    565
    566int
    567gp100_vmm_join(struct nvkm_vmm *vmm, struct nvkm_memory *inst)
    568{
    569	u64 base = BIT_ULL(10) /* VER2 */ | BIT_ULL(11) /* 64KiB */;
    570	if (vmm->replay) {
    571		base |= BIT_ULL(4); /* FAULT_REPLAY_TEX */
    572		base |= BIT_ULL(5); /* FAULT_REPLAY_GCC */
    573	}
    574	return gf100_vmm_join_(vmm, inst, base);
    575}
    576
    577static const struct nvkm_vmm_func
    578gp100_vmm = {
    579	.join = gp100_vmm_join,
    580	.part = gf100_vmm_part,
    581	.aper = gf100_vmm_aper,
    582	.valid = gp100_vmm_valid,
    583	.flush = gp100_vmm_flush,
    584	.mthd = gp100_vmm_mthd,
    585	.invalidate_pdb = gp100_vmm_invalidate_pdb,
    586	.page = {
    587		{ 47, &gp100_vmm_desc_16[4], NVKM_VMM_PAGE_Sxxx },
    588		{ 38, &gp100_vmm_desc_16[3], NVKM_VMM_PAGE_Sxxx },
    589		{ 29, &gp100_vmm_desc_16[2], NVKM_VMM_PAGE_Sxxx },
    590		{ 21, &gp100_vmm_desc_16[1], NVKM_VMM_PAGE_SVxC },
    591		{ 16, &gp100_vmm_desc_16[0], NVKM_VMM_PAGE_SVxC },
    592		{ 12, &gp100_vmm_desc_12[0], NVKM_VMM_PAGE_SVHx },
    593		{}
    594	}
    595};
    596
    597int
    598gp100_vmm_new_(const struct nvkm_vmm_func *func,
    599	       struct nvkm_mmu *mmu, bool managed, u64 addr, u64 size,
    600	       void *argv, u32 argc, struct lock_class_key *key,
    601	       const char *name, struct nvkm_vmm **pvmm)
    602{
    603	union {
    604		struct gp100_vmm_vn vn;
    605		struct gp100_vmm_v0 v0;
    606	} *args = argv;
    607	int ret = -ENOSYS;
    608	bool replay;
    609
    610	if (!(ret = nvif_unpack(ret, &argv, &argc, args->v0, 0, 0, false))) {
    611		replay = args->v0.fault_replay != 0;
    612	} else
    613	if (!(ret = nvif_unvers(ret, &argv, &argc, args->vn))) {
    614		replay = false;
    615	} else
    616		return ret;
    617
    618	ret = nvkm_vmm_new_(func, mmu, 0, managed, addr, size, key, name, pvmm);
    619	if (ret)
    620		return ret;
    621
    622	(*pvmm)->replay = replay;
    623	return 0;
    624}
    625
    626int
    627gp100_vmm_new(struct nvkm_mmu *mmu, bool managed, u64 addr, u64 size,
    628	      void *argv, u32 argc, struct lock_class_key *key,
    629	      const char *name, struct nvkm_vmm **pvmm)
    630{
    631	return gp100_vmm_new_(&gp100_vmm, mmu, managed, addr, size,
    632			      argv, argc, key, name, pvmm);
    633}