From 908087ffbe896c100ed73d5f0ce11a5b7264af4a Mon Sep 17 00:00:00 2001 From: Oded Gabbay Date: Mon, 23 Dec 2019 17:51:48 +0200 Subject: habanalabs: halt the engines before hard-reset The driver must halt the engines before doing hard-reset, otherwise the device can go into undefined state. There is a place where the driver didn't do that and this patch fixes it. Reviewed-by: Tomer Tayar Signed-off-by: Oded Gabbay --- drivers/misc/habanalabs/device.c | 1 + drivers/misc/habanalabs/goya/goya.c | 42 +++++++++++++++++++++++++++++++++++++ 2 files changed, 43 insertions(+) (limited to 'drivers/misc') diff --git a/drivers/misc/habanalabs/device.c b/drivers/misc/habanalabs/device.c index b155e9549076..166883b64725 100644 --- a/drivers/misc/habanalabs/device.c +++ b/drivers/misc/habanalabs/device.c @@ -1189,6 +1189,7 @@ int hl_device_init(struct hl_device *hdev, struct class *hclass) if (hdev->asic_funcs->get_hw_state(hdev) == HL_DEVICE_HW_STATE_DIRTY) { dev_info(hdev->dev, "H/W state is dirty, must reset before initializing\n"); + hdev->asic_funcs->halt_engines(hdev, true); hdev->asic_funcs->hw_fini(hdev, true); } diff --git a/drivers/misc/habanalabs/goya/goya.c b/drivers/misc/habanalabs/goya/goya.c index 7344e8a222ae..f24fe909b88d 100644 --- a/drivers/misc/habanalabs/goya/goya.c +++ b/drivers/misc/habanalabs/goya/goya.c @@ -895,6 +895,11 @@ void goya_init_dma_qmans(struct hl_device *hdev) */ static void goya_disable_external_queues(struct hl_device *hdev) { + struct goya_device *goya = hdev->asic_specific; + + if (!(goya->hw_cap_initialized & HW_CAP_DMA)) + return; + WREG32(mmDMA_QM_0_GLBL_CFG0, 0); WREG32(mmDMA_QM_1_GLBL_CFG0, 0); WREG32(mmDMA_QM_2_GLBL_CFG0, 0); @@ -956,6 +961,11 @@ static int goya_stop_external_queues(struct hl_device *hdev) { int rc, retval = 0; + struct goya_device *goya = hdev->asic_specific; + + if (!(goya->hw_cap_initialized & HW_CAP_DMA)) + return retval; + rc = goya_stop_queue(hdev, mmDMA_QM_0_GLBL_CFG1, mmDMA_QM_0_CP_STS, @@ -1744,9 +1754,18 @@ void goya_init_tpc_qmans(struct hl_device *hdev) */ static void goya_disable_internal_queues(struct hl_device *hdev) { + struct goya_device *goya = hdev->asic_specific; + + if (!(goya->hw_cap_initialized & HW_CAP_MME)) + goto disable_tpc; + WREG32(mmMME_QM_GLBL_CFG0, 0); WREG32(mmMME_CMDQ_GLBL_CFG0, 0); +disable_tpc: + if (!(goya->hw_cap_initialized & HW_CAP_TPC)) + return; + WREG32(mmTPC0_QM_GLBL_CFG0, 0); WREG32(mmTPC0_CMDQ_GLBL_CFG0, 0); @@ -1782,8 +1801,12 @@ static void goya_disable_internal_queues(struct hl_device *hdev) */ static int goya_stop_internal_queues(struct hl_device *hdev) { + struct goya_device *goya = hdev->asic_specific; int rc, retval = 0; + if (!(goya->hw_cap_initialized & HW_CAP_MME)) + goto stop_tpc; + /* * Each queue (QMAN) is a separate H/W logic. That means that each * QMAN can be stopped independently and failure to stop one does NOT @@ -1810,6 +1833,10 @@ static int goya_stop_internal_queues(struct hl_device *hdev) retval = -EIO; } +stop_tpc: + if (!(goya->hw_cap_initialized & HW_CAP_TPC)) + return retval; + rc = goya_stop_queue(hdev, mmTPC0_QM_GLBL_CFG1, mmTPC0_QM_CP_STS, @@ -1975,6 +2002,11 @@ static int goya_stop_internal_queues(struct hl_device *hdev) static void goya_dma_stall(struct hl_device *hdev) { + struct goya_device *goya = hdev->asic_specific; + + if (!(goya->hw_cap_initialized & HW_CAP_DMA)) + return; + WREG32(mmDMA_QM_0_GLBL_CFG1, 1 << DMA_QM_0_GLBL_CFG1_DMA_STOP_SHIFT); WREG32(mmDMA_QM_1_GLBL_CFG1, 1 << DMA_QM_1_GLBL_CFG1_DMA_STOP_SHIFT); WREG32(mmDMA_QM_2_GLBL_CFG1, 1 << DMA_QM_2_GLBL_CFG1_DMA_STOP_SHIFT); @@ -1984,6 +2016,11 @@ static void goya_dma_stall(struct hl_device *hdev) static void goya_tpc_stall(struct hl_device *hdev) { + struct goya_device *goya = hdev->asic_specific; + + if (!(goya->hw_cap_initialized & HW_CAP_TPC)) + return; + WREG32(mmTPC0_CFG_TPC_STALL, 1 << TPC0_CFG_TPC_STALL_V_SHIFT); WREG32(mmTPC1_CFG_TPC_STALL, 1 << TPC1_CFG_TPC_STALL_V_SHIFT); WREG32(mmTPC2_CFG_TPC_STALL, 1 << TPC2_CFG_TPC_STALL_V_SHIFT); @@ -1996,6 +2033,11 @@ static void goya_tpc_stall(struct hl_device *hdev) static void goya_mme_stall(struct hl_device *hdev) { + struct goya_device *goya = hdev->asic_specific; + + if (!(goya->hw_cap_initialized & HW_CAP_MME)) + return; + WREG32(mmMME_STALL, 0xFFFFFFFF); } -- cgit v1.2.3 From a37e47192dfa98f79a0cd5ab991c224b5980c982 Mon Sep 17 00:00:00 2001 From: Omer Shpigelman Date: Sun, 5 Jan 2020 09:05:45 +0000 Subject: habanalabs: do not halt CoreSight during hard reset During hard reset we must not write to the device. Hence avoid halting CoreSight during user context close if it is done during hard reset. In addition, we must not re-enable clock gating afterwards as it was deliberately disabled in the beginning of the hard reset flow. Signed-off-by: Omer Shpigelman Reviewed-by: Oded Gabbay Signed-off-by: Oded Gabbay --- drivers/misc/habanalabs/device.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) (limited to 'drivers/misc') diff --git a/drivers/misc/habanalabs/device.c b/drivers/misc/habanalabs/device.c index 166883b64725..b680b0caa69b 100644 --- a/drivers/misc/habanalabs/device.c +++ b/drivers/misc/habanalabs/device.c @@ -598,7 +598,9 @@ int hl_device_set_debug_mode(struct hl_device *hdev, bool enable) goto out; } - hdev->asic_funcs->halt_coresight(hdev); + if (!hdev->hard_reset_pending) + hdev->asic_funcs->halt_coresight(hdev); + hdev->in_debug = 0; goto out; -- cgit v1.2.3 From cf01514c5c6efa2d521d35e68dff2e0674d08e91 Mon Sep 17 00:00:00 2001 From: Oded Gabbay Date: Thu, 23 Jan 2020 00:43:06 +0200 Subject: habanalabs: patched cb equals user cb in device memset During device memory memset, the driver allocates and use a CB (command buffer). To reuse existing code, it keeps a pointer to the CB in two variables, user_cb and patched_cb. Therefore, there is no need to "put" both the user_cb and patched_cb, as it will cause an underflow of the refcnt of the CB. Signed-off-by: Oded Gabbay --- drivers/misc/habanalabs/goya/goya.c | 2 -- 1 file changed, 2 deletions(-) (limited to 'drivers/misc') diff --git a/drivers/misc/habanalabs/goya/goya.c b/drivers/misc/habanalabs/goya/goya.c index f24fe909b88d..b8a8de24aaf7 100644 --- a/drivers/misc/habanalabs/goya/goya.c +++ b/drivers/misc/habanalabs/goya/goya.c @@ -4690,8 +4690,6 @@ static int goya_memset_device_memory(struct hl_device *hdev, u64 addr, u64 size, rc = goya_send_job_on_qman0(hdev, job); - hl_cb_put(job->patched_cb); - hl_debugfs_remove_job(hdev, job); kfree(job); cb->cs_cnt--; -- cgit v1.2.3