blob: b1a7a8b9b46aa3ea8dcfb84d5af39ba8fa902bfb [file] [log] [blame]
/*
* Copyright © 2010 Daniel Vetter
* Copyright © 2011-2014 Intel Corporation
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* to deal in the Software without restriction, including without limitation
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
* and/or sell copies of the Software, and to permit persons to whom the
* Software is furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice (including the next
* paragraph) shall be included in all copies or substantial portions of the
* Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
* IN THE SOFTWARE.
*
*/
#include <linux/slab.h> /* fault-inject.h is not standalone! */
#include <linux/fault-inject.h>
#include <linux/log2.h>
#include <linux/random.h>
#include <linux/seq_file.h>
#include <linux/stop_machine.h>
#include <asm/set_memory.h>
#include <asm/smp.h>
#include <drm/i915_drm.h>
#include "display/intel_frontbuffer.h"
#include "gt/intel_gt.h"
#include "i915_drv.h"
#include "i915_scatterlist.h"
#include "i915_trace.h"
#include "i915_vgpu.h"
#define I915_GFP_ALLOW_FAIL (GFP_KERNEL | __GFP_RETRY_MAYFAIL | __GFP_NOWARN)
#if IS_ENABLED(CONFIG_DRM_I915_TRACE_GTT)
#define DBG(...) trace_printk(__VA_ARGS__)
#else
#define DBG(...)
#endif
/**
* DOC: Global GTT views
*
* Background and previous state
*
* Historically objects could exists (be bound) in global GTT space only as
* singular instances with a view representing all of the object's backing pages
* in a linear fashion. This view will be called a normal view.
*
* To support multiple views of the same object, where the number of mapped
* pages is not equal to the backing store, or where the layout of the pages
* is not linear, concept of a GGTT view was added.
*
* One example of an alternative view is a stereo display driven by a single
* image. In this case we would have a framebuffer looking like this
* (2x2 pages):
*
* 12
* 34
*
* Above would represent a normal GGTT view as normally mapped for GPU or CPU
* rendering. In contrast, fed to the display engine would be an alternative
* view which could look something like this:
*
* 1212
* 3434
*
* In this example both the size and layout of pages in the alternative view is
* different from the normal view.
*
* Implementation and usage
*
* GGTT views are implemented using VMAs and are distinguished via enum
* i915_ggtt_view_type and struct i915_ggtt_view.
*
* A new flavour of core GEM functions which work with GGTT bound objects were
* added with the _ggtt_ infix, and sometimes with _view postfix to avoid
* renaming in large amounts of code. They take the struct i915_ggtt_view
* parameter encapsulating all metadata required to implement a view.
*
* As a helper for callers which are only interested in the normal view,
* globally const i915_ggtt_view_normal singleton instance exists. All old core
* GEM API functions, the ones not taking the view parameter, are operating on,
* or with the normal GGTT view.
*
* Code wanting to add or use a new GGTT view needs to:
*
* 1. Add a new enum with a suitable name.
* 2. Extend the metadata in the i915_ggtt_view structure if required.
* 3. Add support to i915_get_vma_pages().
*
* New views are required to build a scatter-gather table from within the
* i915_get_vma_pages function. This table is stored in the vma.ggtt_view and
* exists for the lifetime of an VMA.
*
* Core API is designed to have copy semantics which means that passed in
* struct i915_ggtt_view does not need to be persistent (left around after
* calling the core API functions).
*
*/
#define as_pd(x) container_of((x), typeof(struct i915_page_directory), pt)
static int
i915_get_ggtt_vma_pages(struct i915_vma *vma);
static void gen6_ggtt_invalidate(struct i915_ggtt *ggtt)
{
struct intel_uncore *uncore = ggtt->vm.gt->uncore;
/*
* Note that as an uncached mmio write, this will flush the
* WCB of the writes into the GGTT before it triggers the invalidate.
*/
intel_uncore_write_fw(uncore, GFX_FLSH_CNTL_GEN6, GFX_FLSH_CNTL_EN);
}
static void guc_ggtt_invalidate(struct i915_ggtt *ggtt)
{
struct intel_uncore *uncore = ggtt->vm.gt->uncore;
gen6_ggtt_invalidate(ggtt);
intel_uncore_write_fw(uncore, GEN8_GTCR, GEN8_GTCR_INVALIDATE);
}
static void gmch_ggtt_invalidate(struct i915_ggtt *ggtt)
{
intel_gtt_chipset_flush();
}
static int ppgtt_bind_vma(struct i915_vma *vma,
enum i915_cache_level cache_level,
u32 unused)
{
u32 pte_flags;
int err;
if (!(vma->flags & I915_VMA_LOCAL_BIND)) {
err = vma->vm->allocate_va_range(vma->vm,
vma->node.start, vma->size);
if (err)
return err;
}
/* Applicable to VLV, and gen8+ */
pte_flags = 0;
if (i915_gem_object_is_readonly(vma->obj))
pte_flags |= PTE_READ_ONLY;
vma->vm->insert_entries(vma->vm, vma, cache_level, pte_flags);
return 0;
}
static void ppgtt_unbind_vma(struct i915_vma *vma)
{
vma->vm->clear_range(vma->vm, vma->node.start, vma->size);
}
static int ppgtt_set_pages(struct i915_vma *vma)
{
GEM_BUG_ON(vma->pages);
vma->pages = vma->obj->mm.pages;
vma->page_sizes = vma->obj->mm.page_sizes;
return 0;
}
static void clear_pages(struct i915_vma *vma)
{
GEM_BUG_ON(!vma->pages);
if (vma->pages != vma->obj->mm.pages) {
sg_free_table(vma->pages);
kfree(vma->pages);
}
vma->pages = NULL;
memset(&vma->page_sizes, 0, sizeof(vma->page_sizes));
}
static u64 gen8_pte_encode(dma_addr_t addr,
enum i915_cache_level level,
u32 flags)
{
gen8_pte_t pte = addr | _PAGE_PRESENT | _PAGE_RW;
if (unlikely(flags & PTE_READ_ONLY))
pte &= ~_PAGE_RW;
switch (level) {
case I915_CACHE_NONE:
pte |= PPAT_UNCACHED;
break;
case I915_CACHE_WT:
pte |= PPAT_DISPLAY_ELLC;
break;
default:
pte |= PPAT_CACHED;
break;
}
return pte;
}
static u64 gen8_pde_encode(const dma_addr_t addr,
const enum i915_cache_level level)
{
u64 pde = _PAGE_PRESENT | _PAGE_RW;
pde |= addr;
if (level != I915_CACHE_NONE)
pde |= PPAT_CACHED_PDE;
else
pde |= PPAT_UNCACHED;
return pde;
}
static u64 snb_pte_encode(dma_addr_t addr,
enum i915_cache_level level,
u32 flags)
{
gen6_pte_t pte = GEN6_PTE_VALID;
pte |= GEN6_PTE_ADDR_ENCODE(addr);
switch (level) {
case I915_CACHE_L3_LLC:
case I915_CACHE_LLC:
pte |= GEN6_PTE_CACHE_LLC;
break;
case I915_CACHE_NONE:
pte |= GEN6_PTE_UNCACHED;
break;
default:
MISSING_CASE(level);
}
return pte;
}
static u64 ivb_pte_encode(dma_addr_t addr,
enum i915_cache_level level,
u32 flags)
{
gen6_pte_t pte = GEN6_PTE_VALID;
pte |= GEN6_PTE_ADDR_ENCODE(addr);
switch (level) {
case I915_CACHE_L3_LLC:
pte |= GEN7_PTE_CACHE_L3_LLC;
break;
case I915_CACHE_LLC:
pte |= GEN6_PTE_CACHE_LLC;
break;
case I915_CACHE_NONE:
pte |= GEN6_PTE_UNCACHED;
break;
default:
MISSING_CASE(level);
}
return pte;
}
static u64 byt_pte_encode(dma_addr_t addr,
enum i915_cache_level level,
u32 flags)
{
gen6_pte_t pte = GEN6_PTE_VALID;
pte |= GEN6_PTE_ADDR_ENCODE(addr);
if (!(flags & PTE_READ_ONLY))
pte |= BYT_PTE_WRITEABLE;
if (level != I915_CACHE_NONE)
pte |= BYT_PTE_SNOOPED_BY_CPU_CACHES;
return pte;
}
static u64 hsw_pte_encode(dma_addr_t addr,
enum i915_cache_level level,
u32 flags)
{
gen6_pte_t pte = GEN6_PTE_VALID;
pte |= HSW_PTE_ADDR_ENCODE(addr);
if (level != I915_CACHE_NONE)
pte |= HSW_WB_LLC_AGE3;
return pte;
}
static u64 iris_pte_encode(dma_addr_t addr,
enum i915_cache_level level,
u32 flags)
{
gen6_pte_t pte = GEN6_PTE_VALID;
pte |= HSW_PTE_ADDR_ENCODE(addr);
switch (level) {
case I915_CACHE_NONE:
break;
case I915_CACHE_WT:
pte |= HSW_WT_ELLC_LLC_AGE3;
break;
default:
pte |= HSW_WB_ELLC_LLC_AGE3;
break;
}
return pte;
}
static void stash_init(struct pagestash *stash)
{
pagevec_init(&stash->pvec);
spin_lock_init(&stash->lock);
}
static struct page *stash_pop_page(struct pagestash *stash)
{
struct page *page = NULL;
spin_lock(&stash->lock);
if (likely(stash->pvec.nr))
page = stash->pvec.pages[--stash->pvec.nr];
spin_unlock(&stash->lock);
return page;
}
static void stash_push_pagevec(struct pagestash *stash, struct pagevec *pvec)
{
unsigned int nr;
spin_lock_nested(&stash->lock, SINGLE_DEPTH_NESTING);
nr = min_t(typeof(nr), pvec->nr, pagevec_space(&stash->pvec));
memcpy(stash->pvec.pages + stash->pvec.nr,
pvec->pages + pvec->nr - nr,
sizeof(pvec->pages[0]) * nr);
stash->pvec.nr += nr;
spin_unlock(&stash->lock);
pvec->nr -= nr;
}
static struct page *vm_alloc_page(struct i915_address_space *vm, gfp_t gfp)
{
struct pagevec stack;
struct page *page;
if (I915_SELFTEST_ONLY(should_fail(&vm->fault_attr, 1)))
i915_gem_shrink_all(vm->i915);
page = stash_pop_page(&vm->free_pages);
if (page)
return page;
if (!vm->pt_kmap_wc)
return alloc_page(gfp);
/* Look in our global stash of WC pages... */
page = stash_pop_page(&vm->i915->mm.wc_stash);
if (page)
return page;
/*
* Otherwise batch allocate pages to amortize cost of set_pages_wc.
*
* We have to be careful as page allocation may trigger the shrinker
* (via direct reclaim) which will fill up the WC stash underneath us.
* So we add our WB pages into a temporary pvec on the stack and merge
* them into the WC stash after all the allocations are complete.
*/
pagevec_init(&stack);
do {
struct page *page;
page = alloc_page(gfp);
if (unlikely(!page))
break;
stack.pages[stack.nr++] = page;
} while (pagevec_space(&stack));
if (stack.nr && !set_pages_array_wc(stack.pages, stack.nr)) {
page = stack.pages[--stack.nr];
/* Merge spare WC pages to the global stash */
if (stack.nr)
stash_push_pagevec(&vm->i915->mm.wc_stash, &stack);
/* Push any surplus WC pages onto the local VM stash */
if (stack.nr)
stash_push_pagevec(&vm->free_pages, &stack);
}
/* Return unwanted leftovers */
if (unlikely(stack.nr)) {
WARN_ON_ONCE(set_pages_array_wb(stack.pages, stack.nr));
__pagevec_release(&stack);
}
return page;
}
static void vm_free_pages_release(struct i915_address_space *vm,
bool immediate)
{
struct pagevec *pvec = &vm->free_pages.pvec;
struct pagevec stack;
lockdep_assert_held(&vm->free_pages.lock);
GEM_BUG_ON(!pagevec_count(pvec));
if (vm->pt_kmap_wc) {
/*
* When we use WC, first fill up the global stash and then
* only if full immediately free the overflow.
*/
stash_push_pagevec(&vm->i915->mm.wc_stash, pvec);
/*
* As we have made some room in the VM's free_pages,
* we can wait for it to fill again. Unless we are
* inside i915_address_space_fini() and must
* immediately release the pages!
*/
if (pvec->nr <= (immediate ? 0 : PAGEVEC_SIZE - 1))
return;
/*
* We have to drop the lock to allow ourselves to sleep,
* so take a copy of the pvec and clear the stash for
* others to use it as we sleep.
*/
stack = *pvec;
pagevec_reinit(pvec);
spin_unlock(&vm->free_pages.lock);
pvec = &stack;
set_pages_array_wb(pvec->pages, pvec->nr);
spin_lock(&vm->free_pages.lock);
}
__pagevec_release(pvec);
}
static void vm_free_page(struct i915_address_space *vm, struct page *page)
{
/*
* On !llc, we need to change the pages back to WB. We only do so
* in bulk, so we rarely need to change the page attributes here,
* but doing so requires a stop_machine() from deep inside arch/x86/mm.
* To make detection of the possible sleep more likely, use an
* unconditional might_sleep() for everybody.
*/
might_sleep();
spin_lock(&vm->free_pages.lock);
while (!pagevec_space(&vm->free_pages.pvec))
vm_free_pages_release(vm, false);
GEM_BUG_ON(pagevec_count(&vm->free_pages.pvec) >= PAGEVEC_SIZE);
pagevec_add(&vm->free_pages.pvec, page);
spin_unlock(&vm->free_pages.lock);
}
static void i915_address_space_fini(struct i915_address_space *vm)
{
spin_lock(&vm->free_pages.lock);
if (pagevec_count(&vm->free_pages.pvec))
vm_free_pages_release(vm, true);
GEM_BUG_ON(pagevec_count(&vm->free_pages.pvec));
spin_unlock(&vm->free_pages.lock);
drm_mm_takedown(&vm->mm);
mutex_destroy(&vm->mutex);
}
static void ppgtt_destroy_vma(struct i915_address_space *vm)
{
struct list_head *phases[] = {
&vm->bound_list,
&vm->unbound_list,
NULL,
}, **phase;
mutex_lock(&vm->i915->drm.struct_mutex);
for (phase = phases; *phase; phase++) {
struct i915_vma *vma, *vn;
list_for_each_entry_safe(vma, vn, *phase, vm_link)
i915_vma_destroy(vma);
}
mutex_unlock(&vm->i915->drm.struct_mutex);
}
static void __i915_vm_release(struct work_struct *work)
{
struct i915_address_space *vm =
container_of(work, struct i915_address_space, rcu.work);
ppgtt_destroy_vma(vm);
GEM_BUG_ON(!list_empty(&vm->bound_list));
GEM_BUG_ON(!list_empty(&vm->unbound_list));
vm->cleanup(vm);
i915_address_space_fini(vm);
kfree(vm);
}
void i915_vm_release(struct kref *kref)
{
struct i915_address_space *vm =
container_of(kref, struct i915_address_space, ref);
GEM_BUG_ON(i915_is_ggtt(vm));
trace_i915_ppgtt_release(vm);
vm->closed = true;
queue_rcu_work(vm->i915->wq, &vm->rcu);
}
static void i915_address_space_init(struct i915_address_space *vm, int subclass)
{
kref_init(&vm->ref);
INIT_RCU_WORK(&vm->rcu, __i915_vm_release);
/*
* The vm->mutex must be reclaim safe (for use in the shrinker).
* Do a dummy acquire now under fs_reclaim so that any allocation
* attempt holding the lock is immediately reported by lockdep.
*/
mutex_init(&vm->mutex);
lockdep_set_subclass(&vm->mutex, subclass);
i915_gem_shrinker_taints_mutex(vm->i915, &vm->mutex);
GEM_BUG_ON(!vm->total);
drm_mm_init(&vm->mm, 0, vm->total);
vm->mm.head_node.color = I915_COLOR_UNEVICTABLE;
stash_init(&vm->free_pages);
INIT_LIST_HEAD(&vm->unbound_list);
INIT_LIST_HEAD(&vm->bound_list);
}
static int __setup_page_dma(struct i915_address_space *vm,
struct i915_page_dma *p,
gfp_t gfp)
{
p->page = vm_alloc_page(vm, gfp | I915_GFP_ALLOW_FAIL);
if (unlikely(!p->page))
return -ENOMEM;
p->daddr = dma_map_page_attrs(vm->dma,
p->page, 0, PAGE_SIZE,
PCI_DMA_BIDIRECTIONAL,
DMA_ATTR_SKIP_CPU_SYNC |
DMA_ATTR_NO_WARN);
if (unlikely(dma_mapping_error(vm->dma, p->daddr))) {
vm_free_page(vm, p->page);
return -ENOMEM;
}
return 0;
}
static int setup_page_dma(struct i915_address_space *vm,
struct i915_page_dma *p)
{
return __setup_page_dma(vm, p, __GFP_HIGHMEM);
}
static void cleanup_page_dma(struct i915_address_space *vm,
struct i915_page_dma *p)
{
dma_unmap_page(vm->dma, p->daddr, PAGE_SIZE, PCI_DMA_BIDIRECTIONAL);
vm_free_page(vm, p->page);
}
#define kmap_atomic_px(px) kmap_atomic(px_base(px)->page)
static void
fill_page_dma(const struct i915_page_dma *p, const u64 val, unsigned int count)
{
kunmap_atomic(memset64(kmap_atomic(p->page), val, count));
}
#define fill_px(px, v) fill_page_dma(px_base(px), (v), PAGE_SIZE / sizeof(u64))
#define fill32_px(px, v) do { \
u64 v__ = lower_32_bits(v); \
fill_px((px), v__ << 32 | v__); \
} while (0)
static int
setup_scratch_page(struct i915_address_space *vm, gfp_t gfp)
{
unsigned long size;
/*
* In order to utilize 64K pages for an object with a size < 2M, we will
* need to support a 64K scratch page, given that every 16th entry for a
* page-table operating in 64K mode must point to a properly aligned 64K
* region, including any PTEs which happen to point to scratch.
*
* This is only relevant for the 48b PPGTT where we support
* huge-gtt-pages, see also i915_vma_insert(). However, as we share the
* scratch (read-only) between all vm, we create one 64k scratch page
* for all.
*/
size = I915_GTT_PAGE_SIZE_4K;
if (i915_vm_is_4lvl(vm) &&
HAS_PAGE_SIZES(vm->i915, I915_GTT_PAGE_SIZE_64K)) {
size = I915_GTT_PAGE_SIZE_64K;
gfp |= __GFP_NOWARN;
}
gfp |= __GFP_ZERO | __GFP_RETRY_MAYFAIL;
do {
unsigned int order = get_order(size);
struct page *page;
dma_addr_t addr;
page = alloc_pages(gfp, order);
if (unlikely(!page))
goto skip;
addr = dma_map_page_attrs(vm->dma,
page, 0, size,
PCI_DMA_BIDIRECTIONAL,
DMA_ATTR_SKIP_CPU_SYNC |
DMA_ATTR_NO_WARN);
if (unlikely(dma_mapping_error(vm->dma, addr)))
goto free_page;
if (unlikely(!IS_ALIGNED(addr, size)))
goto unmap_page;
vm->scratch[0].base.page = page;
vm->scratch[0].base.daddr = addr;
vm->scratch_order = order;
return 0;
unmap_page:
dma_unmap_page(vm->dma, addr, size, PCI_DMA_BIDIRECTIONAL);
free_page:
__free_pages(page, order);
skip:
if (size == I915_GTT_PAGE_SIZE_4K)
return -ENOMEM;
size = I915_GTT_PAGE_SIZE_4K;
gfp &= ~__GFP_NOWARN;
} while (1);
}
static void cleanup_scratch_page(struct i915_address_space *vm)
{
struct i915_page_dma *p = px_base(&vm->scratch[0]);
unsigned int order = vm->scratch_order;
dma_unmap_page(vm->dma, p->daddr, BIT(order) << PAGE_SHIFT,
PCI_DMA_BIDIRECTIONAL);
__free_pages(p->page, order);
}
static void free_scratch(struct i915_address_space *vm)
{
int i;
if (!px_dma(&vm->scratch[0])) /* set to 0 on clones */
return;
for (i = 1; i <= vm->top; i++) {
if (!px_dma(&vm->scratch[i]))
break;
cleanup_page_dma(vm, px_base(&vm->scratch[i]));
}
cleanup_scratch_page(vm);
}
static struct i915_page_table *alloc_pt(struct i915_address_space *vm)
{
struct i915_page_table *pt;
pt = kmalloc(sizeof(*pt), I915_GFP_ALLOW_FAIL);
if (unlikely(!pt))
return ERR_PTR(-ENOMEM);
if (unlikely(setup_page_dma(vm, &pt->base))) {
kfree(pt);
return ERR_PTR(-ENOMEM);
}
atomic_set(&pt->used, 0);
return pt;
}
static struct i915_page_directory *__alloc_pd(size_t sz)
{
struct i915_page_directory *pd;
pd = kzalloc(sz, I915_GFP_ALLOW_FAIL);
if (unlikely(!pd))
return NULL;
spin_lock_init(&pd->lock);
return pd;
}
static struct i915_page_directory *alloc_pd(struct i915_address_space *vm)
{
struct i915_page_directory *pd;
pd = __alloc_pd(sizeof(*pd));
if (unlikely(!pd))
return ERR_PTR(-ENOMEM);
if (unlikely(setup_page_dma(vm, px_base(pd)))) {
kfree(pd);
return ERR_PTR(-ENOMEM);
}
return pd;
}
static void free_pd(struct i915_address_space *vm, struct i915_page_dma *pd)
{
cleanup_page_dma(vm, pd);
kfree(pd);
}
#define free_px(vm, px) free_pd(vm, px_base(px))
static inline void
write_dma_entry(struct i915_page_dma * const pdma,
const unsigned short idx,
const u64 encoded_entry)
{
u64 * const vaddr = kmap_atomic(pdma->page);
vaddr[idx] = encoded_entry;
kunmap_atomic(vaddr);
}
static inline void
__set_pd_entry(struct i915_page_directory * const pd,
const unsigned short idx,
struct i915_page_dma * const to,
u64 (*encode)(const dma_addr_t, const enum i915_cache_level))
{
/* Each thread pre-pins the pd, and we may have a thread per pde. */
GEM_BUG_ON(atomic_read(px_used(pd)) > 2 * ARRAY_SIZE(pd->entry));
atomic_inc(px_used(pd));
pd->entry[idx] = to;
write_dma_entry(px_base(pd), idx, encode(to->daddr, I915_CACHE_LLC));
}
#define set_pd_entry(pd, idx, to) \
__set_pd_entry((pd), (idx), px_base(to), gen8_pde_encode)
static inline void
clear_pd_entry(struct i915_page_directory * const pd,
const unsigned short idx,
const struct i915_page_scratch * const scratch)
{
GEM_BUG_ON(atomic_read(px_used(pd)) == 0);
write_dma_entry(px_base(pd), idx, scratch->encode);
pd->entry[idx] = NULL;
atomic_dec(px_used(pd));
}
static bool
release_pd_entry(struct i915_page_directory * const pd,
const unsigned short idx,
struct i915_page_table * const pt,
const struct i915_page_scratch * const scratch)
{
bool free = false;
if (atomic_add_unless(&pt->used, -1, 1))
return false;
spin_lock(&pd->lock);
if (atomic_dec_and_test(&pt->used)) {
clear_pd_entry(pd, idx, scratch);
free = true;
}
spin_unlock(&pd->lock);
return free;
}
/*
* PDE TLBs are a pain to invalidate on GEN8+. When we modify
* the page table structures, we mark them dirty so that
* context switching/execlist queuing code takes extra steps
* to ensure that tlbs are flushed.
*/
static void mark_tlbs_dirty(struct i915_ppgtt *ppgtt)
{
ppgtt->pd_dirty_engines = ALL_ENGINES;
}
static void gen8_ppgtt_notify_vgt(struct i915_ppgtt *ppgtt, bool create)
{
struct drm_i915_private *dev_priv = ppgtt->vm.i915;
enum vgt_g2v_type msg;
int i;
if (create)
atomic_inc(px_used(ppgtt->pd)); /* never remove */
else
atomic_dec(px_used(ppgtt->pd));
mutex_lock(&dev_priv->vgpu.lock);
if (i915_vm_is_4lvl(&ppgtt->vm)) {
const u64 daddr = px_dma(ppgtt->pd);
I915_WRITE(vgtif_reg(pdp[0].lo), lower_32_bits(daddr));
I915_WRITE(vgtif_reg(pdp[0].hi), upper_32_bits(daddr));
msg = (create ? VGT_G2V_PPGTT_L4_PAGE_TABLE_CREATE :
VGT_G2V_PPGTT_L4_PAGE_TABLE_DESTROY);
} else {
for (i = 0; i < GEN8_3LVL_PDPES; i++) {
const u64 daddr = i915_page_dir_dma_addr(ppgtt, i);
I915_WRITE(vgtif_reg(pdp[i].lo), lower_32_bits(daddr));
I915_WRITE(vgtif_reg(pdp[i].hi), upper_32_bits(daddr));
}
msg = (create ? VGT_G2V_PPGTT_L3_PAGE_TABLE_CREATE :
VGT_G2V_PPGTT_L3_PAGE_TABLE_DESTROY);
}
/* g2v_notify atomically (via hv trap) consumes the message packet. */
I915_WRITE(vgtif_reg(g2v_notify), msg);
mutex_unlock(&dev_priv->vgpu.lock);
}
/* Index shifts into the pagetable are offset by GEN8_PTE_SHIFT [12] */
#define GEN8_PAGE_SIZE (SZ_4K) /* page and page-directory sizes are the same */
#define GEN8_PTE_SHIFT (ilog2(GEN8_PAGE_SIZE))
#define GEN8_PDES (GEN8_PAGE_SIZE / sizeof(u64))
#define gen8_pd_shift(lvl) ((lvl) * ilog2(GEN8_PDES))
#define gen8_pd_index(i, lvl) i915_pde_index((i), gen8_pd_shift(lvl))
#define __gen8_pte_shift(lvl) (GEN8_PTE_SHIFT + gen8_pd_shift(lvl))
#define __gen8_pte_index(a, lvl) i915_pde_index((a), __gen8_pte_shift(lvl))
static inline unsigned int
gen8_pd_range(u64 start, u64 end, int lvl, unsigned int *idx)
{
const int shift = gen8_pd_shift(lvl);
const u64 mask = ~0ull << gen8_pd_shift(lvl + 1);
GEM_BUG_ON(start >= end);
end += ~mask >> gen8_pd_shift(1);
*idx = i915_pde_index(start, shift);
if ((start ^ end) & mask)
return GEN8_PDES - *idx;
else
return i915_pde_index(end, shift) - *idx;
}
static inline bool gen8_pd_contains(u64 start, u64 end, int lvl)
{
const u64 mask = ~0ull << gen8_pd_shift(lvl + 1);
GEM_BUG_ON(start >= end);
return (start ^ end) & mask && (start & ~mask) == 0;
}
static inline unsigned int gen8_pt_count(u64 start, u64 end)
{
GEM_BUG_ON(start >= end);
if ((start ^ end) >> gen8_pd_shift(1))
return GEN8_PDES - (start & (GEN8_PDES - 1));
else
return end - start;
}
static inline unsigned int gen8_pd_top_count(const struct i915_address_space *vm)
{
unsigned int shift = __gen8_pte_shift(vm->top);
return (vm->total + (1ull << shift) - 1) >> shift;
}
static inline struct i915_page_directory *
gen8_pdp_for_page_index(struct i915_address_space * const vm, const u64 idx)
{
struct i915_ppgtt * const ppgtt = i915_vm_to_ppgtt(vm);
if (vm->top == 2)
return ppgtt->pd;
else
return i915_pd_entry(ppgtt->pd, gen8_pd_index(idx, vm->top));
}
static inline struct i915_page_directory *
gen8_pdp_for_page_address(struct i915_address_space * const vm, const u64 addr)
{
return gen8_pdp_for_page_index(vm, addr >> GEN8_PTE_SHIFT);
}
static void __gen8_ppgtt_cleanup(struct i915_address_space *vm,
struct i915_page_directory *pd,
int count, int lvl)
{
if (lvl) {
void **pde = pd->entry;
do {
if (!*pde)
continue;
__gen8_ppgtt_cleanup(vm, *pde, GEN8_PDES, lvl - 1);
} while (pde++, --count);
}
free_px(vm, pd);
}
static void gen8_ppgtt_cleanup(struct i915_address_space *vm)
{
struct i915_ppgtt *ppgtt = i915_vm_to_ppgtt(vm);
if (intel_vgpu_active(vm->i915))
gen8_ppgtt_notify_vgt(ppgtt, false);
__gen8_ppgtt_cleanup(vm, ppgtt->pd, gen8_pd_top_count(vm), vm->top);
free_scratch(vm);
}
static u64 __gen8_ppgtt_clear(struct i915_address_space * const vm,
struct i915_page_directory * const pd,
u64 start, const u64 end, int lvl)
{
const struct i915_page_scratch * const scratch = &vm->scratch[lvl];
unsigned int idx, len;
GEM_BUG_ON(end > vm->total >> GEN8_PTE_SHIFT);
len = gen8_pd_range(start, end, lvl--, &idx);
DBG("%s(%p):{ lvl:%d, start:%llx, end:%llx, idx:%d, len:%d, used:%d }\n",
__func__, vm, lvl + 1, start, end,
idx, len, atomic_read(px_used(pd)));
GEM_BUG_ON(!len || len >= atomic_read(px_used(pd)));
do {
struct i915_page_table *pt = pd->entry[idx];
if (atomic_fetch_inc(&pt->used) >> gen8_pd_shift(1) &&
gen8_pd_contains(start, end, lvl)) {
DBG("%s(%p):{ lvl:%d, idx:%d, start:%llx, end:%llx } removing pd\n",
__func__, vm, lvl + 1, idx, start, end);
clear_pd_entry(pd, idx, scratch);
__gen8_ppgtt_cleanup(vm, as_pd(pt), I915_PDES, lvl);
start += (u64)I915_PDES << gen8_pd_shift(lvl);
continue;
}
if (lvl) {
start = __gen8_ppgtt_clear(vm, as_pd(pt),
start, end, lvl);
} else {
unsigned int count;
u64 *vaddr;
count = gen8_pt_count(start, end);
DBG("%s(%p):{ lvl:%d, start:%llx, end:%llx, idx:%d, len:%d, used:%d } removing pte\n",
__func__, vm, lvl, start, end,
gen8_pd_index(start, 0), count,
atomic_read(&pt->used));
GEM_BUG_ON(!count || count >= atomic_read(&pt->used));
vaddr = kmap_atomic_px(pt);
memset64(vaddr + gen8_pd_index(start, 0),
vm->scratch[0].encode,
count);
kunmap_atomic(vaddr);
atomic_sub(count, &pt->used);
start += count;
}
if (release_pd_entry(pd, idx, pt, scratch))
free_px(vm, pt);
} while (idx++, --len);
return start;
}
static void gen8_ppgtt_clear(struct i915_address_space *vm,
u64 start, u64 length)
{
GEM_BUG_ON(!IS_ALIGNED(start, BIT_ULL(GEN8_PTE_SHIFT)));
GEM_BUG_ON(!IS_ALIGNED(length, BIT_ULL(GEN8_PTE_SHIFT)));
GEM_BUG_ON(range_overflows(start, length, vm->total));
start >>= GEN8_PTE_SHIFT;
length >>= GEN8_PTE_SHIFT;
GEM_BUG_ON(length == 0);
__gen8_ppgtt_clear(vm, i915_vm_to_ppgtt(vm)->pd,
start, start + length, vm->top);
}
static int __gen8_ppgtt_alloc(struct i915_address_space * const vm,
struct i915_page_directory * const pd,
u64 * const start, const u64 end, int lvl)
{
const struct i915_page_scratch * const scratch = &vm->scratch[lvl];
struct i915_page_table *alloc = NULL;
unsigned int idx, len;
int ret = 0;
GEM_BUG_ON(end > vm->total >> GEN8_PTE_SHIFT);
len = gen8_pd_range(*start, end, lvl--, &idx);
DBG("%s(%p):{ lvl:%d, start:%llx, end:%llx, idx:%d, len:%d, used:%d }\n",
__func__, vm, lvl + 1, *start, end,
idx, len, atomic_read(px_used(pd)));
GEM_BUG_ON(!len || (idx + len - 1) >> gen8_pd_shift(1));
spin_lock(&pd->lock);
GEM_BUG_ON(!atomic_read(px_used(pd))); /* Must be pinned! */
do {
struct i915_page_table *pt = pd->entry[idx];
if (!pt) {
spin_unlock(&pd->lock);
DBG("%s(%p):{ lvl:%d, idx:%d } allocating new tree\n",
__func__, vm, lvl + 1, idx);
pt = fetch_and_zero(&alloc);
if (lvl) {
if (!pt) {
pt = &alloc_pd(vm)->pt;
if (IS_ERR(pt)) {
ret = PTR_ERR(pt);
goto out;
}
}
fill_px(pt, vm->scratch[lvl].encode);
} else {
if (!pt) {
pt = alloc_pt(vm);
if (IS_ERR(pt)) {
ret = PTR_ERR(pt);
goto out;
}
}
if (intel_vgpu_active(vm->i915) ||
gen8_pt_count(*start, end) < I915_PDES)
fill_px(pt, vm->scratch[lvl].encode);
}
spin_lock(&pd->lock);
if (likely(!pd->entry[idx]))
set_pd_entry(pd, idx, pt);
else
alloc = pt, pt = pd->entry[idx];
}
if (lvl) {
atomic_inc(&pt->used);
spin_unlock(&pd->lock);
ret = __gen8_ppgtt_alloc(vm, as_pd(pt),
start, end, lvl);
if (unlikely(ret)) {
if (release_pd_entry(pd, idx, pt, scratch))
free_px(vm, pt);
goto out;
}
spin_lock(&pd->lock);
atomic_dec(&pt->used);
GEM_BUG_ON(!atomic_read(&pt->used));
} else {
unsigned int count = gen8_pt_count(*start, end);
DBG("%s(%p):{ lvl:%d, start:%llx, end:%llx, idx:%d, len:%d, used:%d } inserting pte\n",
__func__, vm, lvl, *start, end,
gen8_pd_index(*start, 0), count,
atomic_read(&pt->used));
atomic_add(count, &pt->used);
/* All other pdes may be simultaneously removed */
GEM_BUG_ON(atomic_read(&pt->used) > 2 * I915_PDES);
*start += count;
}
} while (idx++, --len);
spin_unlock(&pd->lock);
out:
if (alloc)
free_px(vm, alloc);
return ret;
}
static int gen8_ppgtt_alloc(struct i915_address_space *vm,
u64 start, u64 length)
{
u64 from;
int err;
GEM_BUG_ON(!IS_ALIGNED(start, BIT_ULL(GEN8_PTE_SHIFT)));
GEM_BUG_ON(!IS_ALIGNED(length, BIT_ULL(GEN8_PTE_SHIFT)));
GEM_BUG_ON(range_overflows(start, length, vm->total));
start >>= GEN8_PTE_SHIFT;
length >>= GEN8_PTE_SHIFT;
GEM_BUG_ON(length == 0);
from = start;
err = __gen8_ppgtt_alloc(vm, i915_vm_to_ppgtt(vm)->pd,
&start, start + length, vm->top);
if (unlikely(err && from != start))
__gen8_ppgtt_clear(vm, i915_vm_to_ppgtt(vm)->pd,
from, start, vm->top);
return err;
}
static inline struct sgt_dma {
struct scatterlist *sg;
dma_addr_t dma, max;
} sgt_dma(struct i915_vma *vma) {
struct scatterlist *sg = vma->pages->sgl;
dma_addr_t addr = sg_dma_address(sg);
return (struct sgt_dma) { sg, addr, addr + sg->length };
}
static __always_inline u64
gen8_ppgtt_insert_pte(struct i915_ppgtt *ppgtt,
struct i915_page_directory *pdp,
struct sgt_dma *iter,
u64 idx,
enum i915_cache_level cache_level,
u32 flags)
{
struct i915_page_directory *pd;
const gen8_pte_t pte_encode = gen8_pte_encode(0, cache_level, flags);
gen8_pte_t *vaddr;
pd = i915_pd_entry(pdp, gen8_pd_index(idx, 2));
vaddr = kmap_atomic_px(i915_pt_entry(pd, gen8_pd_index(idx, 1)));
do {
vaddr[gen8_pd_index(idx, 0)] = pte_encode | iter->dma;
iter->dma += I915_GTT_PAGE_SIZE;
if (iter->dma >= iter->max) {
iter->sg = __sg_next(iter->sg);
if (!iter->sg) {
idx = 0;
break;
}
iter->dma = sg_dma_address(iter->sg);
iter->max = iter->dma + iter->sg->length;
}
if (gen8_pd_index(++idx, 0) == 0) {
if (gen8_pd_index(idx, 1) == 0) {
/* Limited by sg length for 3lvl */
if (gen8_pd_index(idx, 2) == 0)
break;
pd = pdp->entry[gen8_pd_index(idx, 2)];
}
kunmap_atomic(vaddr);
vaddr = kmap_atomic_px(i915_pt_entry(pd, gen8_pd_index(idx, 1)));
}
} while (1);
kunmap_atomic(vaddr);
return idx;
}
static void gen8_ppgtt_insert_huge(struct i915_vma *vma,
struct sgt_dma *iter,
enum i915_cache_level cache_level,
u32 flags)
{
const gen8_pte_t pte_encode = gen8_pte_encode(0, cache_level, flags);
u64 start = vma->node.start;
dma_addr_t rem = iter->sg->length;
GEM_BUG_ON(!i915_vm_is_4lvl(vma->vm));
do {
struct i915_page_directory * const pdp =
gen8_pdp_for_page_address(vma->vm, start);
struct i915_page_directory * const pd =
i915_pd_entry(pdp, __gen8_pte_index(start, 2));
gen8_pte_t encode = pte_encode;
unsigned int maybe_64K = -1;
unsigned int page_size;
gen8_pte_t *vaddr;
u16 index;
if (vma->page_sizes.sg & I915_GTT_PAGE_SIZE_2M &&
IS_ALIGNED(iter->dma, I915_GTT_PAGE_SIZE_2M) &&
rem >= I915_GTT_PAGE_SIZE_2M &&
!__gen8_pte_index(start, 0)) {
index = __gen8_pte_index(start, 1);
encode |= GEN8_PDE_PS_2M;
page_size = I915_GTT_PAGE_SIZE_2M;
vaddr = kmap_atomic_px(pd);
} else {
struct i915_page_table *pt =
i915_pt_entry(pd, __gen8_pte_index(start, 1));
index = __gen8_pte_index(start, 0);
page_size = I915_GTT_PAGE_SIZE;
if (!index &&
vma->page_sizes.sg & I915_GTT_PAGE_SIZE_64K &&
IS_ALIGNED(iter->dma, I915_GTT_PAGE_SIZE_64K) &&
(IS_ALIGNED(rem, I915_GTT_PAGE_SIZE_64K) ||
rem >= (I915_PDES - index) * I915_GTT_PAGE_SIZE))
maybe_64K = __gen8_pte_index(start, 1);
vaddr = kmap_atomic_px(pt);
}
do {
GEM_BUG_ON(iter->sg->length < page_size);
vaddr[index++] = encode | iter->dma;
start += page_size;
iter->dma += page_size;
rem -= page_size;
if (iter->dma >= iter->max) {
iter->sg = __sg_next(iter->sg);
if (!iter->sg)
break;
rem = iter->sg->length;
iter->dma = sg_dma_address(iter->sg);
iter->max = iter->dma + rem;
if (maybe_64K != -1 && index < I915_PDES &&
!(IS_ALIGNED(iter->dma, I915_GTT_PAGE_SIZE_64K) &&
(IS_ALIGNED(rem, I915_GTT_PAGE_SIZE_64K) ||
rem >= (I915_PDES - index) * I915_GTT_PAGE_SIZE)))
maybe_64K = -1;
if (unlikely(!IS_ALIGNED(iter->dma, page_size)))
break;
}
} while (rem >= page_size && index < I915_PDES);
kunmap_atomic(vaddr);
/*
* Is it safe to mark the 2M block as 64K? -- Either we have
* filled whole page-table with 64K entries, or filled part of
* it and have reached the end of the sg table and we have
* enough padding.
*/
if (maybe_64K != -1 &&
(index == I915_PDES ||
(i915_vm_has_scratch_64K(vma->vm) &&
!iter->sg && IS_ALIGNED(vma->node.start +
vma->node.size,
I915_GTT_PAGE_SIZE_2M)))) {
vaddr = kmap_atomic_px(pd);
vaddr[maybe_64K] |= GEN8_PDE_IPS_64K;
kunmap_atomic(vaddr);
page_size = I915_GTT_PAGE_SIZE_64K;
/*
* We write all 4K page entries, even when using 64K
* pages. In order to verify that the HW isn't cheating
* by using the 4K PTE instead of the 64K PTE, we want
* to remove all the surplus entries. If the HW skipped
* the 64K PTE, it will read/write into the scratch page
* instead - which we detect as missing results during
* selftests.
*/
if (I915_SELFTEST_ONLY(vma->vm->scrub_64K)) {
u16 i;
encode = vma->vm->scratch[0].encode;
vaddr = kmap_atomic_px(i915_pt_entry(pd, maybe_64K));
for (i = 1; i < index; i += 16)
memset64(vaddr + i, encode, 15);
kunmap_atomic(vaddr);
}
}
vma->page_sizes.gtt |= page_size;
} while (iter->sg);
}
static void gen8_ppgtt_insert(struct i915_address_space *vm,
struct i915_vma *vma,
enum i915_cache_level cache_level,
u32 flags)
{
struct i915_ppgtt * const ppgtt = i915_vm_to_ppgtt(vm);
struct sgt_dma iter = sgt_dma(vma);
if (vma->page_sizes.sg > I915_GTT_PAGE_SIZE) {
gen8_ppgtt_insert_huge(vma, &iter, cache_level, flags);
} else {
u64 idx = vma->node.start >> GEN8_PTE_SHIFT;
do {
struct i915_page_directory * const pdp =
gen8_pdp_for_page_index(vm, idx);
idx = gen8_ppgtt_insert_pte(ppgtt, pdp, &iter, idx,
cache_level, flags);
} while (idx);
vma->page_sizes.gtt = I915_GTT_PAGE_SIZE;
}
}
static int gen8_init_scratch(struct i915_address_space *vm)
{
int ret;
int i;
/*
* If everybody agrees to not to write into the scratch page,
* we can reuse it for all vm, keeping contexts and processes separate.
*/
if (vm->has_read_only &&
vm->i915->kernel_context &&
vm->i915->kernel_context->vm) {
struct i915_address_space *clone = vm->i915->kernel_context->vm;
GEM_BUG_ON(!clone->has_read_only);
vm->scratch_order = clone->scratch_order;
memcpy(vm->scratch, clone->scratch, sizeof(vm->scratch));
px_dma(&vm->scratch[0]) = 0; /* no xfer of ownership */
return 0;
}
ret = setup_scratch_page(vm, __GFP_HIGHMEM);
if (ret)
return ret;
vm->scratch[0].encode =
gen8_pte_encode(px_dma(&vm->scratch[0]),
I915_CACHE_LLC, vm->has_read_only);
for (i = 1; i <= vm->top; i++) {
if (unlikely(setup_page_dma(vm, px_base(&vm->scratch[i]))))
goto free_scratch;
fill_px(&vm->scratch[i], vm->scratch[i - 1].encode);
vm->scratch[i].encode =
gen8_pde_encode(px_dma(&vm->scratch[i]),
I915_CACHE_LLC);
}
return 0;
free_scratch:
free_scratch(vm);
return -ENOMEM;
}
static int gen8_preallocate_top_level_pdp(struct i915_ppgtt *ppgtt)
{
struct i915_address_space *vm = &ppgtt->vm;
struct i915_page_directory *pd = ppgtt->pd;
unsigned int idx;
GEM_BUG_ON(vm->top != 2);
GEM_BUG_ON(gen8_pd_top_count(vm) != GEN8_3LVL_PDPES);
for (idx = 0; idx < GEN8_3LVL_PDPES; idx++) {
struct i915_page_directory *pde;
pde = alloc_pd(vm);
if (IS_ERR(pde))
return PTR_ERR(pde);
fill_px(pde, vm->scratch[1].encode);
set_pd_entry(pd, idx, pde);
atomic_inc(px_used(pde)); /* keep pinned */
}
return 0;
}
static void ppgtt_init(struct i915_ppgtt *ppgtt, struct intel_gt *gt)
{
struct drm_i915_private *i915 = gt->i915;
ppgtt->vm.gt = gt;
ppgtt->vm.i915 = i915;
ppgtt->vm.dma = &i915->drm.pdev->dev;
ppgtt->vm.total = BIT_ULL(INTEL_INFO(i915)->ppgtt_size);
i915_address_space_init(&ppgtt->vm, VM_CLASS_PPGTT);
ppgtt->vm.vma_ops.bind_vma = ppgtt_bind_vma;
ppgtt->vm.vma_ops.unbind_vma = ppgtt_unbind_vma;
ppgtt->vm.vma_ops.set_pages = ppgtt_set_pages;
ppgtt->vm.vma_ops.clear_pages = clear_pages;
}
static struct i915_page_directory *
gen8_alloc_top_pd(struct i915_address_space *vm)
{
const unsigned int count = gen8_pd_top_count(vm);
struct i915_page_directory *pd;
GEM_BUG_ON(count > ARRAY_SIZE(pd->entry));
pd = __alloc_pd(offsetof(typeof(*pd), entry[count]));
if (unlikely(!pd))
return ERR_PTR(-ENOMEM);
if (unlikely(setup_page_dma(vm, px_base(pd)))) {
kfree(pd);
return ERR_PTR(-ENOMEM);
}
fill_page_dma(px_base(pd), vm->scratch[vm->top].encode, count);
atomic_inc(px_used(pd)); /* mark as pinned */
return pd;
}
/*
* GEN8 legacy ppgtt programming is accomplished through a max 4 PDP registers
* with a net effect resembling a 2-level page table in normal x86 terms. Each
* PDP represents 1GB of memory 4 * 512 * 512 * 4096 = 4GB legacy 32b address
* space.
*
*/
static struct i915_ppgtt *gen8_ppgtt_create(struct drm_i915_private *i915)
{
struct i915_ppgtt *ppgtt;
int err;
ppgtt = kzalloc(sizeof(*ppgtt), GFP_KERNEL);
if (!ppgtt)
return ERR_PTR(-ENOMEM);
ppgtt_init(ppgtt, &i915->gt);
ppgtt->vm.top = i915_vm_is_4lvl(&ppgtt->vm) ? 3 : 2;
/*
* From bdw, there is hw support for read-only pages in the PPGTT.
*
* Gen11 has HSDES#:1807136187 unresolved. Disable ro support
* for now.
*/
ppgtt->vm.has_read_only = INTEL_GEN(i915) != 11;
/* There are only few exceptions for gen >=6. chv and bxt.
* And we are not sure about the latter so play safe for now.
*/
if (IS_CHERRYVIEW(i915) || IS_BROXTON(i915))
ppgtt->vm.pt_kmap_wc = true;
err = gen8_init_scratch(&ppgtt->vm);
if (err)
goto err_free;
ppgtt->pd = gen8_alloc_top_pd(&ppgtt->vm);
if (IS_ERR(ppgtt->pd)) {
err = PTR_ERR(ppgtt->pd);
goto err_free_scratch;
}
if (!i915_vm_is_4lvl(&ppgtt->vm)) {
if (intel_vgpu_active(i915)) {
err = gen8_preallocate_top_level_pdp(ppgtt);
if (err)
goto err_free_pd;
}
}
ppgtt->vm.insert_entries = gen8_ppgtt_insert;
ppgtt->vm.allocate_va_range = gen8_ppgtt_alloc;
ppgtt->vm.clear_range = gen8_ppgtt_clear;
if (intel_vgpu_active(i915))
gen8_ppgtt_notify_vgt(ppgtt, true);
ppgtt->vm.cleanup = gen8_ppgtt_cleanup;
return ppgtt;
err_free_pd:
__gen8_ppgtt_cleanup(&ppgtt->vm, ppgtt->pd,
gen8_pd_top_count(&ppgtt->vm), ppgtt->vm.top);
err_free_scratch:
free_scratch(&ppgtt->vm);
err_free:
kfree(ppgtt);
return ERR_PTR(err);
}
/* Write pde (index) from the page directory @pd to the page table @pt */
static inline void gen6_write_pde(const struct gen6_ppgtt *ppgtt,
const unsigned int pde,
const struct i915_page_table *pt)
{
/* Caller needs to make sure the write completes if necessary */
iowrite32(GEN6_PDE_ADDR_ENCODE(px_dma(pt)) | GEN6_PDE_VALID,
ppgtt->pd_addr + pde);
}
static void gen7_ppgtt_enable(struct intel_gt *gt)
{
struct drm_i915_private *i915 = gt->i915;
struct intel_uncore *uncore = gt->uncore;
struct intel_engine_cs *engine;
enum intel_engine_id id;
u32 ecochk;
intel_uncore_rmw(uncore, GAC_ECO_BITS, 0, ECOBITS_PPGTT_CACHE64B);
ecochk = intel_uncore_read(uncore, GAM_ECOCHK);
if (IS_HASWELL(i915)) {
ecochk |= ECOCHK_PPGTT_WB_HSW;
} else {
ecochk |= ECOCHK_PPGTT_LLC_IVB;
ecochk &= ~ECOCHK_PPGTT_GFDT_IVB;
}
intel_uncore_write(uncore, GAM_ECOCHK, ecochk);
for_each_engine(engine, i915, id) {
/* GFX_MODE is per-ring on gen7+ */
ENGINE_WRITE(engine,
RING_MODE_GEN7,
_MASKED_BIT_ENABLE(GFX_PPGTT_ENABLE));
}
}
static void gen6_ppgtt_enable(struct intel_gt *gt)
{
struct intel_uncore *uncore = gt->uncore;
intel_uncore_rmw(uncore,
GAC_ECO_BITS,
0,
ECOBITS_SNB_BIT | ECOBITS_PPGTT_CACHE64B);
intel_uncore_rmw(uncore,
GAB_CTL,
0,
GAB_CTL_CONT_AFTER_PAGEFAULT);
intel_uncore_rmw(uncore,
GAM_ECOCHK,
0,
ECOCHK_SNB_BIT | ECOCHK_PPGTT_CACHE64B);
if (HAS_PPGTT(uncore->i915)) /* may be disabled for VT-d */
intel_uncore_write(uncore,
GFX_MODE,
_MASKED_BIT_ENABLE(GFX_PPGTT_ENABLE));
}
/* PPGTT support for Sandybdrige/Gen6 and later */
static void gen6_ppgtt_clear_range(struct i915_address_space *vm,
u64 start, u64 length)
{
struct gen6_ppgtt * const ppgtt = to_gen6_ppgtt(i915_vm_to_ppgtt(vm));
const unsigned int first_entry = start / I915_GTT_PAGE_SIZE;
const gen6_pte_t scratch_pte = vm->scratch[0].encode;
unsigned int pde = first_entry / GEN6_PTES;
unsigned int pte = first_entry % GEN6_PTES;
unsigned int num_entries = length / I915_GTT_PAGE_SIZE;
while (num_entries) {
struct i915_page_table * const pt =
i915_pt_entry(ppgtt->base.pd, pde++);
const unsigned int count = min(num_entries, GEN6_PTES - pte);
gen6_pte_t *vaddr;
GEM_BUG_ON(px_base(pt) == px_base(&vm->scratch[1]));
num_entries -= count;
GEM_BUG_ON(count > atomic_read(&pt->used));
if (!atomic_sub_return(count, &pt->used))
ppgtt->scan_for_unused_pt = true;
/*
* Note that the hw doesn't support removing PDE on the fly
* (they are cached inside the context with no means to
* invalidate the cache), so we can only reset the PTE
* entries back to scratch.
*/
vaddr = kmap_atomic_px(pt);
memset32(vaddr + pte, scratch_pte, count);
kunmap_atomic(vaddr);
pte = 0;
}
}
static void gen6_ppgtt_insert_entries(struct i915_address_space *vm,
struct i915_vma *vma,
enum i915_cache_level cache_level,
u32 flags)
{
struct i915_ppgtt *ppgtt = i915_vm_to_ppgtt(vm);
struct i915_page_directory * const pd = ppgtt->pd;
unsigned first_entry = vma->node.start / I915_GTT_PAGE_SIZE;
unsigned act_pt = first_entry / GEN6_PTES;
unsigned act_pte = first_entry % GEN6_PTES;
const u32 pte_encode = vm->pte_encode(0, cache_level, flags);
struct sgt_dma iter = sgt_dma(vma);
gen6_pte_t *vaddr;
GEM_BUG_ON(pd->entry[act_pt] == &vm->scratch[1]);
vaddr = kmap_atomic_px(i915_pt_entry(pd, act_pt));
do {
vaddr[act_pte] = pte_encode | GEN6_PTE_ADDR_ENCODE(iter.dma);
iter.dma += I915_GTT_PAGE_SIZE;
if (iter.dma == iter.max) {
iter.sg = __sg_next(iter.sg);
if (!iter.sg)
break;
iter.dma = sg_dma_address(iter.sg);
iter.max = iter.dma + iter.sg->length;
}
if (++act_pte == GEN6_PTES) {
kunmap_atomic(vaddr);
vaddr = kmap_atomic_px(i915_pt_entry(pd, ++act_pt));
act_pte = 0;
}
} while (1);
kunmap_atomic(vaddr);
vma->page_sizes.gtt = I915_GTT_PAGE_SIZE;
}
static int gen6_alloc_va_range(struct i915_address_space *vm,
u64 start, u64 length)
{
struct gen6_ppgtt *ppgtt = to_gen6_ppgtt(i915_vm_to_ppgtt(vm));
struct i915_page_directory * const pd = ppgtt->base.pd;
struct i915_page_table *pt, *alloc = NULL;
intel_wakeref_t wakeref;
u64 from = start;
unsigned int pde;
bool flush = false;
int ret = 0;
wakeref = intel_runtime_pm_get(&vm->i915->runtime_pm);
spin_lock(&pd->lock);
gen6_for_each_pde(pt, pd, start, length, pde) {
const unsigned int count = gen6_pte_count(start, length);
if (px_base(pt) == px_base(&vm->scratch[1])) {
spin_unlock(&pd->lock);
pt = fetch_and_zero(&alloc);
if (!pt)
pt = alloc_pt(vm);
if (IS_ERR(pt)) {
ret = PTR_ERR(pt);
goto unwind_out;
}
fill32_px(pt, vm->scratch[0].encode);
spin_lock(&pd->lock);
if (pd->entry[pde] == &vm->scratch[1]) {
pd->entry[pde] = pt;
if (i915_vma_is_bound(ppgtt->vma,
I915_VMA_GLOBAL_BIND)) {
gen6_write_pde(ppgtt, pde, pt);
flush = true;
}
} else {
alloc = pt;
pt = pd->entry[pde];
}
}
atomic_add(count, &pt->used);
}
spin_unlock(&pd->lock);
if (flush) {
mark_tlbs_dirty(&ppgtt->base);
gen6_ggtt_invalidate(vm->gt->ggtt);
}
goto out;
unwind_out:
gen6_ppgtt_clear_range(vm, from, start - from);
out:
if (alloc)
free_px(vm, alloc);
intel_runtime_pm_put(&vm->i915->runtime_pm, wakeref);
return ret;
}
static int gen6_ppgtt_init_scratch(struct gen6_ppgtt *ppgtt)
{
struct i915_address_space * const vm = &ppgtt->base.vm;
struct i915_page_directory * const pd = ppgtt->base.pd;
int ret;
ret = setup_scratch_page(vm, __GFP_HIGHMEM);
if (ret)
return ret;
vm->scratch[0].encode =
vm->pte_encode(px_dma(&vm->scratch[0]),
I915_CACHE_NONE, PTE_READ_ONLY);
if (unlikely(setup_page_dma(vm, px_base(&vm->scratch[1])))) {
cleanup_scratch_page(vm);
return -ENOMEM;
}
fill32_px(&vm->scratch[1], vm->scratch[0].encode);
memset_p(pd->entry, &vm->scratch[1], I915_PDES);
return 0;
}
static void gen6_ppgtt_free_pd(struct gen6_ppgtt *ppgtt)
{
struct i915_page_directory * const pd = ppgtt->base.pd;
struct i915_page_dma * const scratch =
px_base(&ppgtt->base.vm.scratch[1]);
struct i915_page_table *pt;
u32 pde;
gen6_for_all_pdes(pt, pd, pde)
if (px_base(pt) != scratch)
free_px(&ppgtt->base.vm, pt);
}
static void gen6_ppgtt_cleanup(struct i915_address_space *vm)
{
struct gen6_ppgtt *ppgtt = to_gen6_ppgtt(i915_vm_to_ppgtt(vm));
struct drm_i915_private *i915 = vm->i915;
/* FIXME remove the struct_mutex to bring the locking under control */
mutex_lock(&i915->drm.struct_mutex);
i915_vma_destroy(ppgtt->vma);
mutex_unlock(&i915->drm.struct_mutex);
gen6_ppgtt_free_pd(ppgtt);
free_scratch(vm);
kfree(ppgtt->base.pd);
}
static int pd_vma_set_pages(struct i915_vma *vma)
{
vma->pages = ERR_PTR(-ENODEV);
return 0;
}
static void pd_vma_clear_pages(struct i915_vma *vma)
{
GEM_BUG_ON(!vma->pages);
vma->pages = NULL;
}
static int pd_vma_bind(struct i915_vma *vma,
enum i915_cache_level cache_level,
u32 unused)
{
struct i915_ggtt *ggtt = i915_vm_to_ggtt(vma->vm);
struct gen6_ppgtt *ppgtt = vma->private;
u32 ggtt_offset = i915_ggtt_offset(vma) / I915_GTT_PAGE_SIZE;
struct i915_page_table *pt;
unsigned int pde;
px_base(ppgtt->base.pd)->ggtt_offset = ggtt_offset * sizeof(gen6_pte_t);
ppgtt->pd_addr = (gen6_pte_t __iomem *)ggtt->gsm + ggtt_offset;
gen6_for_all_pdes(pt, ppgtt->base.pd, pde)
gen6_write_pde(ppgtt, pde, pt);
mark_tlbs_dirty(&ppgtt->base);
gen6_ggtt_invalidate(ggtt);
return 0;
}
static void pd_vma_unbind(struct i915_vma *vma)
{
struct gen6_ppgtt *ppgtt = vma->private;
struct i915_page_directory * const pd = ppgtt->base.pd;
struct i915_page_dma * const scratch =
px_base(&ppgtt->base.vm.scratch[1]);
struct i915_page_table *pt;
unsigned int pde;
if (!ppgtt->scan_for_unused_pt)
return;
/* Free all no longer used page tables */
gen6_for_all_pdes(pt, ppgtt->base.pd, pde) {
if (px_base(pt) == scratch || atomic_read(&pt->used))
continue;
free_px(&ppgtt->base.vm, pt);
pd->entry[pde] = scratch;
}
ppgtt->scan_for_unused_pt = false;
}
static const struct i915_vma_ops pd_vma_ops = {
.set_pages = pd_vma_set_pages,
.clear_pages = pd_vma_clear_pages,
.bind_vma = pd_vma_bind,
.unbind_vma = pd_vma_unbind,
};
static struct i915_vma *pd_vma_create(struct gen6_ppgtt *ppgtt, int size)
{
struct drm_i915_private *i915 = ppgtt->base.vm.i915;
struct i915_ggtt *ggtt = ppgtt->base.vm.gt->ggtt;
struct i915_vma *vma;
GEM_BUG_ON(!IS_ALIGNED(size, I915_GTT_PAGE_SIZE));
GEM_BUG_ON(size > ggtt->vm.total);
vma = i915_vma_alloc();
if (!vma)
return ERR_PTR(-ENOMEM);
i915_active_init(i915, &vma->active, NULL, NULL);
vma->vm = &ggtt->vm;
vma->ops = &pd_vma_ops;
vma->private = ppgtt;
vma->size = size;
vma->fence_size = size;
vma->flags = I915_VMA_GGTT;
vma->ggtt_view.type = I915_GGTT_VIEW_ROTATED; /* prevent fencing */
INIT_LIST_HEAD(&vma->obj_link);
INIT_LIST_HEAD(&vma->closed_link);
mutex_lock(&vma->vm->mutex);
list_add(&vma->vm_link, &vma->vm->unbound_list);
mutex_unlock(&vma->vm->mutex);
return vma;
}
int gen6_ppgtt_pin(struct i915_ppgtt *base)
{
struct gen6_ppgtt *ppgtt = to_gen6_ppgtt(base);
int err;
GEM_BUG_ON(ppgtt->base.vm.closed);
/*
* Workaround the limited maximum vma->pin_count and the aliasing_ppgtt
* which will be pinned into every active context.
* (When vma->pin_count becomes atomic, I expect we will naturally
* need a larger, unpacked, type and kill this redundancy.)
*/
if (ppgtt->pin_count++)
return 0;
/*
* PPGTT PDEs reside in the GGTT and consists of 512 entries. The
* allocator works in address space sizes, so it's multiplied by page
* size. We allocate at the top of the GTT to avoid fragmentation.
*/
err = i915_vma_pin(ppgtt->vma,
0, GEN6_PD_ALIGN,
PIN_GLOBAL | PIN_HIGH);
if (err)
goto unpin;
return 0;
unpin:
ppgtt->pin_count = 0;
return err;
}
void gen6_ppgtt_unpin(struct i915_ppgtt *base)
{
struct gen6_ppgtt *ppgtt = to_gen6_ppgtt(base);
GEM_BUG_ON(!ppgtt->pin_count);
if (--ppgtt->pin_count)
return;
i915_vma_unpin(ppgtt->vma);
}
void gen6_ppgtt_unpin_all(struct i915_ppgtt *base)
{
struct gen6_ppgtt *ppgtt = to_gen6_ppgtt(base);
if (!ppgtt->pin_count)
return;
ppgtt->pin_count = 0;
i915_vma_unpin(ppgtt->vma);
}
static struct i915_ppgtt *gen6_ppgtt_create(struct drm_i915_private *i915)
{
struct i915_ggtt * const ggtt = &i915->ggtt;
struct gen6_ppgtt *ppgtt;
int err;
ppgtt = kzalloc(sizeof(*ppgtt), GFP_KERNEL);
if (!ppgtt)
return ERR_PTR(-ENOMEM);
ppgtt_init(&ppgtt->base, &i915->gt);
ppgtt->base.vm.top = 1;
ppgtt->base.vm.allocate_va_range = gen6_alloc_va_range;
ppgtt->base.vm.clear_range = gen6_ppgtt_clear_range;
ppgtt->base.vm.insert_entries = gen6_ppgtt_insert_entries;
ppgtt->base.vm.cleanup = gen6_ppgtt_cleanup;
ppgtt->base.vm.pte_encode = ggtt->vm.pte_encode;
ppgtt->base.pd = __alloc_pd(sizeof(*ppgtt->base.pd));
if (!ppgtt->base.pd) {
err = -ENOMEM;
goto err_free;
}
err = gen6_ppgtt_init_scratch(ppgtt);
if (err)
goto err_pd;
ppgtt->vma = pd_vma_create(ppgtt, GEN6_PD_SIZE);
if (IS_ERR(ppgtt->vma)) {
err = PTR_ERR(ppgtt->vma);
goto err_scratch;
}
return &ppgtt->base;
err_scratch:
free_scratch(&ppgtt->base.vm);
err_pd:
kfree(ppgtt->base.pd);
err_free:
kfree(ppgtt);
return ERR_PTR(err);
}
static void gtt_write_workarounds(struct intel_gt *gt)
{
struct drm_i915_private *i915 = gt->i915;
struct intel_uncore *uncore = gt->uncore;
/* This function is for gtt related workarounds. This function is
* called on driver load and after a GPU reset, so you can place
* workarounds here even if they get overwritten by GPU reset.
*/
/* WaIncreaseDefaultTLBEntries:chv,bdw,skl,bxt,kbl,glk,cfl,cnl,icl */
if (IS_BROADWELL(i915))
intel_uncore_write(uncore,
GEN8_L3_LRA_1_GPGPU,
GEN8_L3_LRA_1_GPGPU_DEFAULT_VALUE_BDW);
else if (IS_CHERRYVIEW(i915))
intel_uncore_write(uncore,
GEN8_L3_LRA_1_GPGPU,
GEN8_L3_LRA_1_GPGPU_DEFAULT_VALUE_CHV);
else if (IS_GEN9_LP(i915))
intel_uncore_write(uncore,
GEN8_L3_LRA_1_GPGPU,
GEN9_L3_LRA_1_GPGPU_DEFAULT_VALUE_BXT);
else if (INTEL_GEN(i915) >= 9)
intel_uncore_write(uncore,
GEN8_L3_LRA_1_GPGPU,
GEN9_L3_LRA_1_GPGPU_DEFAULT_VALUE_SKL);
/*
* To support 64K PTEs we need to first enable the use of the
* Intermediate-Page-Size(IPS) bit of the PDE field via some magical
* mmio, otherwise the page-walker will simply ignore the IPS bit. This
* shouldn't be needed after GEN10.
*
* 64K pages were first introduced from BDW+, although technically they
* only *work* from gen9+. For pre-BDW we instead have the option for
* 32K pages, but we don't currently have any support for it in our
* driver.
*/
if (HAS_PAGE_SIZES(i915, I915_GTT_PAGE_SIZE_64K) &&
INTEL_GEN(i915) <= 10)
intel_uncore_rmw(uncore,
GEN8_GAMW_ECO_DEV_RW_IA,
0,
GAMW_ECO_ENABLE_64K_IPS_FIELD);
if (IS_GEN_RANGE(i915, 8, 11)) {
bool can_use_gtt_cache = true;
/*
* According to the BSpec if we use 2M/1G pages then we also
* need to disable the GTT cache. At least on BDW we can see
* visual corruption when using 2M pages, and not disabling the
* GTT cache.
*/
if (HAS_PAGE_SIZES(i915, I915_GTT_PAGE_SIZE_2M))
can_use_gtt_cache = false;
/* WaGttCachingOffByDefault */
intel_uncore_write(uncore,
HSW_GTT_CACHE_EN,
can_use_gtt_cache ? GTT_CACHE_EN_ALL : 0);
WARN_ON_ONCE(can_use_gtt_cache &&
intel_uncore_read(uncore,
HSW_GTT_CACHE_EN) == 0);
}
}
int i915_ppgtt_init_hw(struct intel_gt *gt)
{
struct drm_i915_private *i915 = gt->i915;
gtt_write_workarounds(gt);
if (IS_GEN(i915, 6))
gen6_ppgtt_enable(gt);
else if (IS_GEN(i915, 7))
gen7_ppgtt_enable(gt);
return 0;
}
static struct i915_ppgtt *
__ppgtt_create(struct drm_i915_private *i915)
{
if (INTEL_GEN(i915) < 8)
return gen6_ppgtt_create(i915);
else
return gen8_ppgtt_create(i915);
}
struct i915_ppgtt *
i915_ppgtt_create(struct drm_i915_private *i915)
{
struct i915_ppgtt *ppgtt;
ppgtt = __ppgtt_create(i915);
if (IS_ERR(ppgtt))
return ppgtt;
trace_i915_ppgtt_create(&ppgtt->vm);
return ppgtt;
}
/* Certain Gen5 chipsets require require idling the GPU before
* unmapping anything from the GTT when VT-d is enabled.
*/
static bool needs_idle_maps(struct drm_i915_private *dev_priv)
{
/* Query intel_iommu to see if we need the workaround. Presumably that
* was loaded first.
*/
return IS_GEN(dev_priv, 5) && IS_MOBILE(dev_priv) && intel_vtd_active();
}
static void ggtt_suspend_mappings(struct i915_ggtt *ggtt)
{
struct drm_i915_private *i915 = ggtt->vm.i915;
/* Don't bother messing with faults pre GEN6 as we have little
* documentation supporting that it's a good idea.
*/
if (INTEL_GEN(i915) < 6)
return;
intel_gt_check_and_clear_faults(ggtt->vm.gt);
ggtt->vm.clear_range(&ggtt->vm, 0, ggtt->vm.total);
ggtt->invalidate(ggtt);
}
void i915_gem_suspend_gtt_mappings(struct drm_i915_private *i915)
{
ggtt_suspend_mappings(&i915->ggtt);
}
int i915_gem_gtt_prepare_pages(struct drm_i915_gem_object *obj,
struct sg_table *pages)
{
do {
if (dma_map_sg_attrs(&obj->base.dev->pdev->dev,
pages->sgl, pages->nents,
PCI_DMA_BIDIRECTIONAL,
DMA_ATTR_NO_WARN))
return 0;
/*
* If the DMA remap fails, one cause can be that we have
* too many objects pinned in a small remapping table,
* such as swiotlb. Incrementally purge all other objects and
* try again - if there are no more pages to remove from
* the DMA remapper, i915_gem_shrink will return 0.
*/
GEM_BUG_ON(obj->mm.pages == pages);
} while (i915_gem_shrink(to_i915(obj->base.dev),
obj->base.size >> PAGE_SHIFT, NULL,
I915_SHRINK_BOUND |
I915_SHRINK_UNBOUND));
return -ENOSPC;
}
static void gen8_set_pte(void __iomem *addr, gen8_pte_t pte)
{
writeq(pte, addr);
}
static void gen8_ggtt_insert_page(struct i915_address_space *vm,
dma_addr_t addr,
u64 offset,
enum i915_cache_level level,
u32 unused)
{
struct i915_ggtt *ggtt = i915_vm_to_ggtt(vm);
gen8_pte_t __iomem *pte =
(gen8_pte_t __iomem *)ggtt->gsm + offset / I915_GTT_PAGE_SIZE;
gen8_set_pte(pte, gen8_pte_encode(addr, level, 0));
ggtt->invalidate(ggtt);
}
static void gen8_ggtt_insert_entries(struct i915_address_space *vm,
struct i915_vma *vma,
enum i915_cache_level level,
u32 flags)
{
struct i915_ggtt *ggtt = i915_vm_to_ggtt(vm);
struct sgt_iter sgt_iter;
gen8_pte_t __iomem *gtt_entries;
const gen8_pte_t pte_encode = gen8_pte_encode(0, level, 0);
dma_addr_t addr;
/*
* Note that we ignore PTE_READ_ONLY here. The caller must be careful
* not to allow the user to override access to a read only page.
*/
gtt_entries = (gen8_pte_t __iomem *)ggtt->gsm;
gtt_entries += vma->node.start / I915_GTT_PAGE_SIZE;
for_each_sgt_dma(addr, sgt_iter, vma->pages)
gen8_set_pte(gtt_entries++, pte_encode | addr);
/*
* We want to flush the TLBs only after we're certain all the PTE
* updates have finished.
*/
ggtt->invalidate(ggtt);
}
static void gen6_ggtt_insert_page(struct i915_address_space *vm,
dma_addr_t addr,
u64 offset,
enum i915_cache_level level,
u32 flags)
{
struct i915_ggtt *ggtt = i915_vm_to_ggtt(vm);
gen6_pte_t __iomem *pte =
(gen6_pte_t __iomem *)ggtt->gsm + offset / I915_GTT_PAGE_SIZE;
iowrite32(vm->pte_encode(addr, level, flags), pte);
ggtt->invalidate(ggtt);
}
/*
* Binds an object into the global gtt with the specified cache level. The object
* will be accessible to the GPU via commands whose operands reference offsets
* within the global GTT as well as accessible by the GPU through the GMADR
* mapped BAR (dev_priv->mm.gtt->gtt).
*/
static void gen6_ggtt_insert_entries(struct i915_address_space *vm,
struct i915_vma *vma,
enum i915_cache_level level,
u32 flags)
{
struct i915_ggtt *ggtt = i915_vm_to_ggtt(vm);
gen6_pte_t __iomem *entries = (gen6_pte_t __iomem *)ggtt->gsm;
unsigned int i = vma->node.start / I915_GTT_PAGE_SIZE;
struct sgt_iter iter;
dma_addr_t addr;
for_each_sgt_dma(addr, iter, vma->pages)
iowrite32(vm->pte_encode(addr, level, flags), &entries[i++]);
/*
* We want to flush the TLBs only after we're certain all the PTE
* updates have finished.
*/
ggtt->invalidate(ggtt);
}
static void nop_clear_range(struct i915_address_space *vm,
u64 start, u64 length)
{
}
static void gen8_ggtt_clear_range(struct i915_address_space *vm,
u64 start, u64 length)
{
struct i915_ggtt *ggtt = i915_vm_to_ggtt(vm);
unsigned first_entry = start / I915_GTT_PAGE_SIZE;
unsigned num_entries = length / I915_GTT_PAGE_SIZE;
const gen8_pte_t scratch_pte = vm->scratch[0].encode;
gen8_pte_t __iomem *gtt_base =
(gen8_pte_t __iomem *)ggtt->gsm + first_entry;
const int max_entries = ggtt_total_entries(ggtt) - first_entry;
int i;
if (WARN(num_entries > max_entries,
"First entry = %d; Num entries = %d (max=%d)\n",
first_entry, num_entries, max_entries))
num_entries = max_entries;
for (i = 0; i < num_entries; i++)
gen8_set_pte(&gtt_base[i], scratch_pte);
}
static void bxt_vtd_ggtt_wa(struct i915_address_space *vm)
{
struct drm_i915_private *dev_priv = vm->i915;
/*
* Make sure the internal GAM fifo has been cleared of all GTT
* writes before exiting stop_machine(). This guarantees that
* any aperture accesses waiting to start in another process
* cannot back up behind the GTT writes causing a hang.
* The register can be any arbitrary GAM register.
*/
POSTING_READ(GFX_FLSH_CNTL_GEN6);
}
struct insert_page {
struct i915_address_space *vm;
dma_addr_t addr;
u64 offset;
enum i915_cache_level level;
};
static int bxt_vtd_ggtt_insert_page__cb(void *_arg)
{
struct insert_page *arg = _arg;
gen8_ggtt_insert_page(arg->vm, arg->addr, arg->offset, arg->level, 0);
bxt_vtd_ggtt_wa(arg->vm);
return 0;
}
static void bxt_vtd_ggtt_insert_page__BKL(struct i915_address_space *vm,
dma_addr_t addr,
u64 offset,
enum i915_cache_level level,
u32 unused)
{
struct insert_page arg = { vm, addr, offset, level };
stop_machine(bxt_vtd_ggtt_insert_page__cb, &arg, NULL);
}
struct insert_entries {
struct i915_address_space *vm;
struct i915_vma *vma;
enum i915_cache_level level;
u32 flags;
};
static int bxt_vtd_ggtt_insert_entries__cb(void *_arg)
{
struct insert_entries *arg = _arg;
gen8_ggtt_insert_entries(arg->vm, arg->vma, arg->level, arg->flags);
bxt_vtd_ggtt_wa(arg->vm);
return 0;
}
static void bxt_vtd_ggtt_insert_entries__BKL(struct i915_address_space *vm,
struct i915_vma *vma,
enum i915_cache_level level,
u32 flags)
{
struct insert_entries arg = { vm, vma, level, flags };
stop_machine(bxt_vtd_ggtt_insert_entries__cb, &arg, NULL);
}
struct clear_range {
struct i915_address_space *vm;
u64 start;
u64 length;
};
static int bxt_vtd_ggtt_clear_range__cb(void *_arg)
{
struct clear_range *arg = _arg;
gen8_ggtt_clear_range(arg->vm, arg->start, arg->length);
bxt_vtd_ggtt_wa(arg->vm);
return 0;
}
static void bxt_vtd_ggtt_clear_range__BKL(struct i915_address_space *vm,
u64 start,
u64 length)
{
struct clear_range arg = { vm, start, length };
stop_machine(bxt_vtd_ggtt_clear_range__cb, &arg, NULL);
}
static void gen6_ggtt_clear_range(struct i915_address_space *vm,
u64 start, u64 length)
{
struct i915_ggtt *ggtt = i915_vm_to_ggtt(vm);
unsigned first_entry = start / I915_GTT_PAGE_SIZE;
unsigned num_entries = length / I915_GTT_PAGE_SIZE;
gen6_pte_t scratch_pte, __iomem *gtt_base =
(gen6_pte_t __iomem *)ggtt->gsm + first_entry;
const int max_entries = ggtt_total_entries(ggtt) - first_entry;
int i;
if (WARN(num_entries > max_entries,
"First entry = %d; Num entries = %d (max=%d)\n",
first_entry, num_entries, max_entries))
num_entries = max_entries;
scratch_pte = vm->scratch[0].encode;
for (i = 0; i < num_entries; i++)
iowrite32(scratch_pte, &gtt_base[i]);
}
static void i915_ggtt_insert_page(struct i915_address_space *vm,
dma_addr_t addr,
u64 offset,
enum i915_cache_level cache_level,
u32 unused)
{
unsigned int flags = (cache_level == I915_CACHE_NONE) ?
AGP_USER_MEMORY : AGP_USER_CACHED_MEMORY;
intel_gtt_insert_page(addr, offset >> PAGE_SHIFT, flags);
}
static void i915_ggtt_insert_entries(struct i915_address_space *vm,
struct i915_vma *vma,
enum i915_cache_level cache_level,
u32 unused)
{
unsigned int flags = (cache_level == I915_CACHE_NONE) ?
AGP_USER_MEMORY : AGP_USER_CACHED_MEMORY;
intel_gtt_insert_sg_entries(vma->pages, vma->node.start >> PAGE_SHIFT,
flags);
}
static void i915_ggtt_clear_range(struct i915_address_space *vm,
u64 start, u64 length)
{
intel_gtt_clear_range(start >> PAGE_SHIFT, length >> PAGE_SHIFT);
}
static int ggtt_bind_vma(struct i915_vma *vma,
enum i915_cache_level cache_level,
u32 flags)
{
struct drm_i915_private *i915 = vma->vm->i915;
struct drm_i915_gem_object *obj = vma->obj;
intel_wakeref_t wakeref;
u32 pte_flags;
/* Applicable to VLV (gen8+ do not support RO in the GGTT) */
pte_flags = 0;
if (i915_gem_object_is_readonly(obj))
pte_flags |= PTE_READ_ONLY;
with_intel_runtime_pm(&i915->runtime_pm, wakeref)
vma->vm->insert_entries(vma->vm, vma, cache_level, pte_flags);
vma->page_sizes.gtt = I915_GTT_PAGE_SIZE;
/*
* Without aliasing PPGTT there's no difference between
* GLOBAL/LOCAL_BIND, it's all the same ptes. Hence unconditionally
* upgrade to both bound if we bind either to avoid double-binding.
*/
vma->flags |= I915_VMA_GLOBAL_BIND | I915_VMA_LOCAL_BIND;
return 0;
}
static void ggtt_unbind_vma(struct i915_vma *vma)
{
struct drm_i915_private *i915 = vma->vm->i915;
intel_wakeref_t wakeref;
with_intel_runtime_pm(&i915->runtime_pm, wakeref)
vma->vm->clear_range(vma->vm, vma->node.start, vma->size);
}
static int aliasing_gtt_bind_vma(struct i915_vma *vma,
enum i915_cache_level cache_level,
u32 flags)
{
struct drm_i915_private *i915 = vma->vm->i915;
u32 pte_flags;
int ret;
/* Currently applicable only to VLV */
pte_flags = 0;
if (i915_gem_object_is_readonly(vma->obj))
pte_flags |= PTE_READ_ONLY;
if (flags & I915_VMA_LOCAL_BIND) {
struct i915_ppgtt *alias = i915_vm_to_ggtt(vma->vm)->alias;
if (!(vma->flags & I915_VMA_LOCAL_BIND)) {
ret = alias->vm.allocate_va_range(&alias->vm,
vma->node.start,
vma->size);
if (ret)
return ret;
}
alias->vm.insert_entries(&alias->vm, vma,
cache_level, pte_flags);
}
if (flags & I915_VMA_GLOBAL_BIND) {
intel_wakeref_t wakeref;
with_intel_runtime_pm(&i915->runtime_pm, wakeref) {
vma->vm->insert_entries(vma->vm, vma,
cache_level, pte_flags);
}
}
return 0;
}
static void aliasing_gtt_unbind_vma(struct i915_vma *vma)
{
struct drm_i915_private *i915 = vma->vm->i915;
if (vma->flags & I915_VMA_GLOBAL_BIND) {
struct i915_address_space *vm = vma->vm;
intel_wakeref_t wakeref;
with_intel_runtime_pm(&i915->runtime_pm, wakeref)
vm->clear_range(vm, vma->node.start, vma->size);
}
if (vma->flags & I915_VMA_LOCAL_BIND) {
struct i915_address_space *vm =
&i915_vm_to_ggtt(vma->vm)->alias->vm;
vm->clear_range(vm, vma->node.start, vma->size);
}
}
void i915_gem_gtt_finish_pages(struct drm_i915_gem_object *obj,
struct sg_table *pages)
{
struct drm_i915_private *dev_priv = to_i915(obj->base.dev);
struct device *kdev = &dev_priv->drm.pdev->dev;
struct i915_ggtt *ggtt = &dev_priv->ggtt;
if (unlikely(ggtt->do_idle_maps)) {
if (i915_gem_wait_for_idle(dev_priv, 0, MAX_SCHEDULE_TIMEOUT)) {
DRM_ERROR("Failed to wait for idle; VT'd may hang.\n");
/* Wait a bit, in hopes it avoids the hang */
udelay(10);
}
}
dma_unmap_sg(kdev, pages->sgl, pages->nents, PCI_DMA_BIDIRECTIONAL);
}
static int ggtt_set_pages(struct i915_vma *vma)
{
int ret;
GEM_BUG_ON(vma->pages);
ret = i915_get_ggtt_vma_pages(vma);
if (ret)
return ret;
vma->page_sizes = vma->obj->mm.page_sizes;
return 0;
}
static void i915_gtt_color_adjust(const struct drm_mm_node *node,
unsigned long color,
u64 *start,
u64 *end)
{
if (node->allocated && node->color != color)
*start += I915_GTT_PAGE_SIZE;
/* Also leave a space between the unallocated reserved node after the
* GTT and any objects within the GTT, i.e. we use the color adjustment
* to insert a guard page to prevent prefetches crossing over the
* GTT boundary.
*/
node = list_next_entry(node, node_list);
if (node->color != color)
*end -= I915_GTT_PAGE_SIZE;
}
static int init_aliasing_ppgtt(struct i915_ggtt *ggtt)
{
struct i915_ppgtt *ppgtt;
int err;
ppgtt = i915_ppgtt_create(ggtt->vm.i915);
if (IS_ERR(ppgtt))
return PTR_ERR(ppgtt);
if (GEM_WARN_ON(ppgtt->vm.total < ggtt->vm.total)) {
err = -ENODEV;
goto err_ppgtt;
}
/*
* Note we only pre-allocate as far as the end of the global
* GTT. On 48b / 4-level page-tables, the difference is very,
* very significant! We have to preallocate as GVT/vgpu does
* not like the page directory disappearing.
*/
err = ppgtt->vm.allocate_va_range(&ppgtt->vm, 0, ggtt->vm.total);
if (err)
goto err_ppgtt;
ggtt->alias = ppgtt;
GEM_BUG_ON(ggtt->vm.vma_ops.bind_vma != ggtt_bind_vma);
ggtt->vm.vma_ops.bind_vma = aliasing_gtt_bind_vma;
GEM_BUG_ON(ggtt->vm.vma_ops.unbind_vma != ggtt_unbind_vma);
ggtt->vm.vma_ops.unbind_vma = aliasing_gtt_unbind_vma;
return 0;
err_ppgtt:
i915_vm_put(&ppgtt->vm);
return err;
}
static void fini_aliasing_ppgtt(struct i915_ggtt *ggtt)
{
struct drm_i915_private *i915 = ggtt->vm.i915;
struct i915_ppgtt *ppgtt;
mutex_lock(&i915->drm.struct_mutex);
ppgtt = fetch_and_zero(&ggtt->alias);
if (!ppgtt)
goto out;
i915_vm_put(&ppgtt->vm);
ggtt->vm.vma_ops.bind_vma = ggtt_bind_vma;
ggtt->vm.vma_ops.unbind_vma = ggtt_unbind_vma;
out:
mutex_unlock(&i915->drm.struct_mutex);
}
static int ggtt_reserve_guc_top(struct i915_ggtt *ggtt)
{
u64 size;
int ret;
if (!USES_GUC(ggtt->vm.i915))
return 0;
GEM_BUG_ON(ggtt->vm.total <= GUC_GGTT_TOP);
size = ggtt->vm.total - GUC_GGTT_TOP;
ret = i915_gem_gtt_reserve(&ggtt->vm, &ggtt->uc_fw, size,
GUC_GGTT_TOP, I915_COLOR_UNEVICTABLE,
PIN_NOEVICT);
if (ret)
DRM_DEBUG_DRIVER("Failed to reserve top of GGTT for GuC\n");
return ret;
}
static void ggtt_release_guc_top(struct i915_ggtt *ggtt)
{
if (drm_mm_node_allocated(&ggtt->uc_fw))
drm_mm_remove_node(&ggtt->uc_fw);
}
static void cleanup_init_ggtt(struct i915_ggtt *ggtt)
{
ggtt_release_guc_top(ggtt);
drm_mm_remove_node(&ggtt->error_capture);
}
static int init_ggtt(struct i915_ggtt *ggtt)
{
/* Let GEM Manage all of the aperture.
*
* However, leave one page at the end still bound to the scratch page.
* There are a number of places where the hardware apparently prefetches
* past the end of the object, and we've seen multiple hangs with the
* GPU head pointer stuck in a batchbuffer bound at the last page of the
* aperture. One page should be enough to keep any prefetching inside
* of the aperture.
*/
unsigned long hole_start, hole_end;
struct drm_mm_node *entry;
int ret;
/*
* GuC requires all resources that we're sharing with it to be placed in
* non-WOPCM memory. If GuC is not present or not in use we still need a
* small bias as ring wraparound at offset 0 sometimes hangs. No idea
* why.
*/
ggtt->pin_bias = max_t(u32, I915_GTT_PAGE_SIZE,
intel_wopcm_guc_size(&ggtt->vm.i915->wopcm));
ret = intel_vgt_balloon(ggtt);
if (ret)
return ret;
/* Reserve a mappable slot for our lockless error capture */
ret = drm_mm_insert_node_in_range(&ggtt->vm.mm, &ggtt->error_capture,
PAGE_SIZE, 0, I915_COLOR_UNEVICTABLE,
0, ggtt->mappable_end,
DRM_MM_INSERT_LOW);
if (ret)
return ret;
/*
* The upper portion of the GuC address space has a sizeable hole
* (several MB) that is inaccessible by GuC. Reserve this range within
* GGTT as it can comfortably hold GuC/HuC firmware images.
*/
ret = ggtt_reserve_guc_top(ggtt);
if (ret)
goto err;
/* Clear any non-preallocated blocks */
drm_mm_for_each_hole(entry, &ggtt->vm.mm, hole_start, hole_end) {
DRM_DEBUG_KMS("clearing unused GTT space: [%lx, %lx]\n",
hole_start, hole_end);
ggtt->vm.clear_range(&ggtt->vm, hole_start,
hole_end - hole_start);
}
/* And finally clear the reserved guard page */
ggtt->vm.clear_range(&ggtt->vm, ggtt->vm.total - PAGE_SIZE, PAGE_SIZE);
return 0;
err:
cleanup_init_ggtt(ggtt);
return ret;
}
int i915_init_ggtt(struct drm_i915_private *i915)
{
int ret;
ret = init_ggtt(&i915->ggtt);
if (ret)
return ret;
if (INTEL_PPGTT(i915) == INTEL_PPGTT_ALIASING) {
ret = init_aliasing_ppgtt(&i915->ggtt);
if (ret)
cleanup_init_ggtt(&i915->ggtt);
}
return 0;
}
static void ggtt_cleanup_hw(struct i915_ggtt *ggtt)
{
struct drm_i915_private *i915 = ggtt->vm.i915;
struct i915_vma *vma, *vn;
ggtt->vm.closed = true;
rcu_barrier(); /* flush the RCU'ed__i915_vm_release */
flush_workqueue(i915->wq);
mutex_lock(&i915->drm.struct_mutex);
list_for_each_entry_safe(vma, vn, &ggtt->vm.bound_list, vm_link)
WARN_ON(i915_vma_unbind(vma));
if (drm_mm_node_allocated(&ggtt->error_capture))
drm_mm_remove_node(&ggtt->error_capture);
ggtt_release_guc_top(ggtt);
if (drm_mm_initialized(&ggtt->vm.mm)) {
intel_vgt_deballoon(ggtt);
i915_address_space_fini(&ggtt->vm);
}
ggtt->vm.cleanup(&ggtt->vm);
mutex_unlock(&i915->drm.struct_mutex);
arch_phys_wc_del(ggtt->mtrr);
io_mapping_fini(&ggtt->iomap);
}
/**
* i915_ggtt_driver_release - Clean up GGTT hardware initialization
* @i915: i915 device
*/
void i915_ggtt_driver_release(struct drm_i915_private *i915)
{
struct pagevec *pvec;
fini_aliasing_ppgtt(&i915->ggtt);
ggtt_cleanup_hw(&i915->ggtt);
pvec = &i915->mm