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

gf100.c (7248B)


      1/*
      2 * Copyright 2012 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 * Authors: Ben Skeggs
     23 */
     24#include "priv.h"
     25
     26#include <core/memory.h>
     27#include <subdev/fb.h>
     28#include <subdev/timer.h>
     29
     30void
     31gf100_ltc_cbc_clear(struct nvkm_ltc *ltc, u32 start, u32 limit)
     32{
     33	struct nvkm_device *device = ltc->subdev.device;
     34	nvkm_wr32(device, 0x17e8cc, start);
     35	nvkm_wr32(device, 0x17e8d0, limit);
     36	nvkm_wr32(device, 0x17e8c8, 0x00000004);
     37}
     38
     39void
     40gf100_ltc_cbc_wait(struct nvkm_ltc *ltc)
     41{
     42	struct nvkm_device *device = ltc->subdev.device;
     43	int c, s;
     44	for (c = 0; c < ltc->ltc_nr; c++) {
     45		for (s = 0; s < ltc->lts_nr; s++) {
     46			const u32 addr = 0x1410c8 + (c * 0x2000) + (s * 0x400);
     47			nvkm_msec(device, 2000,
     48				if (!nvkm_rd32(device, addr))
     49					break;
     50			);
     51		}
     52	}
     53}
     54
     55void
     56gf100_ltc_zbc_clear_color(struct nvkm_ltc *ltc, int i, const u32 color[4])
     57{
     58	struct nvkm_device *device = ltc->subdev.device;
     59	nvkm_mask(device, 0x17ea44, 0x0000000f, i);
     60	nvkm_wr32(device, 0x17ea48, color[0]);
     61	nvkm_wr32(device, 0x17ea4c, color[1]);
     62	nvkm_wr32(device, 0x17ea50, color[2]);
     63	nvkm_wr32(device, 0x17ea54, color[3]);
     64}
     65
     66void
     67gf100_ltc_zbc_clear_depth(struct nvkm_ltc *ltc, int i, const u32 depth)
     68{
     69	struct nvkm_device *device = ltc->subdev.device;
     70	nvkm_mask(device, 0x17ea44, 0x0000000f, i);
     71	nvkm_wr32(device, 0x17ea58, depth);
     72}
     73
     74const struct nvkm_bitfield
     75gf100_ltc_lts_intr_name[] = {
     76	{ 0x00000001, "IDLE_ERROR_IQ" },
     77	{ 0x00000002, "IDLE_ERROR_CBC" },
     78	{ 0x00000004, "IDLE_ERROR_TSTG" },
     79	{ 0x00000008, "IDLE_ERROR_DSTG" },
     80	{ 0x00000010, "EVICTED_CB" },
     81	{ 0x00000020, "ILLEGAL_COMPSTAT" },
     82	{ 0x00000040, "BLOCKLINEAR_CB" },
     83	{ 0x00000100, "ECC_SEC_ERROR" },
     84	{ 0x00000200, "ECC_DED_ERROR" },
     85	{ 0x00000400, "DEBUG" },
     86	{ 0x00000800, "ATOMIC_TO_Z" },
     87	{ 0x00001000, "ILLEGAL_ATOMIC" },
     88	{ 0x00002000, "BLKACTIVITY_ERR" },
     89	{}
     90};
     91
     92static void
     93gf100_ltc_lts_intr(struct nvkm_ltc *ltc, int c, int s)
     94{
     95	struct nvkm_subdev *subdev = &ltc->subdev;
     96	struct nvkm_device *device = subdev->device;
     97	u32 base = 0x141000 + (c * 0x2000) + (s * 0x400);
     98	u32 intr = nvkm_rd32(device, base + 0x020);
     99	u32 stat = intr & 0x0000ffff;
    100	char msg[128];
    101
    102	if (stat) {
    103		nvkm_snprintbf(msg, sizeof(msg), gf100_ltc_lts_intr_name, stat);
    104		nvkm_error(subdev, "LTC%d_LTS%d: %08x [%s]\n", c, s, stat, msg);
    105	}
    106
    107	nvkm_wr32(device, base + 0x020, intr);
    108}
    109
    110void
    111gf100_ltc_intr(struct nvkm_ltc *ltc)
    112{
    113	struct nvkm_device *device = ltc->subdev.device;
    114	u32 mask;
    115
    116	mask = nvkm_rd32(device, 0x00017c);
    117	while (mask) {
    118		u32 s, c = __ffs(mask);
    119		for (s = 0; s < ltc->lts_nr; s++)
    120			gf100_ltc_lts_intr(ltc, c, s);
    121		mask &= ~(1 << c);
    122	}
    123}
    124
    125void
    126gf100_ltc_invalidate(struct nvkm_ltc *ltc)
    127{
    128	struct nvkm_device *device = ltc->subdev.device;
    129	s64 taken;
    130
    131	nvkm_wr32(device, 0x70004, 0x00000001);
    132	taken = nvkm_wait_msec(device, 2000, 0x70004, 0x00000003, 0x00000000);
    133
    134	if (taken > 0)
    135		nvkm_debug(&ltc->subdev, "LTC invalidate took %lld ns\n", taken);
    136}
    137
    138void
    139gf100_ltc_flush(struct nvkm_ltc *ltc)
    140{
    141	struct nvkm_device *device = ltc->subdev.device;
    142	s64 taken;
    143
    144	nvkm_wr32(device, 0x70010, 0x00000001);
    145	taken = nvkm_wait_msec(device, 2000, 0x70010, 0x00000003, 0x00000000);
    146
    147	if (taken > 0)
    148		nvkm_debug(&ltc->subdev, "LTC flush took %lld ns\n", taken);
    149}
    150
    151/* TODO: Figure out tag memory details and drop the over-cautious allocation.
    152 */
    153int
    154gf100_ltc_oneinit_tag_ram(struct nvkm_ltc *ltc)
    155{
    156	struct nvkm_device *device = ltc->subdev.device;
    157	struct nvkm_fb *fb = device->fb;
    158	struct nvkm_ram *ram = fb->ram;
    159	u32 bits = (nvkm_rd32(device, 0x100c80) & 0x00001000) ? 16 : 17;
    160	u32 tag_size, tag_margin, tag_align;
    161	int ret;
    162
    163	/* No VRAM, no tags for now. */
    164	if (!ram) {
    165		ltc->num_tags = 0;
    166		goto mm_init;
    167	}
    168
    169	/* tags for 1/4 of VRAM should be enough (8192/4 per GiB of VRAM) */
    170	ltc->num_tags = (ram->size >> 17) / 4;
    171	if (ltc->num_tags > (1 << bits))
    172		ltc->num_tags = 1 << bits; /* we have 16/17 bits in PTE */
    173	ltc->num_tags = (ltc->num_tags + 63) & ~63; /* round up to 64 */
    174
    175	tag_align = ltc->ltc_nr * 0x800;
    176	tag_margin = (tag_align < 0x6000) ? 0x6000 : tag_align;
    177
    178	/* 4 part 4 sub: 0x2000 bytes for 56 tags */
    179	/* 3 part 4 sub: 0x6000 bytes for 168 tags */
    180	/*
    181	 * About 147 bytes per tag. Let's be safe and allocate x2, which makes
    182	 * 0x4980 bytes for 64 tags, and round up to 0x6000 bytes for 64 tags.
    183	 *
    184	 * For 4 GiB of memory we'll have 8192 tags which makes 3 MiB, < 0.1 %.
    185	 */
    186	tag_size  = (ltc->num_tags / 64) * 0x6000 + tag_margin;
    187	tag_size += tag_align;
    188
    189	ret = nvkm_ram_get(device, NVKM_RAM_MM_NORMAL, 0x01, 12, tag_size,
    190			   true, true, &ltc->tag_ram);
    191	if (ret) {
    192		ltc->num_tags = 0;
    193	} else {
    194		u64 tag_base = nvkm_memory_addr(ltc->tag_ram) + tag_margin;
    195
    196		tag_base += tag_align - 1;
    197		do_div(tag_base, tag_align);
    198
    199		ltc->tag_base = tag_base;
    200	}
    201
    202mm_init:
    203	nvkm_mm_fini(&fb->tags.mm);
    204	return nvkm_mm_init(&fb->tags.mm, 0, 0, ltc->num_tags, 1);
    205}
    206
    207int
    208gf100_ltc_oneinit(struct nvkm_ltc *ltc)
    209{
    210	struct nvkm_device *device = ltc->subdev.device;
    211	const u32 parts = nvkm_rd32(device, 0x022438);
    212	const u32  mask = nvkm_rd32(device, 0x022554);
    213	const u32 slice = nvkm_rd32(device, 0x17e8dc) >> 28;
    214	int i;
    215
    216	for (i = 0; i < parts; i++) {
    217		if (!(mask & (1 << i)))
    218			ltc->ltc_nr++;
    219	}
    220	ltc->lts_nr = slice;
    221
    222	return gf100_ltc_oneinit_tag_ram(ltc);
    223}
    224
    225static void
    226gf100_ltc_init(struct nvkm_ltc *ltc)
    227{
    228	struct nvkm_device *device = ltc->subdev.device;
    229	u32 lpg128 = !(nvkm_rd32(device, 0x100c80) & 0x00000001);
    230
    231	nvkm_mask(device, 0x17e820, 0x00100000, 0x00000000); /* INTR_EN &= ~0x10 */
    232	nvkm_wr32(device, 0x17e8d8, ltc->ltc_nr);
    233	nvkm_wr32(device, 0x17e8d4, ltc->tag_base);
    234	nvkm_mask(device, 0x17e8c0, 0x00000002, lpg128 ? 0x00000002 : 0x00000000);
    235}
    236
    237static const struct nvkm_ltc_func
    238gf100_ltc = {
    239	.oneinit = gf100_ltc_oneinit,
    240	.init = gf100_ltc_init,
    241	.intr = gf100_ltc_intr,
    242	.cbc_clear = gf100_ltc_cbc_clear,
    243	.cbc_wait = gf100_ltc_cbc_wait,
    244	.zbc = 16,
    245	.zbc_clear_color = gf100_ltc_zbc_clear_color,
    246	.zbc_clear_depth = gf100_ltc_zbc_clear_depth,
    247	.invalidate = gf100_ltc_invalidate,
    248	.flush = gf100_ltc_flush,
    249};
    250
    251int
    252gf100_ltc_new(struct nvkm_device *device, enum nvkm_subdev_type type, int inst,
    253	      struct nvkm_ltc **pltc)
    254{
    255	return nvkm_ltc_new_(&gf100_ltc, device, type, inst, pltc);
    256}