adapt to ${CC_WNO_IMPLICIT_FALLTHROUGH}

Use ${CC_WNO_IMPLICIT_FALLTHROUGH} instead of
the older style more complex expressions.

drm: Propagate CPPFLAGS.drmkms into all the drm drivers.

drm: Modularize ttm.

Revert part of previous: no separate CPPFLAGS.drmkms.

There are too many drms users outside sys/external/bsd/drm2 for now;
will hold off on this until we find a better way to keep the drm
includes modularized.

Narrow the CPPFLAGS for drm subsystems.

No need for the whole kernel to be built with -I flags for the drm
header files.

Fix up ttm agp.

Build ttm with -Wno-missing-field-initializers, -Wno-shadow.

convert HAVE_GCC == 7 to HAVE_GCC >= 7.

add -Wno-error=int-in-bool-context or -Wno-error=implicit-fallthrough
for various drm files that would be best ignored or upstream fixed.
(could fix savagedrm, no one upstream is that any more.)

branches: 1.1.4; 1.1.20; 1.1.24;
Split drm ttm config into a separate file.

Fix up ttm agp.

Rework PCI and AGP conditionalization.

- Push drm_pci_set_unique into driver callback.
- Eliminate drm_pci_set_unique_hook.
- Gather all drm_agp_* functions into struct drm_agp_hooks.
- Replace the nonsensical old atomic garbage by serious locking.
- Make drm_agpsupport.c its own module.
- Eliminate NDRMKMS_PCI.
- Use NAGP from "agp.h" only in drm_module.c for horrible hack.
=> See comment in file for rationale.
- Always define CONFIG_PCI=1 and CONFIG_AGP=1.
- Always go through the drm_agp_* function hooks.
- Ifdef out nouveau agp stuff that doesn't go through drm_agp_*
for reasons that I'm too frustrated to figure out tonight.
- pci_iomap no longer automagically does agp_i810_borrow.
=> Use drm_agp_borrow instead.

Upstream nixed __OS_HAS_AGP; use CONFIG_AGP instead.

branches: 1.6.12; 1.6.14;
Clarify ttm state transitions tt_unpopulated<->tt_unbound<->tt_bound.
Assert it, too, and don't handle other cases.

We can add the assertion to ttm_agp_tt_unpopulate because it is called by
{nouveau,radeon}_ttm_tt_unpopulate, which is generically called
ttm_tt_unpopulate.

And the sole caller to ttm_tt_unpopulate (ttm_tt_destroy) only does so if the
state is unbound. the other caller is in a !NetBSD block.

We can add the assertion to ttm_agp_tt_populate and avoid handling the
!unpopulated case because the sole callers are {nouveau,radeon}_ttm_tt_populate
both of which return early in the !unpopulated case.

We can change the assertion on ttm_tt_wire because it is solely called by
ttm_bus_dma_populate, which already asserts that it is the unpopulated case.

from riastradh

branches: 1.5.2; 1.5.4;
only build ttm_agp code if __OS_HAS_AGP is defined

kmem_free the container structure, not a substructure.

branches: 1.3.2; 1.3.4;
Zero ttm_agp objects on creation.

Use ttm_dma_tt_init in ttm_agp_tt_create so we can use ttm_bus_dma.

branches: 1.1.2; 1.1.4;
First whack at radeon driver.

No hardware to test yet, but it builds.

ttm: Add XXX about readahead fault failures.

ttm: Sync cacheability flag logic with Linux.

ttm: Respect PGO_ALLPAGES.

Not sure this is useful but it reduces XXX's and makes this match
udv_fault better so it's easier to understand.

ttm: Sync more with Linux.

Add the original copyright and attribution since this is now,
intentionally, a modified copy of the original and not just roughly
the same algorithm.

ttm: Undo mistake in previous.

PR xsrc/58133

ttm: Sync ttm_bo_uvm_fault_idle better with Linux.

PR xsrc/58133

branches: 1.22.4;
drm/ttm: Cast to potentially larger types before << PAGE_SHIFT.

Not sure if this is a problem but it might affect, e.g., i386pae.
Can't hurt, in any case, to cast to what the type will be after the
shift anyway.

drm/ttm: Handle io_mem_pfn, used by amdgpu.

drm/ttm: More careful diagnostics earlier on.

drm/ttm: Compare bytes to bytes, not pages to bytes.

While here, KASSERT -> KASSERTMSG for better diagnostics.

drm/ttm: ttm_bo_put(bo) won't bo=NULL like ttm_bo_unref(&bo) did.

Don't call TTM functions that were removed, try replacements


Author: Maya Rashish <maya@NetBSD.org>

More obvious part of TTM API changes:

- ref/unref to get/put
- some global TTM things are now global
- More indirection to get vma_node


Author: Maya Rashish <maya@NetBSD.org>

Sync ttm_bo_vm.c.

UVM locking changes, proposed on tech-kern:

- Change the lock on uvm_object, vm_amap and vm_anon to be a RW lock.
- Break v_interlock and vmobjlock apart. v_interlock remains a mutex.
- Do partial PV list locking in the x86 pmap. Others to follow later.

Do not extract mmap flags from the bus_space_mmap cookie as they may
conflict with the bo's existing protection flags. This caused VRAM to
be incorrectly mapped as Device-nGnRE on Arm64 instead of Normal-NC. Ok
riastradh@.

branches: 1.12.6;
ttm_bo_uvm_fault() is a netbsd API function and thus should return
netbsd errnos. fix the one "return -ERESTART".

struct ttm_bo_device::fence_lock is no more.

branches: 1.10.10; 1.10.16; 1.10.18;
ttm_bo_unreserve in case of ttm_bo_uvm_fault_idle restart.

Better chance of fixing PR kern/49862 by avoiding leaking a buffer
ww_mutex lock.

Fix error branch: ttm_bo_unreserve on failure after ttm_bo_reserve.

Tiny chance this will fix PR kern/49862 by not leaking a ww_mutex
lock.

branches: 1.8.2;
Update comment to reflect recent change to.

Oops -- fix call to ttm_bo_uvm_fault_idle in last commit.

Simplify uvmfault_unlockall use in ttm_bo_uvm_fault.

Pass NULL for uobj instead of locking uobj->vmobjlock before calling
uvmfault_unlockall.

Should fix lockdebug unlocking-unlocked-mutex crash in
ttm_bo_uvm_fault_idle.

XXX pullup to netbsd-7

We dropped vmobjlock, so reacquire it before uvmfault_unlockall.

Drop take the {ttm,gem} vmobjlock in the fault handler.

- We don't need this lock.
- uvm does nothing between taking it and calling the fault handler.
- Now that the uvm_aobj shares vmobjlock with the {ttm,gem} uvm
object, we must not hold the lock when we call uvm_obj_wirepages on
the uvm_aobj.

XXX pullup to netbsd-7

branches: 1.3.2;
bus_space_mmap cookies are not paddrs, so don't pmap_enter them!

branches: 1.2.2; 1.2.4;
Multiply by PAGE_SIZE, not PAGE_SHIFT!

First whack at radeon driver.

No hardware to test yet, but it builds.

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

UVM locking changes, proposed on tech-kern:

- Change the lock on uvm_object, vm_amap and vm_anon to be a RW lock.
- Break v_interlock and vmobjlock apart. v_interlock remains a mutex.
- Do partial PV list locking in the x86 pmap. Others to follow later.

branches: 1.8.4;
in ttm_bus_dma_populate(), move the clearing of TTM_PAGE_FLAG_SWAPPED to be
earlier to avoid wrongly asserting if bus_dmamap_load_pglist() fails.

branches: 1.7.6; 1.7.14;
Clarify ttm state transitions tt_unpopulated<->tt_unbound<->tt_bound.
Assert it, too, and don't handle other cases.

We can add the assertion to ttm_agp_tt_unpopulate because it is called by
{nouveau,radeon}_ttm_tt_unpopulate, which is generically called
ttm_tt_unpopulate.

And the sole caller to ttm_tt_unpopulate (ttm_tt_destroy) only does so if the
state is unbound. the other caller is in a !NetBSD block.

We can add the assertion to ttm_agp_tt_populate and avoid handling the
!unpopulated case because the sole callers are {nouveau,radeon}_ttm_tt_populate
both of which return early in the !unpopulated case.

We can change the assertion on ttm_tt_wire because it is solely called by
ttm_bus_dma_populate, which already asserts that it is the unpopulated case.

from riastradh

oops, copied the code from the wrong diff. ... unbreak the build, still
doing the same thing.

Restore unpopulated state on fail1 error path. I shouldn't have committed
it separately.

from riastradh@

Remove redundant handling for tt_unbound case, and assert that it is the
unpopulated case.

All callers of ttm_bus_dma_populate (ttm_agp_tt_populate,
radeon_ttm_tt_populate, nouveau_ttm_tt_populate) return early if it isn't
unpopulated.

from riastradh@

Mark the state as unbound a bit earlier, to avoid triggering an assert in
the fail1 error path.

PR kern/52048: radeondrmkms: ttm->state == tt_unbound assertion

branches: 1.2.2; 1.2.4;
Rework ttm tt swapin/swapout logic.

Rather than handling `swapping in/out' here, per se, we let uvm do
that, we interpret `swap out' as `deactivate pages', and we add
generic ttm operations to wire and unwire pages, for the ttm_tt
driver to use.

This fixes certain graphics buffer eviction logic, which enables
nouveau to suspend/resume on one of my machines. (The machine
doesn't resume overall for other reasons, but the nouveau device
suspends and resumes in isolation.)

XXX pullup to netbsd-7 after a couple weeks

branches: 1.1.2; 1.1.4; 1.1.6; 1.1.8; 1.1.12;
First whack at radeon driver.

No hardware to test yet, but it builds.

drm: Modularize ttm.