Commit a00f1f57 authored by Oded Gabbay's avatar Oded Gabbay
Browse files

habanalabs: define soft-reset as inference op 



Soft-reset is the procedure where we reset only the compute/DMA engines
of the device, without requiring the current user-space process to
release the device.

This type of reset can happen if TDR event occurred (a workload got
stuck) or by a root request through sysfs.

This is only relevant for inference ASICs, as there is no real-world
use-case to do that in training, because training runs on multiple
devices.

In addition, we also do (in certain ASICs) a reset upon device release.
That reset uses the same code as the soft-reset.

Therefore, to better differentiate between the two resets, it is better
to rename the soft-reset support as "inference soft-reset", to make
the code more self-explanatory.

Signed-off-by: default avatarOded Gabbay <ogabbay@kernel.org>
parent dd08335f

drivers/misc/habanalabs/common/device.c

+1 −1
Original line number Diff line number Diff line
@@ -954,7 +954,7 @@ int hl_device_reset(struct hl_device *hdev, u32 flags) 
		goto do_reset;

	}



	if (!hard_reset && !hdev->allow_external_soft_reset) {

	if (!hard_reset && !hdev->allow_inference_soft_reset) {

		hard_instead_soft = true;

		hard_reset = true;

	}

drivers/misc/habanalabs/common/habanalabs.h

+7 −3
Original line number Diff line number Diff line
@@ -2440,8 +2440,12 @@ struct multi_cs_data { 
 * @collective_mon_idx: helper index for collective initialization

 * @supports_coresight: is CoreSight supported.

 * @supports_soft_reset: is soft reset supported.

 * @allow_external_soft_reset: true if soft reset initiated by user or TDR is

 *                             allowed.

 * @allow_inference_soft_reset: true if the ASIC supports soft reset that is

 *                              initiated by user or TDR. This is only true

 *                              in inference ASICs, as there is no real-world

 *                              use-case of doing soft-reset in training (due

 *                              to the fact that training runs on multiple

 *                              devices)

 * @supports_cb_mapping: is mapping a CB to the device's MMU supported.

 * @needs_reset: true if reset_on_lockup is false and device should be reset

 *               due to lockup.
@@ -2572,7 +2576,7 @@ struct hl_device { 
	u8				collective_mon_idx;

	u8				supports_coresight;

	u8				supports_soft_reset;

	u8				allow_external_soft_reset;

	u8				allow_inference_soft_reset;

	u8				supports_cb_mapping;

	u8				needs_reset;

	u8				process_kill_trial_cnt;

drivers/misc/habanalabs/common/sysfs.c

+3 −3
Original line number Diff line number Diff line
@@ -206,12 +206,12 @@ static ssize_t soft_reset_store(struct device *dev, 
		goto out;

	}



	if (!hdev->allow_external_soft_reset) {

		dev_err(hdev->dev, "Device does not support soft-reset\n");

	if (!hdev->allow_inference_soft_reset) {

		dev_err(hdev->dev, "Device does not support inference soft-reset\n");

		goto out;

	}



	dev_warn(hdev->dev, "Soft-Reset requested through sysfs\n");

	dev_warn(hdev->dev, "Inference Soft-Reset requested through sysfs\n");



	hl_device_reset(hdev, 0);

drivers/misc/habanalabs/goya/goya.c

+1 −1
Original line number Diff line number Diff line
@@ -959,7 +959,7 @@ static int goya_sw_init(struct hl_device *hdev) 
	spin_lock_init(&goya->hw_queues_lock);

	hdev->supports_coresight = true;

	hdev->supports_soft_reset = true;

	hdev->allow_external_soft_reset = true;

	hdev->allow_inference_soft_reset = true;

	hdev->supports_wait_for_multi_cs = false;



	hdev->asic_funcs->set_pci_memory_regions(hdev);