From: Mikko Perttunen <cyndis@kapsi.fi>
To: Rob Herring <robh+dt@kernel.org>,
Krzysztof Kozlowski <krzysztof.kozlowski+dt@linaro.org>,
Thierry Reding <thierry.reding@gmail.com>,
Jonathan Hunter <jonathanh@nvidia.com>,
Dmitry Osipenko <digetx@gmail.com>
Cc: Mikko Perttunen <mperttunen@nvidia.com>,
Robin Murphy <robin.murphy@arm.com>,
dri-devel@lists.freedesktop.org, devicetree@vger.kernel.org,
linux-tegra@vger.kernel.org, linux-kernel@vger.kernel.org
Subject: [PATCH v7/v3 19/22] gpu: host1x: Rewrite job opcode sequence
Date: Mon, 27 Jun 2022 17:20:05 +0300 [thread overview]
Message-ID: <20220627142008.2072474-20-cyndis@kapsi.fi> (raw)
In-Reply-To: <20220627142008.2072474-1-cyndis@kapsi.fi>
From: Mikko Perttunen <mperttunen@nvidia.com>
For new (Tegra186+) SoCs, use a new ('full-featured') job opcode
sequence that is compatible with virtualization. In particular,
the Host1x hardware in Tegra234 is more strict regarding the sequence,
requiring ACQUIRE_MLOCK-SETCLASS-SETSTREAMID opcodes to occur in
that sequence without gaps (except for SETPAYLOAD), so let's do it
properly in one go now.
Signed-off-by: Mikko Perttunen <mperttunen@nvidia.com>
---
drivers/gpu/host1x/hw/channel_hw.c | 144 +++++++++++++++++------------
1 file changed, 85 insertions(+), 59 deletions(-)
diff --git a/drivers/gpu/host1x/hw/channel_hw.c b/drivers/gpu/host1x/hw/channel_hw.c
index f84caf06621a..4eb7fb2e4f0a 100644
--- a/drivers/gpu/host1x/hw/channel_hw.c
+++ b/drivers/gpu/host1x/hw/channel_hw.c
@@ -47,10 +47,41 @@ static void trace_write_gather(struct host1x_cdma *cdma, struct host1x_bo *bo,
}
}
-static void submit_wait(struct host1x_cdma *cdma, u32 id, u32 threshold,
+static void submit_wait(struct host1x_job *job, u32 id, u32 threshold,
u32 next_class)
{
-#if HOST1X_HW >= 2
+ struct host1x_cdma *cdma = &job->channel->cdma;
+
+#if HOST1X_HW >= 6
+ u32 stream_id;
+
+ /*
+ * If a memory context has been set, use it. Otherwise
+ * (if context isolation is disabled) use the engine's
+ * firmware stream ID.
+ */
+ if (job->memory_context)
+ stream_id = job->memory_context->stream_id;
+ else
+ stream_id = job->engine_fallback_streamid;
+
+ host1x_cdma_push_wide(cdma,
+ host1x_opcode_setclass(
+ HOST1X_CLASS_HOST1X,
+ HOST1X_UCLASS_LOAD_SYNCPT_PAYLOAD_32,
+ /* WAIT_SYNCPT_32 is at SYNCPT_PAYLOAD_32+2 */
+ BIT(0) | BIT(2)
+ ),
+ threshold,
+ id,
+ HOST1X_OPCODE_NOP
+ );
+ host1x_cdma_push_wide(&job->channel->cdma,
+ host1x_opcode_setclass(job->class, 0, 0),
+ host1x_opcode_setpayload(stream_id),
+ host1x_opcode_setstreamid(job->engine_streamid_offset / 4),
+ HOST1X_OPCODE_NOP);
+#elif HOST1X_HW >= 2
host1x_cdma_push_wide(cdma,
host1x_opcode_setclass(
HOST1X_CLASS_HOST1X,
@@ -97,7 +128,7 @@ static void submit_gathers(struct host1x_job *job, u32 job_syncpt_base)
else
threshold = cmd->wait.threshold;
- submit_wait(cdma, cmd->wait.id, threshold, cmd->wait.next_class);
+ submit_wait(job, cmd->wait.id, threshold, cmd->wait.next_class);
} else {
struct host1x_job_gather *g = &cmd->gather;
@@ -180,42 +211,70 @@ static void host1x_enable_gather_filter(struct host1x_channel *ch)
#endif
}
-static void host1x_channel_program_engine_streamid(struct host1x_job *job)
+static void channel_program_cdma(struct host1x_job *job)
{
+ struct host1x_cdma *cdma = &job->channel->cdma;
+ struct host1x_syncpt *sp = job->syncpt;
+
#if HOST1X_HW >= 6
u32 fence;
- if (!job->memory_context)
- return;
+ /* Enter engine class with invalid stream ID. */
+ host1x_cdma_push_wide(cdma,
+ host1x_opcode_acquire_mlock(job->class),
+ host1x_opcode_setclass(job->class, 0, 0),
+ host1x_opcode_setpayload(0),
+ host1x_opcode_setstreamid(job->engine_streamid_offset / 4));
- fence = host1x_syncpt_incr_max(job->syncpt, 1);
+ /* Before switching stream ID to real stream ID, ensure engine is idle. */
+ fence = host1x_syncpt_incr_max(sp, 1);
+ host1x_cdma_push(&job->channel->cdma,
+ host1x_opcode_nonincr(HOST1X_UCLASS_INCR_SYNCPT, 1),
+ HOST1X_UCLASS_INCR_SYNCPT_INDX_F(job->syncpt->id) |
+ HOST1X_UCLASS_INCR_SYNCPT_COND_F(4));
+ submit_wait(job, job->syncpt->id, fence, job->class);
- /* First, increment a syncpoint on OP_DONE condition.. */
+ /* Submit work. */
+ job->syncpt_end = host1x_syncpt_incr_max(sp, job->syncpt_incrs);
+ submit_gathers(job, job->syncpt_end - job->syncpt_incrs);
+ /* Before releasing MLOCK, ensure engine is idle again. */
+ fence = host1x_syncpt_incr_max(sp, 1);
host1x_cdma_push(&job->channel->cdma,
host1x_opcode_nonincr(HOST1X_UCLASS_INCR_SYNCPT, 1),
HOST1X_UCLASS_INCR_SYNCPT_INDX_F(job->syncpt->id) |
- HOST1X_UCLASS_INCR_SYNCPT_COND_F(1));
+ HOST1X_UCLASS_INCR_SYNCPT_COND_F(4));
+ submit_wait(job, job->syncpt->id, fence, job->class);
- /* Wait for syncpoint to increment */
+ /* Release MLOCK. */
+ host1x_cdma_push(cdma,
+ HOST1X_OPCODE_NOP, host1x_opcode_release_mlock(job->class));
+#else
+ if (job->serialize) {
+ /*
+ * Force serialization by inserting a host wait for the
+ * previous job to finish before this one can commence.
+ */
+ host1x_cdma_push(cdma,
+ host1x_opcode_setclass(HOST1X_CLASS_HOST1X,
+ host1x_uclass_wait_syncpt_r(), 1),
+ host1x_class_host_wait_syncpt(job->syncpt->id,
+ host1x_syncpt_read_max(sp)));
+ }
- host1x_cdma_push(&job->channel->cdma,
- host1x_opcode_setclass(HOST1X_CLASS_HOST1X,
- host1x_uclass_wait_syncpt_r(), 1),
- host1x_class_host_wait_syncpt(job->syncpt->id, fence));
+ /* Synchronize base register to allow using it for relative waiting */
+ if (sp->base)
+ synchronize_syncpt_base(job);
- /*
- * Now that we know the engine is idle, return to class and
- * change stream ID.
- */
+ /* add a setclass for modules that require it */
+ if (job->class)
+ host1x_cdma_push(cdma,
+ host1x_opcode_setclass(job->class, 0, 0),
+ HOST1X_OPCODE_NOP);
- host1x_cdma_push(&job->channel->cdma,
- host1x_opcode_setclass(job->class, 0, 0),
- HOST1X_OPCODE_NOP);
+ job->syncpt_end = host1x_syncpt_incr_max(sp, job->syncpt_incrs);
- host1x_cdma_push(&job->channel->cdma,
- host1x_opcode_setpayload(job->memory_context->stream_id),
- host1x_opcode_setstreamid(job->engine_streamid_offset / 4));
+ submit_gathers(job, job->syncpt_end - job->syncpt_incrs);
#endif
}
@@ -223,7 +282,6 @@ static int channel_submit(struct host1x_job *job)
{
struct host1x_channel *ch = job->channel;
struct host1x_syncpt *sp = job->syncpt;
- u32 user_syncpt_incrs = job->syncpt_incrs;
u32 prev_max = 0;
u32 syncval;
int err;
@@ -251,6 +309,7 @@ static int channel_submit(struct host1x_job *job)
host1x_channel_set_streamid(ch);
host1x_enable_gather_filter(ch);
+ host1x_hw_syncpt_assign_to_channel(host, sp, ch);
/* begin a CDMA submit */
err = host1x_cdma_begin(&ch->cdma, job);
@@ -259,40 +318,7 @@ static int channel_submit(struct host1x_job *job)
goto error;
}
- if (job->serialize) {
- /*
- * Force serialization by inserting a host wait for the
- * previous job to finish before this one can commence.
- */
- host1x_cdma_push(&ch->cdma,
- host1x_opcode_setclass(HOST1X_CLASS_HOST1X,
- host1x_uclass_wait_syncpt_r(), 1),
- host1x_class_host_wait_syncpt(job->syncpt->id,
- host1x_syncpt_read_max(sp)));
- }
-
- /* Synchronize base register to allow using it for relative waiting */
- if (sp->base)
- synchronize_syncpt_base(job);
-
- host1x_hw_syncpt_assign_to_channel(host, sp, ch);
-
- /* add a setclass for modules that require it */
- if (job->class)
- host1x_cdma_push(&ch->cdma,
- host1x_opcode_setclass(job->class, 0, 0),
- HOST1X_OPCODE_NOP);
-
- /*
- * Ensure engine DMA is idle and set new stream ID. May increment
- * syncpt max.
- */
- host1x_channel_program_engine_streamid(job);
-
- syncval = host1x_syncpt_incr_max(sp, user_syncpt_incrs);
- job->syncpt_end = syncval;
-
- submit_gathers(job, syncval - user_syncpt_incrs);
+ channel_program_cdma(job);
/* end CDMA submit & stash pinned hMems into sync queue */
host1x_cdma_end(&ch->cdma, job);
--
2.36.1
next prev parent reply other threads:[~2022-06-27 14:20 UTC|newest]
Thread overview: 26+ messages / expand[flat|nested] mbox.gz Atom feed top
2022-06-27 14:19 [PATCH v7/v3 00/22] Host1x context isolation / Tegra234 support Mikko Perttunen
2022-06-27 14:19 ` [PATCH v7/v3 01/22] dt-bindings: host1x: Add iommu-map property Mikko Perttunen
2022-06-27 14:19 ` [PATCH v7/v3 02/22] gpu: host1x: Add context device management code Mikko Perttunen
2022-06-27 14:19 ` [PATCH v7/v3 03/22] gpu: host1x: Program context stream ID on submission Mikko Perttunen
2022-06-27 14:19 ` [PATCH v7/v3 04/22] arm64: tegra: Add Host1x context stream IDs on Tegra186+ Mikko Perttunen
2022-06-27 14:19 ` [PATCH v7/v3 05/22] drm/tegra: falcon: Set DMACTX field on DMA transactions Mikko Perttunen
2022-06-27 14:19 ` [PATCH v7/v3 06/22] drm/tegra: nvdec: Fix TRANSCFG register offset Mikko Perttunen
2022-06-27 14:19 ` [PATCH v7/v3 07/22] drm/tegra: Support context isolation Mikko Perttunen
2022-06-27 14:19 ` [PATCH v7/v3 08/22] drm/tegra: Implement stream ID related callbacks on engines Mikko Perttunen
2022-06-27 14:19 ` [PATCH v7/v3 09/22] dt-bindings: Add bindings for Tegra234 Host1x and VIC Mikko Perttunen
2022-06-30 15:20 ` Rob Herring
2022-06-27 14:19 ` [PATCH v7/v3 10/22] dt-bindings: host1x: Fix bracketing in example Mikko Perttunen
2022-06-30 15:20 ` Rob Herring
2022-06-27 14:19 ` [PATCH v7/v3 11/22] dt-bindings: Add headers for Host1x and VIC on Tegra234 Mikko Perttunen
2022-06-27 14:19 ` [PATCH v7/v3 12/22] arm64: tegra: Add " Mikko Perttunen
2022-06-27 14:19 ` [PATCH v7/v3 13/22] gpu: host1x: Deduplicate hardware headers Mikko Perttunen
2022-06-27 14:20 ` [PATCH v7/v3 14/22] gpu: host1x: Simplify register mapping and add common aperture Mikko Perttunen
2022-06-27 14:20 ` [PATCH v7/v3 15/22] gpu: host1x: Program virtualization tables Mikko Perttunen
2022-06-27 14:20 ` [PATCH v7/v3 16/22] gpu: host1x: Allow reset to be missing Mikko Perttunen
2022-06-27 14:20 ` [PATCH v7/v3 17/22] gpu: host1x: Program interrupt destinations on Tegra234 Mikko Perttunen
2022-06-27 14:20 ` [PATCH v7/v3 18/22] gpu: host1x: Tegra234 device data and headers Mikko Perttunen
2022-06-27 14:20 ` Mikko Perttunen [this message]
2022-06-27 14:20 ` [PATCH v7/v3 20/22] gpu: host1x: Add MLOCK release code on Tegra234 Mikko Perttunen
2022-06-27 14:20 ` [PATCH v7/v3 21/22] gpu: host1x: Use RESTART_W to skip timed out jobs on Tegra186+ Mikko Perttunen
2022-06-27 14:20 ` [PATCH v7/v3 22/22] drm/tegra: vic: Add Tegra234 support Mikko Perttunen
2022-06-30 15:19 ` [PATCH v7/v3 00/22] Host1x context isolation / " Rob Herring
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20220627142008.2072474-20-cyndis@kapsi.fi \
--to=cyndis@kapsi.fi \
--cc=devicetree@vger.kernel.org \
--cc=digetx@gmail.com \
--cc=dri-devel@lists.freedesktop.org \
--cc=jonathanh@nvidia.com \
--cc=krzysztof.kozlowski+dt@linaro.org \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-tegra@vger.kernel.org \
--cc=mperttunen@nvidia.com \
--cc=robh+dt@kernel.org \
--cc=robin.murphy@arm.com \
--cc=thierry.reding@gmail.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).