Skip to contentSkip to conversation

drm/i915/lrc: Stop writing to ELSP until HW has processed the previous write

PATCH In review Michel Thierry 3 participants 4 messages Nov 18, 2017

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 drm/i915/lrc: Stop writing to ELSP until HW has processed the previous write

The hardware needs some time to process the information received in the ExecList Submission Port, and expects us to don't write anything new until it has 'acknowledged' this new execlist by sending an IDLE_ACTIVE or PREEMPTED CSB event.

If we do not follow this, the driver could write new data into the ELSP before HW had finishing fetching the previous one, putting us in 'undefined behaviour' space.

This seems to be the problem causing the spurious PREEMPTED & COMPLETE events after a COMPLETE like the one below:

[] vcs0: sw rd pointer = 2, hw wr pointer = 0, current 'head' = 3.

[] vcs0:  Execlist CSB[0]: 0x00000018 _ 0x00000007
[] vcs0:  Execlist CSB[1]: 0x00000001 _ 0x00000000
[] vcs0:  Execlist CSB[2]: 0x00000018 _ 0x00000007  <<< COMPLETE
[] vcs0:  Execlist CSB[3]: 0x00000012 _ 0x00000007  <<< PREEMPTED & COMPLETE
[] vcs0:  Execlist CSB[4]: 0x00008002 _ 0x00000006
[] vcs0:  Execlist CSB[5]: 0x00000014 _ 0x00000006

The ELSP writes that lead to this CSB sequence show that the HW hadn't started executing the previous execlist (the one with only ctx 0x6) by the time the new one was submitted; this is a bit more clear in the data show in the EXECLIST_STATUS register at the time of the ELSP write.

[] vcs0: ELSP[0] = 0x0_0 [execlist1] - status_reg = 0x0_302
[] vcs0: ELSP[1] = 0x6_fedb2119 [execlist0] - status_reg = 0x0_8302

[] vcs0: ELSP[2] = 0x7_fedaf119 [execlist1] - status_reg = 0x0_8308 [] vcs0: ELSP[3] = 0x6_fedb2119 [execlist0] - status_reg = 0x7_8308

Note that having to wait for this ack does not disable lite-restores, although it may reduce their numbers.

And take this as a RFC, since there are probably better ways to still respect this HW requirement.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=102035

---
 drivers/gpu/drm/i915/intel_lrc.c        | 16 +++++++++++++++-
 drivers/gpu/drm/i915/intel_ringbuffer.h |  6 ++++++
 2 files changed, 21 insertions(+), 1 deletion(-)
drivers/gpu/drm/i915/intel_lrc.c +15 -1
index af41165e3da4..10b7eb64f169 100644
@@ -449,11 +449,16 @@ static inline void elsp_write(u64 desc, u32 __iomem *elsp)
 
 static void execlists_submit_ports(struct intel_engine_cs *engine)
 {
+	struct intel_engine_execlists * const execlists = &engine->execlists;
 	struct execlist_port *port = engine->execlists.port;
 	u32 __iomem *elsp =
 		engine->i915->regs + i915_mmio_reg_offset(RING_ELSP(engine));
 	unsigned int n;
 
+	if (wait_for_atomic(!READ_ONCE(execlists->outstanding_submission), 10))
+		GEM_TRACE("%s outstanding submission stuck\n",
+			  engine->name);
+
 	for (n = execlists_num_ports(&engine->execlists); n--; ) {
 		struct drm_i915_gem_request *rq;
 		unsigned int count;
@@ -479,6 +484,8 @@ static void execlists_submit_ports(struct intel_engine_cs *engine)
 
 		elsp_write(desc, elsp);
 	}
+
+	WRITE_ONCE(execlists->outstanding_submission, 1);
 }
 
 static bool ctx_single_port_submission(const struct i915_gem_context *ctx)
@@ -889,6 +896,11 @@ static void execlists_submission_tasklet(unsigned long data)
 			GEM_TRACE("%s csb[%dd]: status=0x%08x:0x%08x\n",
 				  engine->name, head,
 				  status, buf[2*head + 1]);
+
+			if ((status & GEN8_CTX_STATUS_IDLE_ACTIVE) ||
+			    (status & GEN8_CTX_STATUS_PREEMPTED))
+				WRITE_ONCE(execlists->outstanding_submission, 0);
+
 			if (!(status & GEN8_CTX_STATUS_COMPLETED_MASK))
 				continue;
 
@@ -944,9 +956,11 @@ static void execlists_submission_tasklet(unsigned long data)
 			/* After the final element, the hw should be idle */
 			GEM_BUG_ON(port_count(port) == 0 &&
 				   !(status & GEN8_CTX_STATUS_ACTIVE_IDLE));
-			if (port_count(port) == 0)
+			if (port_count(port) == 0) {
 				execlists_clear_active(execlists,
 						       EXECLISTS_ACTIVE_USER);
+				WRITE_ONCE(execlists->outstanding_submission, 0);
+			}
 		}
 
 		if (head != execlists->csb_head) {
drivers/gpu/drm/i915/intel_ringbuffer.h +6 -0
index e5c62f8ef0da..2c8e1a74c266 100644
@@ -288,6 +288,12 @@ struct intel_engine_execlists {
 	 * @csb_use_mmio: access csb through mmio, instead of hwsp
 	 */
 	bool csb_use_mmio;
+
+	/**
+	 * @outstanding_submission: prevent further ELSP writes until HW
+	 * has ack'd one (with IDLE_ACTIVE or PREEMPT CSB events)
+	 */
+	bool outstanding_submission;
 };
 
 #define INTEL_ENGINE_CS_MAX_NAME 8
Signature · 6 lines
2.15.0

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
#1 lore raw

Quoting Michel Thierry (2017-11-18 00:30:38)

Quoted 52 lines

The hardware needs some time to process the information received in the ExecList Submission Port, and expects us to don't write anything new until it has 'acknowledged' this new execlist by sending an IDLE_ACTIVE or PREEMPTED CSB event.

If we do not follow this, the driver could write new data into the ELSP before HW had finishing fetching the previous one, putting us in 'undefined behaviour' space.

This seems to be the problem causing the spurious PREEMPTED & COMPLETE events after a COMPLETE like the one below:

[] vcs0: sw rd pointer = 2, hw wr pointer = 0, current 'head' = 3.

[] vcs0:  Execlist CSB[0]: 0x00000018 _ 0x00000007
[] vcs0:  Execlist CSB[1]: 0x00000001 _ 0x00000000
[] vcs0:  Execlist CSB[2]: 0x00000018 _ 0x00000007  <<< COMPLETE
[] vcs0:  Execlist CSB[3]: 0x00000012 _ 0x00000007  <<< PREEMPTED & COMPLETE
[] vcs0:  Execlist CSB[4]: 0x00008002 _ 0x00000006
[] vcs0:  Execlist CSB[5]: 0x00000014 _ 0x00000006

The ELSP writes that lead to this CSB sequence show that the HW hadn't started executing the previous execlist (the one with only ctx 0x6) by the time the new one was submitted; this is a bit more clear in the data show in the EXECLIST_STATUS register at the time of the ELSP write.

[] vcs0: ELSP[0] = 0x0_0 [execlist1] - status_reg = 0x0_302
[] vcs0: ELSP[1] = 0x6_fedb2119 [execlist0] - status_reg = 0x0_8302

[] vcs0: ELSP[2] = 0x7_fedaf119 [execlist1] - status_reg = 0x0_8308 [] vcs0: ELSP[3] = 0x6_fedb2119 [execlist0] - status_reg = 0x7_8308

Note that having to wait for this ack does not disable lite-restores, although it may reduce their numbers.

And take this as a RFC, since there are probably better ways to still respect this HW requirement.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=102035

---
 drivers/gpu/drm/i915/intel_lrc.c        | 16 +++++++++++++++-
 drivers/gpu/drm/i915/intel_ringbuffer.h |  6 ++++++
 2 files changed, 21 insertions(+), 1 deletion(-)
drivers/gpu/drm/i915/intel_lrc.c +0 -0
index af41165e3da4..10b7eb64f169 100644
@@ -449,11 +449,16 @@ static inline void elsp_write(u64 desc, u32 __iomem *elsp)
 
 static void execlists_submit_ports(struct intel_engine_cs *engine)

drivers/gpu/drm/i915/intel_lrc.c static inline void elsp_write(u64 desc, u32 __iomem *elsp) · line 460

Quoted 10 lines
drivers/gpu/drm/i915/intel_lrc.c +4 -0
 {
+       struct intel_engine_execlists * const execlists = &engine->execlists;
        struct execlist_port *port = engine->execlists.port;
        u32 __iomem *elsp =
                engine->i915->regs + i915_mmio_reg_offset(RING_ELSP(engine));
        unsigned int n;
 
+       if (wait_for_atomic(!READ_ONCE(execlists->outstanding_submission), 10))
+               GEM_TRACE("%s outstanding submission stuck\n",
+                         engine->name);

That can never succeed. Processing events and submitting ports is serialised to the same thread. All the READ_ONCE/WRITE_ONCE are incorrect.

I actually did try tracking something like this; the problem is that we do not always get the IDLE_ACTIVE indicator so actually tracking when the hw is awake is tricky.
-Chris
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

#2 lore raw

✓ Fi.CI.BAT: success for drm/i915/lrc: Stop writing to ELSP until HW has processed the previous write

In reply to Michel Thierry
== Series Details ==

== Series Details ==

Series: drm/i915/lrc: Stop writing to ELSP until HW has processed the previous write
URL : https://patchwork.freedesktop.org/series/34043/
State : success

== Summary ==

Series 34043v1 drm/i915/lrc: Stop writing to ELSP until HW has processed the previous write
https://patchwork.freedesktop.org/api/1.0/series/34043/revisions/1/mbox/

fi-bdw-5557u     total:289  pass:268  dwarn:0   dfail:0   fail:0   skip:21  time:446s
fi-bdw-gvtdvm    total:289  pass:265  dwarn:0   dfail:0   fail:0   skip:24  time:453s
fi-blb-e6850     total:289  pass:223  dwarn:1   dfail:0   fail:0   skip:65  time:383s
fi-bsw-n3050     total:289  pass:243  dwarn:0   dfail:0   fail:0   skip:46  time:549s
fi-bwr-2160      total:289  pass:183  dwarn:0   dfail:0   fail:0   skip:106 time:279s
fi-bxt-dsi       total:289  pass:259  dwarn:0   dfail:0   fail:0   skip:30  time:542s
fi-bxt-j4205     total:289  pass:260  dwarn:0   dfail:0   fail:0   skip:29  time:545s
fi-byt-j1900     total:289  pass:254  dwarn:0   dfail:0   fail:0   skip:35  time:505s
fi-byt-n2820     total:289  pass:250  dwarn:0   dfail:0   fail:0   skip:39  time:488s
fi-elk-e7500     total:289  pass:229  dwarn:0   dfail:0   fail:0   skip:60  time:428s
fi-gdg-551       total:289  pass:178  dwarn:1   dfail:0   fail:1   skip:109 time:267s
fi-glk-1         total:289  pass:261  dwarn:0   dfail:0   fail:0   skip:28  time:581s
fi-hsw-4770      total:289  pass:262  dwarn:0   dfail:0   fail:0   skip:27  time:435s
fi-hsw-4770r     total:289  pass:262  dwarn:0   dfail:0   fail:0   skip:27  time:440s
fi-ilk-650       total:289  pass:228  dwarn:0   dfail:0   fail:0   skip:61  time:430s
fi-ivb-3520m     total:289  pass:260  dwarn:0   dfail:0   fail:0   skip:29  time:495s
fi-ivb-3770      total:289  pass:260  dwarn:0   dfail:0   fail:0   skip:29  time:464s
fi-kbl-7500u     total:289  pass:264  dwarn:1   dfail:0   fail:0   skip:24  time:525s
fi-kbl-7560u     total:289  pass:270  dwarn:0   dfail:0   fail:0   skip:19  time:588s
fi-kbl-7567u     total:289  pass:269  dwarn:0   dfail:0   fail:0   skip:20  time:531s
fi-kbl-r         total:289  pass:262  dwarn:0   dfail:0   fail:0   skip:27  time:573s
fi-pnv-d510      total:289  pass:222  dwarn:1   dfail:0   fail:0   skip:66  time:579s
fi-skl-6260u     total:289  pass:269  dwarn:0   dfail:0   fail:0   skip:20  time:506s
fi-skl-6600u     total:289  pass:262  dwarn:0   dfail:0   fail:0   skip:27  time:590s
fi-skl-6700hq    total:289  pass:263  dwarn:0   dfail:0   fail:0   skip:26  time:605s
fi-skl-6700k     total:289  pass:265  dwarn:0   dfail:0   fail:0   skip:24  time:568s
fi-skl-6770hq    total:289  pass:269  dwarn:0   dfail:0   fail:0   skip:20  time:539s
fi-skl-gvtdvm    total:289  pass:266  dwarn:0   dfail:0   fail:0   skip:23  time:513s
fi-snb-2520m     total:246  pass:212  dwarn:0   dfail:0   fail:0   skip:33 
fi-snb-2600      total:289  pass:249  dwarn:0   dfail:0   fail:0   skip:40  time:427s
Blacklisted hosts:
fi-cfl-s2        total:289  pass:263  dwarn:0   dfail:0   fail:0   skip:26  time:650s
fi-cnl-y         total:241  pass:216  dwarn:0   dfail:0   fail:0   skip:24 
fi-glk-dsi       total:289  pass:258  dwarn:0   dfail:0   fail:1   skip:30  time:533s

72ffd011e980c8131f0c63e08c576d09977a2c54 drm-tip: 2017y-11m-17d-18h-38m-47s UTC integration manifest 1012245361ae drm/i915/lrc: Stop writing to ELSP until HW has processed the previous write

== Logs ==

For more details see: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_7182/
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

#3 lore raw

✗ Fi.CI.IGT: warning for drm/i915/lrc: Stop writing to ELSP until HW has processed the previous write

In reply to Michel Thierry
== Series Details ==

== Series Details ==

Series: drm/i915/lrc: Stop writing to ELSP until HW has processed the previous write
URL : https://patchwork.freedesktop.org/series/34043/
State : warning

== Summary ==

Warning: bzip CI_DRM_3359/shard-glkb6/results1.json.bz2 wasn't in correct JSON format Test kms_flip:

        Subgroup flip-vs-modeset-interruptible:
                dmesg-warn -> PASS       (shard-hsw)
Test kms_cursor_crc:
        Subgroup cursor-128x128-sliding:
                pass       -> SKIP       (shard-hsw)
Test pm_rpm:
        Subgroup basic-rte:
                pass       -> SKIP       (shard-hsw)
shard-hsw        total:2585 pass:1473 dwarn:1   dfail:1   fail:9   skip:1101 time:9405s
shard-snb        total:2585 pass:1261 dwarn:1   dfail:1   fail:11  skip:1311 time:8013s
Blacklisted hosts:
shard-apl        total:2547 pass:1591 dwarn:2   dfail:0   fail:23  skip:929 time:13018s
shard-kbl        total:2442 pass:1629 dwarn:4   dfail:0   fail:23  skip:783 time:10521s

== Logs ==

For more details see: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_7182/shards.html
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx