Re: [PATCH] drm/i915: Reserve space improvements

[Tomas Elf <tomas.elf@xxxxxxxxx>]

On 30/06/2015 17:15, Tomas Elf wrote:
> On 30/06/2015 16:53, John Harrison wrote:
> > On 30/06/2015 15:43, Tomas Elf wrote:
> > > On 30/06/2015 12:40, John.C.Harrison@xxxxxxxxx wrote:
> > > > From: John Harrison <John.C.Harrison@xxxxxxxxx>
> > > >
> > > > An earlier patch was added to reserve space in the ring buffer for the
> > > > commands issued during 'add_request()'. The initial version was
> > > > pessimistic in the way it handled buffer wrapping and would cause
> > > > premature wraps and thus waste ring space.
> > > >
> > > > This patch updates the code to better handle the wrap case. It no
> > > > longer enforces that the space being asked for and the reserved space
> > > > are a single contiguous block. Instead, it allows the reserve to be on
> > > > the far end of a wrap operation. It still guarantees that the space is
> > > > available so when the wrap occurs, no wait will happen. Thus the wrap
> > > > cannot fail which is the whole point of the exercise.
> > > >
> > > > Also fixed a merge failure with some comments from the original patch.
> > > >
> > > > v2: Incorporated suggestion by David Gordon to move the wrap code
> > > > inside the prepare function and thus allow a single combined
> > > > wait_for_space() call rather than doing one before the wrap and
> > > > another after. This also makes the prepare code much simpler and
> > > > easier to follow.
> > > >
> > > > v3: Fix for 'effective_size' vs 'size' during ring buffer remainder
> > > > calculations (spotted by Tomas Elf).
> > > >
> > > > For: VIZ-5115
> > > > CC: Daniel Vetter <daniel@xxxxxxxx>
> > > > Signed-off-by: John Harrison <John.C.Harrison@xxxxxxxxx>
> > > > ---
> > > >   drivers/gpu/drm/i915/intel_lrc.c        | 73
> > > > +++++++++++++-------------
> > > >   drivers/gpu/drm/i915/intel_ringbuffer.c | 90
> > > > +++++++++++++++++++--------------
> > > >   drivers/gpu/drm/i915/intel_ringbuffer.h |  4 +-
> > > >   3 files changed, 90 insertions(+), 77 deletions(-)
> > > >
> > > > diff --git a/drivers/gpu/drm/i915/intel_lrc.c
> > > > b/drivers/gpu/drm/i915/intel_lrc.c
> > > > index b36e064..7d9515d 100644
> > > > --- a/drivers/gpu/drm/i915/intel_lrc.c
> > > > +++ b/drivers/gpu/drm/i915/intel_lrc.c
> > > > @@ -663,12 +663,12 @@ static int logical_ring_wait_for_space(struct
> > > > drm_i915_gem_request *req,
> > > >       unsigned space;
> > > >       int ret;
> > > >
> > > > -    /* The whole point of reserving space is to not wait! */
> > > > -    WARN_ON(ringbuf->reserved_in_use);
> > > > -
> > > >       if (intel_ring_space(ringbuf) >= bytes)
> > > >           return 0;
> > > >
> > > > +    /* The whole point of reserving space is to not wait! */
> > > > +    WARN_ON(ringbuf->reserved_in_use);
> > > > +
> > > >       list_for_each_entry(target, &ring->request_list, list) {
> > > >           /*
> > > >            * The request queue is per-engine, so can contain requests
> > > > @@ -718,22 +718,11 @@ intel_logical_ring_advance_and_submit(struct
> > > > drm_i915_gem_request *request)
> > > >       execlists_context_queue(request);
> > > >   }
> > > >
> > > > -static int logical_ring_wrap_buffer(struct drm_i915_gem_request *req)
> > > > +static void __wrap_ring_buffer(struct intel_ringbuffer *ringbuf)
> > > >   {
> > > > -    struct intel_ringbuffer *ringbuf = req->ringbuf;
> > > >       uint32_t __iomem *virt;
> > > >       int rem = ringbuf->size - ringbuf->tail;
> > > >
> > > > -    /* Can't wrap if space has already been reserved! */
> > > > -    WARN_ON(ringbuf->reserved_in_use);
> > > > -
> > > > -    if (ringbuf->space < rem) {
> > > > -        int ret = logical_ring_wait_for_space(req, rem);
> > > > -
> > > > -        if (ret)
> > > > -            return ret;
> > > > -    }
> > > > -
> > > >       virt = ringbuf->virtual_start + ringbuf->tail;
> > > >       rem /= 4;
> > > >       while (rem--)
> > > > @@ -741,40 +730,50 @@ static int logical_ring_wrap_buffer(struct
> > > > drm_i915_gem_request *req)
> > > >
> > > >       ringbuf->tail = 0;
> > > >       intel_ring_update_space(ringbuf);
> > > > -
> > > > -    return 0;
> > > >   }
> > > >
> > > >   static int logical_ring_prepare(struct drm_i915_gem_request *req,
> > > > int bytes)
> > > >   {
> > > >       struct intel_ringbuffer *ringbuf = req->ringbuf;
> > > > -    int ret;
> > > > -
> > > > -    /*
> > > > -     * Add on the reserved size to the request to make sure that after
> > > > -     * the intended commands have been emitted, there is guaranteed to
> > > > -     * still be enough free space to send them to the hardware.
> > > > -     */
> > > > -    if (!ringbuf->reserved_in_use)
> > > > -        bytes += ringbuf->reserved_size;
> > > > -
> > > > -    if (unlikely(ringbuf->tail + bytes > ringbuf->effective_size)) {
> > > > -        ret = logical_ring_wrap_buffer(req);
> > > > -        if (unlikely(ret))
> > > > -            return ret;
> > > > +    int remain_usable = ringbuf->effective_size - ringbuf->tail;
> > > > +    int remain_actual = ringbuf->size - ringbuf->tail;
> > >
> > > You could add a comment here (and in the legacy implementation)
> > > explaining why we make the distinction between remain_usable and
> > > remain_actual. Or you could add the comment when we actually use
> > > remain_actual further down in the function. It's up to you.
> > >
> > > We first need to be pessimistic about how much space is left in the
> > > ring buffer when determining the need for wrapping - therefore we use
> > > remain_usable, which disregards the end-of-buffer padding - and then
> > > we need to be pessimistic about how much ring space we need to wait
> > > for - therefore we use remain_actual, which takes the end-of-buffer
> > > padding into consideration to make sure that we're actually not
> > > waiting for too little space.
> >
> > It's not about being pessimistic or optimistic. That implies there is
> > some choice, that one version is a little bit better in one situation
> > but either would do. Whereas this is about functional correctness. The
> > difference between the actual ring size and the usable ring size is not
> > some reserved space thing to make something go faster. This is about the
> > hardware locking up if the entire buffer is used. I think 'usable' and
> > 'actual' are fairly obvious names. If you want to know the details about
> > why there is an 'effective_size' in the first place then look up
> > 'effective_size' in the code and read the comment about i830 hangs.
>
> You're probably right. Let's just forget about it.
>
> > > If you add those comments maybe you could also rename the variables to
> > > something like "remaining_space_usable" etc. since "remain_usable" is
> > > more of a verb than a noun. But that's seriously nitpicking.
> > Or maybe
> > 'remaining_space_that_is_usable_without_causing_an_i830_to_hang_because_it_skips_the_last_two_cachelines'?
> >
> > There is such a thing as being too verbose and making the code hard to
> > read.
>
> That is indeed a very long variable name. It's 104 characters long so
> that wouldn't clear checkpatch.pl :). My nitpicky suggestion was 9
> characters longer. And, yeah, it's possible to get too verbose. This
> driver is not even remotely close to that point as it stands today :).
>
> Thanks,
> Tomas

Anyway, I think we're done here.

Reviewed-by: Tomas Elf <tomas.elf@xxxxxxxxx>

Thanks,
Tomas


> > > If you at least add the comments to make it crystal clear why we do it
> > > this way then I'm fine.
> >
> > I don't see what comment could be added that would actually make things
> > clearer without being a long and unnecessary description of the i830
> > issue. There are two variables declared on consecutive lines that cache
> > pretty simple calculations and have fairly clear names. It looks quite
> > self explanatory to me! The code then tests to see if it can use the
> > usable space, if not then it wraps around the actual buffer size. Again,
> > seems pretty obvious as to what is going on and why - you can only use
> > the usable bit, but when traversing the whole buffer you pretty
> > obviously need to traverse the whole buffer not just the part that may
> > or may not have been used.
> >
> >
> > > Reviewed-by: Tomas Elf <tomas.elf@xxxxxxxxx>
> > >
> > > Thanks,
> > > Tomas
> > >
> > > > +    int ret, total_bytes, wait_bytes = 0;
> > > > +    bool need_wrap = false;
> > > >
> > > > -        if(ringbuf->reserved_size) {
> > > > -            uint32_t size = ringbuf->reserved_size;
> > > > +    if (ringbuf->reserved_in_use)
> > > > +        total_bytes = bytes;
> > > > +    else
> > > > +        total_bytes = bytes + ringbuf->reserved_size;
> > > >
> > > > -            intel_ring_reserved_space_cancel(ringbuf);
> > > > -            intel_ring_reserved_space_reserve(ringbuf, size);
> > > > +    if (unlikely(bytes > remain_usable)) {
> > > > +        /*
> > > > +         * Not enough space for the basic request. So need to flush
> > > > +         * out the remainder and then wait for base + reserved.
> > > > +         */
> > > > +        wait_bytes = remain_actual + total_bytes;
> > > > +        need_wrap = true;
> > > > +    } else {
> > > > +        if (unlikely(total_bytes > remain_usable)) {
> > > > +            /*
> > > > +             * The base request will fit but the reserved space
> > > > +             * falls off the end. So only need to to wait for the
> > > > +             * reserved size after flushing out the remainder.
> > > > +             */
> > > > +            wait_bytes = remain_actual + ringbuf->reserved_size;
> > > > +            need_wrap = true;
> > > > +        } else if (total_bytes > ringbuf->space) {
> > > > +            /* No wrapping required, just waiting. */
> > > > +            wait_bytes = total_bytes;
> > > >           }
> > > >       }
> > > >
> > > > -    if (unlikely(ringbuf->space < bytes)) {
> > > > -        ret = logical_ring_wait_for_space(req, bytes);
> > > > +    if (wait_bytes) {
> > > > +        ret = logical_ring_wait_for_space(req, wait_bytes);
> > > >           if (unlikely(ret))
> > > >               return ret;
> > > > +
> > > > +        if (need_wrap)
> > > > +            __wrap_ring_buffer(ringbuf);
> > > >       }
> > > >
> > > >       return 0;
> > > > diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.c
> > > > b/drivers/gpu/drm/i915/intel_ringbuffer.c
> > > > index af7c12e..e39c891 100644
> > > > --- a/drivers/gpu/drm/i915/intel_ringbuffer.c
> > > > +++ b/drivers/gpu/drm/i915/intel_ringbuffer.c
> > > > @@ -2121,12 +2121,12 @@ static int ring_wait_for_space(struct
> > > > intel_engine_cs *ring, int n)
> > > >       unsigned space;
> > > >       int ret;
> > > >
> > > > -    /* The whole point of reserving space is to not wait! */
> > > > -    WARN_ON(ringbuf->reserved_in_use);
> > > > -
> > > >       if (intel_ring_space(ringbuf) >= n)
> > > >           return 0;
> > > >
> > > > +    /* The whole point of reserving space is to not wait! */
> > > > +    WARN_ON(ringbuf->reserved_in_use);
> > > > +
> > > >       list_for_each_entry(request, &ring->request_list, list) {
> > > >           space = __intel_ring_space(request->postfix, ringbuf->tail,
> > > >                          ringbuf->size);
> > > > @@ -2145,21 +2145,11 @@ static int ring_wait_for_space(struct
> > > > intel_engine_cs *ring, int n)
> > > >       return 0;
> > > >   }
> > > >
> > > > -static int intel_wrap_ring_buffer(struct intel_engine_cs *ring)
> > > > +static void __wrap_ring_buffer(struct intel_ringbuffer *ringbuf)
> > > >   {
> > > >       uint32_t __iomem *virt;
> > > > -    struct intel_ringbuffer *ringbuf = ring->buffer;
> > > >       int rem = ringbuf->size - ringbuf->tail;
> > > >
> > > > -    /* Can't wrap if space has already been reserved! */
> > > > -    WARN_ON(ringbuf->reserved_in_use);
> > > > -
> > > > -    if (ringbuf->space < rem) {
> > > > -        int ret = ring_wait_for_space(ring, rem);
> > > > -        if (ret)
> > > > -            return ret;
> > > > -    }
> > > > -
> > > >       virt = ringbuf->virtual_start + ringbuf->tail;
> > > >       rem /= 4;
> > > >       while (rem--)
> > > > @@ -2167,8 +2157,6 @@ static int intel_wrap_ring_buffer(struct
> > > > intel_engine_cs *ring)
> > > >
> > > >       ringbuf->tail = 0;
> > > >       intel_ring_update_space(ringbuf);
> > > > -
> > > > -    return 0;
> > > >   }
> > > >
> > > >   int intel_ring_idle(struct intel_engine_cs *ring)
> > > > @@ -2238,9 +2226,21 @@ void intel_ring_reserved_space_use(struct
> > > > intel_ringbuffer *ringbuf)
> > > >   void intel_ring_reserved_space_end(struct intel_ringbuffer *ringbuf)
> > > >   {
> > > >       WARN_ON(!ringbuf->reserved_in_use);
> > > > -    WARN(ringbuf->tail > ringbuf->reserved_tail +
> > > > ringbuf->reserved_size,
> > > > -         "request reserved size too small: %d vs %d!\n",
> > > > -         ringbuf->tail - ringbuf->reserved_tail,
> > > > ringbuf->reserved_size);
> > > > +    if (ringbuf->tail > ringbuf->reserved_tail) {
> > > > +        WARN(ringbuf->tail > ringbuf->reserved_tail +
> > > > ringbuf->reserved_size,
> > > > +             "request reserved size too small: %d vs %d!\n",
> > > > +             ringbuf->tail - ringbuf->reserved_tail,
> > > > ringbuf->reserved_size);
> > > > +    } else {
> > > > +        /*
> > > > +         * The ring was wrapped while the reserved space was in use.
> > > > +         * That means that some unknown amount of the ring tail was
> > > > +         * no-op filled and skipped. Thus simply adding the ring size
> > > > +         * to the tail and doing the above space check will not work.
> > > > +         * Rather than attempt to track how much tail was skipped,
> > > > +         * it is much simpler to say that also skipping the sanity
> > > > +         * check every once in a while is not a big issue.
> > > > +         */
> > > > +    }
> > > >
> > > >       ringbuf->reserved_size   = 0;
> > > >       ringbuf->reserved_in_use = false;
> > > > @@ -2249,33 +2249,45 @@ void intel_ring_reserved_space_end(struct
> > > > intel_ringbuffer *ringbuf)
> > > >   static int __intel_ring_prepare(struct intel_engine_cs *ring, int
> > > > bytes)
> > > >   {
> > > >       struct intel_ringbuffer *ringbuf = ring->buffer;
> > > > -    int ret;
> > > > -
> > > > -    /*
> > > > -     * Add on the reserved size to the request to make sure that after
> > > > -     * the intended commands have been emitted, there is guaranteed to
> > > > -     * still be enough free space to send them to the hardware.
> > > > -     */
> > > > -    if (!ringbuf->reserved_in_use)
> > > > -        bytes += ringbuf->reserved_size;
> > > > -
> > > > -    if (unlikely(ringbuf->tail + bytes > ringbuf->effective_size)) {
> > > > -        ret = intel_wrap_ring_buffer(ring);
> > > > -        if (unlikely(ret))
> > > > -            return ret;
> > > > +    int remain_usable = ringbuf->effective_size - ringbuf->tail;
> > > > +    int remain_actual = ringbuf->size - ringbuf->tail;
> > > > +    int ret, total_bytes, wait_bytes = 0;
> > > > +    bool need_wrap = false;
> > > >
> > > > -        if(ringbuf->reserved_size) {
> > > > -            uint32_t size = ringbuf->reserved_size;
> > > > +    if (ringbuf->reserved_in_use)
> > > > +        total_bytes = bytes;
> > > > +    else
> > > > +        total_bytes = bytes + ringbuf->reserved_size;
> > > >
> > > > -            intel_ring_reserved_space_cancel(ringbuf);
> > > > -            intel_ring_reserved_space_reserve(ringbuf, size);
> > > > +    if (unlikely(bytes > remain_usable)) {
> > > > +        /*
> > > > +         * Not enough space for the basic request. So need to flush
> > > > +         * out the remainder and then wait for base + reserved.
> > > > +         */
> > > > +        wait_bytes = remain_actual + total_bytes;
> > > > +        need_wrap = true;
> > > > +    } else {
> > > > +        if (unlikely(total_bytes > remain_usable)) {
> > > > +            /*
> > > > +             * The base request will fit but the reserved space
> > > > +             * falls off the end. So only need to to wait for the
> > > > +             * reserved size after flushing out the remainder.
> > > > +             */
> > > > +            wait_bytes = remain_actual + ringbuf->reserved_size;
> > > > +            need_wrap = true;
> > > > +        } else if (total_bytes > ringbuf->space) {
> > > > +            /* No wrapping required, just waiting. */
> > > > +            wait_bytes = total_bytes;
> > > >           }
> > > >       }
> > > >
> > > > -    if (unlikely(ringbuf->space < bytes)) {
> > > > -        ret = ring_wait_for_space(ring, bytes);
> > > > +    if (wait_bytes) {
> > > > +        ret = ring_wait_for_space(ring, wait_bytes);
> > > >           if (unlikely(ret))
> > > >               return ret;
> > > > +
> > > > +        if (need_wrap)
> > > > +            __wrap_ring_buffer(ringbuf);
> > > >       }
> > > >
> > > >       return 0;
> > > > diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.h
> > > > b/drivers/gpu/drm/i915/intel_ringbuffer.h
> > > > index 0e2bbc6..304cac4 100644
> > > > --- a/drivers/gpu/drm/i915/intel_ringbuffer.h
> > > > +++ b/drivers/gpu/drm/i915/intel_ringbuffer.h
> > > > @@ -473,7 +473,6 @@ static inline u32 intel_ring_get_tail(struct
> > > > intel_ringbuffer *ringbuf)
> > > >    * will always have sufficient room to do its stuff. The request
> > > > creation
> > > >    * code calls this automatically.
> > > >    */
> > > > -int intel_ring_reserve_space(struct drm_i915_gem_request *request);
> > > >   void intel_ring_reserved_space_reserve(struct intel_ringbuffer
> > > > *ringbuf, int size);
> > > >   /* Cancel the reservation, e.g. because the request is being
> > > > discarded. */
> > > >   void intel_ring_reserved_space_cancel(struct intel_ringbuffer
> > > > *ringbuf);
> > > > @@ -482,4 +481,7 @@ void intel_ring_reserved_space_use(struct
> > > > intel_ringbuffer *ringbuf);
> > > >   /* Finish with the reserved space - for use by i915_add_request()
> > > > only. */
> > > >   void intel_ring_reserved_space_end(struct intel_ringbuffer *ringbuf);
> > > >
> > > > +/* Legacy ringbuffer specific portion of reservation code: */
> > > > +int intel_ring_reserve_space(struct drm_i915_gem_request *request);
> > > > +
> > > >   #endif /* _INTEL_RINGBUFFER_H_ */

_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
http://lists.freedesktop.org/mailman/listinfo/intel-gfx