From: Oded Gabbay <ogabbay@kernel.org>
To: dri-devel@lists.freedesktop.org
Cc: Ofir Bitton <obitton@habana.ai>
Subject: [PATCH 03/12] accel/habanalabs: notify user about undefined opcode event
Date: Thu, 8 Jun 2023 16:38:40 +0300 [thread overview]
Message-ID: <20230608133849.2739411-3-ogabbay@kernel.org> (raw)
In-Reply-To: <20230608133849.2739411-1-ogabbay@kernel.org>
From: Ofir Bitton <obitton@habana.ai>
In order for user to be aware of undefined opcode events, we must
store all relevant information and notify user about the failure.
The user will fetch the stored info via info ioctl.
Signed-off-by: Ofir Bitton <obitton@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
---
drivers/accel/habanalabs/gaudi2/gaudi2.c | 142 ++++++++++++++++++++++-
1 file changed, 137 insertions(+), 5 deletions(-)
diff --git a/drivers/accel/habanalabs/gaudi2/gaudi2.c b/drivers/accel/habanalabs/gaudi2/gaudi2.c
index 20c4583f12b0..ed3b0b6225d2 100644
--- a/drivers/accel/habanalabs/gaudi2/gaudi2.c
+++ b/drivers/accel/habanalabs/gaudi2/gaudi2.c
@@ -993,6 +993,111 @@ gaudi2_pcie_addr_dec_error_cause[GAUDI2_NUM_OF_PCIE_ADDR_DEC_ERR_CAUSE] = {
"TLP is blocked by RR"
};
+static const int gaudi2_queue_id_to_engine_id[] = {
+ [GAUDI2_QUEUE_ID_PDMA_0_0...GAUDI2_QUEUE_ID_PDMA_0_3] = GAUDI2_ENGINE_ID_PDMA_0,
+ [GAUDI2_QUEUE_ID_PDMA_1_0...GAUDI2_QUEUE_ID_PDMA_1_3] = GAUDI2_ENGINE_ID_PDMA_1,
+ [GAUDI2_QUEUE_ID_DCORE0_EDMA_0_0...GAUDI2_QUEUE_ID_DCORE0_EDMA_0_3] =
+ GAUDI2_DCORE0_ENGINE_ID_EDMA_0,
+ [GAUDI2_QUEUE_ID_DCORE0_EDMA_1_0...GAUDI2_QUEUE_ID_DCORE0_EDMA_1_3] =
+ GAUDI2_DCORE0_ENGINE_ID_EDMA_1,
+ [GAUDI2_QUEUE_ID_DCORE1_EDMA_0_0...GAUDI2_QUEUE_ID_DCORE1_EDMA_0_3] =
+ GAUDI2_DCORE1_ENGINE_ID_EDMA_0,
+ [GAUDI2_QUEUE_ID_DCORE1_EDMA_1_0...GAUDI2_QUEUE_ID_DCORE1_EDMA_1_3] =
+ GAUDI2_DCORE1_ENGINE_ID_EDMA_1,
+ [GAUDI2_QUEUE_ID_DCORE2_EDMA_0_0...GAUDI2_QUEUE_ID_DCORE2_EDMA_0_3] =
+ GAUDI2_DCORE2_ENGINE_ID_EDMA_0,
+ [GAUDI2_QUEUE_ID_DCORE2_EDMA_1_0...GAUDI2_QUEUE_ID_DCORE2_EDMA_1_3] =
+ GAUDI2_DCORE2_ENGINE_ID_EDMA_1,
+ [GAUDI2_QUEUE_ID_DCORE3_EDMA_0_0...GAUDI2_QUEUE_ID_DCORE3_EDMA_0_3] =
+ GAUDI2_DCORE3_ENGINE_ID_EDMA_0,
+ [GAUDI2_QUEUE_ID_DCORE3_EDMA_1_0...GAUDI2_QUEUE_ID_DCORE3_EDMA_1_3] =
+ GAUDI2_DCORE3_ENGINE_ID_EDMA_1,
+ [GAUDI2_QUEUE_ID_DCORE0_MME_0_0...GAUDI2_QUEUE_ID_DCORE0_MME_0_3] =
+ GAUDI2_DCORE0_ENGINE_ID_MME,
+ [GAUDI2_QUEUE_ID_DCORE1_MME_0_0...GAUDI2_QUEUE_ID_DCORE1_MME_0_3] =
+ GAUDI2_DCORE1_ENGINE_ID_MME,
+ [GAUDI2_QUEUE_ID_DCORE2_MME_0_0...GAUDI2_QUEUE_ID_DCORE2_MME_0_3] =
+ GAUDI2_DCORE2_ENGINE_ID_MME,
+ [GAUDI2_QUEUE_ID_DCORE3_MME_0_0...GAUDI2_QUEUE_ID_DCORE3_MME_0_3] =
+ GAUDI2_DCORE3_ENGINE_ID_MME,
+ [GAUDI2_QUEUE_ID_DCORE0_TPC_0_0...GAUDI2_QUEUE_ID_DCORE0_TPC_0_3] =
+ GAUDI2_DCORE0_ENGINE_ID_TPC_0,
+ [GAUDI2_QUEUE_ID_DCORE0_TPC_1_0...GAUDI2_QUEUE_ID_DCORE0_TPC_1_3] =
+ GAUDI2_DCORE0_ENGINE_ID_TPC_1,
+ [GAUDI2_QUEUE_ID_DCORE0_TPC_2_0...GAUDI2_QUEUE_ID_DCORE0_TPC_2_3] =
+ GAUDI2_DCORE0_ENGINE_ID_TPC_2,
+ [GAUDI2_QUEUE_ID_DCORE0_TPC_3_0...GAUDI2_QUEUE_ID_DCORE0_TPC_3_3] =
+ GAUDI2_DCORE0_ENGINE_ID_TPC_3,
+ [GAUDI2_QUEUE_ID_DCORE0_TPC_4_0...GAUDI2_QUEUE_ID_DCORE0_TPC_4_3] =
+ GAUDI2_DCORE0_ENGINE_ID_TPC_4,
+ [GAUDI2_QUEUE_ID_DCORE0_TPC_5_0...GAUDI2_QUEUE_ID_DCORE0_TPC_5_3] =
+ GAUDI2_DCORE0_ENGINE_ID_TPC_5,
+ [GAUDI2_QUEUE_ID_DCORE0_TPC_6_0...GAUDI2_QUEUE_ID_DCORE0_TPC_6_3] =
+ GAUDI2_DCORE0_ENGINE_ID_TPC_6,
+ [GAUDI2_QUEUE_ID_DCORE1_TPC_0_0...GAUDI2_QUEUE_ID_DCORE1_TPC_0_3] =
+ GAUDI2_DCORE1_ENGINE_ID_TPC_0,
+ [GAUDI2_QUEUE_ID_DCORE1_TPC_1_0...GAUDI2_QUEUE_ID_DCORE1_TPC_1_3] =
+ GAUDI2_DCORE1_ENGINE_ID_TPC_1,
+ [GAUDI2_QUEUE_ID_DCORE1_TPC_2_0...GAUDI2_QUEUE_ID_DCORE1_TPC_2_3] =
+ GAUDI2_DCORE1_ENGINE_ID_TPC_2,
+ [GAUDI2_QUEUE_ID_DCORE1_TPC_3_0...GAUDI2_QUEUE_ID_DCORE1_TPC_3_3] =
+ GAUDI2_DCORE1_ENGINE_ID_TPC_3,
+ [GAUDI2_QUEUE_ID_DCORE1_TPC_4_0...GAUDI2_QUEUE_ID_DCORE1_TPC_4_3] =
+ GAUDI2_DCORE1_ENGINE_ID_TPC_4,
+ [GAUDI2_QUEUE_ID_DCORE1_TPC_5_0...GAUDI2_QUEUE_ID_DCORE1_TPC_5_3] =
+ GAUDI2_DCORE1_ENGINE_ID_TPC_5,
+ [GAUDI2_QUEUE_ID_DCORE2_TPC_0_0...GAUDI2_QUEUE_ID_DCORE2_TPC_0_3] =
+ GAUDI2_DCORE2_ENGINE_ID_TPC_0,
+ [GAUDI2_QUEUE_ID_DCORE2_TPC_1_0...GAUDI2_QUEUE_ID_DCORE2_TPC_1_3] =
+ GAUDI2_DCORE2_ENGINE_ID_TPC_1,
+ [GAUDI2_QUEUE_ID_DCORE2_TPC_2_0...GAUDI2_QUEUE_ID_DCORE2_TPC_2_3] =
+ GAUDI2_DCORE2_ENGINE_ID_TPC_2,
+ [GAUDI2_QUEUE_ID_DCORE2_TPC_3_0...GAUDI2_QUEUE_ID_DCORE2_TPC_3_3] =
+ GAUDI2_DCORE2_ENGINE_ID_TPC_3,
+ [GAUDI2_QUEUE_ID_DCORE2_TPC_4_0...GAUDI2_QUEUE_ID_DCORE2_TPC_4_3] =
+ GAUDI2_DCORE2_ENGINE_ID_TPC_4,
+ [GAUDI2_QUEUE_ID_DCORE2_TPC_5_0...GAUDI2_QUEUE_ID_DCORE2_TPC_5_3] =
+ GAUDI2_DCORE2_ENGINE_ID_TPC_5,
+ [GAUDI2_QUEUE_ID_DCORE3_TPC_0_0...GAUDI2_QUEUE_ID_DCORE3_TPC_0_3] =
+ GAUDI2_DCORE3_ENGINE_ID_TPC_0,
+ [GAUDI2_QUEUE_ID_DCORE3_TPC_1_0...GAUDI2_QUEUE_ID_DCORE3_TPC_1_3] =
+ GAUDI2_DCORE3_ENGINE_ID_TPC_1,
+ [GAUDI2_QUEUE_ID_DCORE3_TPC_2_0...GAUDI2_QUEUE_ID_DCORE3_TPC_2_3] =
+ GAUDI2_DCORE3_ENGINE_ID_TPC_2,
+ [GAUDI2_QUEUE_ID_DCORE3_TPC_3_0...GAUDI2_QUEUE_ID_DCORE3_TPC_3_3] =
+ GAUDI2_DCORE3_ENGINE_ID_TPC_3,
+ [GAUDI2_QUEUE_ID_DCORE3_TPC_4_0...GAUDI2_QUEUE_ID_DCORE3_TPC_4_3] =
+ GAUDI2_DCORE3_ENGINE_ID_TPC_4,
+ [GAUDI2_QUEUE_ID_DCORE3_TPC_5_0...GAUDI2_QUEUE_ID_DCORE3_TPC_5_3] =
+ GAUDI2_DCORE3_ENGINE_ID_TPC_5,
+ [GAUDI2_QUEUE_ID_NIC_0_0...GAUDI2_QUEUE_ID_NIC_0_3] = GAUDI2_ENGINE_ID_NIC0_0,
+ [GAUDI2_QUEUE_ID_NIC_1_0...GAUDI2_QUEUE_ID_NIC_1_3] = GAUDI2_ENGINE_ID_NIC0_1,
+ [GAUDI2_QUEUE_ID_NIC_2_0...GAUDI2_QUEUE_ID_NIC_2_3] = GAUDI2_ENGINE_ID_NIC1_0,
+ [GAUDI2_QUEUE_ID_NIC_3_0...GAUDI2_QUEUE_ID_NIC_3_3] = GAUDI2_ENGINE_ID_NIC1_1,
+ [GAUDI2_QUEUE_ID_NIC_4_0...GAUDI2_QUEUE_ID_NIC_4_3] = GAUDI2_ENGINE_ID_NIC2_0,
+ [GAUDI2_QUEUE_ID_NIC_5_0...GAUDI2_QUEUE_ID_NIC_5_3] = GAUDI2_ENGINE_ID_NIC2_1,
+ [GAUDI2_QUEUE_ID_NIC_6_0...GAUDI2_QUEUE_ID_NIC_6_3] = GAUDI2_ENGINE_ID_NIC3_0,
+ [GAUDI2_QUEUE_ID_NIC_7_0...GAUDI2_QUEUE_ID_NIC_7_3] = GAUDI2_ENGINE_ID_NIC3_1,
+ [GAUDI2_QUEUE_ID_NIC_8_0...GAUDI2_QUEUE_ID_NIC_8_3] = GAUDI2_ENGINE_ID_NIC4_0,
+ [GAUDI2_QUEUE_ID_NIC_9_0...GAUDI2_QUEUE_ID_NIC_9_3] = GAUDI2_ENGINE_ID_NIC4_1,
+ [GAUDI2_QUEUE_ID_NIC_10_0...GAUDI2_QUEUE_ID_NIC_10_3] = GAUDI2_ENGINE_ID_NIC5_0,
+ [GAUDI2_QUEUE_ID_NIC_11_0...GAUDI2_QUEUE_ID_NIC_11_3] = GAUDI2_ENGINE_ID_NIC5_1,
+ [GAUDI2_QUEUE_ID_NIC_12_0...GAUDI2_QUEUE_ID_NIC_12_3] = GAUDI2_ENGINE_ID_NIC6_0,
+ [GAUDI2_QUEUE_ID_NIC_13_0...GAUDI2_QUEUE_ID_NIC_13_3] = GAUDI2_ENGINE_ID_NIC6_1,
+ [GAUDI2_QUEUE_ID_NIC_14_0...GAUDI2_QUEUE_ID_NIC_14_3] = GAUDI2_ENGINE_ID_NIC7_0,
+ [GAUDI2_QUEUE_ID_NIC_15_0...GAUDI2_QUEUE_ID_NIC_15_3] = GAUDI2_ENGINE_ID_NIC7_1,
+ [GAUDI2_QUEUE_ID_NIC_16_0...GAUDI2_QUEUE_ID_NIC_16_3] = GAUDI2_ENGINE_ID_NIC8_0,
+ [GAUDI2_QUEUE_ID_NIC_17_0...GAUDI2_QUEUE_ID_NIC_17_3] = GAUDI2_ENGINE_ID_NIC8_1,
+ [GAUDI2_QUEUE_ID_NIC_18_0...GAUDI2_QUEUE_ID_NIC_18_3] = GAUDI2_ENGINE_ID_NIC9_0,
+ [GAUDI2_QUEUE_ID_NIC_19_0...GAUDI2_QUEUE_ID_NIC_19_3] = GAUDI2_ENGINE_ID_NIC9_1,
+ [GAUDI2_QUEUE_ID_NIC_20_0...GAUDI2_QUEUE_ID_NIC_20_3] = GAUDI2_ENGINE_ID_NIC10_0,
+ [GAUDI2_QUEUE_ID_NIC_21_0...GAUDI2_QUEUE_ID_NIC_21_3] = GAUDI2_ENGINE_ID_NIC10_1,
+ [GAUDI2_QUEUE_ID_NIC_22_0...GAUDI2_QUEUE_ID_NIC_22_3] = GAUDI2_ENGINE_ID_NIC11_0,
+ [GAUDI2_QUEUE_ID_NIC_23_0...GAUDI2_QUEUE_ID_NIC_23_3] = GAUDI2_ENGINE_ID_NIC11_1,
+ [GAUDI2_QUEUE_ID_ROT_0_0...GAUDI2_QUEUE_ID_ROT_0_3] = GAUDI2_ENGINE_ID_ROT_0,
+ [GAUDI2_QUEUE_ID_ROT_1_0...GAUDI2_QUEUE_ID_ROT_1_3] = GAUDI2_ENGINE_ID_ROT_1,
+};
+
const u32 gaudi2_qm_blocks_bases[GAUDI2_QUEUE_ID_SIZE] = {
[GAUDI2_QUEUE_ID_PDMA_0_0] = mmPDMA0_QM_BASE,
[GAUDI2_QUEUE_ID_PDMA_0_1] = mmPDMA0_QM_BASE,
@@ -7753,7 +7858,7 @@ static bool gaudi2_handle_ecc_event(struct hl_device *hdev, u16 event_type,
return !!ecc_data->is_critical;
}
-static void print_lower_qman_data_on_err(struct hl_device *hdev, u64 qman_base)
+static void handle_lower_qman_data_on_err(struct hl_device *hdev, u64 qman_base, u64 event_mask)
{
u32 lo, hi, cq_ptr_size, arc_cq_ptr_size;
u64 cq_ptr, arc_cq_ptr, cp_current_inst;
@@ -7775,10 +7880,22 @@ static void print_lower_qman_data_on_err(struct hl_device *hdev, u64 qman_base)
dev_info(hdev->dev,
"LowerQM. CQ: {ptr %#llx, size %u}, ARC_CQ: {ptr %#llx, size %u}, CP: {instruction %#llx}\n",
cq_ptr, cq_ptr_size, arc_cq_ptr, arc_cq_ptr_size, cp_current_inst);
+
+ if (event_mask & HL_NOTIFIER_EVENT_UNDEFINED_OPCODE) {
+ if (arc_cq_ptr) {
+ hdev->captured_err_info.undef_opcode.cq_addr = arc_cq_ptr;
+ hdev->captured_err_info.undef_opcode.cq_size = arc_cq_ptr_size;
+ } else {
+ hdev->captured_err_info.undef_opcode.cq_addr = cq_ptr;
+ hdev->captured_err_info.undef_opcode.cq_size = cq_ptr_size;
+ }
+
+ hdev->captured_err_info.undef_opcode.stream_id = QMAN_STREAMS;
+ }
}
static int gaudi2_handle_qman_err_generic(struct hl_device *hdev, u16 event_type,
- u64 qman_base, u32 qid_base)
+ u64 qman_base, u32 qid_base, u64 *event_mask)
{
u32 i, j, glbl_sts_val, arb_err_val, num_error_causes, error_count = 0;
u64 glbl_sts_addr, arb_err_addr;
@@ -7812,8 +7929,22 @@ static int gaudi2_handle_qman_err_generic(struct hl_device *hdev, u16 event_type
error_count++;
}
- if (i == QMAN_STREAMS)
- print_lower_qman_data_on_err(hdev, qman_base);
+ if (i == QMAN_STREAMS && error_count) {
+ /* check for undefined opcode */
+ if (glbl_sts_val & PDMA0_QM_GLBL_ERR_STS_CP_UNDEF_CMD_ERR_MASK &&
+ hdev->captured_err_info.undef_opcode.write_enable) {
+ memset(&hdev->captured_err_info.undef_opcode, 0,
+ sizeof(hdev->captured_err_info.undef_opcode));
+
+ hdev->captured_err_info.undef_opcode.write_enable = false;
+ hdev->captured_err_info.undef_opcode.timestamp = ktime_get();
+ hdev->captured_err_info.undef_opcode.engine_id =
+ gaudi2_queue_id_to_engine_id[qid_base];
+ *event_mask |= HL_NOTIFIER_EVENT_UNDEFINED_OPCODE;
+ }
+
+ handle_lower_qman_data_on_err(hdev, qman_base, *event_mask);
+ }
}
arb_err_val = RREG32(arb_err_addr);
@@ -8475,7 +8606,8 @@ static int gaudi2_handle_qman_err(struct hl_device *hdev, u16 event_type, u64 *e
return 0;
}
- error_count = gaudi2_handle_qman_err_generic(hdev, event_type, qman_base, qid_base);
+ error_count = gaudi2_handle_qman_err_generic(hdev, event_type, qman_base,
+ qid_base, event_mask);
/* Handle EDMA QM SEI here because there is no AXI error response event for EDMA */
if (event_type >= GAUDI2_EVENT_HDMA2_QM && event_type <= GAUDI2_EVENT_HDMA5_QM) {
--
2.40.1
next prev parent reply other threads:[~2023-06-08 13:39 UTC|newest]
Thread overview: 12+ messages / expand[flat|nested] mbox.gz Atom feed top
2023-06-08 13:38 [PATCH 01/12] accel/habanalabs: prevent immediate hard reset due to 2 adjacent H/W events Oded Gabbay
2023-06-08 13:38 ` [PATCH 02/12] accel/habanalabs: update pending reset flags with new reset requests Oded Gabbay
2023-06-08 13:38 ` Oded Gabbay [this message]
2023-06-08 13:38 ` [PATCH 04/12] accel/habanalabs: print task name and request code upon ioctl failure Oded Gabbay
2023-06-08 13:38 ` [PATCH 05/12] accel/habanalabs: print task name upon creation of a user context Oded Gabbay
2023-06-08 13:38 ` [PATCH 06/12] accel/habanalabs: set device status 'malfunction' while in rmmod Oded Gabbay
2023-06-08 13:38 ` [PATCH 07/12] accel/habanalabs: stop fetching MME SBTE error cause Oded Gabbay
2023-06-08 13:38 ` [PATCH 08/12] accel/habanalabs: handle arc farm razwi Oded Gabbay
2023-06-08 13:38 ` [PATCH 09/12] accel/habanalabs: fix standalone preboot descriptor request Oded Gabbay
2023-06-08 13:38 ` [PATCH 10/12] accel/habanalabs: print return code when process termination fails Oded Gabbay
2023-06-08 13:38 ` [PATCH 11/12] accel/habanalabs: call put_pid after hpriv list is updated Oded Gabbay
2023-06-08 13:38 ` [PATCH 12/12] accel/habanalabs: rename fd_list to hpriv_list Oded Gabbay
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=20230608133849.2739411-3-ogabbay@kernel.org \
--to=ogabbay@kernel.org \
--cc=dri-devel@lists.freedesktop.org \
--cc=obitton@habana.ai \
/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