From e53e004e346062e15df9511bd4b5a19e34701384 Mon Sep 17 00:00:00 2001 From: Karol Wachowski Date: Wed, 16 Apr 2025 12:26:16 +0200 Subject: [PATCH 01/16] accel/ivpu: Correct DCT interrupt handling Fix improper use of dct_active_percent field in DCT interrupt handler causing DCT to never get enabled. Set dct_active_percent internally before IPC to ensure correct driver value even if IPC fails. Set default DCT value to 30 accordingly to HW architecture specification. Fixes: a19bffb10c46 ("accel/ivpu: Implement DCT handling") Signed-off-by: Karol Wachowski Signed-off-by: Maciej Falkowski Reviewed-by: Jeff Hugo Signed-off-by: Jacek Lawrynowicz Link: https://lore.kernel.org/r/20250416102616.384577-1-maciej.falkowski@linux.intel.com --- drivers/accel/ivpu/ivpu_hw_btrs.h | 2 +- drivers/accel/ivpu/ivpu_pm.c | 18 ++++++++++-------- 2 files changed, 11 insertions(+), 9 deletions(-) diff --git a/drivers/accel/ivpu/ivpu_hw_btrs.h b/drivers/accel/ivpu/ivpu_hw_btrs.h index 300f749971d4..d2d82651976d 100644 --- a/drivers/accel/ivpu/ivpu_hw_btrs.h +++ b/drivers/accel/ivpu/ivpu_hw_btrs.h @@ -14,7 +14,7 @@ #define PLL_PROFILING_FREQ_DEFAULT 38400000 #define PLL_PROFILING_FREQ_HIGH 400000000 -#define DCT_DEFAULT_ACTIVE_PERCENT 15u +#define DCT_DEFAULT_ACTIVE_PERCENT 30u #define DCT_PERIOD_US 35300u int ivpu_hw_btrs_info_init(struct ivpu_device *vdev); diff --git a/drivers/accel/ivpu/ivpu_pm.c b/drivers/accel/ivpu/ivpu_pm.c index b5891e91f7ab..ac0e22454596 100644 --- a/drivers/accel/ivpu/ivpu_pm.c +++ b/drivers/accel/ivpu/ivpu_pm.c @@ -428,16 +428,17 @@ int ivpu_pm_dct_enable(struct ivpu_device *vdev, u8 active_percent) active_us = (DCT_PERIOD_US * active_percent) / 100; inactive_us = DCT_PERIOD_US - active_us; + vdev->pm->dct_active_percent = active_percent; + + ivpu_dbg(vdev, PM, "DCT requested %u%% (D0: %uus, D0i2: %uus)\n", + active_percent, active_us, inactive_us); + ret = ivpu_jsm_dct_enable(vdev, active_us, inactive_us); if (ret) { ivpu_err_ratelimited(vdev, "Failed to enable DCT: %d\n", ret); return ret; } - vdev->pm->dct_active_percent = active_percent; - - ivpu_dbg(vdev, PM, "DCT set to %u%% (D0: %uus, D0i2: %uus)\n", - active_percent, active_us, inactive_us); return 0; } @@ -445,15 +446,16 @@ int ivpu_pm_dct_disable(struct ivpu_device *vdev) { int ret; + vdev->pm->dct_active_percent = 0; + + ivpu_dbg(vdev, PM, "DCT requested to be disabled\n"); + ret = ivpu_jsm_dct_disable(vdev); if (ret) { ivpu_err_ratelimited(vdev, "Failed to disable DCT: %d\n", ret); return ret; } - vdev->pm->dct_active_percent = 0; - - ivpu_dbg(vdev, PM, "DCT disabled\n"); return 0; } @@ -466,7 +468,7 @@ void ivpu_pm_irq_dct_work_fn(struct work_struct *work) if (ivpu_hw_btrs_dct_get_request(vdev, &enable)) return; - if (vdev->pm->dct_active_percent) + if (enable) ret = ivpu_pm_dct_enable(vdev, DCT_DEFAULT_ACTIVE_PERCENT); else ret = ivpu_pm_dct_disable(vdev); -- 2.51.0 From 759ee400d1d95ac903d81a1a229a117be822d077 Mon Sep 17 00:00:00 2001 From: Andrzej Kacprowski Date: Wed, 16 Apr 2025 12:26:29 +0200 Subject: [PATCH 02/16] accel/ivpu: Fix the D0i2 disable test mode Correct setup of D0i2 disable which was by mistake set up to value 1 and use BIT(1) instead. Fixes: 011529fe8112 ("accel/ivpu: Implement D0i2 disable test mode") Signed-off-by: Andrzej Kacprowski Signed-off-by: Maciej Falkowski Reviewed-by: Jeff Hugo Signed-off-by: Jacek Lawrynowicz Link: https://lore.kernel.org/r/20250416102629.384626-1-maciej.falkowski@linux.intel.com --- drivers/accel/ivpu/ivpu_fw.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/drivers/accel/ivpu/ivpu_fw.c b/drivers/accel/ivpu/ivpu_fw.c index 5e1d709c6a46..ccaaf6c100c0 100644 --- a/drivers/accel/ivpu/ivpu_fw.c +++ b/drivers/accel/ivpu/ivpu_fw.c @@ -544,7 +544,7 @@ static void ivpu_fw_boot_params_print(struct ivpu_device *vdev, struct vpu_boot_ boot_params->d0i3_entry_vpu_ts); ivpu_dbg(vdev, FW_BOOT, "boot_params.system_time_us = %llu\n", boot_params->system_time_us); - ivpu_dbg(vdev, FW_BOOT, "boot_params.power_profile = %u\n", + ivpu_dbg(vdev, FW_BOOT, "boot_params.power_profile = 0x%x\n", boot_params->power_profile); } @@ -646,7 +646,7 @@ void ivpu_fw_boot_params_setup(struct ivpu_device *vdev, struct vpu_boot_params boot_params->d0i3_residency_time_us = 0; boot_params->d0i3_entry_vpu_ts = 0; if (IVPU_WA(disable_d0i2)) - boot_params->power_profile = 1; + boot_params->power_profile |= BIT(1); boot_params->system_time_us = ktime_to_us(ktime_get_real()); wmb(); /* Flush WC buffers after writing bootparams */ -- 2.51.0 From 76047483fe94414edf409dc498498abf346e22f1 Mon Sep 17 00:00:00 2001 From: Sunil Khatri Date: Wed, 23 Apr 2025 09:54:42 +0530 Subject: [PATCH 03/16] drm/ttm: fix the warning for hit_low and evict_low MIME-Version: 1.0 Content-Type: text/plain; charset=utf8 Content-Transfer-Encoding: 8bit fix the below warning messages: ttm/ttm_bo.c:1098: warning: Function parameter or struct member 'hit_low' not described in 'ttm_bo_swapout_walk' ttm/ttm_bo.c:1098: warning: Function parameter or struct member 'evict_low' not described in 'ttm_bo_swapout_walk' Cc: Maarten Lankhorst Cc: Tvrtko Ursulin Signed-off-by: Sunil Khatri Reviewed-by: Maarten Lankhorst Signed-off-by: Christian König Link: https://lore.kernel.org/r/20250423042442.762108-1-sunil.khatri@amd.com --- drivers/gpu/drm/ttm/ttm_bo.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c index 95b86003c50d..5bf3c969907c 100644 --- a/drivers/gpu/drm/ttm/ttm_bo.c +++ b/drivers/gpu/drm/ttm/ttm_bo.c @@ -1093,7 +1093,8 @@ struct ttm_bo_swapout_walk { struct ttm_lru_walk walk; /** @gfp_flags: The gfp flags to use for ttm_tt_swapout() */ gfp_t gfp_flags; - + /** @hit_low: Whether we should attempt to swap BO's with low watermark threshold */ + /** @evict_low: If we cannot swap a bo when @try_low is false (first pass) */ bool hit_low, evict_low; }; -- 2.51.0 From 5b1834d6202f86180e451ad1a2a8a193a1da18fc Mon Sep 17 00:00:00 2001 From: Tvrtko Ursulin Date: Fri, 18 Apr 2025 17:25:12 +0100 Subject: [PATCH 04/16] drm/fdinfo: Protect against driver unbind MIME-Version: 1.0 Content-Type: text/plain; charset=utf8 Content-Transfer-Encoding: 8bit If we unbind a driver from the PCI device with an active DRM client, subsequent read of the fdinfo data associated with the file descriptor in question will not end well. Protect the path with a drm_dev_enter/exit() pair. Signed-off-by: Tvrtko Ursulin Cc: Christian König Cc: Lucas De Marchi Cc: Rodrigo Vivi Cc: Umesh Nerlige Ramappa Reviewed-by: Christian König Fixes: 3f09a0cd4ea3 ("drm: Add common fdinfo helper") Cc: # v6.5+ Signed-off-by: Christian König Link: https://lore.kernel.org/r/20250418162512.72324-1-tvrtko.ursulin@igalia.com --- drivers/gpu/drm/drm_file.c | 6 ++++++ 1 file changed, 6 insertions(+) diff --git a/drivers/gpu/drm/drm_file.c b/drivers/gpu/drm/drm_file.c index c299cd94d3f7..cf2463090d3a 100644 --- a/drivers/gpu/drm/drm_file.c +++ b/drivers/gpu/drm/drm_file.c @@ -964,6 +964,10 @@ void drm_show_fdinfo(struct seq_file *m, struct file *f) struct drm_file *file = f->private_data; struct drm_device *dev = file->minor->dev; struct drm_printer p = drm_seq_file_printer(m); + int idx; + + if (!drm_dev_enter(dev, &idx)) + return; drm_printf(&p, "drm-driver:\t%s\n", dev->driver->name); drm_printf(&p, "drm-client-id:\t%llu\n", file->client_id); @@ -983,6 +987,8 @@ void drm_show_fdinfo(struct seq_file *m, struct file *f) if (dev->driver->show_fdinfo) dev->driver->show_fdinfo(&p, file); + + drm_dev_exit(idx); } EXPORT_SYMBOL(drm_show_fdinfo); -- 2.51.0 From bc43f7114a0e8173968085b21535d57b8030d571 Mon Sep 17 00:00:00 2001 From: Janne Grunau Date: Mon, 28 Apr 2025 13:37:13 +0200 Subject: [PATCH 05/16] drm: adp: Use spin_lock_irqsave for drm device event_lock The lock is used in the interrupt handler so use spin_lock_irqsave to disable interrupts and avoid deadlocks with the irq handler. Fixes: 332122eba628 ("drm: adp: Add Apple Display Pipe driver") Reviewed-by: Alyssa Rosenzweig Signed-off-by: Janne Grunau Link: https://lore.kernel.org/r/20250428-drm_adp_fixes-v2-1-912e081e55d8@jannau.net Signed-off-by: Alyssa Rosenzweig --- drivers/gpu/drm/adp/adp_drv.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-) diff --git a/drivers/gpu/drm/adp/adp_drv.c b/drivers/gpu/drm/adp/adp_drv.c index c98c647f981d..157298a8ff42 100644 --- a/drivers/gpu/drm/adp/adp_drv.c +++ b/drivers/gpu/drm/adp/adp_drv.c @@ -310,6 +310,7 @@ static void adp_crtc_atomic_flush(struct drm_crtc *crtc, struct drm_atomic_state *state) { u32 frame_num = 1; + unsigned long flags; struct adp_drv_private *adp = crtc_to_adp(crtc); struct drm_crtc_state *new_state = drm_atomic_get_new_crtc_state(state, crtc); u64 new_size = ALIGN(new_state->mode.hdisplay * @@ -330,13 +331,13 @@ static void adp_crtc_atomic_flush(struct drm_crtc *crtc, } writel(ADBE_FIFO_SYNC | frame_num, adp->be + ADBE_FIFO); //FIXME: use adbe flush interrupt - spin_lock_irq(&crtc->dev->event_lock); + spin_lock_irqsave(&crtc->dev->event_lock, flags); if (crtc->state->event) { drm_crtc_vblank_get(crtc); adp->event = crtc->state->event; } crtc->state->event = NULL; - spin_unlock_irq(&crtc->dev->event_lock); + spin_unlock_irqrestore(&crtc->dev->event_lock, flags); } static const struct drm_crtc_funcs adp_crtc_funcs = { -- 2.51.0 From 7a7d6681d5adde7dc7e648dcc6b9e9be6ca93d5d Mon Sep 17 00:00:00 2001 From: Janne Grunau Date: Mon, 28 Apr 2025 13:37:14 +0200 Subject: [PATCH 06/16] drm: adp: Handle drm_crtc_vblank_get() errors drm_crtc_vblank_get() may fail when it's called before drm_crtc_vblank_on() on a resetted CRTC. This occurs in drm_crtc_helper_funcs' atomic_flush() calls after drm_atomic_helper_crtc_reset() for example directly after probe. Send the vblank event directly in such cases. Avoids following warning in the subsequent drm_crtc_vblank_put() call from the vblank irq handler as below: adp 228200000.display-pipe: [drm] drm_WARN_ON(atomic_read(&vblank->refcount) == 0) WARNING: CPU: 5 PID: 1206 at drivers/gpu/drm/drm_vblank.c:1247 drm_vblank_put+0x158/0x170 ... Call trace: drm_vblank_put+0x158/0x170 (P) drm_crtc_vblank_put+0x24/0x38 adp_fe_irq+0xd8/0xe8 [adpdrm] __handle_irq_event_percpu+0x94/0x318 handle_irq_event+0x54/0xd0 handle_fasteoi_irq+0xa8/0x240 handle_irq_desc+0x3c/0x68 generic_handle_domain_irq+0x24/0x40 Modifying `crtc->state->event` here is fine as crtc->mutex is locked by the non-async atomic commit. In retrospect this looks so obvious that it doesn't warrant a comment in the file. Signed-off-by: Janne Grunau Reviewed-by: Alyssa Rosenzweig Link: https://lore.kernel.org/r/20250428-drm_adp_fixes-v2-2-912e081e55d8@jannau.net Signed-off-by: Alyssa Rosenzweig --- drivers/gpu/drm/adp/adp_drv.c | 16 +++++++++++----- 1 file changed, 11 insertions(+), 5 deletions(-) diff --git a/drivers/gpu/drm/adp/adp_drv.c b/drivers/gpu/drm/adp/adp_drv.c index 157298a8ff42..bdf27ee742ea 100644 --- a/drivers/gpu/drm/adp/adp_drv.c +++ b/drivers/gpu/drm/adp/adp_drv.c @@ -331,13 +331,19 @@ static void adp_crtc_atomic_flush(struct drm_crtc *crtc, } writel(ADBE_FIFO_SYNC | frame_num, adp->be + ADBE_FIFO); //FIXME: use adbe flush interrupt - spin_lock_irqsave(&crtc->dev->event_lock, flags); if (crtc->state->event) { - drm_crtc_vblank_get(crtc); - adp->event = crtc->state->event; + struct drm_pending_vblank_event *event = crtc->state->event; + + crtc->state->event = NULL; + spin_lock_irqsave(&crtc->dev->event_lock, flags); + + if (drm_crtc_vblank_get(crtc) != 0) + drm_crtc_send_vblank_event(crtc, event); + else + adp->event = event; + + spin_unlock_irqrestore(&crtc->dev->event_lock, flags); } - crtc->state->event = NULL; - spin_unlock_irqrestore(&crtc->dev->event_lock, flags); } static const struct drm_crtc_funcs adp_crtc_funcs = { -- 2.51.0 From c082a52125d9007b488d590c412fd126aa78c345 Mon Sep 17 00:00:00 2001 From: Janne Grunau Date: Mon, 28 Apr 2025 13:37:15 +0200 Subject: [PATCH 07/16] drm: adp: Enable vblank interrupts in crtc's .atomic_enable Calling drm_crtc_vblank_on() drm_crtc_helper_funcs' atomic_enable is expected to enable vblank interrupts. It may have been avoided here to due to drm_crtc_vblank_get()'s error behavior after drm_crtc_vblank_reset(). With that fixed in the preceding change the driver can call drm_crtc_vblank_on() from adp_crtc_atomic_enable(). Reviewed-by: Alyssa Rosenzweig Signed-off-by: Janne Grunau Link: https://lore.kernel.org/r/20250428-drm_adp_fixes-v2-3-912e081e55d8@jannau.net Signed-off-by: Alyssa Rosenzweig --- drivers/gpu/drm/adp/adp_drv.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/drivers/gpu/drm/adp/adp_drv.c b/drivers/gpu/drm/adp/adp_drv.c index bdf27ee742ea..50d26c73301c 100644 --- a/drivers/gpu/drm/adp/adp_drv.c +++ b/drivers/gpu/drm/adp/adp_drv.c @@ -288,6 +288,7 @@ static void adp_crtc_atomic_enable(struct drm_crtc *crtc, writel(BIT(0), adp->be + ADBE_BLEND_EN3); writel(BIT(0), adp->be + ADBE_BLEND_BYPASS); writel(BIT(0), adp->be + ADBE_BLEND_EN4); + drm_crtc_vblank_on(crtc); } static void adp_crtc_atomic_disable(struct drm_crtc *crtc, @@ -519,8 +520,7 @@ static int adp_drm_bind(struct device *dev) struct adp_drv_private *adp = to_adp(drm); int err; - adp_disable_vblank(adp); - writel(ADP_CTRL_FIFO_ON | ADP_CTRL_VBLANK_ON, adp->fe + ADP_CTRL); + writel(ADP_CTRL_FIFO_ON, adp->fe + ADP_CTRL); adp->next_bridge = drmm_of_get_bridge(&adp->drm, dev->of_node, 0, 0); if (IS_ERR(adp->next_bridge)) { -- 2.51.0 From 8f6dfc4d7037e88cc0a4be4f290829946999341f Mon Sep 17 00:00:00 2001 From: Janne Grunau Date: Mon, 28 Apr 2025 13:37:16 +0200 Subject: [PATCH 08/16] drm: adp: Remove pointless irq_lock spin lock Interrupt handlers run with interrupts disabled so it is not necessary to protect them against reentrancy. Reviewed-by: Alyssa Rosenzweig Signed-off-by: Janne Grunau Link: https://lore.kernel.org/r/20250428-drm_adp_fixes-v2-4-912e081e55d8@jannau.net Signed-off-by: Alyssa Rosenzweig --- drivers/gpu/drm/adp/adp_drv.c | 6 ------ 1 file changed, 6 deletions(-) diff --git a/drivers/gpu/drm/adp/adp_drv.c b/drivers/gpu/drm/adp/adp_drv.c index 50d26c73301c..54cde090c3f4 100644 --- a/drivers/gpu/drm/adp/adp_drv.c +++ b/drivers/gpu/drm/adp/adp_drv.c @@ -121,7 +121,6 @@ struct adp_drv_private { dma_addr_t mask_iova; int be_irq; int fe_irq; - spinlock_t irq_lock; struct drm_pending_vblank_event *event; }; @@ -490,8 +489,6 @@ static irqreturn_t adp_fe_irq(int irq, void *arg) u32 int_status; u32 int_ctl; - spin_lock(&adp->irq_lock); - int_status = readl(adp->fe + ADP_INT_STATUS); if (int_status & ADP_INT_STATUS_VBLANK) { drm_crtc_handle_vblank(&adp->crtc); @@ -509,7 +506,6 @@ static irqreturn_t adp_fe_irq(int irq, void *arg) writel(int_status, adp->fe + ADP_INT_STATUS); - spin_unlock(&adp->irq_lock); return IRQ_HANDLED; } @@ -574,8 +570,6 @@ static int adp_probe(struct platform_device *pdev) if (IS_ERR(adp)) return PTR_ERR(adp); - spin_lock_init(&adp->irq_lock); - dev_set_drvdata(&pdev->dev, &adp->drm); err = adp_parse_of(pdev, adp); -- 2.51.0 From 32dce6b1949a696dc7abddc04de8cbe35c260217 Mon Sep 17 00:00:00 2001 From: Janne Grunau Date: Tue, 4 Mar 2025 20:12:14 +0100 Subject: [PATCH 09/16] drm: Select DRM_KMS_HELPER from DRM_DEBUG_DP_MST_TOPOLOGY_REFS Using "depends on" and "select" for the same Kconfig symbol is known to cause circular dependencies (cmp. "Kconfig recursive dependency limitations" in Documentation/kbuild/kconfig-language.rst. DRM drivers are selecting drm helpers so do the same for DRM_DEBUG_DP_MST_TOPOLOGY_REFS. Fixes following circular dependency reported on x86 for the downstream Asahi Linux tree: error: recursive dependency detected! symbol DRM_KMS_HELPER is selected by DRM_GEM_SHMEM_HELPER symbol DRM_GEM_SHMEM_HELPER is selected by RUST_DRM_GEM_SHMEM_HELPER symbol RUST_DRM_GEM_SHMEM_HELPER is selected by DRM_ASAHI symbol DRM_ASAHI depends on RUST symbol RUST depends on CALL_PADDING symbol CALL_PADDING depends on OBJTOOL symbol OBJTOOL is selected by STACK_VALIDATION symbol STACK_VALIDATION depends on UNWINDER_FRAME_POINTER symbol UNWINDER_FRAME_POINTER is part of choice block at arch/x86/Kconfig.debug:224 symbol unknown is visible depending on UNWINDER_GUESS symbol UNWINDER_GUESS prompt is visible depending on STACKDEPOT symbol STACKDEPOT is selected by DRM_DEBUG_DP_MST_TOPOLOGY_REFS symbol DRM_DEBUG_DP_MST_TOPOLOGY_REFS depends on DRM_KMS_HELPER Fixes: 12a280c72868 ("drm/dp_mst: Add topology ref history tracking for debugging") Cc: stable@vger.kernel.org Signed-off-by: Janne Grunau Acked-by: Thomas Zimmermann Link: https://lore.kernel.org/r/20250304-drm_debug_dp_mst_topo_kconfig-v1-1-e16fd152f258@jannau.net Signed-off-by: Alyssa Rosenzweig --- drivers/gpu/drm/Kconfig | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/drivers/gpu/drm/Kconfig b/drivers/gpu/drm/Kconfig index 2cba2b6ebe1c..f01925ed8176 100644 --- a/drivers/gpu/drm/Kconfig +++ b/drivers/gpu/drm/Kconfig @@ -188,7 +188,7 @@ config DRM_DEBUG_DP_MST_TOPOLOGY_REFS bool "Enable refcount backtrace history in the DP MST helpers" depends on STACKTRACE_SUPPORT select STACKDEPOT - depends on DRM_KMS_HELPER + select DRM_KMS_HELPER depends on DEBUG_KERNEL depends on EXPERT help -- 2.51.0 From 48ccf21fa8dc595c8aa4f1d347b593dcae0727d0 Mon Sep 17 00:00:00 2001 From: Maxime Ripard Date: Tue, 8 Apr 2025 16:07:58 +0200 Subject: [PATCH 10/16] drm/tests: shmem: Fix memleak The drm_gem_shmem_test_get_pages_sgt() gets a scatter-gather table using the drm_gem_shmem_get_sg_table() function and rightfully calls sg_free_table() on it. However, it's also supposed to kfree() the returned sg_table, but doesn't. This leads to a memory leak, reported by kmemleak. Fix it by adding a kunit action to kfree the sgt when the test ends. Reported-by: Philipp Stanner Closes: https://lore.kernel.org/dri-devel/a7655158a6367ac46194d57f4b7433ef0772a73e.camel@mailbox.org/ Fixes: 93032ae634d4 ("drm/test: add a test suite for GEM objects backed by shmem") Reviewed-by: Javier Martinez Canillas Link: https://lore.kernel.org/r/20250408140758.1831333-1-mripard@kernel.org Signed-off-by: Maxime Ripard --- drivers/gpu/drm/tests/drm_gem_shmem_test.c | 3 +++ 1 file changed, 3 insertions(+) diff --git a/drivers/gpu/drm/tests/drm_gem_shmem_test.c b/drivers/gpu/drm/tests/drm_gem_shmem_test.c index fd4215e2f982..925fbc2cda70 100644 --- a/drivers/gpu/drm/tests/drm_gem_shmem_test.c +++ b/drivers/gpu/drm/tests/drm_gem_shmem_test.c @@ -216,6 +216,9 @@ static void drm_gem_shmem_test_get_pages_sgt(struct kunit *test) KUNIT_ASSERT_NOT_ERR_OR_NULL(test, sgt); KUNIT_EXPECT_NULL(test, shmem->sgt); + ret = kunit_add_action_or_reset(test, kfree_wrapper, sgt); + KUNIT_ASSERT_EQ(test, ret, 0); + ret = kunit_add_action_or_reset(test, sg_free_table_wrapper, sgt); KUNIT_ASSERT_EQ(test, ret, 0); -- 2.51.0 From 1a8bc0fe8039e1e57f68c4a588f0403d98bfeb1f Mon Sep 17 00:00:00 2001 From: Russell Cloran Date: Mon, 14 Apr 2025 22:32:59 -0700 Subject: [PATCH 11/16] drm/mipi-dbi: Fix blanking for non-16 bit formats On r6x2b6x2g6x2 displays not enough blank data is sent to blank the entire screen. When support for these displays was added, the dirty function was updated to handle the different amount of data, but blanking was not, and remained hardcoded as 2 bytes per pixel. This change applies almost the same algorithm used in the dirty function to the blank function, but there is no fb available at that point, and no concern about having to transform any data, so the dbidev pixel format is always used for calculating the length. Fixes: 4aebb79021f3 ("drm/mipi-dbi: Add support for DRM_FORMAT_RGB888") Signed-off-by: Russell Cloran Link: https://lore.kernel.org/r/20250415053259.79572-1-rcloran@gmail.com Signed-off-by: Maxime Ripard --- drivers/gpu/drm/drm_mipi_dbi.c | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-) diff --git a/drivers/gpu/drm/drm_mipi_dbi.c b/drivers/gpu/drm/drm_mipi_dbi.c index 89e05a5bed1d..a4cd476f9b30 100644 --- a/drivers/gpu/drm/drm_mipi_dbi.c +++ b/drivers/gpu/drm/drm_mipi_dbi.c @@ -404,12 +404,16 @@ static void mipi_dbi_blank(struct mipi_dbi_dev *dbidev) u16 height = drm->mode_config.min_height; u16 width = drm->mode_config.min_width; struct mipi_dbi *dbi = &dbidev->dbi; - size_t len = width * height * 2; + const struct drm_format_info *dst_format; + size_t len; int idx; if (!drm_dev_enter(drm, &idx)) return; + dst_format = drm_format_info(dbidev->pixel_format); + len = drm_format_info_min_pitch(dst_format, 0, width) * height; + memset(dbidev->tx_buf, 0, len); mipi_dbi_set_window_address(dbidev, 0, width - 1, 0, height - 1); -- 2.51.0 From c4eb2f88d2796ab90c5430e11c48709716181364 Mon Sep 17 00:00:00 2001 From: Jacek Lawrynowicz Date: Fri, 25 Apr 2025 11:28:22 +0200 Subject: [PATCH 12/16] accel/ivpu: Increase state dump msg timeout Increase JMS message state dump command timeout to 100 ms. On some platforms, the FW may take a bit longer than 50 ms to dump its state to the log buffer and we don't want to miss any debug info during TDR. Fixes: 5e162f872d7a ("accel/ivpu: Add FW state dump on TDR") Cc: stable@vger.kernel.org # v6.13+ Reviewed-by: Jeff Hugo Signed-off-by: Jacek Lawrynowicz Link: https://lore.kernel.org/r/20250425092822.2194465-1-jacek.lawrynowicz@linux.intel.com --- drivers/accel/ivpu/ivpu_hw.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/drivers/accel/ivpu/ivpu_hw.c b/drivers/accel/ivpu/ivpu_hw.c index ec9a3629da3a..633160470c93 100644 --- a/drivers/accel/ivpu/ivpu_hw.c +++ b/drivers/accel/ivpu/ivpu_hw.c @@ -119,7 +119,7 @@ static void timeouts_init(struct ivpu_device *vdev) else vdev->timeout.autosuspend = 100; vdev->timeout.d0i3_entry_msg = 5; - vdev->timeout.state_dump_msg = 10; + vdev->timeout.state_dump_msg = 100; } } -- 2.51.0 From f2ecc700d1ef53127c14a3463442a8ac2a6a5cf5 Mon Sep 17 00:00:00 2001 From: Jacek Lawrynowicz Date: Fri, 25 Apr 2025 11:33:40 +0200 Subject: [PATCH 13/16] accel/ivpu: Fix pm related deadlocks in cmdq ioctls Fix deadlocks in ivpu_cmdq_create_ioctl() and ivpu_cmdq_destroy_ioctl() related to runtime suspend. Runtime suspend acquires file_priv->lock mutex by calling ivpu_cmdq_reset_all_contexts(). The same lock is acquired in the cmdq ioctls. If one of the cmdq ioctls is called while runtime suspend is in progress, it can lead to a deadlock. Call stacks from example deadlock below. Runtime suspend thread: [ 3443.179717] Call Trace: [ 3443.179724] __schedule+0x4b6/0x16b0 [ 3443.179732] ? __mod_timer+0x27d/0x3a0 [ 3443.179738] schedule+0x2f/0x140 [ 3443.179741] schedule_preempt_disabled+0x19/0x30 [ 3443.179743] __mutex_lock.constprop.0+0x335/0x7d0 [ 3443.179745] ? xas_find+0x1ed/0x260 [ 3443.179747] ? xa_find+0x8e/0xf0 [ 3443.179749] __mutex_lock_slowpath+0x13/0x20 [ 3443.179751] mutex_lock+0x41/0x60 [ 3443.179757] ivpu_cmdq_reset_all_contexts+0x82/0x150 [intel_vpu a9bd091a97f28f0235f161316b29f8234f437295] [ 3443.179786] ivpu_pm_runtime_suspend_cb+0x1f1/0x3f0 [intel_vpu a9bd091a97f28f0235f161316b29f8234f437295] [ 3443.179850] pci_pm_runtime_suspend+0x6e/0x1f0 [ 3443.179870] ? __pfx_pci_pm_runtime_suspend+0x10/0x10 [ 3443.179886] __rpm_callback+0x48/0x130 [ 3443.179899] rpm_callback+0x64/0x70 [ 3443.179911] rpm_suspend+0x12c/0x630 [ 3443.179922] ? __schedule+0x4be/0x16b0 [ 3443.179941] pm_runtime_work+0xca/0xf0 [ 3443.179955] process_one_work+0x188/0x3d0 [ 3443.179971] worker_thread+0x2b9/0x3c0 [ 3443.179984] kthread+0xfb/0x220 [ 3443.180001] ? __pfx_worker_thread+0x10/0x10 [ 3443.180013] ? __pfx_kthread+0x10/0x10 [ 3443.180029] ret_from_fork+0x47/0x70 [ 3443.180044] ? __pfx_kthread+0x10/0x10 [ 3443.180059] ret_from_fork_asm+0x1a/0x30 User space thread: [ 3443.180128] Call Trace: [ 3443.180138] __schedule+0x4b6/0x16b0 [ 3443.180159] schedule+0x2f/0x140 [ 3443.180163] rpm_resume+0x1a7/0x6a0 [ 3443.180165] ? __pfx_autoremove_wake_function+0x10/0x10 [ 3443.180169] __pm_runtime_resume+0x56/0x90 [ 3443.180171] ivpu_rpm_get+0x28/0xb0 [intel_vpu a9bd091a97f28f0235f161316b29f8234f437295] [ 3443.180181] ivpu_ipc_send_receive+0x6d/0x120 [intel_vpu a9bd091a97f28f0235f161316b29f8234f437295] [ 3443.180193] ? free_frozen_pages+0x395/0x670 [ 3443.180199] ? __free_pages+0xa7/0xc0 [ 3443.180202] ivpu_jsm_hws_destroy_cmdq+0x76/0xf0 [intel_vpu a9bd091a97f28f0235f161316b29f8234f437295] [ 3443.180213] ? locks_dispose_list+0x6c/0xa0 [ 3443.180219] ? kmem_cache_free+0x342/0x470 [ 3443.180222] ? vm_area_free+0x19/0x30 [ 3443.180225] ? xas_load+0x17/0xf0 [ 3443.180229] ? xa_load+0x72/0xb0 [ 3443.180230] ivpu_cmdq_unregister.isra.0+0xb1/0x100 [intel_vpu a9bd091a97f28f0235f161316b29f8234f437295] [ 3443.180241] ivpu_cmdq_destroy_ioctl+0x8d/0x130 [intel_vpu a9bd091a97f28f0235f161316b29f8234f437295] [ 3443.180251] ? __pfx_ivpu_cmdq_destroy_ioctl+0x10/0x10 [intel_vpu a9bd091a97f28f0235f161316b29f8234f437295] [ 3443.180260] drm_ioctl_kernel+0xb3/0x110 [ 3443.180265] drm_ioctl+0x2ca/0x580 [ 3443.180266] ? __pfx_ivpu_cmdq_destroy_ioctl+0x10/0x10 [intel_vpu a9bd091a97f28f0235f161316b29f8234f437295] [ 3443.180275] ? __fput+0x1ae/0x2f0 [ 3443.180279] ? kmem_cache_free+0x342/0x470 [ 3443.180282] __x64_sys_ioctl+0xa9/0xe0 [ 3443.180286] x64_sys_call+0x13b7/0x26f0 [ 3443.180289] do_syscall_64+0x62/0x180 [ 3443.180291] entry_SYSCALL_64_after_hwframe+0x71/0x79 Fixes: 465a3914b254 ("accel/ivpu: Add API for command queue create/destroy/submit") Reviewed-by: Jeff Hugo Signed-off-by: Jacek Lawrynowicz Link: https://lore.kernel.org/r/20250425093341.2202895-1-jacek.lawrynowicz@linux.intel.com --- drivers/accel/ivpu/ivpu_job.c | 33 ++++++++++++++++++++++++--------- 1 file changed, 24 insertions(+), 9 deletions(-) diff --git a/drivers/accel/ivpu/ivpu_job.c b/drivers/accel/ivpu/ivpu_job.c index 863e3cd6ace5..e17b3deda201 100644 --- a/drivers/accel/ivpu/ivpu_job.c +++ b/drivers/accel/ivpu/ivpu_job.c @@ -874,15 +874,21 @@ int ivpu_cmdq_submit_ioctl(struct drm_device *dev, void *data, struct drm_file * int ivpu_cmdq_create_ioctl(struct drm_device *dev, void *data, struct drm_file *file) { struct ivpu_file_priv *file_priv = file->driver_priv; + struct ivpu_device *vdev = file_priv->vdev; struct drm_ivpu_cmdq_create *args = data; struct ivpu_cmdq *cmdq; + int ret; - if (!ivpu_is_capable(file_priv->vdev, DRM_IVPU_CAP_MANAGE_CMDQ)) + if (!ivpu_is_capable(vdev, DRM_IVPU_CAP_MANAGE_CMDQ)) return -ENODEV; if (args->priority > DRM_IVPU_JOB_PRIORITY_REALTIME) return -EINVAL; + ret = ivpu_rpm_get(vdev); + if (ret < 0) + return ret; + mutex_lock(&file_priv->lock); cmdq = ivpu_cmdq_create(file_priv, ivpu_job_to_jsm_priority(args->priority), false); @@ -891,6 +897,8 @@ int ivpu_cmdq_create_ioctl(struct drm_device *dev, void *data, struct drm_file * mutex_unlock(&file_priv->lock); + ivpu_rpm_put(vdev); + return cmdq ? 0 : -ENOMEM; } @@ -900,28 +908,35 @@ int ivpu_cmdq_destroy_ioctl(struct drm_device *dev, void *data, struct drm_file struct ivpu_device *vdev = file_priv->vdev; struct drm_ivpu_cmdq_destroy *args = data; struct ivpu_cmdq *cmdq; - u32 cmdq_id; + u32 cmdq_id = 0; int ret; if (!ivpu_is_capable(vdev, DRM_IVPU_CAP_MANAGE_CMDQ)) return -ENODEV; + ret = ivpu_rpm_get(vdev); + if (ret < 0) + return ret; + mutex_lock(&file_priv->lock); cmdq = xa_load(&file_priv->cmdq_xa, args->cmdq_id); if (!cmdq || cmdq->is_legacy) { ret = -ENOENT; - goto err_unlock; + } else { + cmdq_id = cmdq->id; + ivpu_cmdq_destroy(file_priv, cmdq); + ret = 0; } - cmdq_id = cmdq->id; - ivpu_cmdq_destroy(file_priv, cmdq); mutex_unlock(&file_priv->lock); - ivpu_cmdq_abort_all_jobs(vdev, file_priv->ctx.id, cmdq_id); - return 0; -err_unlock: - mutex_unlock(&file_priv->lock); + /* Abort any pending jobs only if cmdq was destroyed */ + if (!ret) + ivpu_cmdq_abort_all_jobs(vdev, file_priv->ctx.id, cmdq_id); + + ivpu_rpm_put(vdev); + return ret; } -- 2.51.0 From 75680b7cd461b169c7ccd2a0fba7542868b7fce2 Mon Sep 17 00:00:00 2001 From: Karol Wachowski Date: Fri, 25 Apr 2025 11:36:56 +0200 Subject: [PATCH 14/16] accel/ivpu: Correct mutex unlock order in job submission The mutex unlock for vdev->submitted_jobs_lock was incorrectly placed before unlocking file_priv->lock. Change order of unlocks to avoid potential race conditions. Fixes: 5bbccadaf33e ("accel/ivpu: Abort all jobs after command queue unregister") Signed-off-by: Karol Wachowski Reviewed-by: Jeff Hugo Signed-off-by: Jacek Lawrynowicz Link: https://lore.kernel.org/r/20250425093656.2228168-1-jacek.lawrynowicz@linux.intel.com --- drivers/accel/ivpu/ivpu_job.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/drivers/accel/ivpu/ivpu_job.c b/drivers/accel/ivpu/ivpu_job.c index e17b3deda201..b28da35c30b6 100644 --- a/drivers/accel/ivpu/ivpu_job.c +++ b/drivers/accel/ivpu/ivpu_job.c @@ -681,8 +681,8 @@ static int ivpu_job_submit(struct ivpu_job *job, u8 priority, u32 cmdq_id) err_erase_xa: xa_erase(&vdev->submitted_jobs_xa, job->job_id); err_unlock: - mutex_unlock(&vdev->submitted_jobs_lock); mutex_unlock(&file_priv->lock); + mutex_unlock(&vdev->submitted_jobs_lock); ivpu_rpm_put(vdev); return ret; } -- 2.51.0 From b662b162c3d06f120749eea0351ec9317d9dd905 Mon Sep 17 00:00:00 2001 From: Feng Jiang Date: Wed, 9 Apr 2025 09:46:33 +0800 Subject: [PATCH 15/16] drm: Fix potential overflow issue in event_string array MIME-Version: 1.0 Content-Type: text/plain; charset=utf8 Content-Transfer-Encoding: 8bit When calling scnprintf() to append recovery method to event_string, the second argument should be `sizeof(event_string) - len`, otherwise there is a potential overflow problem. Fixes: b7cf9f4ac1b8 ("drm: Introduce device wedged event") Signed-off-by: Feng Jiang Reviewed-by: André Almeida Reviewed-by: Raag Jadav Link: https://lore.kernel.org/r/20250409014633.31303-1-jiangfeng@kylinos.cn Signed-off-by: Rodrigo Vivi --- drivers/gpu/drm/drm_drv.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/drivers/gpu/drm/drm_drv.c b/drivers/gpu/drm/drm_drv.c index 17fc5dc708f4..60e5ac179c15 100644 --- a/drivers/gpu/drm/drm_drv.c +++ b/drivers/gpu/drm/drm_drv.c @@ -549,7 +549,7 @@ int drm_dev_wedged_event(struct drm_device *dev, unsigned long method) if (drm_WARN_ONCE(dev, !recovery, "invalid recovery method %u\n", opt)) break; - len += scnprintf(event_string + len, sizeof(event_string), "%s,", recovery); + len += scnprintf(event_string + len, sizeof(event_string) - len, "%s,", recovery); } if (recovery) -- 2.51.0 From 35e4079bf1a2570abffce6ababa631afcf8ea0e5 Mon Sep 17 00:00:00 2001 From: =?utf8?q?Ma=C3=ADra=20Canal?= Date: Wed, 30 Apr 2025 17:51:52 -0300 Subject: [PATCH 16/16] drm/v3d: Add job to pending list if the reset was skipped MIME-Version: 1.0 Content-Type: text/plain; charset=utf8 Content-Transfer-Encoding: 8bit When a CL/CSD job times out, we check if the GPU has made any progress since the last timeout. If so, instead of resetting the hardware, we skip the reset and let the timer get rearmed. This gives long-running jobs a chance to complete. However, when `timedout_job()` is called, the job in question is removed from the pending list, which means it won't be automatically freed through `free_job()`. Consequently, when we skip the reset and keep the job running, the job won't be freed when it finally completes. This situation leads to a memory leak, as exposed in [1] and [2]. Similarly to commit 704d3d60fec4 ("drm/etnaviv: don't block scheduler when GPU is still active"), this patch ensures the job is put back on the pending list when extending the timeout. Cc: stable@vger.kernel.org # 6.0 Reported-by: Daivik Bhatia Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/12227 [1] Closes: https://github.com/raspberrypi/linux/issues/6817 [2] Reviewed-by: Iago Toral Quiroga Acked-by: Tvrtko Ursulin Link: https://lore.kernel.org/r/20250430210643.57924-1-mcanal@igalia.com Signed-off-by: Maíra Canal --- drivers/gpu/drm/v3d/v3d_sched.c | 28 +++++++++++++++++++++------- 1 file changed, 21 insertions(+), 7 deletions(-) diff --git a/drivers/gpu/drm/v3d/v3d_sched.c b/drivers/gpu/drm/v3d/v3d_sched.c index 4a7701a33cf8..eb35482f6fb5 100644 --- a/drivers/gpu/drm/v3d/v3d_sched.c +++ b/drivers/gpu/drm/v3d/v3d_sched.c @@ -744,11 +744,16 @@ v3d_gpu_reset_for_timeout(struct v3d_dev *v3d, struct drm_sched_job *sched_job) return DRM_GPU_SCHED_STAT_NOMINAL; } -/* If the current address or return address have changed, then the GPU - * has probably made progress and we should delay the reset. This - * could fail if the GPU got in an infinite loop in the CL, but that - * is pretty unlikely outside of an i-g-t testcase. - */ +static void +v3d_sched_skip_reset(struct drm_sched_job *sched_job) +{ + struct drm_gpu_scheduler *sched = sched_job->sched; + + spin_lock(&sched->job_list_lock); + list_add(&sched_job->list, &sched->pending_list); + spin_unlock(&sched->job_list_lock); +} + static enum drm_gpu_sched_stat v3d_cl_job_timedout(struct drm_sched_job *sched_job, enum v3d_queue q, u32 *timedout_ctca, u32 *timedout_ctra) @@ -758,9 +763,16 @@ v3d_cl_job_timedout(struct drm_sched_job *sched_job, enum v3d_queue q, u32 ctca = V3D_CORE_READ(0, V3D_CLE_CTNCA(q)); u32 ctra = V3D_CORE_READ(0, V3D_CLE_CTNRA(q)); + /* If the current address or return address have changed, then the GPU + * has probably made progress and we should delay the reset. This + * could fail if the GPU got in an infinite loop in the CL, but that + * is pretty unlikely outside of an i-g-t testcase. + */ if (*timedout_ctca != ctca || *timedout_ctra != ctra) { *timedout_ctca = ctca; *timedout_ctra = ctra; + + v3d_sched_skip_reset(sched_job); return DRM_GPU_SCHED_STAT_NOMINAL; } @@ -800,11 +812,13 @@ v3d_csd_job_timedout(struct drm_sched_job *sched_job) struct v3d_dev *v3d = job->base.v3d; u32 batches = V3D_CORE_READ(0, V3D_CSD_CURRENT_CFG4(v3d->ver)); - /* If we've made progress, skip reset and let the timer get - * rearmed. + /* If we've made progress, skip reset, add the job to the pending + * list, and let the timer get rearmed. */ if (job->timedout_batches != batches) { job->timedout_batches = batches; + + v3d_sched_skip_reset(sched_job); return DRM_GPU_SCHED_STAT_NOMINAL; } -- 2.51.0