Re: [PATCH 5/6] habanalabs: use notifications and graceful reset for decoder

[Stanislaw Gruszka]

On Mon, Feb 27, 2023 at 01:13:05PM +0200, Oded Gabbay wrote:
> From: Tomer Tayar 
> 
> Add notifications to user in case of decoder abnormal interrupts, and
> use the graceful reset mechanism if reset is required.
> 
> Signed-off-by: Tomer Tayar 
> Reviewed-by: Oded Gabbay 
> Signed-off-by: Oded Gabbay 
Reviewed-by: Stanislaw Gruszka 

[PATCH 5/6] habanalabs: use notifications and graceful reset for decoder

[Oded Gabbay]

From: Tomer Tayar 

Add notifications to user in case of decoder abnormal interrupts, and
use the graceful reset mechanism if reset is required.

Signed-off-by: Tomer Tayar 
Reviewed-by: Oded Gabbay 
Signed-off-by: Oded Gabbay 
---
 drivers/accel/habanalabs/common/decoder.c | 22 --
 1 file changed, 16 insertions(+), 6 deletions(-)

diff --git a/drivers/accel/habanalabs/common/decoder.c 
b/drivers/accel/habanalabs/common/decoder.c
index 2aab14d74b53..69c78c1784b4 100644
--- a/drivers/accel/habanalabs/common/decoder.c
+++ b/drivers/accel/habanalabs/common/decoder.c
@@ -46,7 +46,7 @@ static void dec_print_abnrm_intr_source(struct hl_device 
*hdev, u32 irq_status)
 static void dec_error_intr_work(struct hl_device *hdev, u32 base_addr, u32 
core_id)
 {
bool reset_required = false;
-   u32 irq_status;
+   u32 irq_status, event_mask;
 
irq_status = RREG32(base_addr + VCMD_IRQ_STATUS_OFFSET);
 
@@ -54,17 +54,27 @@ static void dec_error_intr_work(struct hl_device *hdev, u32 
base_addr, u32 core_
 
dec_print_abnrm_intr_source(hdev, irq_status);
 
-   if (irq_status & VCMD_IRQ_STATUS_TIMEOUT_MASK)
-   reset_required = true;
-
/* Clear the interrupt */
WREG32(base_addr + VCMD_IRQ_STATUS_OFFSET, irq_status);
 
/* Flush the interrupt clear */
RREG32(base_addr + VCMD_IRQ_STATUS_OFFSET);
 
-   if (reset_required)
-   hl_device_reset(hdev, HL_DRV_RESET_HARD);
+   if (irq_status & VCMD_IRQ_STATUS_TIMEOUT_MASK) {
+   reset_required = true;
+   event_mask = HL_NOTIFIER_EVENT_GENERAL_HW_ERR;
+   } else if (irq_status & VCMD_IRQ_STATUS_CMDERR_MASK) {
+   event_mask = HL_NOTIFIER_EVENT_UNDEFINED_OPCODE;
+   } else {
+   event_mask = HL_NOTIFIER_EVENT_USER_ENGINE_ERR;
+   }
+
+   if (reset_required) {
+   event_mask |= HL_NOTIFIER_EVENT_DEVICE_RESET;
+   hl_device_cond_reset(hdev, 0, event_mask);
+   } else {
+   hl_notifier_event_send_all(hdev, event_mask);
+   }
 }
 
 static void dec_completion_abnrm(struct work_struct *work)
-- 
2.39.2