Now that all framebuffer usage is properly refcounted, we are no
longer required to hold the modeset locks while dropping the last
reference. Hence implemented a fastpath which avoids the potential
stalls associated with grabbing mode_config.lock for the case where
there's no other reference around.
Explain in a big comment why it is safe. Also update kerneldocs with
the new locking rules around drm_framebuffer_remove.
Reviewed-by: Rob Clark <rob@ti.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Afact vmwgfx already has all the right refcounting implemented on the
backing storage, and we only need to ensure that the drm fb doesn't
disappear untimely. So holding onto the fb reference from _lookup
until vmw_kms_present has completed should be enough.
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
With the prep patch to encapsulate ->set_crtc calls, this is now
rather easy. Hooray for inconsistent semantics between ->set_crtc and
->page_flip, where the driver callback is supposed to update the fb
pointer, and ->update_plane, where the drm core does the same.
Also, since the drm core functions check crtc->fb before calling into
driver callbacks, we can't really reduce the critical sections
protected by the mode_config locks.
Reviewed-by: Rob Clark <rob@ti.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Now plane->fb holds a reference onto it's framebuffer. Nothing too
fancy going on here:
- Extract __drm_framebuffer_unreference to be called when we know
we're not dropping the last reference, e.g. useful in the fb cleanup
code.
- Reduce the locked sections in the set_plane ioctl to only protect
plane->fb/plane->crtc and the driver callback (i.e. hw state).
Everything either doesn't disappear (crtc, plane) or is refcounted
(fb), and all the data we check is invariant over the respective
object's lifetimes.
Reviewed-by: Rob Clark <rob@ti.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
We only need to ensure that the fb stays around for long enough. While
at it, only grab the modeset locks when we need them (since most
drivers don't implement the dirty callback, this should help jitter
and stalls when using the generic modeset driver).
Reviewed-by: Rob Clark <rob@ti.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
We only need to push the fb unreference a bit down. While at it,
properly pass the return value from ->create_handle back to userspace.
Most drivers either return -ENODEV if they don't have a concept of
buffer objects (ast, cirrus, ...) or just install a handle for the
underlying gem object (which is ok since we hold a reference on that
through the framebuffer).
v2: Split out the ->create_handle rework in the individual drivers.
Reviewed-by: Rob Clark <rob@ti.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
And drop it where it's not needed. Most driver just lookup the gem
object, allocate an fb struct, fill in all the useful fields and then
register it with drm_framebuffer_init.
All of these operations are already separately locked, and since we
only put the fb into the fpriv->fbs list _after_ having called
->fb_create, we can't also race with rmfb. We can otoh race with other
ioctls that put the framebuffer to use, but all drivers have been
reorganized already to call drm_framebuffer_init last in the fb
creation sequence.
So essentially, we can completely remove any modeset locks from the
addfb ioctl paths. Yeah!
Also, reference-counting is solid - we get a reference from fb_create
which we transfer to the fpriv->fbs list. And after unlocking the
fpriv->fbs_lock we don't touch the framebuffer any longer. Furthermore
drm_framebuffer_init has added a 2nd reference for the idr lookup, and
any access through that table will do it's own refcounting.
Reviewed-by: Rob Clark <rob@ti.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Atm we still need to unconditionally take the modeset locks in the
rmfb paths. But eventually we only want to take them if there are
other users around as a slow-path. This way sane userspace avoids
blocking on edid reads and other stuff in rmfb if it ensures that the
fb isn't used anywhere by a crtc/plane.
We can do a quick check for such other users once framebuffers are
properly refcounting by locking at the refcount - if it's more than 1,
there are other users left. Again, rmfb racing against other ioctls
isn't a real problem, userspace is allowed to shoot its foot.
This patch just prepares this by moving the modeset locks to nest
within fpriv->fbs_lock. Now the distinction between the fbs_lock and
the device-global fb_lock is clear, since we need to hold the fbs_lock
outside of any modeset_locks in fb_release.
Reviewed-by: Rob Clark <rob@ti.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Since otherwise looking and reference-counting around
drm_framebuffer_lookup will be an unmanageable mess. With this change,
an object can either be found in the idr and will stay around once we
incremented the reference counter. Or it will be gone for good and
can't be looked up using its id any more.
Atomicity is guaranteed by the dev->mode_config.fb_lock. The
newly-introduce fpriv->fbs_lock looks a bit redundant, but the next
patch will shuffle the locking order between these two locks and all
the modeset locks taken in modeset_lock_all, so we'll need it.
Also, since userspace could do really funky stuff and race e.g. a
getresources with an rmfb, we need to make sure that the kernel
doesn't fall over trying to look-up an inexistent fb, or causing
confusion by having two fbs around with the same id. Simply reset the
framebuffer id to 0, which marks it as reaped. Any lookups of that id
will fail, so the object is really gone for good from userspace's pov.
Note that we still need to protect the "remove framebuffer from all
use-cases" and the final unreference with the modeset-lock, since most
framebuffer use-sites don't implement proper reference counting yet.
We can only lift this once _all_ users are converted.
With this change, two references are held on alife, but unused
framebuffers:
- The reference for the idr lookup, created in this patch.
- For user-created framebuffers the fpriv->fbs reference, for
driver-private fbs the driver is supposed to hold it's own last
reference.
Note that the dev->mode_config.fb_list itself does _not_ hold a
reference onto the framebuffers (this list is essentially only used
for debugfs files). Hence if there's anything left there when the
driver has cleaned up all it's modeset resources, this is a ref-leak.
WARN about it.
Now we only need to fix up all other places to properly reference
count framebuffers.
v2: Fix spelling fail in a comment spotted by Rob Clark.
Reviewed-by: Rob Clark <rob@ti.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
We have two classes of framebuffer
- Created by the driver (atm only for fbdev), and the driver holds
onto the last reference count until destruction.
- Created by userspace and associated with a given fd. These
framebuffers will be reaped when their assoiciated fb is closed.
Now these two cases are set up differently, the framebuffers are on
different lists and hence destruction needs to clean up different
things. Also, for userspace framebuffers we remove them from any
current usage, whereas for internal framebuffers it is assumed that
the driver has done this already.
Long story short, we need two different ways to cleanup such drivers.
Three functions are involved in total:
- drm_framebuffer_remove: Convenience function which removes the fb
from all active usage and then drops the passed-in reference.
- drm_framebuffer_unregister_private: Will remove driver-private
framebuffers from relevant lists and drop the corresponding
references. Should be called for driver-private framebuffers before
dropping the last reference (or like for a lot of the drivers where
the fbdev is embedded someplace else, before doing the cleanup
manually).
- drm_framebuffer_cleanup: Final cleanup for both classes of fbs,
should be called by the driver's ->destroy callback once the last
reference is gone.
This patch just rolls out the new interfaces and updates all drivers
(by adding calls to drm_framebuffer_unregister_private at all the
right places)- no functional changes yet. Follow-on patches will move
drm core code around and update the lifetime management for
framebuffers, so that we are no longer required to keep framebuffers
alive by locking mode_config.mutex.
I've also updated the kerneldoc already.
vmwgfx seems to again be a bit special, at least I haven't figured out
how the fbdev support in that driver works. It smells like it's
external though.
v2: The i915 driver creates another private framebuffer in the
load-detect code. Adjust its cleanup code, too.
Reviewed-by: Rob Clark <rob@ti.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
And replace all fb lookups with it. Also add a WARN to
drm_mode_object_find since that is now no longer the blessed interface
to look up an fb. And add kerneldoc to both functions.
This only updates all callsites, but immediately drops the acquired
refence again. Hence all callers still rely on the fact that a mode fb
can't disappear while they're holding the struct mutex. Subsequent
patches will instate proper use of refcounts, and then rework the rmfb
and unref code to no longer serialize fb destruction with the
mode_config lock. We don't want that since otherwise a compositor
might end up stalling for a few frames in rmfb.
v2: Don't use kref_get_unless_zero - Greg KH doesn't like that kind of
interface.
Reviewed-by: Rob Clark <rob@ti.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Well, at least step 1. The goal here is that framebuffer objects can
survive outside of the mode_config lock, with just a reference held
as protection. The first step to get there is to introduce a special
fb_lock which protects fb lookup, creation and destruction, to make
them appear atomic.
This new fb_lock can nest within the mode_config lock. But the idea is
(once the reference counting part is completed) that we only quickly
take that fb_lock to lookup a framebuffer and grab a reference,
without any other locks involved.
vmwgfx is the only driver which does framebuffer lookups itself, also
wrap those calls to drm_mode_object_find with the new lock.
Also protect the fb_list walking in i915 and omapdrm with the new lock.
As a slight complication there's also the list of user-created fbs
attached to the file private. The problem now is that at fclose() time
we need to walk that list, eventually do a modeset call to remove the
fb from active usage (and are required to be able to take the
mode_config lock), but in the end we need to grab the new fb_lock to
remove the fb from the list. The easiest solution is to add another
mutex to protect this per-file list.
Currently that new fbs_lock nests within the modeset locks and so
appears redudant. But later patches will switch around this sequence
so that taking the modeset locks in the fb destruction path is
optional in the fastpath. Ultimately the goal is that addfb and rmfb
do not require the mode_config lock, since otherwise they have the
potential to introduce stalls in the pageflip sequence of a compositor
(if the compositor e.g. switches to a fullscreen client or if it
enables a plane). But that requires a few more steps and hoops to jump
through.
Note that framebuffer creation/destruction is now double-protected -
once by the fb_lock and in parts by the idr_lock. The later would be
unnecessariy if framebuffers would have their own idr allocator. But
that's material for another patch (series).
v2: Properly initialize the fb->filp_head list in _init, otherwise the
newly added WARN to check whether the fb isn't on a fpriv list any
more will fail for driver-private objects.
v3: Fixup two error-case unlock bugs spotted by Richard Wilbur.
Reviewed-by: Rob Clark <rob@ti.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
->cursor_move uses mostly the same facilities in drivers as
->cursor_set, so pretty much nothing to fix up:
- ast/gma500/i915: They all use per-crtc registers to update the
cursor position. ast again touches the global cursor cache, but
that's ok since there's only one crtc.
- nouveau: nv50+ is again special, updates happen through the per-crtc
channel (without pushbufs), so it's not protected by the new evo
lock introduced earlier. But since this channel is per-crtc, we
should be fine anyway.
- radeon: A bit a mess: avivo asics need a workaround when both output
pipes are enabled, which means it'll access the crtc list. Just
reading that flag is ok though as long as radeon _always_ grabs all
locks when changing the crtc configuration. Which means with the
current scheme it cannot do an optimized modeset which only locks
the relevant crtcs. This can be fixed though by introducing a bit of
global state with separate locks and ensure in the modeset code that
the cursor will be updated appropriately when enabling the 2nd pipe
(on affected asics).
- vmwgfx: I still don't understand what it's doing exactly, so apply
the same trick for now.
v2: Fixup unlocking for the error cases, spotted by Richard Wilbur.
v3: Another error-case fixup.
Reviewed-by: Rob Clark <rob@ti.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
First convert ->cursor_set to only take the crtc lock, since that
seems to be the function with the least amount of state - the core
ioctl function doesn't check anything which can change at runtime, so
we don't have any object lifetime issues to contend.
The only thing which is important is that the driver's implementation
doesn't touch any state outside of that single crtc which is not yet
properly protected by other locking:
- ast: access the global ast->cache_kmap. Luckily we only have on crtc
on this driver, so this is fine. Add a comment.
- gma500: calls gma_power_begin|and and psb_gtt_pin|unpin, both which
have their own locking to protect their state. Everything else is
crtc-local.
- i915: touches a bit of global gem state, all protected by the One
Lock to Rule Them All (dev->struct_mutex).
- nouveau: Pre-nv50 is all nice, nv50+ uses the evo channels to queue
up all display changes. And some of these channels are device
global. But this is fine now since the previous patch introduced an
evo channel mutex.
- radeon: Uses some indirect register access for cursor updates, but
with the previous patches to protect these indirect 2-register
access patterns with a spinlock, this should be fine now, too.
- vmwgfx: I have no idea how that works - update_cursor_position
doesn't take any per-crtc argument and I haven't figured out any
other place where this could be set in some form of a side-channel.
But vmwgfx definitely has more than one crtc (or at least can
register more than one), so I have no idea how this is supposed to
not fail with the current code already. Hence take the easy way out
and simply acquire all locks (which requires dropping the crtc lock
the core acquired for us). That way it's not worse off for
consistency than the old code.
Reviewed-by: Rob Clark <rob@ti.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
*drumroll*
The basic idea is to protect per-crtc state which can change without
touching the output configuration with separate mutexes, i.e. all the
input side state to a crtc like framebuffers, cursor settings or plane
configuration. Holding such a crtc lock gives a read-lock on all the
other crtc state which can be changed by e.g. a modeset.
All non-crtc state is still protected by the mode_config mutex.
Callers that need to change modeset state of a crtc (e.g. dpms or
set_mode) need to grab both the mode_config lock and nested within any
crtc locks.
Note that since there can only ever be one holder of the mode_config
lock we can grab the subordinate crtc locks in any order (if we need
to grab more than one of them). Lockdep can handle such nesting with
the mutex_lock_nest_lock call correctly.
With this functions that only touch connectors/encoders but not crtcs
only need to take the mode_config lock. The biggest such case is the
output probing, which means that we can now pageflip and move cursors
while the output probe code is reading an edid.
Most cases neatly fall into the three buckets:
- Only touches connectors and similar output state and so only needs
the mode_config lock.
- Touches the global configuration and so needs all locks.
- Only touches the crtc input side and so only needs the crtc lock.
But a few cases that need special consideration:
- Load detection which requires a crtc. The mode_config lock already
prevents a modeset change, so we can use any unused crtc as we like
to do load detection. The only thing to consider is that such
temporary state changes don't leak out to userspace through ioctls
that only take the crtc look (like a pageflip). Hence the load
detect code needs to grab the crtc of any output pipes it touches
(but only if it touches state used by the pageflip or cursor
ioctls).
- Atomic pageflip when moving planes. The first case is sane hw, where
planes have a fixed association with crtcs - nothing needs to be
done there. More insane^Wflexible hw needs to have plane->crtc
mapping which is separately protect with a lock that nests within
the crtc lock. If the plane is unused we can just assign it to the
current crtc and continue. But if a plane is already in use by
another crtc we can't just reassign it.
Two solution present themselves: Either go back to a slow-path which
takes all modeset locks, potentially incure quite a hefty delay. Or
simply disallowing such changes in one atomic pageflip - in general
the vblanks of two crtcs are not synced, so there's no sane way to
atomically flip such plane changes accross more than one crtc. I'd
heavily favour the later approach, going as far as mandating it as
part of the ABI of such a new a nuclear pageflip.
And if we _really_ want such semantics, we can always get them by
introducing another pageflip mutex between the mode_config.mutex and
the individual crtc locks. Pageflips crossing more than one crtc
would then need to take that lock first, to lock out concurrent
multi-crtc pageflips.
- Optimized global modeset operations: We could just take the
mode_config lock and then lazily lock all crtc which are affected by
a modeset operation. This has the advantage that pageflip could
continue unhampered on unaffected crtc. But if e.g. global resources
like plls need to be reassigned and so affect unrelated crtcs we can
still do that - nested locking works in any order.
This patch just adds the locks and takes them in drm_modeset_lock_all,
no real locking changes yet.
v2: Need to initialize the new lock in crtc_init and lock it righ
away, for otherwise the modeset_unlock_all below will try to unlock a
not-locked mutex.
Reviewed-by: Rob Clark <rob@ti.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Ok, this one here is a bit more complicated, and I can't really claim
to fully understand the locking and lifetime rules of the vmwgfx
driver. So just convert ever mutex_lock call, including the
interruptible one. Since other places (e.g. in the execbuf ioctl) take
the mode_config.mutex without bothering with interruptible handling,
I've figured I should be able to get away with this in a few more
places ...
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Only two places:
- suspend/resume
- Some really strange mode validation tool with too much funny-lucking
hand-rolled conversion code.
- The recently-added lastclose fbdev restore code.
Better safe than sorry, so convert both places to keep the locking
semantics as much as possible.
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Two exceptions:
- debugfs files only read information which is not related to crtc, so
can stay on the modeset_config lock.
- Same holds for the edp vdd work in intel_dp.c. Add a corresponding
WARN_ON and a comment next to the intel_dp struct fields for
documentation.
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
This is the first step towards introducing the new modeset locking
scheme. The plan is to put helper functions into place at all the
right places step-by-step, so that the final patch to switch on the
new locking scheme doesn't need to touch every single driver.
This helper here will serve as the shotgun solutions for all places
where a more fine-grained locking isn't (yet) implemented.
v2: Fixup kerneldoc for unlock_all.
Reviewed-by: Rob Clark <rob@ti.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
With refcounting we need to adjust framebuffer refcounts at each
callsite - much easier to do if they all call the same little helper
function.
Reviewed-by: Rob Clark <rob@ti.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Some drivers don't have real ->create_handle callbacks.
- cirrus/ast/mga200: Returns either 0 or -EINVAL.
- udl: Didn't even bother with a callback, leading to a nice
userspace-triggerable OOPS.
- vmwgfx: This driver bothered with an implementation to return 0 as
the handle (which is the canonical no-obj gem handle).
All have in common that ->create_handle doesn't really make too much
sense for them - that ioctl is used only for seamless fb takeover in
the radeon/nouveau/i915 ddx drivers. So allow drivers to not implement
this and return a consistent -ENODEV.
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
... by moving the bo_pin/bo_unpin manipulation of the pin_refcount
under the protection of the ttm reservation lock. pin/unpin seems
to get called from all over the place, so atm this is completely racy.
After this patch there are only a few places in cleanup functions
left which access ->pin_refcount without locking. But I'm hoping that
those are safe and some other code invariant guarantees that this
won't blow up.
In any case, I only need to fix up pin/unpin to make ->pageflip work
safely, so let's keep it at that.
Add a comment to the header to explain the new locking rule.
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
With per-crtc locks modeset operations can run in parallel, and the
cursor code uses the device-global evo master channel for hw frobbing.
But the pageflip code can also sync with the master under some
circumstances. Hence just wrap things up in a mutex to ensure that
pushbuf access doesn't intermingle.
The approach here is a bit overkill since the per-crtc channels used
to schedule the pageflips could probably be used without this pushbuf
locking, but I'm not familiar enough with the nouveau codebase to be
sure of that.
v2: Add missing mutex_init to avoid angering lockdep.
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Doing this within the fb->destroy callback leads to a locking
nightmare. And all other drm drivers that restore the fbcon do
it in lastclose, too.
With this adjustments all fb->destroy callbacks optionally drop
references to any gem objects used as backing storage, call
drm_framebuffer_cleanup and then kfree the struct. Which nicely
simplifies the locking for framebuffer unreferencing and freeing,
since this doesn't require that we hold the mode_config lock. A
slight exception is the vmwgfx surface backed framebuffer, it also
calls drm_master_put and removes the object from a device-private
framebuffer list. Both seem to have solid locking in place already.
Conclusion is that now it is no longer required to hold the
mode_config lock while freeing a framebuffer.
v2: Drop the corresponding mutex_lock WARN check from
drm_framebuffer_unreference.
v3: Use just the mode_config lock not modeset_lock_all, due to patch
reordering.
Acked-by: Alan Cox <alan@lxorguk.ukuu.org.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
vmwgfx has an oddity, when failing to reference the surface it'll
return 0, since that's what the successfull drm_framebuffer_init will
leave behind in ret. Fix this up by returning -EINVAL.
Split out from all the other driver updates due to the above tiny
semantic change. Shouldn't matter though since the reference grabbing
seemingly can't fail.
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
With more fine-grained locking we can no longer rely on the big
mode_config lock to prevent concurrent access to mode resources
like framebuffers. Instead a framebuffer becomes accessible to
other threads as soon as it is added to the relevant lookup
structures. Hence it needs to be fully set up by the time drivers
call drm_framebuffer_init.
This patch here is the drivers part of that reorg. Nothing really fancy
going on safe for three special cases.
- exynos needs to be careful to properly unref all handles.
- nouveau gets a resource leak fixed for free: one of the error
cases didn't cleanup the framebuffer, which is now moot since
the framebuffer is only registered once it is fully set up.
- vmwgfx requires a slight reordering of operations, I'm hoping I didn't
break anything (but it's refcount management only, so should be safe).
v2: Split out exynos, since it's a bit more hairy than expected.
v3: Drop bogus cirrus hunk noticed by Richard Wilbur.
v4: Split out vmwgfx since there's a small change in return values.
Reviewed-by: Rob Clark <rob@ti.com> (core + omapdrm)
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
And do a quick pass to adjust them to the last few (years?) of changes
...
This time actually compile-tested ;-)
Reviewed-by: Rob Clark <rob@ti.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
- config_cleanup was confused: It claimed that callers need to hold
the modeset lock, but the connector|encoder_cleanup helpers grabbed
that themselves (note that crtc_cleanup did _not_ grab the modeset
lock). Which resulted in all drivers _not_ hodling the lock. Since
this is for single-threaded cleanup code, drop the requirement from
docs and also drop the lock_grabbing from all _cleanup functions.
- Kill the LOCKING section in the doctype, since clearly we're not
good enough to keep them up-to-date. And misleading locking
documentation is worse than useless (see e.g. the comment in the
vmgfx driver about the cleanup mess). And since for most functions
the very first line either grabs the lock or has a WARN_ON(!locked)
the documentation doesn't really add anything.
- Instead put in some effort into explaining the only two special
cases a bit better: config_init and config_cleanup are both called
from single-threaded setup/teardown code, so don't do any locking.
It's the driver's job though to enforce this.
- Where lacking, add a WARN_ON(!is_locked). Not many places though,
since locking around fbdev setup/teardown is through-roughly screwed
up, and so will break almost every single WARN annotation I've tried
to add.
- Add a drm_modeset_is_locked helper - the Grate Modset Locking Rework
will use the compiler to assist in the big reorg by renaming the
mode lock, so start encapsulating things. Unfortunately this ended
up in the "wrong" header file since it needs the definition of
struct drm_device.
v2: Drop most WARNS again - we hit them all over the place, mostly in
the setup and teardown sequences. And trying to fix it up leads to
nice deadlocks, since the locking in the setup code is really
inconsistent.
Reviewed-by: Rob Clark <rob@ti.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Now that we seem to have brought order to the GTT barriers, the last one
to review is the terminal barrier before we unbind the buffer from the
GTT. This needs to only be performed if the buffer still resides in the
GTT domain, and so we can skip some needless barriers otherwise.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
With a fence, we only need to insert a memory barrier around the actual
fence alteration for CPU accesses through the GTT. Performing the
barrier in flush-fence was inserting unnecessary and expensive barriers
for never fenced objects.
Note removing the barriers from flush-fence, which was effectively a
barrier before every direct access through the GTT, revealed that we
where missing a barrier before the first access through the GTT. Lack of
that barrier was sufficient to cause GPU hangs.
v2: Add a couple more comments to explain the new barriers
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
We have two important transitions of the wedged state in the current
code:
- 0 -> 1: This means a hang has been detected, and signals to everyone
that they please get of any locks, so that the reset work item can
do its job.
- 1 -> 0: The reset handler has completed.
Now the last transition mixes up two states: "Reset completed and
successful" and "Reset failed". To distinguish these two we do some
tricks with the reset completion, but I simply could not convince
myself that this doesn't race under odd circumstances.
Hence split this up, and add a new terminal state indicating that the
hw is gone for good.
Also add explicit #defines for both states, update comments.
v2: Split out the reset handling bugfix for the throttle ioctl.
v3: s/tmp/wedged/ sugested by Chris Wilson. Also fixup up a rebase
error which prevented this patch from actually compiling.
v4: To unify the wedged state with the reset counter, keep the
reset-in-progress state just as a flag. The terminally-wedged state is
now denoted with a big number.
v5: Add a comment to the reset_counter special values explaining that
WEDGED & RESET_IN_PROGRESS needs to be true for the code to be
correct.
v6: Fixup logic errors introduced with the wedged+reset_counter
unification. Since WEDGED implies reset-in-progress (in a way we're
terminally stuck in the dead-but-reset-not-completed state), we need
ensure that we check for this everywhere. The specific bug was in
wait_for_error, which would simply have timed out.
v7: Extract an inline i915_reset_in_progress helper to make the code
more readable. Also annote the reset-in-progress case with an
unlikely, to help the compiler optimize the fastpath. Do the same for
the terminally wedged case with i915_terminally_wedged.
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-Off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
While auditing the code I've noticed one place (the throttle ioctl)
which does not yet wait for the reset handler to complete and doesn't
properly decode the wedge state into -EAGAIN/-EIO. Fix this up by
calling the right helpers. This might explain the oddball "my
compositor just died in a successfull gpu reset" reports. Or maybe not, since
current mesa doesn't use this ioctl to throttle command submission.
The throttle ioctl doesn't take the struct_mutex, so to avoid busy-looping
with -EAGAIN while a reset is in process, check for errors first and wait
for the handler to complete if a reset is pending by calling
i915_gem_wait_for_error.
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
And to make Ben Widawsky happier, use the gpu_error instead of
the entire device as the argument in some functions.
Drop the outdated comment on ->wedged for now, a follow-up patch will
change the semantics and add a proper comment again.
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
This has been sprinkled all over the place in dev_priv. I think
it'd be good to also move all the code into a separate file like
i915_gem_error.c, but that's for another patch.
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Tha one is really big, since it contains tons of comments explaining
how things work. Which is nice ;-)
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
The reasoning behind our code taking two paths depending upon whether or
not we may have been configured for IOMMU isn't clear to me. It should
always be safe to use the pci mapping functions as they are designed to
abstract the decision we were handling in i915.
Aside from simpler code, removing another member for the intel_gtt
struct is a nice motivation.
I ran this by Chris, and he wasn't concerned about the extra kzalloc,
and memory references vs. page_to_phys calculation in the case without
IOMMU.
v2: Update commit message
v3: Remove needs_dmar addition from Zhenyu upstream
This reverts (and then other stuff)
commit 20652097da
Author: Zhenyu Wang <zhenyuw@linux.intel.com>
Date: Thu Dec 13 23:47:47 2012 +0800
drm/i915: Fix missed needs_dmar setting
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com> (v2)
Cc: Zhenyu Wang <zhenyuw@linux.intel.com>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
[danvet: Squash in follow-up fix to remove the bogus hunk which
deleted the dma_mask configuration for gen6+.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
We already had a mapping in both (minus the phys_addr in AGP).
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
And, move it to where the rest of the logic is.
There is some slight functionality changes. There was extra paranoid
checks in AGP code making sure we never do idle maps on gen2 parts. That
was not duplicated as the simple PCI id check should do the right thing.
v2: use IS_GEN5 && IS_MOBILE check instead. For now, this is the same as
IS_IRONLAKE_M but is more future proof. The workaround docs hint that
more than one platform may be effected, but we've never seen such a
platform in the wild. (Rodrigo, Daniel)
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com> (v1)
Cc: Dave Airlie <airlied@redhat.com>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
The AVI infoframe is able to inform the display whether the source is
sending full or limited range RGB data.
As per CEA-861 [1] we must first check whether the display reports the
quantization range as selectable, and if so we can set the approriate
bits in the AVI inforframe.
[1] CEA-861-E - 6.4 Format of Version 2 AVI InfoFrame
v2: Give the Q bits better names, add spec chapter information
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
drm_rgb_quant_range_selectable() will report whether the monitor
claims to support for RGB quantization range selection.
The information can be found in the CEA Video capability block.
v2: s/quantzation/quantization/ in the comment
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Acked-by: David Airlie <airlied@linux.ie>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Add a new "Automatic" mode to the "Broadcast RGB" range property.
When selected the driver automagically selects between full range and
limited range output.
Based on CEA-861 [1] guidelines, limited range output is selected if the
mode is a CEA mode, except 640x480. Otherwise full range output is used.
Additionally DVI monitors should most likely default to full range
always.
As per DP1.2a [2] DisplayPort should always use full range for 18bpp, and
otherwise will follow CEA-861 rules.
NOTE: The default value for the property will now be "Automatic"
so some people may be affected in case they're relying on the
current full range default.
[1] CEA-861-E - 5.1 Default Encoding Parameters
[2] VESA DisplayPort Ver.1.2a - 5.1.1.1 Video Colorimetry
v2: Use has_hdmi_sink to check if a HDMI monitor is present
v3: Add information about relevant spec chapters
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
The RGB color range select bit on the DP/SDVO/HDMI registers
disappeared when PCH was introduced, and instead a new PIPECONF bit
was added that performs the same function.
Add a new INTEL_MODE_LIMITED_COLOR_RANGE private mode flag, and set
it in the encoder mode_fixup if limited color range is requested.
Set the the PIPECONF bit 13 based on the flag.
Experimentation showed that simply toggling the bit while the pipe is
active doesn't work. We need to restart the pipe, which luckily already
happens.
The DP/SDVO/HDMI bit 8 is marked MBZ in the docs, so avoid setting it,
although it doesn't seem to do any harm in practice.
TODO:
- the PIPECONF bit too seems to have disappeared from HSW. Need a
volunteer to test if it's just a documentation issue or if it's really
gone. If the bit is gone and no easy replacement is found, then I suppose
we may need to use the pipe CSC unit to perform the range compression.
v2: Use mode private_flags instead of intel_encoder virtual functions
v3: Moved the intel_dp color_range handling after bpc check to help
later patches
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=46800
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Mappable_end, ie. size is almost always what you want as opposed to the
number of entries. Since we already have that information, we can scrap
the number of entries and only calculate it when needed.
If gtt_start is !0, this will have slightly different behavior. This
difference can only occur in DRI1, and exists when we try to kick out
the firmware fb. The new code seems like a bugfix to me.
The other case where we've changed the behavior is during init we check
the mappable region against our current known upper and lower limits
(64MB, and 512MB). This now matches the comment, and makes things more
convenient after removing gtt_mappable_entries.
Also worth noting is the setting of mappable_end is taken out of setup
because we do it earlier now in the DRI2 case and therefore need to add
that tiny hunk to support the DRI1 IOCTL.
v2: Move up mappable end to before legacy AGP init
v3: Add the dev_priv inclusion here from previous rebase error in patch
5
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com> (v2)
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
[danvet: squash in fix for a printk format flag mismatch warning.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
This is to fix up a build problem with a wireless driver due to the
dynamic-debug patches in this branch.
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
We have enough info to not use the intel_gtt bridge stuff.
v2: Move setup of mappable_base above the legacy init stuff because we
still need that on older platforms. (Daniel)
v3: Remove the dev_priv hunk which was rebased in by accident
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com> (v2)
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
The purpose of the gtt structure is to help isolate our gtt specific
properties from the rest of the code (in doing so it help us finish the
isolation from the AGP connection).
The following members are pulled out (and renamed):
gtt_start
gtt_total
gtt_mappable_end
gtt_mappable
gtt_base_addr
gsm
The gtt structure will serve as a nice place to put gen specific gtt
routines in upcoming patches. As far as what else I feel belongs in this
structure: it is meant to encapsulate the GTT's physical properties.
This is why I've not added fields which track various drm_mm properties,
or things like gtt_mtrr (which is itself a pretty transient field).
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
[Ben modified commit messages]
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
With the assertion from the previous patch in place, it should be safe
to get rid gtt_mappable_total. Keeps things saner to not have to track
the same info in two places.
In order to keep the diff as simple as possible and keep with the
existing gtt_setup semantics we opt to keep gtt_mappable_end. It's not
as consistent with the 'total' used in the previous patch, but that can
be fixed later.
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
[Ben modified commit message]
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
