Commit b291e4f1 authored by Lizhi Hou's avatar Lizhi Hou
Browse files

accel/amdxdna: Support getting last hardware error



Add new parameter DRM_AMDXDNA_HW_LAST_ASYNC_ERR to get array IOCTL. When
hardware reports an error, the driver save the error information and
timestamp. This new get array parameter retrieves the last error.

Reviewed-by: default avatarMario Limonciello (AMD) <superm1@kernel.org>
Signed-off-by: default avatarLizhi Hou <lizhi.hou@amd.com>
Link: https://lore.kernel.org/r/20251014234119.628453-1-lizhi.hou@amd.com
parent 83f81f54
Loading
Loading
Loading
Loading
+78 −17
Original line number Diff line number Diff line
@@ -13,6 +13,7 @@

#include "aie2_msg_priv.h"
#include "aie2_pci.h"
#include "amdxdna_error.h"
#include "amdxdna_mailbox.h"
#include "amdxdna_pci_drv.h"

@@ -46,6 +47,7 @@ enum aie_module_type {
	AIE_MEM_MOD = 0,
	AIE_CORE_MOD,
	AIE_PL_MOD,
	AIE_UNKNOWN_MOD,
};

enum aie_error_category {
@@ -143,6 +145,31 @@ static const struct aie_event_category aie_ml_shim_tile_event_cat[] = {
	EVENT_CATEGORY(74U, AIE_ERROR_LOCK),
};

static const enum amdxdna_error_num aie_cat_err_num_map[] = {
	[AIE_ERROR_SATURATION] = AMDXDNA_ERROR_NUM_AIE_SATURATION,
	[AIE_ERROR_FP] = AMDXDNA_ERROR_NUM_AIE_FP,
	[AIE_ERROR_STREAM] = AMDXDNA_ERROR_NUM_AIE_STREAM,
	[AIE_ERROR_ACCESS] = AMDXDNA_ERROR_NUM_AIE_ACCESS,
	[AIE_ERROR_BUS] = AMDXDNA_ERROR_NUM_AIE_BUS,
	[AIE_ERROR_INSTRUCTION] = AMDXDNA_ERROR_NUM_AIE_INSTRUCTION,
	[AIE_ERROR_ECC] = AMDXDNA_ERROR_NUM_AIE_ECC,
	[AIE_ERROR_LOCK] = AMDXDNA_ERROR_NUM_AIE_LOCK,
	[AIE_ERROR_DMA] = AMDXDNA_ERROR_NUM_AIE_DMA,
	[AIE_ERROR_MEM_PARITY] = AMDXDNA_ERROR_NUM_AIE_MEM_PARITY,
	[AIE_ERROR_UNKNOWN] = AMDXDNA_ERROR_NUM_UNKNOWN,
};

static_assert(ARRAY_SIZE(aie_cat_err_num_map) == AIE_ERROR_UNKNOWN + 1);

static const enum amdxdna_error_module aie_err_mod_map[] = {
	[AIE_MEM_MOD] = AMDXDNA_ERROR_MODULE_AIE_MEMORY,
	[AIE_CORE_MOD] = AMDXDNA_ERROR_MODULE_AIE_CORE,
	[AIE_PL_MOD] = AMDXDNA_ERROR_MODULE_AIE_PL,
	[AIE_UNKNOWN_MOD] = AMDXDNA_ERROR_MODULE_UNKNOWN,
};

static_assert(ARRAY_SIZE(aie_err_mod_map) == AIE_UNKNOWN_MOD + 1);

static enum aie_error_category
aie_get_error_category(u8 row, u8 event_id, enum aie_module_type mod_type)
{
@@ -176,12 +203,40 @@ aie_get_error_category(u8 row, u8 event_id, enum aie_module_type mod_type)
		if (event_id != lut[i].event_id)
			continue;

		if (lut[i].category > AIE_ERROR_UNKNOWN)
			return AIE_ERROR_UNKNOWN;

		return lut[i].category;
	}

	return AIE_ERROR_UNKNOWN;
}

static void aie2_update_last_async_error(struct amdxdna_dev_hdl *ndev, void *err_info, u32 num_err)
{
	struct aie_error *errs = err_info;
	enum amdxdna_error_module err_mod;
	enum aie_error_category aie_err;
	enum amdxdna_error_num err_num;
	struct aie_error *last_err;

	last_err = &errs[num_err - 1];
	if (last_err->mod_type >= AIE_UNKNOWN_MOD) {
		err_num = aie_cat_err_num_map[AIE_ERROR_UNKNOWN];
		err_mod = aie_err_mod_map[AIE_UNKNOWN_MOD];
	} else {
		aie_err = aie_get_error_category(last_err->row,
						 last_err->event_id,
						 last_err->mod_type);
		err_num = aie_cat_err_num_map[aie_err];
		err_mod = aie_err_mod_map[last_err->mod_type];
	}

	ndev->last_async_err.err_code = AMDXDNA_ERROR_ENCODE(err_num, err_mod);
	ndev->last_async_err.ts_us = ktime_to_us(ktime_get_real());
	ndev->last_async_err.ex_err_code = AMDXDNA_EXTRA_ERR_ENCODE(last_err->row, last_err->col);
}

static u32 aie2_error_backtrack(struct amdxdna_dev_hdl *ndev, void *err_info, u32 num_err)
{
	struct aie_error *errs = err_info;
@@ -264,29 +319,14 @@ static void aie2_error_worker(struct work_struct *err_work)
	}

	mutex_lock(&xdna->dev_lock);
	aie2_update_last_async_error(e->ndev, info->payload, info->err_cnt);

	/* Re-sent this event to firmware */
	if (aie2_error_event_send(e))
		XDNA_WARN(xdna, "Unable to register async event");
	mutex_unlock(&xdna->dev_lock);
}

int aie2_error_async_events_send(struct amdxdna_dev_hdl *ndev)
{
	struct amdxdna_dev *xdna = ndev->xdna;
	struct async_event *e;
	int i, ret;

	drm_WARN_ON(&xdna->ddev, !mutex_is_locked(&xdna->dev_lock));
	for (i = 0; i < ndev->async_events->event_cnt; i++) {
		e = &ndev->async_events->event[i];
		ret = aie2_error_event_send(e);
		if (ret)
			return ret;
	}

	return 0;
}

void aie2_error_async_events_free(struct amdxdna_dev_hdl *ndev)
{
	struct amdxdna_dev *xdna = ndev->xdna;
@@ -341,6 +381,10 @@ int aie2_error_async_events_alloc(struct amdxdna_dev_hdl *ndev)
		e->size = ASYNC_BUF_SIZE;
		e->resp.status = MAX_AIE2_STATUS_CODE;
		INIT_WORK(&e->work, aie2_error_worker);

		ret = aie2_error_event_send(e);
		if (ret)
			goto free_wq;
	}

	ndev->async_events = events;
@@ -349,6 +393,8 @@ int aie2_error_async_events_alloc(struct amdxdna_dev_hdl *ndev)
		 events->event_cnt, events->size);
	return 0;

free_wq:
	destroy_workqueue(events->wq);
free_buf:
	dma_free_noncoherent(xdna->ddev.dev, events->size, events->buf,
			     events->addr, DMA_FROM_DEVICE);
@@ -356,3 +402,18 @@ int aie2_error_async_events_alloc(struct amdxdna_dev_hdl *ndev)
	kfree(events);
	return ret;
}

int aie2_get_array_async_error(struct amdxdna_dev_hdl *ndev, struct amdxdna_drm_get_array *args)
{
	struct amdxdna_dev *xdna = ndev->xdna;

	drm_WARN_ON(&xdna->ddev, !mutex_is_locked(&xdna->dev_lock));

	args->num_element = 1;
	args->element_size = sizeof(ndev->last_async_err);
	if (copy_to_user(u64_to_user_ptr(args->buffer),
			 &ndev->last_async_err, args->element_size))
		return -EFAULT;

	return 0;
}
+3 −0
Original line number Diff line number Diff line
@@ -924,6 +924,9 @@ static int aie2_get_array(struct amdxdna_client *client,
	case DRM_AMDXDNA_HW_CONTEXT_ALL:
		ret = aie2_query_ctx_status_array(client, args);
		break;
	case DRM_AMDXDNA_HW_LAST_ASYNC_ERR:
		ret = aie2_get_array_async_error(xdna->dev_handle, args);
		break;
	default:
		XDNA_ERR(xdna, "Not supported request parameter %u", args->param);
		ret = -EOPNOTSUPP;
+4 −1
Original line number Diff line number Diff line
@@ -190,6 +190,8 @@ struct amdxdna_dev_hdl {

	enum aie2_dev_status		dev_status;
	u32				hwctx_num;

	struct amdxdna_async_error	last_async_err;
};

#define DEFINE_BAR_OFFSET(reg_name, bar, reg_addr) \
@@ -253,8 +255,9 @@ void aie2_psp_stop(struct psp_device *psp);
/* aie2_error.c */
int aie2_error_async_events_alloc(struct amdxdna_dev_hdl *ndev);
void aie2_error_async_events_free(struct amdxdna_dev_hdl *ndev);
int aie2_error_async_events_send(struct amdxdna_dev_hdl *ndev);
int aie2_error_async_msg_thread(void *data);
int aie2_get_array_async_error(struct amdxdna_dev_hdl *ndev,
			       struct amdxdna_drm_get_array *args);

/* aie2_message.c */
int aie2_suspend_fw(struct amdxdna_dev_hdl *ndev);
+59 −0
Original line number Diff line number Diff line
/* SPDX-License-Identifier: GPL-2.0 */
/*
 * Copyright (C) 2025, Advanced Micro Devices, Inc.
 */

#ifndef _AMDXDNA_ERROR_H_
#define _AMDXDNA_ERROR_H_

#include <linux/bitfield.h>
#include <linux/bits.h>

#define AMDXDNA_ERR_DRV_AIE		4
#define AMDXDNA_ERR_SEV_CRITICAL	3
#define AMDXDNA_ERR_CLASS_AIE		2

#define AMDXDNA_ERR_NUM_MASK		GENMASK_U64(15, 0)
#define AMDXDNA_ERR_DRV_MASK		GENMASK_U64(23, 16)
#define AMDXDNA_ERR_SEV_MASK		GENMASK_U64(31, 24)
#define AMDXDNA_ERR_MOD_MASK		GENMASK_U64(39, 32)
#define AMDXDNA_ERR_CLASS_MASK		GENMASK_U64(47, 40)

enum amdxdna_error_num {
	AMDXDNA_ERROR_NUM_AIE_SATURATION = 3,
	AMDXDNA_ERROR_NUM_AIE_FP,
	AMDXDNA_ERROR_NUM_AIE_STREAM,
	AMDXDNA_ERROR_NUM_AIE_ACCESS,
	AMDXDNA_ERROR_NUM_AIE_BUS,
	AMDXDNA_ERROR_NUM_AIE_INSTRUCTION,
	AMDXDNA_ERROR_NUM_AIE_ECC,
	AMDXDNA_ERROR_NUM_AIE_LOCK,
	AMDXDNA_ERROR_NUM_AIE_DMA,
	AMDXDNA_ERROR_NUM_AIE_MEM_PARITY,
	AMDXDNA_ERROR_NUM_UNKNOWN = 15,
};

enum amdxdna_error_module {
	AMDXDNA_ERROR_MODULE_AIE_CORE = 3,
	AMDXDNA_ERROR_MODULE_AIE_MEMORY,
	AMDXDNA_ERROR_MODULE_AIE_SHIM,
	AMDXDNA_ERROR_MODULE_AIE_NOC,
	AMDXDNA_ERROR_MODULE_AIE_PL,
	AMDXDNA_ERROR_MODULE_UNKNOWN = 8,
};

#define AMDXDNA_ERROR_ENCODE(err_num, err_mod)				\
	(FIELD_PREP(AMDXDNA_ERR_NUM_MASK, err_num) |			\
	 FIELD_PREP_CONST(AMDXDNA_ERR_DRV_MASK, AMDXDNA_ERR_DRV_AIE) |	\
	 FIELD_PREP_CONST(AMDXDNA_ERR_SEV_MASK, AMDXDNA_ERR_SEV_CRITICAL) | \
	 FIELD_PREP(AMDXDNA_ERR_MOD_MASK, err_mod) |			\
	 FIELD_PREP_CONST(AMDXDNA_ERR_CLASS_MASK, AMDXDNA_ERR_CLASS_AIE))

#define AMDXDNA_EXTRA_ERR_COL_MASK	GENMASK_U64(7, 0)
#define AMDXDNA_EXTRA_ERR_ROW_MASK	GENMASK_U64(15, 8)

#define AMDXDNA_EXTRA_ERR_ENCODE(row, col)				\
	(FIELD_PREP(AMDXDNA_EXTRA_ERR_COL_MASK, col) |			\
	 FIELD_PREP(AMDXDNA_EXTRA_ERR_ROW_MASK, row))

#endif /* _AMDXDNA_ERROR_H_ */
+2 −1
Original line number Diff line number Diff line
@@ -27,9 +27,10 @@ MODULE_FIRMWARE("amdnpu/17f0_20/npu.sbin");
/*
 * 0.0: Initial version
 * 0.1: Support getting all hardware contexts by DRM_IOCTL_AMDXDNA_GET_ARRAY
 * 0.2: Support getting last error hardware error
 */
#define AMDXDNA_DRIVER_MAJOR		0
#define AMDXDNA_DRIVER_MINOR		1
#define AMDXDNA_DRIVER_MINOR		2

/*
 * Bind the driver base on (vendor_id, device_id) pair and later use the
Loading