From f61145f1a4bd7966aa0b15c5cd3950835b284f55 Mon Sep 17 00:00:00 2001
From: Eric Anholt <eric@anholt.net>
Date: Thu, 21 Dec 2017 14:17:22 -0800
Subject: [PATCH 1/3] drm/vc4: Flush the caches before the bin jobs, as well.
If the frame samples from a render target that was just written, its
cache flush during the binning step may have occurred before the
previous frame's RCL was completed. Flush the texture caches again
before starting each RCL job to make sure that the sampling of the
previous RCL's output is correct.
Fixes flickering in the top left of 3DMMES Taiji.
Signed-off-by: Eric Anholt <eric@anholt.net>
Fixes: ca26d28bbaa3 ("drm/vc4: improve throughput by pipelining binning and rendering jobs")
Link: https://patchwork.freedesktop.org/patch/msgid/20171221221722.23809-1-eric@anholt.net
Reviewed-by: Boris Brezillon <boris.brezillon@free-electrons.com>
---
drivers/gpu/drm/vc4/vc4_gem.c | 21 +++++++++++++++++++++
1 file changed, 21 insertions(+)
diff --git a/drivers/gpu/drm/vc4/vc4_gem.c b/drivers/gpu/drm/vc4/vc4_gem.c
index 638540943c61..e3e868cdee79 100644
--- a/drivers/gpu/drm/vc4/vc4_gem.c
+++ b/drivers/gpu/drm/vc4/vc4_gem.c
@@ -436,6 +436,19 @@ vc4_flush_caches(struct drm_device *dev)
VC4_SET_FIELD(0xf, V3D_SLCACTL_ICC));
}
+static void
+vc4_flush_texture_caches(struct drm_device *dev)
+{
+ struct vc4_dev *vc4 = to_vc4_dev(dev);
+
+ V3D_WRITE(V3D_L2CACTL,
+ V3D_L2CACTL_L2CCLR);
+
+ V3D_WRITE(V3D_SLCACTL,
+ VC4_SET_FIELD(0xf, V3D_SLCACTL_T1CC) |
+ VC4_SET_FIELD(0xf, V3D_SLCACTL_T0CC));
+}
+
/* Sets the registers for the next job to be actually be executed in
* the hardware.
*
@@ -474,6 +487,14 @@ vc4_submit_next_render_job(struct drm_device *dev)
if (!exec)
return;
+ /* A previous RCL may have written to one of our textures, and
+ * our full cache flush at bin time may have occurred before
+ * that RCL completed. Flush the texture cache now, but not
+ * the instructions or uniforms (since we don't write those
+ * from an RCL).
+ */
+ vc4_flush_texture_caches(dev);
+
submit_cl(dev, 1, exec->ct1ca, exec->ct1ea);
}
From 17b11b76b87afe9f8be199d7a5f442497133e2b0 Mon Sep 17 00:00:00 2001
From: Boris Brezillon <boris.brezillon@free-electrons.com>
Date: Thu, 18 Jan 2018 15:58:21 +0100
Subject: [PATCH 2/3] drm/vc4: Fix NULL pointer dereference in
vc4_save_hang_state()
When saving BOs in the hang state we skip one entry of the
kernel_state->bo[] array, thus leaving it to NULL. This leads to a NULL
pointer dereference when, later in this function, we iterate over all
BOs to check their ->madv state.
Fixes: ca26d28bbaa3 ("drm/vc4: improve throughput by pipelining binning and rendering jobs")
Cc: <stable@vger.kernel.org>
Signed-off-by: Boris Brezillon <boris.brezillon@free-electrons.com>
Signed-off-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Eric Anholt <eric@anholt.net>
Link: https://patchwork.freedesktop.org/patch/msgid/20180118145821.22344-1-boris.brezillon@free-electrons.com
---
drivers/gpu/drm/vc4/vc4_gem.c | 12 ++++++------
1 file changed, 6 insertions(+), 6 deletions(-)
diff --git a/drivers/gpu/drm/vc4/vc4_gem.c b/drivers/gpu/drm/vc4/vc4_gem.c
index e3e868cdee79..c94cce96544c 100644
--- a/drivers/gpu/drm/vc4/vc4_gem.c
+++ b/drivers/gpu/drm/vc4/vc4_gem.c
@@ -146,7 +146,7 @@ vc4_save_hang_state(struct drm_device *dev)
struct vc4_exec_info *exec[2];
struct vc4_bo *bo;
unsigned long irqflags;
- unsigned int i, j, unref_list_count, prev_idx;
+ unsigned int i, j, k, unref_list_count;
kernel_state = kcalloc(1, sizeof(*kernel_state), GFP_KERNEL);
if (!kernel_state)
@@ -182,7 +182,7 @@ vc4_save_hang_state(struct drm_device *dev)
return;
}
- prev_idx = 0;
+ k = 0;
for (i = 0; i < 2; i++) {
if (!exec[i])
continue;
@@ -197,7 +197,7 @@ vc4_save_hang_state(struct drm_device *dev)
WARN_ON(!refcount_read(&bo->usecnt));
refcount_inc(&bo->usecnt);
drm_gem_object_get(&exec[i]->bo[j]->base);
- kernel_state->bo[j + prev_idx] = &exec[i]->bo[j]->base;
+ kernel_state->bo[k++] = &exec[i]->bo[j]->base;
}
list_for_each_entry(bo, &exec[i]->unref_list, unref_head) {
@@ -205,12 +205,12 @@ vc4_save_hang_state(struct drm_device *dev)
* because they are naturally unpurgeable.
*/
drm_gem_object_get(&bo->base.base);
- kernel_state->bo[j + prev_idx] = &bo->base.base;
- j++;
+ kernel_state->bo[k++] = &bo->base.base;
}
- prev_idx = j + 1;
}
+ WARN_ON_ONCE(k != state->bo_count);
+
if (exec[0])
state->start_bin = exec[0]->ct0ca;
if (exec[1])
From 0fd189a95fdbc631737df5f27a0fc0a3dd31b75e Mon Sep 17 00:00:00 2001
From: Lyude Paul <lyude@redhat.com>
Date: Thu, 25 Jan 2018 18:29:53 -0500
Subject: [PATCH 3/3] drm/nouveau: Move irq setup/teardown to pci ctor/dtor
For a while we've been having issues with seemingly random interrupts
coming from nvidia cards when resuming them. Originally the fix for this
was thought to be just re-arming the MSI interrupt registers right after
re-allocating our IRQs, however it seems a lot of what we do is both
wrong and not even nessecary.
This was made apparent by what appeared to be a regression in the
mainline kernel that started introducing suspend/resume issues for
nouveau:
a0c9259dc4e1 (irq/matrix: Spread interrupts on allocation)
After this commit was introduced, we started getting interrupts from the
GPU before we actually re-allocated our own IRQ (see references below)
and assigned the IRQ handler. Investigating this turned out that the
problem was not with the commit, but the fact that nouveau even
free/allocates it's irqs before and after suspend/resume.
For starters: drivers in the linux kernel haven't had to handle
freeing/re-allocating their IRQs during suspend/resume cycles for quite
a while now. Nouveau seems to be one of the few drivers left that still
does this, despite the fact there's no reason we actually need to since
disabling interrupts from the device side should be enough, as the
kernel is already smart enough to know to disable host-side interrupts
for us before going into suspend. Since we were tearing down our IRQs by
hand however, that means there was a short period during resume where
interrupts could be received before we re-allocated our IRQ which would
lead to us getting an unhandled IRQ. Since we never handle said IRQ and
re-arm the interrupt registers, this would cause us to miss all of the
interrupts from the GPU and cause our init process to start timing out
on anything requiring interrupts.
So, since this whole setup/teardown every suspend/resume cycle is
useless anyway, move irq setup/teardown into the pci subdev's ctor/dtor
functions instead so they're only called at driver load and driver
unload. This should fix most of the issues with pending interrupts on
resume, along with getting suspend/resume for nouveau to work again.
As well, this probably means we can also just remove the msi rearm call
inside nvkm_pci_init(). But since our main focus here is to fix
suspend/resume before 4.15, we'll save that for a later patch.
Signed-off-by: Lyude Paul <lyude@redhat.com>
Cc: Karol Herbst <kherbst@redhat.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Mike Galbraith <efault@gmx.de>
Cc: stable@vger.kernel.org
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
---
.../gpu/drm/nouveau/nvkm/subdev/pci/base.c | 46 +++++++++++++------
1 file changed, 31 insertions(+), 15 deletions(-)
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/pci/base.c b/drivers/gpu/drm/nouveau/nvkm/subdev/pci/base.c
index deb96de54b00..ee2431a7804e 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/pci/base.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/pci/base.c
@@ -71,6 +71,10 @@ nvkm_pci_intr(int irq, void *arg)
struct nvkm_pci *pci = arg;
struct nvkm_device *device = pci->subdev.device;
bool handled = false;
+
+ if (pci->irq < 0)
+ return IRQ_HANDLED;
+
nvkm_mc_intr_unarm(device);
if (pci->msi)
pci->func->msi_rearm(pci);
@@ -84,11 +88,6 @@ nvkm_pci_fini(struct nvkm_subdev *subdev, bool suspend)
{
struct nvkm_pci *pci = nvkm_pci(subdev);
- if (pci->irq >= 0) {
- free_irq(pci->irq, pci);
- pci->irq = -1;
- }
-
if (pci->agp.bridge)
nvkm_agp_fini(pci);
@@ -108,8 +107,20 @@ static int
nvkm_pci_oneinit(struct nvkm_subdev *subdev)
{
struct nvkm_pci *pci = nvkm_pci(subdev);
- if (pci_is_pcie(pci->pdev))
- return nvkm_pcie_oneinit(pci);
+ struct pci_dev *pdev = pci->pdev;
+ int ret;
+
+ if (pci_is_pcie(pci->pdev)) {
+ ret = nvkm_pcie_oneinit(pci);
+ if (ret)
+ return ret;
+ }
+
+ ret = request_irq(pdev->irq, nvkm_pci_intr, IRQF_SHARED, "nvkm", pci);
+ if (ret)
+ return ret;
+
+ pci->irq = pdev->irq;
return 0;
}
@@ -117,7 +128,6 @@ static int
nvkm_pci_init(struct nvkm_subdev *subdev)
{
struct nvkm_pci *pci = nvkm_pci(subdev);
- struct pci_dev *pdev = pci->pdev;
int ret;
if (pci->agp.bridge) {
@@ -131,28 +141,34 @@ nvkm_pci_init(struct nvkm_subdev *subdev)
if (pci->func->init)
pci->func->init(pci);
- ret = request_irq(pdev->irq, nvkm_pci_intr, IRQF_SHARED, "nvkm", pci);
- if (ret)
- return ret;
-
- pci->irq = pdev->irq;
-
/* Ensure MSI interrupts are armed, for the case where there are
* already interrupts pending (for whatever reason) at load time.
*/
if (pci->msi)
pci->func->msi_rearm(pci);
- return ret;
+ return 0;
}
static void *
nvkm_pci_dtor(struct nvkm_subdev *subdev)
{
struct nvkm_pci *pci = nvkm_pci(subdev);
+
nvkm_agp_dtor(pci);
+
+ if (pci->irq >= 0) {
+ /* freq_irq() will call the handler, we use pci->irq == -1
+ * to signal that it's been torn down and should be a noop.
+ */
+ int irq = pci->irq;
+ pci->irq = -1;
+ free_irq(irq, pci);
+ }
+
if (pci->msi)
pci_disable_msi(pci->pdev);
+
return nvkm_pci(subdev);
}