i915: Avoid dereferencing null fence if resv has changed.

PR kern/57402

XXX pullup-10

branches: 1.3.4;
i915: container_of(const) -> const_container_of

Merge linux-drm-v5-6-rc3

branches: 1.1.1;
Initial revision

drm: Another pass over i915 and some supporting logic.

This makes a shim around sg_table, which essentially represents two
things:

1. an array of pages (roughly corresponding to bus_dma_segment_t[])
2. an array of DMA addresses stored in a bus_dmamap_t

Both parts are optional; different parts of i915 use sg_tables to
pass around one or both of the two parts. This helps to reduce the
ifdefs by quite a bit, although it's not always clear which part of
an sg_table any particular interface is actually using which is why I
was reluctant to do this before.

drm: Nix use of uvm pglist. Just use arrays of page pointers.

i915: machete

Merge linux-drm-v5-6-rc3

branches: 1.1.1;
Initial revision

Merge linux-drm-v5-6-rc3

branches: 1.1.1;
Initial revision

drm/i915: Unhook i915_gem_client_blt.c and revert changes.

Only used by self-tests, which we're not using (for now...).

i915: machete

Merge linux-drm-v5-6-rc3

branches: 1.1.1;
Initial revision

Merge linux-drm-v5-6-rc3

branches: 1.1.1;
Initial revision

i915: Stop pretending the GTT size is zero as workaround.

Whatever bug this was I don't know how to reproduce any more.
Convinces Mesa iris_dri.so to work on this kernel.

i915: For now, pretend gtt size is zero.

Something goes wrong with EXEC_OBJECT_PINNED in Mesa i965 brw bufmgr
if we expose this parameter; this is a sleazy workaround pending
diagnosis of the underlying problem.

i915: mutex_init/destroy and spin_lock_init/destroy audit

drm/i915: Misc build fixes.

First pass at i915, far from complete.

HEAVILY MODIFIED IN MERGE - maya

Merge linux-drm-v5-6-rc3

branches: 1.1.1;
Initial revision

Merge linux-drm-v5-6-rc3

branches: 1.1.1;
Initial revision

Merge linux-drm-v5-6-rc3

branches: 1.1.1;
Initial revision

i915: Fix dmabuf mmap object.

branches: 1.6.4;
drm: Another pass over i915 and some supporting logic.

This makes a shim around sg_table, which essentially represents two
things:

1. an array of pages (roughly corresponding to bus_dma_segment_t[])
2. an array of DMA addresses stored in a bus_dmamap_t

Both parts are optional; different parts of i915 use sg_tables to
pass around one or both of the two parts. This helps to reduce the
ifdefs by quite a bit, although it's not always clear which part of
an sg_table any particular interface is actually using which is why I
was reluctant to do this before.

drm: Nix use of uvm pglist. Just use arrays of page pointers.

drm/i915: Misc build fixes.

First pass at i915, far from complete.

HEAVILY MODIFIED IN MERGE - maya

Merge linux-drm-v5-6-rc3

branches: 1.1.1;
Initial revision

Merge linux-drm-v5-6-rc3

branches: 1.1.1;
Initial revision

i915: Fix error branches of execbuffer vma pinning.

PR kern/56591

drm: Another pass over i915 and some supporting logic.

This makes a shim around sg_table, which essentially represents two
things:

1. an array of pages (roughly corresponding to bus_dma_segment_t[])
2. an array of DMA addresses stored in a bus_dmamap_t

Both parts are optional; different parts of i915 use sg_tables to
pass around one or both of the two parts. This helps to reduce the
ifdefs by quite a bit, although it's not always clear which part of
an sg_table any particular interface is actually using which is why I
was reluctant to do this before.

Hack it up: disable no-fault fast paths, fix fd API, ifdef out stuff.

Merge linux-drm-v5-6-rc3

branches: 1.1.1;
Initial revision

i915: Fix destroy/free order for i915 gem stub fence.

drm/i915: Need a separate spin lock for i915 stub_fence.

Nothing built-in to the dma fence wait queue that we can abuse here.

Merge linux-drm-v5-6-rc3

branches: 1.1.1;
Initial revision

drm: Another pass over i915 and some supporting logic.

This makes a shim around sg_table, which essentially represents two
things:

1. an array of pages (roughly corresponding to bus_dma_segment_t[])
2. an array of DMA addresses stored in a bus_dmamap_t

Both parts are optional; different parts of i915 use sg_tables to
pass around one or both of the two parts. This helps to reduce the
ifdefs by quite a bit, although it's not always clear which part of
an sg_table any particular interface is actually using which is why I
was reluctant to do this before.

Tell i915_gem_internal.c the good news about bus_dma.

Merge linux-drm-v5-6-rc3

branches: 1.1.1;
Initial revision

Merge linux-drm-v5-6-rc3

branches: 1.1.1;
Initial revision

Merge linux-drm-v5-6-rc3

branches: 1.1.1;
Initial revision

Merge linux-drm-v5-6-rc3

branches: 1.1.1;
Initial revision

i915: Fix sense of conditional for i915_gem_mmap_ioctl ABI hack.

Original issue and fix described here:

https://mail-index.netbsd.org/source-changes/2018/08/27/msg098501.html

This was broken in the merge in 2021, and I tried to restore it but
got the sense of the conditional wrong:

https://mail-index.netbsd.org/source-changes/2021/12/19/msg135215.html

branches: 1.27.2;
i915: Match Linux fault routine return code actions.

Omit needless EINTR interception -- this is now handled by
i915_error_to_vmf_fault.

Earlier revert was over a false alarm -- bisection shows the new
warnings arose from linux_pci.c 1.29 here:

https://mail-index.netbsd.org/source-changes/2024/06/23/msg151929.html

i915: Omit needless i915_gem_object_pin/unpin_pages cycle in fault.

vm_fault_cpu and vm_fault_gtt, called by i915_gem_fault, already do
the pinning and unpinning internally, so there is no need for
i915_gem_fault to do it.

No functional change intended, except that the transient pin count
will be one lower than before during the fault routine (but it will
still be positive).

Revert "i915: Reduce diff in fault routine."

Evidently this had more side effects than I thought:

warning: ../../../../external/bsd/drm2/dist/drm/i915/intel_uncore.c:1197: Unclaimed read from register 0x61204
...
{drm:netbsd:wait_panel_status+0x74} *ERROR* PPS state mismatch
...
warning: ../../../../external/bsd/drm2/dist/drm/i915/display/intel_dp.c:1175: eDP powered off while attempting aux channel communication.
...
warning: ../../../../external/bsd/drm2/dist/drm/i915/display/intel_dp.c:5616: Missing case (((&(dev_priv)->__info)->gen) == 9)

So let's roll it back until I have a better understanding of what
changed.

Between making the change and writing the commit message, I forgot
that the change affected the errno values returned by the fault
routine, which may affect how uvm handles the fault failure --
whether to retry, wait for memory and retry, or fail, or what.
Perhaps that explains the issue.

i915: Reduce diff in fault routine.

- Omit needless i915_gem_object_pin/unpin_pages cycle in
i915_gem_fault. This is already done by vm_fault_cpu and
vm_fault_gtt.

- Omit needless EINTR interception. This is now already handled by
i915_error_to_vmf_fault.

No functional change intended, except that the transient pin count
will be one lower than before during the fault routine.

i915_gem_mman.c: Apply mmap types via pmap flags.

This way, userland gets buffers mapped write-combining or uncached as
needed.

PR xsrc/58307

i915_gem: Avoid walking off end of sg_pgs.

sg_npgs currently fails to match obj->base.size/PAGE_SIZE only due to
bugs in the construction of sg_pgs in various i915 gem object types,
which we should also fix, but let's avoid compounding it here.

Related to PR kern/57833.

XXX pullup-10

branches: 1.21.4;
i915: Rearrange how mmap offsets work.

Rather than use the magic offsets in each GEM object's uvm object
itself, just create a separate uvm object for each mapping type and
start at zero within each uvm object. This should avoid problems
with things like ubc_uiomove which don't know to adjust the starting
offset.

i915: Use i915_gem_object_has_pages, not a simple null test.

The pages pointer may be ERR_PTR(-EWHATEVER) instead of null.

i915: Draft intel ggtt partial page view.

i915: Disable partial views on objects for now.

Needs some logic in i915_get_gtt_vma_pages to be wired up.

i915: Release vmobjlock on fault.

We need to take this later, but it's not helpful for uvm_fault to
have taken it immediately prior to calling pgo_fault -- still not
clear on why uvm_fault does that.

i915: Use global uoffset in i915 address space, not per-object.

This is a little sleazy but it enables us to pass the page range
through to the fault routine, which wants it in order to find the
memory mapping type.

i915: Avoid leaking reference count when mapping objects.

We already did i915_gem_object_get_rcu above, which bumps the
reference count; no need to additionally do drm_gem_object_get.

i915: Fix indexing in gem fault lookups.

i915: Simplify releasing mmap offsets; don't release if not mapped.

i915: Fix i915_gem_object_release_mmap_offset locking.

i915: Must unmake vma nodes we have made.

i915: Draft i915_gem_mmap_object.

i915: Draft gem fault.

i915: Define i915_gem_uvm_ops.

drm: Another pass over i915 and some supporting logic.

This makes a shim around sg_table, which essentially represents two
things:

1. an array of pages (roughly corresponding to bus_dma_segment_t[])
2. an array of DMA addresses stored in a bus_dmamap_t

Both parts are optional; different parts of i915 use sg_tables to
pass around one or both of the two parts. This helps to reduce the
ifdefs by quite a bit, although it's not always clear which part of
an sg_table any particular interface is actually using which is why I
was reluctant to do this before.

i915: obj->base.filp, not obj->filp, now.

drm: Nix use of uvm pglist. Just use arrays of page pointers.

i915: machete

Restore mmap ioctl version ABI hack that was lost in merge.

Merge linux-drm-v5-6-rc3

branches: 1.1.1;
Initial revision

i915: Define i915_gem_uvm_ops.

i915: Declare, don't define, object in header file.

drm: Another pass over i915 and some supporting logic.

This makes a shim around sg_table, which essentially represents two
things:

1. an array of pages (roughly corresponding to bus_dma_segment_t[])
2. an array of DMA addresses stored in a bus_dmamap_t

Both parts are optional; different parts of i915 use sg_tables to
pass around one or both of the two parts. This helps to reduce the
ifdefs by quite a bit, although it's not always clear which part of
an sg_table any particular interface is actually using which is why I
was reluctant to do this before.

i915: Custom i915_gem_mmap_object, not fault routine, here.

Provide declarations, we use these from different files.


Author: Maya Rashish <maya@NetBSD.org>
Committer: Taylor R Campbell <riastradh@NetBSD.org>

no struct vm_are_struct, ifdef out declaration


Author: Maya Rashish <maya@NetBSD.org>
Committer: Taylor R Campbell <riastradh@NetBSD.org>

Merge linux-drm-v5-6-rc3

branches: 1.1.1;
Initial revision

i915: Don't need obj->mmo.lock on destruction.

i915: Destroy locks when done.

drm: Destroy struct drm_gem_object::vma_node after use.

i915: Must unmake vma nodes we have made.

i915: More progress.

drm: Another pass over i915 and some supporting logic.

This makes a shim around sg_table, which essentially represents two
things:

1. an array of pages (roughly corresponding to bus_dma_segment_t[])
2. an array of DMA addresses stored in a bus_dmamap_t

Both parts are optional; different parts of i915 use sg_tables to
pass around one or both of the two parts. This helps to reduce the
ifdefs by quite a bit, although it's not always clear which part of
an sg_table any particular interface is actually using which is why I
was reluctant to do this before.

Sync i915_gem_fault. Hack up the rest of i915_gem.c enough to build.

Merge linux-drm-v5-6-rc3

branches: 1.1.1;
Initial revision

drm: Another pass over i915 and some supporting logic.

This makes a shim around sg_table, which essentially represents two
things:

1. an array of pages (roughly corresponding to bus_dma_segment_t[])
2. an array of DMA addresses stored in a bus_dmamap_t

Both parts are optional; different parts of i915 use sg_tables to
pass around one or both of the two parts. This helps to reduce the
ifdefs by quite a bit, although it's not always clear which part of
an sg_table any particular interface is actually using which is why I
was reluctant to do this before.

Switch #ifdef CONFIG_WHATEVER to #if IS_ENABLED(CONFIG_WHATEVER)

Our IS_ENABLED should probably be #if defined(option) && (option),
but that becomes tricky with evaluation order. If option gets
expanded earlier, this errors out (e.g. with CONFIG_AGP)


Author: Maya Rashish <maya@NetBSD.org>
Committer: Taylor R Campbell <riastradh@NetBSD.org>

First pass at i915, far from complete.

HEAVILY MODIFIED IN MERGE - maya

Merge linux-drm-v5-6-rc3

branches: 1.1.1;
Initial revision

Merge linux-drm-v5-6-rc3

branches: 1.1.1;
Initial revision

Merge linux-drm-v5-6-rc3

branches: 1.1.1;
Initial revision

i915: Rearrange how mmap offsets work.

Rather than use the magic offsets in each GEM object's uvm object
itself, just create a separate uvm object for each mapping type and
start at zero within each uvm object. This should avoid problems
with things like ubc_uiomove which don't know to adjust the starting
offset.

drm: Another pass over i915 and some supporting logic.

This makes a shim around sg_table, which essentially represents two
things:

1. an array of pages (roughly corresponding to bus_dma_segment_t[])
2. an array of DMA addresses stored in a bus_dmamap_t

Both parts are optional; different parts of i915 use sg_tables to
pass around one or both of the two parts. This helps to reduce the
ifdefs by quite a bit, although it's not always clear which part of
an sg_table any particular interface is actually using which is why I
was reluctant to do this before.

drm: Nix use of uvm pglist. Just use arrays of page pointers.

drm/i915: Forgot i915_gem_object_types.h change for rb->array.

Tell i915_gem_internal.c the good news about bus_dma.

Merge linux-drm-v5-6-rc3

branches: 1.1.1;
Initial revision

i915_gem: Assert page array size.

Let's detect the bug of sg_npgs failing to match
obj->base.size/PAGE_SIZE earlier.

Related to PR kern/57833.

XXX pullup-10

branches: 1.6.4;
i915: Zero the kernel's ptes before freeing the va.

drm: Another pass over i915 and some supporting logic.

This makes a shim around sg_table, which essentially represents two
things:

1. an array of pages (roughly corresponding to bus_dma_segment_t[])
2. an array of DMA addresses stored in a bus_dmamap_t

Both parts are optional; different parts of i915 use sg_tables to
pass around one or both of the two parts. This helps to reduce the
ifdefs by quite a bit, although it's not always clear which part of
an sg_table any particular interface is actually using which is why I
was reluctant to do this before.

Sync i915_gem_fault. Hack up the rest of i915_gem.c enough to build.

First pass at i915, far from complete.

HEAVILY MODIFIED IN MERGE - maya

Merge linux-drm-v5-6-rc3

branches: 1.1.1;
Initial revision

i915_gem_phys: Fill sg_pgs.

This is needed by i915 gem fault, which maps user virtual addresses
to those pages, and by i915 gem object destruction, which does
pmap_page_protect on the pages to remove any of those user virtual
mappings.

This needs pmap_kenter_pa rather than pmap_enter(pmap_kernel(), ...)
in order to preserve the _kernel's_ mapping of the pages after
pmap_page_protect.

But bus_dmamem_map currently uses pmap_enter(pmap_kernel(), ...)
instead which creates a mapping that is removed by pmap_page_protect.

So we use a variant of bus_dmamem_map that uses pmap_kenter_pa
instead. Perhaps bus_dmamem_map should do this itself, but this
change is less risky to pull up than a change to bus_dmamem_map
itself.

PR kern/57833: kernel panic on xorg exit

XXX pullup-10

branches: 1.8.4;
i915: Use BUS_DMA_COHERENT where Linux uses dma_alloc_coherent.

drm: Destroy dma map in sg_table_free.

Helps to reduce a little diff.

i915: Fix error branches in i915_gem_object_get_pages_phys.

drm: Another pass over i915. Most of i915 gem builds now.

drm: Another pass over i915 and some supporting logic.

This makes a shim around sg_table, which essentially represents two
things:

1. an array of pages (roughly corresponding to bus_dma_segment_t[])
2. an array of DMA addresses stored in a bus_dmamap_t

Both parts are optional; different parts of i915 use sg_tables to
pass around one or both of the two parts. This helps to reduce the
ifdefs by quite a bit, although it's not always clear which part of
an sg_table any particular interface is actually using which is why I
was reluctant to do this before.

Sync i915_gem_fault. Hack up the rest of i915_gem.c enough to build.

Merge linux-drm-v5-6-rc3

branches: 1.1.1;
Initial revision

drm: Another pass over i915 and some supporting logic.

This makes a shim around sg_table, which essentially represents two
things:

1. an array of pages (roughly corresponding to bus_dma_segment_t[])
2. an array of DMA addresses stored in a bus_dmamap_t

Both parts are optional; different parts of i915 use sg_tables to
pass around one or both of the two parts. This helps to reduce the
ifdefs by quite a bit, although it's not always clear which part of
an sg_table any particular interface is actually using which is why I
was reluctant to do this before.

Merge linux-drm-v5-6-rc3

branches: 1.1.1;
Initial revision

Merge linux-drm-v5-6-rc3

branches: 1.1.1;
Initial revision

i915_gem_region: Fill sg_pgs, with size/PAGE_SIZE entries.

Use sg_alloc_table_from_bus_dmamem to do this.

i915_gem_mman.c vm_fault_cpu and i915_gem_object_release_mmap_offset
both rely on sg_pgs to be a page array, so using something else like
size >> ilog2(mem->mm.chunk_size) entries doesn't work. And they
rely on the sg_pgs entries to be initialized, which we weren't doing
before, and which sg_alloc_table_from_bus_dmamem does for us.

Related to PR kern/57833.

XXX pullup-10

i915_gem_region: Reduce diff from upstream a little.

No functional change intended.

Prompted by upcoming nearby changes related to PR kern/57833.

XXX pullup-10 (to make subsequent pullups easier)

branches: 1.4.4;
drm: Destroy dma map in sg_table_free.

Helps to reduce a little diff.

drm: Another pass over i915. Most of i915 gem builds now.

Merge linux-drm-v5-6-rc3

branches: 1.1.1;
Initial revision

Merge linux-drm-v5-6-rc3

branches: 1.1.1;
Initial revision

i915: Restore uao_set_pgfl for i915 gem objects, lost in the merge.

Needed to ensure we allocate paddrs that the GPU can cope with.

i915_gem_shmem.c: shmem_get_pages: fix uninitialize use of "noreclaim"

i915: Mark shmem object pages dirty if needed before unwiring them.

Unlikely to fix the graphics corruption we're seeing, though --
should be relevant only if we're swapping.

i915: Set sg_page_sizes in the #ifdef NetBSD code.

drm: Another pass over i915. Most of i915 gem builds now.

drm: Another pass over i915 and some supporting logic.

This makes a shim around sg_table, which essentially represents two
things:

1. an array of pages (roughly corresponding to bus_dma_segment_t[])
2. an array of DMA addresses stored in a bus_dmamap_t

Both parts are optional; different parts of i915 use sg_tables to
pass around one or both of the two parts. This helps to reduce the
ifdefs by quite a bit, although it's not always clear which part of
an sg_table any particular interface is actually using which is why I
was reluctant to do this before.

drm: Nix use of uvm pglist. Just use arrays of page pointers.

Hack i915_vma.c: rbtree, unmap by pa rather than vm object, PRIx64.

Sync i915_gem_fault. Hack up the rest of i915_gem.c enough to build.

First pass at i915, far from complete.

HEAVILY MODIFIED IN MERGE - maya

Merge linux-drm-v5-6-rc3

branches: 1.1.1;
Initial revision

drm: Another pass over i915. Most of i915 gem builds now.

Merge linux-drm-v5-6-rc3

branches: 1.1.1;
Initial revision

Merge linux-drm-v5-6-rc3

branches: 1.1.1;
Initial revision

i915_gem_stolen: Fill sg_pgs, with size/PAGE_SIZE entries.

Use sg_alloc_table_from_bus_dmamem to do this.

i915_gem_mman.c vm_fault_cpu and i915_gem_object_release_mmap_offset
both rely on sg_pgs to be a page array, so providing a table with
only one entry doesn't work (except by accident, if the object is
page-sized anyway). And they rely on the sg_pgs entries to be
initialized, which we weren't doing before, and which
sg_alloc_table_from_bus_dmamem does for us.

Related to PR kern/57833.

XXX pullup-10

i915_gem_stolen: Fix memory leak.

Found while trying to address the PR 57833 class of problems.

XXX pullup-10

branches: 1.5.4;
drm: Destroy dma map in sg_table_free.

Helps to reduce a little diff.

drm: Another pass over i915 and some supporting logic.

This makes a shim around sg_table, which essentially represents two
things:

1. an array of pages (roughly corresponding to bus_dma_segment_t[])
2. an array of DMA addresses stored in a bus_dmamap_t

Both parts are optional; different parts of i915 use sg_tables to
pass around one or both of the two parts. This helps to reduce the
ifdefs by quite a bit, although it's not always clear which part of
an sg_table any particular interface is actually using which is why I
was reluctant to do this before.

Make i915_gem_stolen.c build again.

Merge linux-drm-v5-6-rc3

branches: 1.1.1;
Initial revision

Merge linux-drm-v5-6-rc3

branches: 1.1.1;
Initial revision

Merge linux-drm-v5-6-rc3

branches: 1.1.1;
Initial revision

drm: Another pass over i915. Most of i915 gem builds now.

Merge linux-drm-v5-6-rc3

branches: 1.1.1;
Initial revision

i915: mutex_init/destroy and spin_lock_init/destroy audit

drm: Another pass over i915. Most of i915 gem builds now.

Destroy some mutexes when done.

Merge linux-drm-v5-6-rc3

branches: 1.1.1;
Initial revision

Sync i915_gem_fault. Hack up the rest of i915_gem.c enough to build.

Merge linux-drm-v5-6-rc3

branches: 1.1.1;
Initial revision

Merge linux-drm-v5-6-rc3

branches: 1.1.1;
Initial revision

Merge linux-drm-v5-6-rc3

branches: 1.1.1;
Initial revision

Merge linux-drm-v5-6-rc3

branches: 1.1.1;
Initial revision

Merge linux-drm-v5-6-rc3

branches: 1.1.1;
Initial revision

Merge linux-drm-v5-6-rc3

branches: 1.1.1;
Initial revision

Merge linux-drm-v5-6-rc3

branches: 1.1.1;
Initial revision

Merge linux-drm-v5-6-rc3

branches: 1.1.1;
Initial revision

Merge linux-drm-v5-6-rc3

branches: 1.1.1;
Initial revision

Merge linux-drm-v5-6-rc3

branches: 1.1.1;
Initial revision

Merge linux-drm-v5-6-rc3

branches: 1.1.1;
Initial revision

Merge linux-drm-v5-6-rc3

branches: 1.1.1;
Initial revision

Merge linux-drm-v5-6-rc3

branches: 1.1.1;
Initial revision

Merge linux-drm-v5-6-rc3

branches: 1.1.1;
Initial revision

Merge linux-drm-v5-6-rc3

branches: 1.1.1;
Initial revision

Merge linux-drm-v5-6-rc3

branches: 1.1.1;
Initial revision

Merge linux-drm-v5-6-rc3

branches: 1.1.1;
Initial revision

Merge linux-drm-v5-6-rc3

branches: 1.1.1;
Initial revision

Merge linux-drm-v5-6-rc3

branches: 1.1.1;
Initial revision

Merge linux-drm-v5-6-rc3

branches: 1.1.1;
Initial revision

Merge linux-drm-v5-6-rc3

branches: 1.1.1;
Initial revision