openkylin
/
linux
mirror of https://gitee.com/openkylin/linux.git
9
0
Fork 0
Code Issues Projects Releases Wiki Activity

drm/i915: make dsm struct resource centric 

Now that we are using struct resource to track the stolen region, it is
more convenient if we track dsm in a resource as well.

v2: check range_overflow when writing to 32b registers (Chris)
    pepper in some comments (Chris)
v3: refit i915_stolen_to_dma()
v4: kill ggtt->stolen_size
v5: some more polish

Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20171211151822.20953-6-matthew.auld@intel.com
This commit is contained in:
Matthew Auld 2017-12-11 15:18:18 +00:00 committed by Joonas Lahtinen
parent f773568b6f
commit 7789422665
8 changed files with 87 additions and 86 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
drivers/char/agp/intel-gtt.c
View File

@ -1424,12 +1424,10 @@ int intel_gmch_probe(struct pci_dev *bridge_pdev, struct pci_dev *gpu_pdev,
EXPORT_SYMBOL(intel_gmch_probe);
void intel_gtt_get(u64 *gtt_total,
u32 *stolen_size,
phys_addr_t *mappable_base,
u64 *mappable_end)
{
*gtt_total = intel_private.gtt_total_entries << PAGE_SHIFT;
*stolen_size = intel_private.stolen_size;
*mappable_base = intel_private.gma_bus_addr;
*mappable_end = intel_private.gtt_mappable_entries << PAGE_SHIFT;
}

12
drivers/gpu/drm/i915/i915_drv.h
View File

@ -1538,9 +1538,6 @@ struct i915_gem_mm {
*/
struct pagevec wc_stash;
/** Usable portion of the GTT for GEM */
dma_addr_t stolen_base; /* limited to low memory (32-bit) */
/**
* tmpfs instance used for shmem backed objects
*/
@ -2254,6 +2251,15 @@ struct drm_i915_private {
const struct intel_device_info info;
/**
* Data Stolen Memory - aka "i915 stolen memory" gives us the start and
* end of stolen which we can optionally use to create GEM objects
* backed by stolen memory. Note that ggtt->stolen_usable_size tells us
* exactly how much of this we are actually allowed to use, given that
* some portion of it is in fact reserved for use by hardware functions.
*/
struct resource dsm;
void __iomem *regs;
struct intel_uncore uncore;

8
drivers/gpu/drm/i915/i915_gem_gtt.c
View File

@ -3299,8 +3299,6 @@ static int gen8_gmch_probe(struct i915_ggtt *ggtt)
pci_read_config_word(pdev, SNB_GMCH_CTRL, &snb_gmch_ctl);
ggtt->stolen_size = resource_size(&intel_graphics_stolen_res);
if (INTEL_GEN(dev_priv) >= 9) {
size = gen8_get_total_gtt_size(snb_gmch_ctl);
} else if (IS_CHERRYVIEW(dev_priv)) {
@ -3363,8 +3361,6 @@ static int gen6_gmch_probe(struct i915_ggtt *ggtt)
DRM_ERROR("Can't set DMA mask/consistent mask (%d)\n", err);
pci_read_config_word(pdev, SNB_GMCH_CTRL, &snb_gmch_ctl);
ggtt->stolen_size = resource_size(&intel_graphics_stolen_res);
size = gen6_get_total_gtt_size(snb_gmch_ctl);
ggtt->base.total = (size / sizeof(gen6_pte_t)) << PAGE_SHIFT;
@ -3410,7 +3406,6 @@ static int i915_gmch_probe(struct i915_ggtt *ggtt)
}
intel_gtt_get(&ggtt->base.total,
&ggtt->stolen_size,
&ggtt->mappable_base,
&ggtt->mappable_end);
@ -3482,7 +3477,8 @@ int i915_ggtt_probe_hw(struct drm_i915_private *dev_priv)
DRM_INFO("Memory usable by graphics device = %lluM\n",
ggtt->base.total >> 20);
DRM_DEBUG_DRIVER("GMADR size = %lldM\n", ggtt->mappable_end >> 20);
DRM_DEBUG_DRIVER("GTT stolen size = %uM\n", ggtt->stolen_size >> 20);
DRM_DEBUG_DRIVER("GTT stolen size = %lluM\n",
(u64)resource_size(&intel_graphics_stolen_res) >> 20);
if (intel_vtd_active())
DRM_INFO("VT-d active for gfx access\n");

1
drivers/gpu/drm/i915/i915_gem_gtt.h
View File

@ -381,7 +381,6 @@ struct i915_ggtt {
* avoid the first page! The upper end of stolen memory is reserved for
* hardware functions and similarly removed from the accessible range.
*/
u32 stolen_size; /* Total size of stolen memory */
u32 stolen_usable_size; /* Total size minus reserved ranges */
u32 stolen_reserved_base;
u32 stolen_reserved_size;

121
drivers/gpu/drm/i915/i915_gem_stolen.c
View File

@ -76,27 +76,26 @@ void i915_gem_stolen_remove_node(struct drm_i915_private *dev_priv,
mutex_unlock(&dev_priv->mm.stolen_lock);
}
static dma_addr_t i915_stolen_to_dma(struct drm_i915_private *dev_priv)
static int i915_adjust_stolen(struct drm_i915_private *dev_priv,
struct resource *dsm)
{
struct i915_ggtt *ggtt = &dev_priv->ggtt;
dma_addr_t base = intel_graphics_stolen_res.start;
struct resource *r;
GEM_BUG_ON(overflows_type(intel_graphics_stolen_res.start, base));
if (dsm->start == 0 || dsm->end <= dsm->start)
return -EINVAL;
if (base == 0 || add_overflows(base, ggtt->stolen_size))
return 0;
/*
* TODO: We have yet too encounter the case where the GTT wasn't at the
* end of stolen. With that assumption we could simplify this.
*/
/* make sure we don't clobber the GTT if it's within stolen memory */
/* Make sure we don't clobber the GTT if it's within stolen memory */
if (INTEL_GEN(dev_priv) <= 4 &&
!IS_G33(dev_priv) && !IS_PINEVIEW(dev_priv) && !IS_G4X(dev_priv)) {
struct {
dma_addr_t start, end;
} stolen[2] = {
{ .start = base, .end = base + ggtt->stolen_size, },
{ .start = base, .end = base + ggtt->stolen_size, },
};
u64 ggtt_start, ggtt_end;
struct resource stolen[2] = {*dsm, *dsm};
struct resource ggtt_res;
u64 ggtt_start;
ggtt_start = I915_READ(PGTBL_CTL);
if (IS_GEN4(dev_priv))
@ -104,70 +103,64 @@ static dma_addr_t i915_stolen_to_dma(struct drm_i915_private *dev_priv)
(ggtt_start & PGTBL_ADDRESS_HI_MASK) << 28;
else
ggtt_start &= PGTBL_ADDRESS_LO_MASK;
ggtt_end = ggtt_start + ggtt_total_entries(ggtt) * 4;
if (ggtt_start >= stolen[0].start && ggtt_start < stolen[0].end)
stolen[0].end = ggtt_start;
if (ggtt_end > stolen[1].start && ggtt_end <= stolen[1].end)
stolen[1].start = ggtt_end;
ggtt_res =
(struct resource) DEFINE_RES_MEM(ggtt_start,
ggtt_total_entries(ggtt) * 4);
/* pick the larger of the two chunks */
if (stolen[0].end - stolen[0].start >
stolen[1].end - stolen[1].start) {
base = stolen[0].start;
ggtt->stolen_size = stolen[0].end - stolen[0].start;
} else {
base = stolen[1].start;
ggtt->stolen_size = stolen[1].end - stolen[1].start;
}
if (ggtt_res.start >= stolen[0].start && ggtt_res.start < stolen[0].end)
stolen[0].end = ggtt_res.start;
if (ggtt_res.end > stolen[1].start && ggtt_res.end <= stolen[1].end)
stolen[1].start = ggtt_res.end;
/* Pick the larger of the two chunks */
if (resource_size(&stolen[0]) > resource_size(&stolen[1]))
*dsm = stolen[0];
else
*dsm = stolen[1];
if (stolen[0].start != stolen[1].start ||
stolen[0].end != stolen[1].end) {
dma_addr_t end = base + ggtt->stolen_size - 1;
DRM_DEBUG_KMS("GTT within stolen memory at 0x%llx-0x%llx\n",
(unsigned long long)ggtt_start,
(unsigned long long)ggtt_end - 1);
DRM_DEBUG_KMS("Stolen memory adjusted to %pad - %pad\n",
&base, &end);
DRM_DEBUG_KMS("GTT within stolen memory at %pR\n", &ggtt_res);
DRM_DEBUG_KMS("Stolen memory adjusted to %pR\n", dsm);
}
}
/* Verify that nothing else uses this physical address. Stolen
/*
* Verify that nothing else uses this physical address. Stolen
* memory should be reserved by the BIOS and hidden from the
* kernel. So if the region is already marked as busy, something
* is seriously wrong.
*/
r = devm_request_mem_region(dev_priv->drm.dev, base, ggtt->stolen_size,
r = devm_request_mem_region(dev_priv->drm.dev, dsm->start,
resource_size(dsm),
"Graphics Stolen Memory");
if (r == NULL) {
/*
* One more attempt but this time requesting region from
* base + 1, as we have seen that this resolves the region
* start + 1, as we have seen that this resolves the region
* conflict with the PCI Bus.
* This is a BIOS w/a: Some BIOS wrap stolen in the root
* PCI bus, but have an off-by-one error. Hence retry the
* reservation starting from 1 instead of 0.
* There's also BIOS with off-by-one on the other end.
*/
r = devm_request_mem_region(dev_priv->drm.dev, base + 1,
ggtt->stolen_size - 2,
r = devm_request_mem_region(dev_priv->drm.dev, dsm->start + 1,
resource_size(dsm) - 2,
"Graphics Stolen Memory");
/*
* GEN3 firmware likes to smash pci bridges into the stolen
* range. Apparently this works.
*/
if (r == NULL && !IS_GEN3(dev_priv)) {
dma_addr_t end = base + ggtt->stolen_size;
DRM_ERROR("conflict detected with stolen region: %pR\n",
dsm);
DRM_ERROR("conflict detected with stolen region: [%pad - %pad]\n",
&base, &end);
base = 0;
return -EBUSY;
}
}
return base;
return 0;
}
void i915_gem_cleanup_stolen(struct drm_device *dev)
@ -183,11 +176,10 @@ void i915_gem_cleanup_stolen(struct drm_device *dev)
static void g4x_get_stolen_reserved(struct drm_i915_private *dev_priv,
dma_addr_t *base, u32 *size)
{
struct i915_ggtt *ggtt = &dev_priv->ggtt;
uint32_t reg_val = I915_READ(IS_GM45(dev_priv) ?
CTG_STOLEN_RESERVED :
ELK_STOLEN_RESERVED);
dma_addr_t stolen_top = dev_priv->mm.stolen_base + ggtt->stolen_size;
dma_addr_t stolen_top = dev_priv->dsm.end + 1;
if ((reg_val & G4X_STOLEN_RESERVED_ENABLE) == 0) {
*base = 0;
@ -308,7 +300,6 @@ static void chv_get_stolen_reserved(struct drm_i915_private *dev_priv,
static void bdw_get_stolen_reserved(struct drm_i915_private *dev_priv,
dma_addr_t *base, u32 *size)
{
struct i915_ggtt *ggtt = &dev_priv->ggtt;
uint32_t reg_val = I915_READ(GEN6_STOLEN_RESERVED);
dma_addr_t stolen_top;
@ -318,7 +309,7 @@ static void bdw_get_stolen_reserved(struct drm_i915_private *dev_priv,
return;
}
stolen_top = dev_priv->mm.stolen_base + ggtt->stolen_size;
stolen_top = dev_priv->dsm.end + 1;
*base = reg_val & GEN6_STOLEN_RESERVED_ADDR_MASK;
@ -351,14 +342,18 @@ int i915_gem_init_stolen(struct drm_i915_private *dev_priv)
return 0;
}
if (ggtt->stolen_size == 0)
if (resource_size(&intel_graphics_stolen_res) == 0)
return 0;
dev_priv->mm.stolen_base = i915_stolen_to_dma(dev_priv);
if (dev_priv->mm.stolen_base == 0)
dev_priv->dsm = intel_graphics_stolen_res;
if (i915_adjust_stolen(dev_priv, &dev_priv->dsm))
return 0;
stolen_top = dev_priv->mm.stolen_base + ggtt->stolen_size;
GEM_BUG_ON(dev_priv->dsm.start == 0);
GEM_BUG_ON(dev_priv->dsm.end <= dev_priv->dsm.start);
stolen_top = dev_priv->dsm.end + 1;
reserved_base = 0;
reserved_size = 0;
@ -399,12 +394,11 @@ int i915_gem_init_stolen(struct drm_i915_private *dev_priv)
reserved_base = stolen_top;
}
if (reserved_base < dev_priv->mm.stolen_base ||
if (reserved_base < dev_priv->dsm.start ||
reserved_base + reserved_size > stolen_top) {
dma_addr_t reserved_top = reserved_base + reserved_size;
DRM_ERROR("Stolen reserved area [%pad - %pad] outside stolen memory [%pad - %pad]\n",
&reserved_base, &reserved_top,
&dev_priv->mm.stolen_base, &stolen_top);
DRM_ERROR("Stolen reserved area [%pad - %pad] outside stolen memory %pR\n",
&reserved_base, &reserved_top, &dev_priv->dsm);
return 0;
}
@ -415,9 +409,9 @@ int i915_gem_init_stolen(struct drm_i915_private *dev_priv)
* memory, so just consider the start. */
reserved_total = stolen_top - reserved_base;
DRM_DEBUG_KMS("Memory reserved for graphics device: %uK, usable: %uK\n",
ggtt->stolen_size >> 10,
(ggtt->stolen_size - reserved_total) >> 10);
DRM_DEBUG_KMS("Memory reserved for graphics device: %lluK, usable: %lluK\n",
(u64)resource_size(&dev_priv->dsm) >> 10,
((u64)resource_size(&dev_priv->dsm) - reserved_total) >> 10);
stolen_usable_start = 0;
/* WaSkipStolenMemoryFirstPage:bdw+ */
@ -425,7 +419,7 @@ int i915_gem_init_stolen(struct drm_i915_private *dev_priv)
stolen_usable_start = 4096;
ggtt->stolen_usable_size =
ggtt->stolen_size - reserved_total - stolen_usable_start;
resource_size(&dev_priv->dsm) - reserved_total - stolen_usable_start;
/* Basic memrange allocator for stolen space. */
drm_mm_init(&dev_priv->mm.stolen, stolen_usable_start,
@ -442,7 +436,8 @@ i915_pages_create_for_stolen(struct drm_device *dev,
struct sg_table *st;
struct scatterlist *sg;
GEM_BUG_ON(range_overflows(offset, size, dev_priv->ggtt.stolen_size));
GEM_BUG_ON(range_overflows_t(resource_size_t, offset, size,
resource_size(&dev_priv->dsm)));
/* We hide that we have no struct page backing our stolen object
* by wrapping the contiguous physical allocation with a fake
@ -462,7 +457,7 @@ i915_pages_create_for_stolen(struct drm_device *dev,
sg->offset = 0;
sg->length = size;
sg_dma_address(sg) = (dma_addr_t)dev_priv->mm.stolen_base + offset;
sg_dma_address(sg) = (dma_addr_t)dev_priv->dsm.start + offset;
sg_dma_len(sg) = size;
return st;

13
drivers/gpu/drm/i915/intel_fbc.c
View File

@ -531,7 +531,6 @@ static int find_compression_threshold(struct drm_i915_private *dev_priv,
int size,
int fb_cpp)
{
struct i915_ggtt *ggtt = &dev_priv->ggtt;
int compression_threshold = 1;
int ret;
u64 end;
@ -541,7 +540,7 @@ static int find_compression_threshold(struct drm_i915_private *dev_priv,
* If we enable FBC using a CFB on that memory range we'll get FIFO
* underruns, even if that range is not reserved by the BIOS. */
if (IS_BROADWELL(dev_priv) || IS_GEN9_BC(dev_priv))
end = ggtt->stolen_size - 8 * 1024 * 1024;
end = resource_size(&dev_priv->dsm) - 8 * 1024 * 1024;
else
end = U64_MAX;
@ -615,10 +614,16 @@ static int intel_fbc_alloc_cfb(struct intel_crtc *crtc)
fbc->compressed_llb = compressed_llb;
GEM_BUG_ON(range_overflows_t(u64, dev_priv->dsm.start,
fbc->compressed_fb.start,
U32_MAX));
GEM_BUG_ON(range_overflows_t(u64, dev_priv->dsm.start,
fbc->compressed_llb->start,
U32_MAX));
I915_WRITE(FBC_CFB_BASE,
dev_priv->mm.stolen_base + fbc->compressed_fb.start);
dev_priv->dsm.start + fbc->compressed_fb.start);
I915_WRITE(FBC_LL_BASE,
dev_priv->mm.stolen_base + compressed_llb->start);
dev_priv->dsm.start + compressed_llb->start);
}
DRM_DEBUG_KMS("reserved %llu bytes of contiguous stolen space for FBC, threshold: %d\n",

15
drivers/gpu/drm/i915/intel_pm.c
View File

@ -7020,7 +7020,7 @@ static void valleyview_check_pctx(struct drm_i915_private *dev_priv)
{
unsigned long pctx_addr = I915_READ(VLV_PCBR) & ~4095;
WARN_ON(pctx_addr != dev_priv->mm.stolen_base +
WARN_ON(pctx_addr != dev_priv->dsm.start +
dev_priv->vlv_pctx->stolen->start);
}
@ -7035,7 +7035,6 @@ static void cherryview_check_pctx(struct drm_i915_private *dev_priv)
static void cherryview_setup_pctx(struct drm_i915_private *dev_priv)
{
struct i915_ggtt *ggtt = &dev_priv->ggtt;
unsigned long pctx_paddr, paddr;
u32 pcbr;
int pctx_size = 32*1024;
@ -7043,8 +7042,8 @@ static void cherryview_setup_pctx(struct drm_i915_private *dev_priv)
pcbr = I915_READ(VLV_PCBR);
if ((pcbr >> VLV_PCBR_ADDR_SHIFT) == 0) {
DRM_DEBUG_DRIVER("BIOS didn't set up PCBR, fixing up\n");
paddr = (dev_priv->mm.stolen_base +
(ggtt->stolen_size - pctx_size));
paddr = dev_priv->dsm.end + 1 - pctx_size;
GEM_BUG_ON(paddr > U32_MAX);
pctx_paddr = (paddr & (~4095));
I915_WRITE(VLV_PCBR, pctx_paddr);
@ -7065,7 +7064,7 @@ static void valleyview_setup_pctx(struct drm_i915_private *dev_priv)
/* BIOS set it up already, grab the pre-alloc'd space */
int pcbr_offset;
pcbr_offset = (pcbr & (~4095)) - dev_priv->mm.stolen_base;
pcbr_offset = (pcbr & (~4095)) - dev_priv->dsm.start;
pctx = i915_gem_object_create_stolen_for_preallocated(dev_priv,
pcbr_offset,
I915_GTT_OFFSET_NONE,
@ -7089,7 +7088,11 @@ static void valleyview_setup_pctx(struct drm_i915_private *dev_priv)
goto out;
}
pctx_paddr = dev_priv->mm.stolen_base + pctx->stolen->start;
GEM_BUG_ON(range_overflows_t(u64,
dev_priv->dsm.start,
pctx->stolen->start,
U32_MAX));
pctx_paddr = dev_priv->dsm.start + pctx->stolen->start;
I915_WRITE(VLV_PCBR, pctx_paddr);
out:

1
include/drm/intel-gtt.h
View File

@ -5,7 +5,6 @@
#define _DRM_INTEL_GTT_H
void intel_gtt_get(u64 *gtt_total,
u32 *stolen_size,
phys_addr_t *mappable_base,
u64 *mappable_end);