The patch before this changed the way in which we allocate space for the
PPGTT PDEs. It began carving out the PPGTT PDEs (which live in the
Global GTT) from the GGTT's drm_mm. Prior to that patch, the PDEs were
hidden from the drm_mm, and therefore could never fail to be allocated.
In unfortunate cases, the drm_mm may be full when we want to allocate
the space. This can technically occur whenever we try to allocate, which
happens in two places currently. Practically, it can only really ever
happen at GPU reset.
Later, when we allocate more PDEs for multiple PPGTTs this will
potentially even more useful.
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
When PPGTT support was originally enabled, it was only designed to
support 1 PPGTT. It therefore made sense to simply hide the GGTT space
required to enable this from the drm_mm allocator.
Since we intend to support full PPGTT, which means more than 1, and they
can be created and destroyed ad hoc it will be required to use the
proper allocation techniques we already have.
The first step here is to make the existing single PPGTT use the
allocator.
The astute observer will notice that we are reserving space in the GGTT
for the PDEs for the lifetime of the address space, and would be right
to question whether or not this is a good idea. It does not make a
difference with this current patch only the aliasing PPGTT (indeed the
PDEs should still be hidden from the shrinker). For the future, we are
allocating from top to bottom to avoid using the precious "gtt
space" The GGTT space at that point should only be used for scanout, HW
contexts, ringbuffers, HWSP, PDEs, and a couple of other small buffers
(potentially) used by the kernel. Everything else should be mapped into
a PPGTT. To put the consumption in more tangible terms, it takes
approximately 4 sets of PDEs to equal one 19x10 framebuffer (with no
fancy stride or alignment constraints). 3/4 of the total [average] GGTT
can be used for PDEs, and hopefully never touch the 1/4 that the
framebuffer needs.
The astute, and persistent observer might ask about the page tables
which are also pinned for the address space. This waste is unfortunate.
We use 2MB of memory per address space. We leave wrapping the PDEs as a
real GEM object as a TODO.
v2: Align PDEs to 64b in GTT
Allocate the node dynamically so we can use drm_mm_put_block
Now tested on IGT
Allocate node at the top to avoid fragmentation (Chris)
v3: Use Chris' top down allocator
v4: Embed drm_mm_node into ppgtt struct (Jesse)
Remove hunks which didn't belong (Jesse)
v5: Don't subtract guard page since we now killed the guard page prior
to this patch. (Ben)
v6: Rebased and removed guard page stuff.
Added a chunk to the commit message
Allow adding a context to mappable region
v7: Undo v3, so we can make the drm patch last in the series
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org> (v4)
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
squash: drm/i915: allow PPGTT to use mappable
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
To sum up what goes on here, we abstract the vma binding, similarly to
the previous object binding. This helps for distinguishing legacy
binding, versus modern binding. To keep the code churn as minimal as
possible, I am leaving in insert_entries(). It serves as the per
platform pte writing basically. bind_vma and insert_entries do share a
lot of similarities, and I did have designs to combine the two, but as
mentioned already... too much churn in an already massive patchset.
What follows are the 3 commits which existed discretely in the original
submissions. Upon rebasing on Broadwell support, it became clear that
separation was not good, and only made for more error prone code. Below
are the 3 commit messages with all their history.
drm/i915: Add bind/unbind object functions to VMA
drm/i915: Use the new vm [un]bind functions
drm/i915: reduce vm->insert_entries() usage
drm/i915: Add bind/unbind object functions to VMA
As we plumb the code with more VM information, it has become more
obvious that the easiest way to deal with bind and unbind is to simply
put the function pointers in the vm, and let those choose the correct
way to handle the page table updates. This change allows many places in
the code to simply be vm->bind, and not have to worry about
distinguishing PPGTT vs GGTT.
Notice that this patch has no impact on functionality. I've decided to
save the actual change until the next patch because I think it's easier
to review that way. I'm happy to squash the two, or let Daniel do it on
merge.
v2:
Make ggtt handle the quirky aliasing ppgtt
Add flags to bind object to support above
Don't ever call bind/unbind directly for PPGTT until we have real, full
PPGTT (use NULLs to assert this)
Make sure we rebind the ggtt if there already is a ggtt binding. This
happens on set cache levels.
Use VMA for bind/unbind (Daniel, Ben)
v3: Reorganize ggtt_vma_bind to be more concise and easier to read
(Ville). Change logic in unbind to only unbind ggtt when there is a
global mapping, and to remove a redundant check if the aliasing ppgtt
exists.
v4: Make the bind function a bit smarter about the cache levels to avoid
unnecessary multiple remaps. "I accept it is a wart, I think unifying
the pin_vma / bind_vma could be unified later" (Chris)
Removed the git notes, and put version info here. (Daniel)
v5: Update the comment to not suck (Chris)
v6:
Move bind/unbind to the VMA. It makes more sense in the VMA structure
(always has, but I was previously lazy). With this change, it will allow
us to keep a distinct insert_entries.
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
drm/i915: Use the new vm [un]bind functions
Building on the last patch which created the new function pointers in
the VM for bind/unbind, here we actually put those new function pointers
to use.
Split out as a separate patch to aid in review. I'm fine with squashing
into the previous patch if people request it.
v2: Updated to address the smart ggtt which can do aliasing as needed
Make sure we bind to global gtt when mappable and fenceable. I thought
we could get away without this initialy, but we cannot.
v3: Make the global GTT binding explicitly use the ggtt VM for
bind_vma(). While at it, use the new ggtt_vma helper (Chris)
At this point the original mailing list thread diverges. ie.
v4^:
use target_obj instead of obj for gen6 relocate_entry
vma->bind_vma() can be called safely during pin. So simply do that
instead of the complicated conditionals.
Don't restore PPGTT bound objects on resume path
Bug fix in resume path for globally bound Bos
Properly handle secure dispatch
Rebased on vma bind/unbind conversion
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
drm/i915: reduce vm->insert_entries() usage
FKA: drm/i915: eliminate vm->insert_entries()
With bind/unbind function pointers in place, we no longer need
insert_entries. We could, and want, to remove clear_range, however it's
not totally easy at this point. Since it's used in a couple of place
still that don't only deal in objects: setup, ppgtt init, and restore
gtt mappings.
v2: Don't actually remove insert_entries, just limit its usage. It will
be useful when we introduce gen8. It will always be called from the vma
bind/unbind.
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> (v1)
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
The initial implementation of this function used MMIO to write the PDPs.
Upon review it was determined (correctly) that the docs say to use LRI.
The issue is there are times where we want to do a synchronous write
(GPU reset).
I've tested this, and it works. I've verified with as many people as
possible that it should work.
This should fix the failing reset problems.
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Our VM code already has a cleanup function, and this is a nice place to
put the drm_mm_takedown. This should have no functional impact, it just
leaves the unload function a bit cleaer, and is more logical IMO
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
This should really have been added in BDW integration, as well as:
commit 93bd8649db
Author: Ben Widawsky <ben@bwidawsk.net>
Date: Tue Jul 16 16:50:06 2013 -0700
drm/i915: Put the mm in the parent address space
It didn't really matter before, but it will in the future.
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
When we fail for some reason on loading the PDPs, it would be wise to
disable the PPGTT in the ring registers. If we do not do this, we have
undefined results.
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Pull in Jani's backlight rework branch. This was merged through a
separate branch to be able to sort out the Broadwell conflicts
properly before pulling it into the main development branch.
Conflicts:
drivers/gpu/drm/i915/intel_display.c
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Because of the way in which we're allocating the pages for the Aliasing
PPGTT, we cannot actually successfully alloc enough space for anything
greater than 2GB.
Instead of a quick hack to fix this, we should defer until we have the
real solution in place (allocating much less contiguous space).
This wasn't found sooner because we didn't not have any systems
supporting more than a 2GB GTT.
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
I am unclear how this got messed up in the shuffle, but it did.
Cc: Imre Deak <imre.deak@intel.com>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Preallocated objects will already have been added to the vma_list when
creating their ggtt vma entry, and coincidentally also marked as holding
a ggtt mapping. Repeating the vma_list manipulation when setting up the
ggtt after preallocation is a recipe for an unhappy kernel.
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
[danvet: Use the improve commit message suggest by Chris.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
v2: Resolve rebase conflicts and switch to gen < 8 color for GenX
checking.
v3: Rebase on top of the address space refactoring.
Reviewed-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com> (v1)
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
v2: Squash in fix from Ben: Set PPGTT batches as necessary
This fixes the regression in the last couple of days when we enabled
PPGTT.
v3: Squash in fixup to still use GTT for secure batches from Ville:
BDW doesn't have a separate secure vs. non-secure bit in
MI_BATCH_BUFFER_START. So for secure batches we have to simply
leave the PPGTT bit unset. Fortunately older generations (except
HSW) had similar limitations so execbuffer already creates a GTT
mapping for all secure batches.
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Legacy PPGTT on GEN8 requires programming 4 PDP registers per ring.
Since all rings are using the same address space with the current code
the logic is simply to program all the tables we've setup for the PPGTT.
v2: Turn on PPGTT in GFX_MODE
v3: v2 was the wrong patch
v4: Resolve conflicts due to patch series reordering.
v5: Squash in fixup from Ben: Use LRI to write PDPs
The docs (and simulator seems to back up) suggest that we can only
program legacy PPGTT PDPs with LRI commands.
v6: Rebase around context differences conflicts.
v7: Use #defines for per ring PDPs. (Damien)
v8: Don't use typede'f private_t.
Signed-off-by: Ben Widawsky <ben@bwidawsk.net> (up to v3 and v7)
Reviewed-by: Imre Deak <imre.deak@intel.com>
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
GEN8 insertion is very similar to GEN6.
v2: Rebase on top of Imre's for_each_sg_page helpers.
v3: Fixup my conversion (spotted by Ville).
v4: Rebase on top of the address space refactoring.
Signed-off-by: Ben Widawsky <ben@bwidawsk.net> (v1)
Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
GEN8 PPGTT range clearing is very similar to GEN6 if we assume that our
PDEs are all valid, which they should be.
v2: Rebase on top of the address space refactoring.
v3: Rebase on top of the bool use_scratch addition to the clear_range interface.
Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net> (v1)
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
The upcoming clear and insert routines will expect that PDEs all point
to valid Page Directories. Doing that lazily doesn't really buy us
anything.
The page allocation is done regardless earlier in init so it shouldn't
hurt set the PDEs.
v2: Squash in patches to implement fixed PDE write function:
- If I had done this in the first place, the bug that's going to be
fixed in an upcoming patch would have been much easier to find.
- Use WB for PDEs.
The PAT bit is used for page size. 2ME PDEs aren't even supported in
BDW, so this was completely invalid. The solution is to make our
PDEs WB+LLC instead of the pervious WB+eLLC. As far as I can guess,
this change won't matter for performance.
Thanks to Ville for the quick correction when discussing on IRC.
v3: Return the pde type for pde encoding (Damien)
Reviewed-by: Imre Deak <imre.deak@intel.com>
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Aside from the potential size increase of the PPGTT, the primary
difference from previous hardware is the Page Directories are no longer
carved out of the Global GTT.
Note that the PDE allocation is done as a 8MB contiguous allocation,
this needs to be eventually fixed (since driver reloading will be a
pain otherwise). Also, this will be a no-go for real PPGTT support.
v2: Move vtable initialization
v3: Resolve conflicts due to patch series reordering.
v4: Rebase on top of the address space refactoring of the PPGTT
support. Drop Imre's r-b tag for v2, too outdated by now.
v5: Free the correct amount of memory, "get_order takes size not a page
count." (Imre)
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
BDW caching works differently than the previous generations. Instead of
having bits in the PTE which directly control how the page is cached,
the 3 PTE bits PWT PCD and PAT provide an index into a PAT defined by
register 0x40e0. This style of caching is functionally equivalent to how
it works on HSW and before.
v2: Tiny bikeshed as discussed on internal irc.
v3: Squash in patch from Ville to mirror the x86 PAT setup more like
in arch/x86/mm/pat.c. Primarily, the 0th index will be WB, and not
uncached.
v4: Comment for reason to not use a 64b write on the PPAT.
v5: Add a FIXME comment that the caching bits in the PAT registers
might be wrong due to doc confusion.
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net> (v1)
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
With the PTE clarifications, the bind and clear functions can now be
added for gen8.
v2: Use for_each_sg_pages in gen8_ggtt_insert_entries.
v3: Drop dev argument to pte encode functions, upstream lost it. Also
rebase on top of the scratch page movement.
v4: Rebase on top of the new address space vfuncs.
v5: Add the bool use_scratch argument to clear_range and the bool valid argument
to the PTE encode function to follow upstream changes.
v6: Add a FIXME(BDW) about the size mismatch of the readback check
that Jon Bloomfield spotted.
v7: Squash in fixup patch from Ben for the posting read to match the
64bit ptes and so shut up the WARN.
Signed-off-by: Ben Widawsky <ben@bwidawsk.net> (v1)
Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
With gen6 PTE type in place, pave the way for the new gen8 type.
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Probing gen8 is similar to gen6. To make the code cleaner and more
maintainable however we can use the probe functions to split it out.
v2: Rebased on top of update gtt probe infrastructure.
v3: Rebased on top of Kenneth' Graunke's ->pte_encode refactoring.
V4: Resolve conflicts with Ben's latest ppgtt patches, also switch to
gen < 8 testing instead of gen <= 7.
v5: Resolve conflicts with address space vfunc changes in upstream.
v6: Use 39b DMA mask. At least, for this mode, it is the correct mask.
(Imre)
Cc: Imre Deak <imre.deak@intel.com>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
All the BARs have the ability to grow.
v2: Pulled out the simulator workaround to a separate patch.
Rebased.
v3: Rebase onto latest vlv patches from Jesse.
v4: Rebased on top of the early stolen quirk patch from Jesse.
v5: Use the new macro names.
s/INTEL_BDW_PCI_IDS_D/INTEL_BDW_D_IDS
s/INTEL_BDW_PCI_IDS_M/INTEL_BDW_M_IDS
It's Jesse's fault for not following the convention I originally set.
Cc: Ingo Molnar <mingo@kernel.org>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
This will be changed once the gen8 code is fully implemented.
v2: Use ENOSYS instead of ENXIO as suggested by Chris.
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
I want to merge in the new Broadwell support as a late hw enabling
pull request. But since the internal branch was based upon our
drm-intel-nightly integration branch I need to resolve all the
oustanding conflicts in drm/i915 with a backmerge to make the 60+
patches apply properly.
We'll propably have some fun because Linus will come up with a
slightly different merge solution.
Conflicts:
drivers/gpu/drm/i915/i915_dma.c
drivers/gpu/drm/i915/i915_drv.c
drivers/gpu/drm/i915/intel_crt.c
drivers/gpu/drm/i915/intel_ddi.c
drivers/gpu/drm/i915/intel_display.c
drivers/gpu/drm/i915/intel_dp.c
drivers/gpu/drm/i915/intel_drv.h
All rather simple adjacent lines changed or partial backports from
-next to -fixes, with the exception of the thaw code in i915_dma.c.
That one needed a bit of shuffling to restore the intent.
Oh and the massive header file reordering in intel_drv.h is a bit
trouble. But not much.
v2: Also don't forget the fixup for the silent conflict that results
in compile fail ...
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Once the machine gets to a certain point in the suspend process, we
expect the GPU to be idle. If it is not, we might corrupt memory.
Empirically (with an early version of this patch) we have seen this is
not the case. We cannot currently explain why the latent GPU writes
occur.
In the technical sense, this patch is a workaround in that we have an
issue we can't explain, and the patch indirectly solves the issue.
However, it's really better than a workaround because we understand why
it works, and it really should be a safe thing to do in all cases.
The noticeable effect other than the debug messages would be an increase
in the suspend time. I have not measure how expensive it actually is.
I think it would be good to spend further time to root cause why we're
seeing these latent writes, but it shouldn't preclude preventing the
fallout.
NOTE: It should be safe (and makes some sense IMO) to also keep the
VALID bit unset on resume when we clear_range(). I've opted not to do
this as properly clearing those bits at some later point would be extra
work.
v2: Fix bugzilla link
Bugzilla: http://bugs.freedesktop.org/show_bug.cgi?id=65496
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=59321
Tested-by: Takashi Iwai <tiwai@suse.de>
Tested-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Tested-By: Todd Previte <tprevite@gmail.com>
Cc: stable@vger.kernel.org
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
We need this to work around a corruption when the boot kernel image
loads the hibernated kernel image from swap on Haswell systems -
somehow not everything is properly shut off.
This is just the prep work, the next patch will implement the actual
workaround.
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
[danvet: Add a commit message suitable for -fixes and add cc: stable]
Cc: stable@vger.kernel.org
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
No buffer overflows here, but better safe than sorry.
v2:
- Fixup the sizeof conversion, I've missed the pointer deref (Jani).
- Drop the redundant GFP_ZERO, kcalloc alreads memsets (Jani).
- Use kmalloc_array for the execbuf fastpath to avoid the memset
(Chris). I've opted to leave all other conversions as-is since they
aren't in a fastpath and dealing with cleared memory instead of
random garbage is just generally nicer.
Cc: Jani Nikula <jani.nikula@linux.intel.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
[danvet: Drop the contentious kmalloc_array hunk in execbuf.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Haswell GT3e has the unique feature of supporting Write-Through cacheing
of objects within the eLLC/LLC. The purpose of this is to enable the display
plane to remain coherent whilst objects lie resident in the eLLC/LLC - so
that we, in theory, get the best of both worlds, perfect display and fast
access.
However, we still need to be careful as the CPU does not see the WT when
accessing the cache. In particular, this means that we need to flush the
cache lines after writing to an object through the CPU, and on
transitioning from a cached state to WT.
v2: Actually do the clflush on transition to WT, nagging by Ville.
v3: Flush the CPU cache after writes into WT objects.
v4: Rease onto LLC updates and report WT as "uncached" for
get_cache_level_ioctl to remain symmetric with set_cache_level_ioctl.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
As mentioned in the previous commit, reads and writes from both the CPU
and GPU go through the LLC. This gives us coherency between the CPU and
GPU irrespective of the attribute settings either device sets. We can
use to avoid having to clflush even uncached memory.
Except for the scanout.
The scanout resides within another functional block that does not use
the LLC but reads directly from main memory. So in order to maintain
coherency with the scanout, writes to uncached memory must be flushed.
In order to optimize writes elsewhere, we start tracking whether an
framebuffer is attached to an object.
v2: Use pin_display tracking rather than fb_count (to ensure we flush
cursors as well etc) and only force the clflush along explicit writes to
the scanout paths (i.e. pin_to_display_plane and pwrite into scanout).
v3: Force the flush after hitting the slowpath in pwrite, as after
dropping the lock the object's cache domain may be invalidated. (Ville)
Based on a patch by Ville Syrjälä.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
MLC_LLC was never validated for Sandybridge and was superseded by a new
level of cacheing for the GPU in Ivybridge. Update our names to be
consistent with usage, and in the process stop setting the unwanted bit
on Sandybridge.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
[danvet: s/BUG/WARN_ON(1) bikeshed.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Just some small cleanups, and a rename of vm->ggtt_vm requested by
Daniel.
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Earlier in the conversion sequence we attempted to quickly wedge in the
transitional interface as static inlines.
Now that we're sure these interfaces are sane, for easier debug and to
decrease code size (since many of these functions may be called quite a
bit), make them real functions
While at it, kill off the set_color interface. We'll always have the
VMA, or easily get to it.
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
The default LLC age was changed:
commit 0d8ff15e9a
Author: Ben Widawsky <benjamin.widawsky@intel.com>
Date: Thu Jul 4 11:02:03 2013 -0700
drm/i915/hsw: Set correct Haswell PTE encodings.
On the surface it would seem setting a default age wouldn't matter
because all GEM BOs are aged similarly, so the order in which objects
are evicted would not be subject to aging. The current working theory as
to why this caused a regression though is that LLC is a bit special in
that it is shared with the CPU. Presumably (not verified) the CPU
fetches cachelines with age 3, and therefore recently cached GPU objects
would be evicted before similar CPU object first when the LLC is full.
It stands to reason therefore that this would negatively impact CPU
bound benchmarks - but those seem to be low on the priority list.
eLLC OTOH does not have this same property as LLC. It should be used
entirely for the GPU, and so the age really shouldn't matter.
Furthermore, we have no evidence to suggest one is better than another
on eLLC. Since we've never properly supported eLLC before no, there
should be no regression. If the GPU client really wants "younger"
objects, they should use MOCS.
v2: Drop the extra #define (Chad)
v3: Actually git add
v4: Pimped commit message
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=67062
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Chad Versace <chad.versace@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
The PTE layouts are the same for both ppgtt and gtt, so we can simplify
the setup for ppgtt by copying the encoding function pointer from gtt.
This prevents bugs where we update one function pointer, but forget the
other.
For instance,
commit 4d15c145a6
Author: Ben Widawsky <ben@bwidawsk.net>
Date: Thu Jul 4 11:02:06 2013 -0700
drm/i915: Use eLLC/LLC by default when available
only extends the gtt to use eLLC/LLC cacheing and forgets to also update
the ppgtt function pointer.
v2: Actually mention the bug being fixed (Kenneth)
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Formerly: "drm/i915: Create VMAs (part 1)"
In a previous patch, the notion of a VM was introduced. A VMA describes
an area of part of the VM address space. A VMA is similar to the concept
in the linux mm. However, instead of representing regular memory, a VMA
is backed by a GEM BO. There may be many VMAs for a given object, one
for each VM the object is to be used in. This may occur through flink,
dma-buf, or a number of other transient states.
Currently the code depends on only 1 VMA per object, for the global GTT
(and aliasing PPGTT). The following patches will address this and make
the rest of the infrastructure more suited
v2: s/i915_obj/i915_gem_obj (Chris)
v3: Only move an object to the now global unbound list if there are no
more VMAs for the object which are bound into a VM (ie. the list is
empty).
v4: killed obj->gtt_space
some reworks due to rebase
v5: Free vma on error path (Imre)
v6: Another missed vma free in i915_gem_object_bind_to_gtt error path
(Imre)
Fixed vma freeing in stolen preallocation (Imre)
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Imre Deak <imre.deak@intel.com>
[danvet: Squash in fixup from Ben to not deref a non-existing vma in
set_cache_level, reported by Chris.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Every address space should support object allocation. It therefore makes
sense to have the allocator be part of the "superclass" which GGTT and
PPGTT will derive.
Since our maximum address space size is only 2GB we're not yet able to
avoid doing allocation/eviction; but we'd hope one day this becomes
almost irrelvant.
v2: Rebased
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
The GTT and PPGTT can be thought of more generally as GPU address
spaces. Many of their actions (insert entries), state (LRU lists), and
many of their characteristics (size) can be shared. Do that.
The change itself doesn't actually impact most of the VMA/VM rework
coming up, it just fits in with the grand scheme of abstracting the GPU
VM operations. GGTT will usually be a special case where we either know
an object must be in the GGTT (dislay engine, workarounds, etc.).
The scratch page is left as part of the VM (even though it's currently
shared with the ppgtt code) because in the future when we have Full
PPGTT, I intend to create a separate scratch page for each.
v2: Drop usage of i915_gtt_vm (Daniel)
Make cleanup also part of the parent class (Ben)
Modified commit msg
Rebased
v3: Properly share scratch page (Imre)
Finish commit message (Daniel, Imre)
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
DRI clients really should be using MOCS to get fine grained streaming
cache controls. With that note, I *hope* that this patch doesn't improve
performance overwhelmingly, because if it does - it means there is a
problem elsewhere.
In any case, the kernel, and old userspace should get some benefit from
this, so let's do it. eLLC is always a good default, and really not
using it is the special case for MOCS.
References: http://www.intel.com/newsroom/kits/restricted/ha$well!/pdfs/4th_Gen_Intel_Core_PressBriefing_5-29.pdf (page 57)
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
The cacheability controls have changed, and the bits have been
rearranged in general.
Note that age 0 is the oldest (most likely to get evicted) and age 3
is the youngest (most likely to stick around for a bit). We've picked
0 for no reason, but atm it shouldn't matter anyway (since we don't
yet try to differentiate between different objects).
v2: Remove comments for snb/ivb cache leves, that's a separate change.
v3: Resolve conflicts due to patch series reordering.
v4: Rebased on top of Kenneth Graunke's ->pte_encode refactoring.
v5: Removed eLLC bits for separate patch.
In the internal repository this was:
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
[danvet: Add comment about cache ages as requested by Ben provoked due
to a question from Damien.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Embedding the node in the obj is more natural in the transition to VMAs
which will also have embedded nodes. This change also helps transition
away from put_block to remove node.
Though it's quite an uncommon occurrence, it's somewhat convenient to not
fail at bind time because we cannot allocate the node. Though in
practice there are other allocations (like the request structure) which
would probably make this point not terribly useful.
Quoting Daniel:
Note that the only difference between put_block and remove_node is
that the former fills up the preallocation cache. Which we don't need
anyway and hence is just wasted space.
v2: Clean up the stolen preallocation code.
Rebased on the reserve_node patches
renames ggtt_ stuff to gtt_ stuff
WARN_ON if the object is already bound (which doesn't mean it's in the
bound list, tricky)
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
With the getters in place from the previous patch this members serves no
purpose other than saving one spare pointer chase, which will be killed
in the next patch anyway.
Moving to VMAs, this members adds unnecessary confusion since an object
may exist at different offsets in different VMs.
v2: Properly preserve the stolen offset. This code is a bit hacky but it
all goes away when we embed the drm_mm_node and removes the need for the
incorrect patch I submitted previously: "Use gtt_space->start for stolen
reservation"
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Soon we want to gut a lot of our existing assumptions how many address
spaces an object can live in, and in doing so, embed the drm_mm_node in
the object (and later the VMA).
It's possible in the future we'll want to add more getter/setter
methods, but for now this is enough to enable the VMAs.
v2: Reworked commit message (Ben)
Added comments to the main functions (Ben)
sed -i "s/i915_gem_obj_set_color/i915_gem_obj_ggtt_set_color/" drivers/gpu/drm/i915/*.[ch]
sed -i "s/i915_gem_obj_bound/i915_gem_obj_ggtt_bound/" drivers/gpu/drm/i915/*.[ch]
sed -i "s/i915_gem_obj_size/i915_gem_obj_ggtt_size/" drivers/gpu/drm/i915/*.[ch]
sed -i "s/i915_gem_obj_offset/i915_gem_obj_ggtt_offset/" drivers/gpu/drm/i915/*.[ch]
(Daniel)
v3: Rebased on new reserve_node patch
Changed DRM_DEBUG_KMS to actually work (will need fixing later)
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
With the previous patch we no longer actually create a node, we simply
find the correct hole and occupy it. This very well could have been
squashed with the last patch, but since I already had David's review, I
figured it's easiest to keep it distinct.
Also update the users in i915. Conveniently this is the only user of the
interface.
CC: David Airlie <airlied@linux.ie>
CC: <dri-devel@lists.freedesktop.org>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Acked-by: David Airlie <airlied@linux.ie>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
For an upcoming patch where we introduce the i915 VMA, it's ideal to
have the drm_mm_node as part of the VMA struct (ie. it's pre-allocated).
Part of the conversion to VMAs is to kill off obj->gtt_space. Doing this
will break a bunch of code, but amongst them are 2 callers of
drm_mm_create_block(), both related to stolen memory.
It also allows us to embed the drm_mm_node into the object currently
which provides a nice transition over to the new code.
v2: Reordered to do before ripping out obj->gtt_offset.
Some minor cleanups made available because of reordering.
v3: s/continue/break on failed stolen node allocation (David)
Set obj->gtt_space on failed node allocation (David)
Only unref stolen (fix double free) on failed create_stolen (David)
Free node, and NULL it in failed create_stolen (David)
Add back accidentally removed newline (David)
CC: <dri-devel@lists.freedesktop.org>
Reviewed-by: David Herrmann <dh.herrmann@gmail.com>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Acked-by: David Airlie <airlied@linux.ie>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
The original pte_encode function needed the dev argument so we could do
platform specific handling via IS_GENX, etc. With the merging of a pte
encoding function there should never been a need to quirk away gen
specific details.
The patch doesn't do much but makes the upcoming reworks in gtt/ppgtt/mm
slightly (albeit, ever so) easier.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
There isn't any special reason to do this other than it makes it obvious
that the two members are connected.
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Ben Widawsky