diff options
author | Aingara Paramakuru <aparamakuru@nvidia.com> | 2016-02-22 12:35:49 -0500 |
---|---|---|
committer | Terje Bergstrom <tbergstrom@nvidia.com> | 2016-03-15 19:23:44 -0400 |
commit | 2a58d3c27b45ca9d0d9dc2136377b7a41b9ed82d (patch) | |
tree | 9d7464bfd0eea8e4b65f591996db59a98f4070e2 /drivers/gpu/nvgpu/gk20a/channel_gk20a.c | |
parent | f07a046a52e7a8074bd1572a12ac65747d3f827d (diff) |
gpu: nvgpu: improve channel interleave support
Previously, only "high" priority bare channels were interleaved
between all other bare channels and TSGs. This patch decouples
priority from interleaving and introduces 3 levels for interleaving
a bare channel or TSG: high, medium, and low. The levels define
the number of times a channel or TSG will appear on a runlist (see
nvgpu.h for details).
By default, all bare channels and TSGs are set to interleave level
low. Userspace can then request the interleave level to be increased
via the CHANNEL_SET_RUNLIST_INTERLEAVE ioctl (TSG-specific ioctl will
be added later).
As timeslice settings will soon be coming from userspace, the default
timeslice for "high" priority channels has been restored.
JIRA VFND-1302
Bug 1729664
Change-Id: I178bc1cecda23f5002fec6d791e6dcaedfa05c0c
Signed-off-by: Aingara Paramakuru <aparamakuru@nvidia.com>
Reviewed-on: http://git-master/r/1014962
Reviewed-by: Terje Bergstrom <tbergstrom@nvidia.com>
Tested-by: Terje Bergstrom <tbergstrom@nvidia.com>
Diffstat (limited to 'drivers/gpu/nvgpu/gk20a/channel_gk20a.c')
-rw-r--r-- | drivers/gpu/nvgpu/gk20a/channel_gk20a.c | 85 |
1 files changed, 41 insertions, 44 deletions
diff --git a/drivers/gpu/nvgpu/gk20a/channel_gk20a.c b/drivers/gpu/nvgpu/gk20a/channel_gk20a.c index 2c2850c6..6eecebf5 100644 --- a/drivers/gpu/nvgpu/gk20a/channel_gk20a.c +++ b/drivers/gpu/nvgpu/gk20a/channel_gk20a.c | |||
@@ -177,7 +177,7 @@ int gk20a_channel_get_timescale_from_timeslice(struct gk20a *g, | |||
177 | } | 177 | } |
178 | 178 | ||
179 | static int channel_gk20a_set_schedule_params(struct channel_gk20a *c, | 179 | static int channel_gk20a_set_schedule_params(struct channel_gk20a *c, |
180 | u32 timeslice_period, bool interleave) | 180 | u32 timeslice_period) |
181 | { | 181 | { |
182 | void *inst_ptr; | 182 | void *inst_ptr; |
183 | int shift = 0, value = 0; | 183 | int shift = 0, value = 0; |
@@ -205,30 +205,6 @@ static int channel_gk20a_set_schedule_params(struct channel_gk20a *c, | |||
205 | gk20a_readl(c->g, ccsr_channel_r(c->hw_chid)) | | 205 | gk20a_readl(c->g, ccsr_channel_r(c->hw_chid)) | |
206 | ccsr_channel_enable_set_true_f()); | 206 | ccsr_channel_enable_set_true_f()); |
207 | 207 | ||
208 | if (c->interleave != interleave) { | ||
209 | mutex_lock(&c->g->interleave_lock); | ||
210 | c->interleave = interleave; | ||
211 | if (interleave) | ||
212 | if (c->g->num_interleaved_channels >= | ||
213 | MAX_INTERLEAVED_CHANNELS) { | ||
214 | gk20a_err(dev_from_gk20a(c->g), | ||
215 | "Change of priority would exceed runlist length, only changing timeslice\n"); | ||
216 | c->interleave = false; | ||
217 | } else | ||
218 | c->g->num_interleaved_channels += 1; | ||
219 | else | ||
220 | c->g->num_interleaved_channels -= 1; | ||
221 | |||
222 | mutex_unlock(&c->g->interleave_lock); | ||
223 | gk20a_dbg_info("Set channel %d to interleave %d", | ||
224 | c->hw_chid, c->interleave); | ||
225 | |||
226 | gk20a_fifo_set_channel_priority( | ||
227 | c->g, 0, c->hw_chid, c->interleave); | ||
228 | c->g->ops.fifo.update_runlist( | ||
229 | c->g, 0, ~0, true, false); | ||
230 | } | ||
231 | |||
232 | return 0; | 208 | return 0; |
233 | } | 209 | } |
234 | 210 | ||
@@ -711,6 +687,32 @@ static int gk20a_channel_set_wdt_status(struct channel_gk20a *ch, | |||
711 | return 0; | 687 | return 0; |
712 | } | 688 | } |
713 | 689 | ||
690 | static int gk20a_channel_set_runlist_interleave(struct channel_gk20a *ch, | ||
691 | u32 level) | ||
692 | { | ||
693 | struct gk20a *g = ch->g; | ||
694 | int ret; | ||
695 | |||
696 | if (gk20a_is_channel_marked_as_tsg(ch)) { | ||
697 | gk20a_err(dev_from_gk20a(g), "invalid operation for TSG!\n"); | ||
698 | return -EINVAL; | ||
699 | } | ||
700 | |||
701 | switch (level) { | ||
702 | case NVGPU_RUNLIST_INTERLEAVE_LEVEL_LOW: | ||
703 | case NVGPU_RUNLIST_INTERLEAVE_LEVEL_MEDIUM: | ||
704 | case NVGPU_RUNLIST_INTERLEAVE_LEVEL_HIGH: | ||
705 | ret = g->ops.fifo.set_runlist_interleave(g, ch->hw_chid, | ||
706 | false, 0, level); | ||
707 | break; | ||
708 | default: | ||
709 | ret = -EINVAL; | ||
710 | break; | ||
711 | } | ||
712 | |||
713 | return ret ? ret : g->ops.fifo.update_runlist(g, 0, ~0, true, true); | ||
714 | } | ||
715 | |||
714 | static int gk20a_init_error_notifier(struct channel_gk20a *ch, | 716 | static int gk20a_init_error_notifier(struct channel_gk20a *ch, |
715 | struct nvgpu_set_error_notifier *args) | 717 | struct nvgpu_set_error_notifier *args) |
716 | { | 718 | { |
@@ -899,17 +901,6 @@ static void gk20a_free_channel(struct channel_gk20a *ch) | |||
899 | } | 901 | } |
900 | mutex_unlock(&f->deferred_reset_mutex); | 902 | mutex_unlock(&f->deferred_reset_mutex); |
901 | 903 | ||
902 | if (ch->interleave) { | ||
903 | ch->interleave = false; | ||
904 | gk20a_fifo_set_channel_priority( | ||
905 | ch->g, 0, ch->hw_chid, ch->interleave); | ||
906 | |||
907 | mutex_lock(&f->g->interleave_lock); | ||
908 | WARN_ON(f->g->num_interleaved_channels == 0); | ||
909 | f->g->num_interleaved_channels -= 1; | ||
910 | mutex_unlock(&f->g->interleave_lock); | ||
911 | } | ||
912 | |||
913 | if (!ch->bound) | 904 | if (!ch->bound) |
914 | goto release; | 905 | goto release; |
915 | 906 | ||
@@ -1154,11 +1145,8 @@ struct channel_gk20a *gk20a_open_new_channel(struct gk20a *g) | |||
1154 | ch->has_timedout = false; | 1145 | ch->has_timedout = false; |
1155 | ch->wdt_enabled = true; | 1146 | ch->wdt_enabled = true; |
1156 | ch->obj_class = 0; | 1147 | ch->obj_class = 0; |
1157 | ch->interleave = false; | ||
1158 | ch->clean_up.scheduled = false; | 1148 | ch->clean_up.scheduled = false; |
1159 | gk20a_fifo_set_channel_priority( | 1149 | ch->interleave_level = NVGPU_RUNLIST_INTERLEAVE_LEVEL_LOW; |
1160 | ch->g, 0, ch->hw_chid, ch->interleave); | ||
1161 | |||
1162 | 1150 | ||
1163 | /* The channel is *not* runnable at this point. It still needs to have | 1151 | /* The channel is *not* runnable at this point. It still needs to have |
1164 | * an address space bound and allocate a gpfifo and grctx. */ | 1152 | * an address space bound and allocate a gpfifo and grctx. */ |
@@ -2613,7 +2601,6 @@ unsigned int gk20a_channel_poll(struct file *filep, poll_table *wait) | |||
2613 | int gk20a_channel_set_priority(struct channel_gk20a *ch, u32 priority) | 2601 | int gk20a_channel_set_priority(struct channel_gk20a *ch, u32 priority) |
2614 | { | 2602 | { |
2615 | u32 timeslice_timeout; | 2603 | u32 timeslice_timeout; |
2616 | bool interleave = false; | ||
2617 | 2604 | ||
2618 | if (gk20a_is_channel_marked_as_tsg(ch)) { | 2605 | if (gk20a_is_channel_marked_as_tsg(ch)) { |
2619 | gk20a_err(dev_from_gk20a(ch->g), | 2606 | gk20a_err(dev_from_gk20a(ch->g), |
@@ -2630,8 +2617,6 @@ int gk20a_channel_set_priority(struct channel_gk20a *ch, u32 priority) | |||
2630 | timeslice_timeout = ch->g->timeslice_medium_priority_us; | 2617 | timeslice_timeout = ch->g->timeslice_medium_priority_us; |
2631 | break; | 2618 | break; |
2632 | case NVGPU_PRIORITY_HIGH: | 2619 | case NVGPU_PRIORITY_HIGH: |
2633 | if (ch->g->interleave_high_priority) | ||
2634 | interleave = true; | ||
2635 | timeslice_timeout = ch->g->timeslice_high_priority_us; | 2620 | timeslice_timeout = ch->g->timeslice_high_priority_us; |
2636 | break; | 2621 | break; |
2637 | default: | 2622 | default: |
@@ -2640,7 +2625,7 @@ int gk20a_channel_set_priority(struct channel_gk20a *ch, u32 priority) | |||
2640 | } | 2625 | } |
2641 | 2626 | ||
2642 | return channel_gk20a_set_schedule_params(ch, | 2627 | return channel_gk20a_set_schedule_params(ch, |
2643 | timeslice_timeout, interleave); | 2628 | timeslice_timeout); |
2644 | } | 2629 | } |
2645 | 2630 | ||
2646 | static int gk20a_channel_zcull_bind(struct channel_gk20a *ch, | 2631 | static int gk20a_channel_zcull_bind(struct channel_gk20a *ch, |
@@ -3045,6 +3030,18 @@ long gk20a_channel_ioctl(struct file *filp, | |||
3045 | err = gk20a_channel_set_wdt_status(ch, | 3030 | err = gk20a_channel_set_wdt_status(ch, |
3046 | (struct nvgpu_channel_wdt_args *)buf); | 3031 | (struct nvgpu_channel_wdt_args *)buf); |
3047 | break; | 3032 | break; |
3033 | case NVGPU_IOCTL_CHANNEL_SET_RUNLIST_INTERLEAVE: | ||
3034 | err = gk20a_busy(dev); | ||
3035 | if (err) { | ||
3036 | dev_err(&dev->dev, | ||
3037 | "%s: failed to host gk20a for ioctl cmd: 0x%x", | ||
3038 | __func__, cmd); | ||
3039 | break; | ||
3040 | } | ||
3041 | err = gk20a_channel_set_runlist_interleave(ch, | ||
3042 | ((struct nvgpu_runlist_interleave_args *)buf)->level); | ||
3043 | gk20a_idle(dev); | ||
3044 | break; | ||
3048 | default: | 3045 | default: |
3049 | dev_dbg(&dev->dev, "unrecognized ioctl cmd: 0x%x", cmd); | 3046 | dev_dbg(&dev->dev, "unrecognized ioctl cmd: 0x%x", cmd); |
3050 | err = -ENOTTY; | 3047 | err = -ENOTTY; |