habanalabs/gaudi2: reset device upon critical ECC event

Correctable ECC events are not fatal, but as they accumulate, the f/w
can decide that a hard-rest is required. This indication is
propagated to the host using the existing ECC event interface.

Signed-off-by: Ofir Bitton <obitton@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
This commit is contained in:
Ofir Bitton 2022-06-28 08:34:28 +03:00 committed by Oded Gabbay
parent 6b4e8a12b2
commit a85e389a84
2 changed files with 16 additions and 11 deletions

View file

@ -6637,7 +6637,7 @@ static void gaudi2_print_irq_info(struct hl_device *hdev, u16 event_type)
event_type, desc);
}
static void gaudi2_handle_ecc_event(struct hl_device *hdev, u16 event_type,
static bool gaudi2_handle_ecc_event(struct hl_device *hdev, u16 event_type,
struct hl_eq_ecc_data *ecc_data)
{
u64 ecc_address = 0, ecc_syndrom = 0;
@ -6647,8 +6647,11 @@ static void gaudi2_handle_ecc_event(struct hl_device *hdev, u16 event_type,
ecc_syndrom = le64_to_cpu(ecc_data->ecc_syndrom);
memory_wrapper_idx = ecc_data->memory_wrapper_idx;
dev_err(hdev->dev, "ECC error detected. address: %#llx. Syndrom: %#llx. block id %u\n",
ecc_address, ecc_syndrom, memory_wrapper_idx);
dev_err(hdev->dev,
"ECC error detected. address: %#llx. Syndrom: %#llx. block id %u. critical %u.\n",
ecc_address, ecc_syndrom, memory_wrapper_idx, ecc_data->is_critical);
return !!ecc_data->is_critical;
}
/*
@ -7991,9 +7994,9 @@ static bool gaudi2_handle_hbm_mc_sei_err(struct hl_device *hdev, u16 event_type,
}
dev_err_ratelimited(hdev->dev,
"System Error Interrupt - HBM(%u) MC(%u) MC_CH(%u) MC_PC(%u). Error cause: %s\n",
hbm_id, mc_id, sei_data->hdr.mc_channel, sei_data->hdr.mc_pseudo_channel,
hbm_mc_sei_cause[cause_idx]);
"System Error Interrupt - HBM(%u) MC(%u) MC_CH(%u) MC_PC(%u). Critical(%u). Error cause: %s\n",
hbm_id, mc_id, sei_data->hdr.mc_channel, sei_data->hdr.mc_pseudo_channel,
sei_data->hdr.is_critical, hbm_mc_sei_cause[cause_idx]);
/* Print error-specific info */
switch (cause_idx) {
@ -8032,6 +8035,8 @@ static bool gaudi2_handle_hbm_mc_sei_err(struct hl_device *hdev, u16 event_type,
break;
}
require_hard_reset |= !!sei_data->hdr.is_critical;
return require_hard_reset;
}
@ -8199,7 +8204,7 @@ static void gaudi2_handle_eqe(struct hl_device *hdev, struct hl_eq_entry *eq_ent
{
u32 ctl, reset_flags = HL_DRV_RESET_HARD | HL_DRV_RESET_DELAY;
struct gaudi2_device *gaudi2 = hdev->asic_specific;
bool hbm_require_reset = false, skip_reset = false;
bool reset_required = false, skip_reset = false;
int index, sbte_index;
u16 event_type;
@ -8222,7 +8227,7 @@ static void gaudi2_handle_eqe(struct hl_device *hdev, struct hl_eq_entry *eq_ent
fallthrough;
case GAUDI2_EVENT_ROTATOR0_SERR ... GAUDI2_EVENT_ROTATOR1_DERR:
reset_flags |= HL_DRV_RESET_FW_FATAL_ERR;
gaudi2_handle_ecc_event(hdev, event_type, &eq_entry->ecc_data);
reset_required = gaudi2_handle_ecc_event(hdev, event_type, &eq_entry->ecc_data);
break;
case GAUDI2_EVENT_TPC0_QM ... GAUDI2_EVENT_PDMA1_QM:
@ -8387,7 +8392,7 @@ static void gaudi2_handle_eqe(struct hl_device *hdev, struct hl_eq_entry *eq_ent
case GAUDI2_EVENT_HBM0_MC0_SEI_SEVERE ... GAUDI2_EVENT_HBM5_MC1_SEI_NON_SEVERE:
if (gaudi2_handle_hbm_mc_sei_err(hdev, event_type, &eq_entry->sei_data)) {
reset_flags |= HL_DRV_RESET_FW_FATAL_ERR;
hbm_require_reset = true;
reset_required = true;
}
break;
@ -8539,7 +8544,7 @@ static void gaudi2_handle_eqe(struct hl_device *hdev, struct hl_eq_entry *eq_ent
event_type);
}
if ((gaudi2_irq_map_table[event_type].reset || hbm_require_reset) && !skip_reset)
if ((gaudi2_irq_map_table[event_type].reset || reset_required) && !skip_reset)
goto reset_device;
/* Send unmask irq only for interrupts not classified as MSG */

View file

@ -192,7 +192,7 @@ struct hl_hbm_sei_header {
__u8 sei_cause; /* enum hl_hbm_sei_cause */
__u8 mc_channel; /* range: 0-3 */
__u8 mc_pseudo_channel; /* range: 0-7 */
__u8 pad[1];
__u8 is_critical;
};
#define HBM_RD_ADDR_SID_SHIFT 0