drm/i915: Use remap_io_mapping() to prefault all PTE in a single pass

Very old numbers indicate this is a 66% improvement when remapping the
entire object for fence contention - due to the elimination of
track_pfn_insert and its strcmp.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Testcase: igt/gem_fence_upload/performance
Testcase: igt/gem_mmap_gtt
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20160819155428.1670-6-chris@chris-wilson.co.uk
This commit is contained in:
Chris Wilson 2016-08-19 16:54:28 +01:00
parent f7bbe7883c
commit c58305af18
5 changed files with 99 additions and 45 deletions

View File

@ -46,7 +46,7 @@ obj-$(CONFIG_DRM_RADEON)+= radeon/
obj-$(CONFIG_DRM_AMDGPU)+= amd/amdgpu/
obj-$(CONFIG_DRM_MGA) += mga/
obj-$(CONFIG_DRM_I810) += i810/
obj-$(CONFIG_DRM_I915) += i915/
obj-$(CONFIG_DRM_I915) += i915/
obj-$(CONFIG_DRM_MGAG200) += mgag200/
obj-$(CONFIG_DRM_VC4) += vc4/
obj-$(CONFIG_DRM_CIRRUS_QEMU) += cirrus/

View File

@ -12,6 +12,7 @@ subdir-ccflags-y += \
i915-y := i915_drv.o \
i915_irq.o \
i915_memcpy.o \
i915_mm.o \
i915_params.o \
i915_pci.o \
i915_suspend.o \
@ -113,6 +114,6 @@ i915-y += intel_gvt.o
include $(src)/gvt/Makefile
endif
obj-$(CONFIG_DRM_I915) += i915.o
obj-$(CONFIG_DRM_I915) += i915.o
CFLAGS_i915_trace_points.o := -I$(src)

View File

@ -3931,6 +3931,11 @@ static inline bool __i915_request_irq_complete(struct drm_i915_gem_request *req)
void i915_memcpy_init_early(struct drm_i915_private *dev_priv);
bool i915_memcpy_from_wc(void *dst, const void *src, unsigned long len);
/* i915_mm.c */
int remap_io_mapping(struct vm_area_struct *vma,
unsigned long addr, unsigned long pfn, unsigned long size,
struct io_mapping *iomap);
#define ptr_mask_bits(ptr) ({ \
unsigned long __v = (unsigned long)(ptr); \
(typeof(ptr))(__v & PAGE_MASK); \

View File

@ -1705,7 +1705,6 @@ int i915_gem_fault(struct vm_area_struct *area, struct vm_fault *vmf)
bool write = !!(vmf->flags & FAULT_FLAG_WRITE);
struct i915_vma *vma;
pgoff_t page_offset;
unsigned long pfn;
unsigned int flags;
int ret;
@ -1790,48 +1789,13 @@ int i915_gem_fault(struct vm_area_struct *area, struct vm_fault *vmf)
goto err_unpin;
/* Finally, remap it using the new GTT offset */
pfn = ggtt->mappable_base + i915_ggtt_offset(vma);
pfn >>= PAGE_SHIFT;
if (vma->ggtt_view.type == I915_GGTT_VIEW_NORMAL) {
if (!obj->fault_mappable) {
unsigned long size =
min_t(unsigned long,
area->vm_end - area->vm_start,
obj->base.size) >> PAGE_SHIFT;
unsigned long base = area->vm_start;
int i;
for (i = 0; i < size; i++) {
ret = vm_insert_pfn(area,
base + i * PAGE_SIZE,
pfn + i);
if (ret)
break;
}
} else
ret = vm_insert_pfn(area,
(unsigned long)vmf->virtual_address,
pfn + page_offset);
} else {
/* Overriding existing pages in partial view does not cause
* us any trouble as TLBs are still valid because the fault
* is due to userspace losing part of the mapping or never
* having accessed it before (at this partials' range).
*/
const struct i915_ggtt_view *view = &vma->ggtt_view;
unsigned long base = area->vm_start +
(view->params.partial.offset << PAGE_SHIFT);
unsigned int i;
for (i = 0; i < view->params.partial.size; i++) {
ret = vm_insert_pfn(area,
base + i * PAGE_SIZE,
pfn + i);
if (ret)
break;
}
}
ret = remap_io_mapping(area,
area->vm_start + (vma->ggtt_view.params.partial.offset << PAGE_SHIFT),
(ggtt->mappable_base + vma->node.start) >> PAGE_SHIFT,
min_t(u64, vma->size, area->vm_end - area->vm_start),
&ggtt->mappable);
if (ret)
goto err_unpin;
obj->fault_mappable = true;
err_unpin:

View File

@ -0,0 +1,84 @@
/*
* Copyright © 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/mm.h>
#include <linux/io-mapping.h>
#include <asm/pgtable.h>
#include "i915_drv.h"
struct remap_pfn {
struct mm_struct *mm;
unsigned long pfn;
pgprot_t prot;
};
static int remap_pfn(pte_t *pte, pgtable_t token,
unsigned long addr, void *data)
{
struct remap_pfn *r = data;
/* Special PTE are not associated with any struct page */
set_pte_at(r->mm, addr, pte, pte_mkspecial(pfn_pte(r->pfn, r->prot)));
r->pfn++;
return 0;
}
/**
* remap_io_mapping - remap an IO mapping to userspace
* @vma: user vma to map to
* @addr: target user address to start at
* @pfn: physical address of kernel memory
* @size: size of map area
* @iomap: the source io_mapping
*
* Note: this is only safe if the mm semaphore is held when called.
*/
int remap_io_mapping(struct vm_area_struct *vma,
unsigned long addr, unsigned long pfn, unsigned long size,
struct io_mapping *iomap)
{
struct remap_pfn r;
int err;
GEM_BUG_ON((vma->vm_flags &
(VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP)) !=
(VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP));
/* We rely on prevalidation of the io-mapping to skip track_pfn(). */
r.mm = vma->vm_mm;
r.pfn = pfn;
r.prot = __pgprot((pgprot_val(iomap->prot) & _PAGE_CACHE_MASK) |
(pgprot_val(vma->vm_page_prot) & ~_PAGE_CACHE_MASK));
err = apply_to_page_range(r.mm, addr, size, remap_pfn, &r);
if (unlikely(err)) {
zap_vma_ptes(vma, addr, (r.pfn - pfn) << PAGE_SHIFT);
return err;
}
return 0;
}