Contributors: 18
Author Tokens Token Proportion Commits Commit Proportion
Matthew Brost 635 35.14% 1 2.86%
Michal Wajdeczko 513 28.39% 5 14.29%
Michael J. Ruhl 291 16.10% 4 11.43%
Himal Prasad Ghimiray 179 9.91% 1 2.86%
Matthew Auld 48 2.66% 3 8.57%
Matt Roper 35 1.94% 2 5.71%
Maarten Lankhorst 32 1.77% 1 2.86%
Balasubramani Vivekanandan 17 0.94% 1 2.86%
Lucas De Marchi 15 0.83% 5 14.29%
Oak Zeng 15 0.83% 2 5.71%
Rodrigo Vivi 11 0.61% 2 5.71%
Arnd Bergmann 4 0.22% 1 2.86%
Bommithi Sakeena 3 0.17% 1 2.86%
Gustavo Sousa 3 0.17% 1 2.86%
Tejas Upadhyay 2 0.11% 2 5.71%
Jani Nikula 2 0.11% 1 2.86%
Michał Winiarski 1 0.06% 1 2.86%
Brian Welty 1 0.06% 1 2.86%
Total 1807 35 
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368
// SPDX-License-Identifier: MIT
/*
 * Copyright © 2021-2024 Intel Corporation
 */

#include <linux/pci.h>

#include <drm/drm_managed.h>
#include <drm/drm_print.h>

#include "regs/xe_bars.h"
#include "regs/xe_gt_regs.h"
#include "regs/xe_regs.h"
#include "xe_assert.h"
#include "xe_device.h"
#include "xe_force_wake.h"
#include "xe_gt_mcr.h"
#include "xe_gt_sriov_vf.h"
#include "xe_mmio.h"
#include "xe_module.h"
#include "xe_sriov.h"
#include "xe_vram.h"

#define BAR_SIZE_SHIFT 20

static void
_resize_bar(struct xe_device *xe, int resno, resource_size_t size)
{
	struct pci_dev *pdev = to_pci_dev(xe->drm.dev);
	int bar_size = pci_rebar_bytes_to_size(size);
	int ret;

	if (pci_resource_len(pdev, resno))
		pci_release_resource(pdev, resno);

	ret = pci_resize_resource(pdev, resno, bar_size);
	if (ret) {
		drm_info(&xe->drm, "Failed to resize BAR%d to %dM (%pe). Consider enabling 'Resizable BAR' support in your BIOS\n",
			 resno, 1 << bar_size, ERR_PTR(ret));
		return;
	}

	drm_info(&xe->drm, "BAR%d resized to %dM\n", resno, 1 << bar_size);
}

/*
 * if force_vram_bar_size is set, attempt to set to the requested size
 * else set to maximum possible size
 */
static void resize_vram_bar(struct xe_device *xe)
{
	u64 force_vram_bar_size = xe_modparam.force_vram_bar_size;
	struct pci_dev *pdev = to_pci_dev(xe->drm.dev);
	struct pci_bus *root = pdev->bus;
	resource_size_t current_size;
	resource_size_t rebar_size;
	struct resource *root_res;
	u32 bar_size_mask;
	u32 pci_cmd;
	int i;

	/* gather some relevant info */
	current_size = pci_resource_len(pdev, LMEM_BAR);
	bar_size_mask = pci_rebar_get_possible_sizes(pdev, LMEM_BAR);

	if (!bar_size_mask)
		return;

	/* set to a specific size? */
	if (force_vram_bar_size) {
		u32 bar_size_bit;

		rebar_size = force_vram_bar_size * (resource_size_t)SZ_1M;

		bar_size_bit = bar_size_mask & BIT(pci_rebar_bytes_to_size(rebar_size));

		if (!bar_size_bit) {
			drm_info(&xe->drm,
				 "Requested size: %lluMiB is not supported by rebar sizes: 0x%x. Leaving default: %lluMiB\n",
				 (u64)rebar_size >> 20, bar_size_mask, (u64)current_size >> 20);
			return;
		}

		rebar_size = 1ULL << (__fls(bar_size_bit) + BAR_SIZE_SHIFT);

		if (rebar_size == current_size)
			return;
	} else {
		rebar_size = 1ULL << (__fls(bar_size_mask) + BAR_SIZE_SHIFT);

		/* only resize if larger than current */
		if (rebar_size <= current_size)
			return;
	}

	drm_info(&xe->drm, "Attempting to resize bar from %lluMiB -> %lluMiB\n",
		 (u64)current_size >> 20, (u64)rebar_size >> 20);

	while (root->parent)
		root = root->parent;

	pci_bus_for_each_resource(root, root_res, i) {
		if (root_res && root_res->flags & (IORESOURCE_MEM | IORESOURCE_MEM_64) &&
		    (u64)root_res->start > 0x100000000ul)
			break;
	}

	if (!root_res) {
		drm_info(&xe->drm, "Can't resize VRAM BAR - platform support is missing. Consider enabling 'Resizable BAR' support in your BIOS\n");
		return;
	}

	pci_read_config_dword(pdev, PCI_COMMAND, &pci_cmd);
	pci_write_config_dword(pdev, PCI_COMMAND, pci_cmd & ~PCI_COMMAND_MEMORY);

	_resize_bar(xe, LMEM_BAR, rebar_size);

	pci_assign_unassigned_bus_resources(pdev->bus);
	pci_write_config_dword(pdev, PCI_COMMAND, pci_cmd);
}

static bool resource_is_valid(struct pci_dev *pdev, int bar)
{
	if (!pci_resource_flags(pdev, bar))
		return false;

	if (pci_resource_flags(pdev, bar) & IORESOURCE_UNSET)
		return false;

	if (!pci_resource_len(pdev, bar))
		return false;

	return true;
}

static int determine_lmem_bar_size(struct xe_device *xe)
{
	struct pci_dev *pdev = to_pci_dev(xe->drm.dev);

	if (!resource_is_valid(pdev, LMEM_BAR)) {
		drm_err(&xe->drm, "pci resource is not valid\n");
		return -ENXIO;
	}

	resize_vram_bar(xe);

	xe->mem.vram.io_start = pci_resource_start(pdev, LMEM_BAR);
	xe->mem.vram.io_size = pci_resource_len(pdev, LMEM_BAR);
	if (!xe->mem.vram.io_size)
		return -EIO;

	/* XXX: Need to change when xe link code is ready */
	xe->mem.vram.dpa_base = 0;

	/* set up a map to the total memory area. */
	xe->mem.vram.mapping = ioremap_wc(xe->mem.vram.io_start, xe->mem.vram.io_size);

	return 0;
}

static inline u64 get_flat_ccs_offset(struct xe_gt *gt, u64 tile_size)
{
	struct xe_device *xe = gt_to_xe(gt);
	u64 offset;
	u32 reg;

	if (GRAPHICS_VER(xe) >= 20) {
		u64 ccs_size = tile_size / 512;
		u64 offset_hi, offset_lo;
		u32 nodes, num_enabled;

		reg = xe_mmio_read32(gt, MIRROR_FUSE3);
		nodes = REG_FIELD_GET(XE2_NODE_ENABLE_MASK, reg);
		num_enabled = hweight32(nodes); /* Number of enabled l3 nodes */

		reg = xe_gt_mcr_unicast_read_any(gt, XE2_FLAT_CCS_BASE_RANGE_LOWER);
		offset_lo = REG_FIELD_GET(XE2_FLAT_CCS_BASE_LOWER_ADDR_MASK, reg);

		reg = xe_gt_mcr_unicast_read_any(gt, XE2_FLAT_CCS_BASE_RANGE_UPPER);
		offset_hi = REG_FIELD_GET(XE2_FLAT_CCS_BASE_UPPER_ADDR_MASK, reg);

		offset = offset_hi << 32; /* HW view bits 39:32 */
		offset |= offset_lo << 6; /* HW view bits 31:6 */
		offset *= num_enabled; /* convert to SW view */

		/* We don't expect any holes */
		xe_assert_msg(xe, offset == (xe_mmio_read64_2x32(gt, GSMBASE) - ccs_size),
			      "Hole between CCS and GSM.\n");
	} else {
		reg = xe_gt_mcr_unicast_read_any(gt, XEHP_FLAT_CCS_BASE_ADDR);
		offset = (u64)REG_FIELD_GET(XEHP_FLAT_CCS_PTR, reg) * SZ_64K;
	}

	return offset;
}

/*
 * tile_vram_size() - Collect vram size and offset information
 * @tile: tile to get info for
 * @vram_size: available vram (size - device reserved portions)
 * @tile_size: actual vram size
 * @tile_offset: physical start point in the vram address space
 *
 * There are 4 places for size information:
 * - io size (from pci_resource_len of LMEM bar) (only used for small bar and DG1)
 * - TILEx size (actual vram size)
 * - GSMBASE offset (TILEx - "stolen")
 * - CSSBASE offset (TILEx - CSS space necessary)
 *
 * CSSBASE is always a lower/smaller offset then GSMBASE.
 *
 * The actual available size of memory is to the CCS or GSM base.
 * NOTE: multi-tile bases will include the tile offset.
 *
 */
static int tile_vram_size(struct xe_tile *tile, u64 *vram_size,
			  u64 *tile_size, u64 *tile_offset)
{
	struct xe_device *xe = tile_to_xe(tile);
	struct xe_gt *gt = tile->primary_gt;
	u64 offset;
	int err;
	u32 reg;

	if (IS_SRIOV_VF(xe)) {
		struct xe_tile *t;
		int id;

		offset = 0;
		for_each_tile(t, xe, id)
			for_each_if(t->id < tile->id)
				offset += xe_gt_sriov_vf_lmem(t->primary_gt);

		*tile_size = xe_gt_sriov_vf_lmem(gt);
		*vram_size = *tile_size;
		*tile_offset = offset;

		return 0;
	}

	err = xe_force_wake_get(gt_to_fw(gt), XE_FW_GT);
	if (err)
		return err;

	/* actual size */
	if (unlikely(xe->info.platform == XE_DG1)) {
		*tile_size = pci_resource_len(to_pci_dev(xe->drm.dev), LMEM_BAR);
		*tile_offset = 0;
	} else {
		reg = xe_gt_mcr_unicast_read_any(gt, XEHP_TILE_ADDR_RANGE(gt->info.id));
		*tile_size = (u64)REG_FIELD_GET(GENMASK(14, 8), reg) * SZ_1G;
		*tile_offset = (u64)REG_FIELD_GET(GENMASK(7, 1), reg) * SZ_1G;
	}

	/* minus device usage */
	if (xe->info.has_flat_ccs) {
		offset = get_flat_ccs_offset(gt, *tile_size);
	} else {
		offset = xe_mmio_read64_2x32(gt, GSMBASE);
	}

	/* remove the tile offset so we have just the available size */
	*vram_size = offset - *tile_offset;

	return xe_force_wake_put(gt_to_fw(gt), XE_FW_GT);
}

static void vram_fini(void *arg)
{
	struct xe_device *xe = arg;
	struct xe_tile *tile;
	int id;

	if (xe->mem.vram.mapping)
		iounmap(xe->mem.vram.mapping);

	xe->mem.vram.mapping = NULL;

	for_each_tile(tile, xe, id)
		tile->mem.vram.mapping = NULL;
}

/**
 * xe_vram_probe() - Probe VRAM configuration
 * @xe: the &xe_device
 *
 * Collect VRAM size and offset information for all tiles.
 *
 * Return: 0 on success, error code on failure
 */
int xe_vram_probe(struct xe_device *xe)
{
	struct xe_tile *tile;
	resource_size_t io_size;
	u64 available_size = 0;
	u64 total_size = 0;
	u64 tile_offset;
	u64 tile_size;
	u64 vram_size;
	int err;
	u8 id;

	if (!IS_DGFX(xe))
		return 0;

	/* Get the size of the root tile's vram for later accessibility comparison */
	tile = xe_device_get_root_tile(xe);
	err = tile_vram_size(tile, &vram_size, &tile_size, &tile_offset);
	if (err)
		return err;

	err = determine_lmem_bar_size(xe);
	if (err)
		return err;

	drm_info(&xe->drm, "VISIBLE VRAM: %pa, %pa\n", &xe->mem.vram.io_start,
		 &xe->mem.vram.io_size);

	io_size = xe->mem.vram.io_size;

	/* tile specific ranges */
	for_each_tile(tile, xe, id) {
		err = tile_vram_size(tile, &vram_size, &tile_size, &tile_offset);
		if (err)
			return err;

		tile->mem.vram.actual_physical_size = tile_size;
		tile->mem.vram.io_start = xe->mem.vram.io_start + tile_offset;
		tile->mem.vram.io_size = min_t(u64, vram_size, io_size);

		if (!tile->mem.vram.io_size) {
			drm_err(&xe->drm, "Tile without any CPU visible VRAM. Aborting.\n");
			return -ENODEV;
		}

		tile->mem.vram.dpa_base = xe->mem.vram.dpa_base + tile_offset;
		tile->mem.vram.usable_size = vram_size;
		tile->mem.vram.mapping = xe->mem.vram.mapping + tile_offset;

		if (tile->mem.vram.io_size < tile->mem.vram.usable_size)
			drm_info(&xe->drm, "Small BAR device\n");
		drm_info(&xe->drm, "VRAM[%u, %u]: Actual physical size %pa, usable size exclude stolen %pa, CPU accessible size %pa\n", id,
			 tile->id, &tile->mem.vram.actual_physical_size, &tile->mem.vram.usable_size, &tile->mem.vram.io_size);
		drm_info(&xe->drm, "VRAM[%u, %u]: DPA range: [%pa-%llx], io range: [%pa-%llx]\n", id, tile->id,
			 &tile->mem.vram.dpa_base, tile->mem.vram.dpa_base + (u64)tile->mem.vram.actual_physical_size,
			 &tile->mem.vram.io_start, tile->mem.vram.io_start + (u64)tile->mem.vram.io_size);

		/* calculate total size using tile size to get the correct HW sizing */
		total_size += tile_size;
		available_size += vram_size;

		if (total_size > xe->mem.vram.io_size) {
			drm_info(&xe->drm, "VRAM: %pa is larger than resource %pa\n",
				 &total_size, &xe->mem.vram.io_size);
		}

		io_size -= min_t(u64, tile_size, io_size);
	}

	xe->mem.vram.actual_physical_size = total_size;

	drm_info(&xe->drm, "Total VRAM: %pa, %pa\n", &xe->mem.vram.io_start,
		 &xe->mem.vram.actual_physical_size);
	drm_info(&xe->drm, "Available VRAM: %pa, %pa\n", &xe->mem.vram.io_start,
		 &available_size);

	return devm_add_action_or_reset(xe->drm.dev, vram_fini, xe);
}