From: Oded Gabbay <ogabbay@kernel.org>
To: linux-kernel@vger.kernel.org
Subject: [PATCH 3/6] habanalabs: define soft-reset as inference op
Date: Mon, 4 Oct 2021 13:08:27 +0300 [thread overview]
Message-ID: <20211004100830.27573-3-ogabbay@kernel.org> (raw)
In-Reply-To: <20211004100830.27573-1-ogabbay@kernel.org>
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: Oded Gabbay <ogabbay@kernel.org>
---
drivers/misc/habanalabs/common/device.c | 2 +-
drivers/misc/habanalabs/common/habanalabs.h | 10 +++++++---
drivers/misc/habanalabs/common/sysfs.c | 6 +++---
drivers/misc/habanalabs/goya/goya.c | 2 +-
4 files changed, 12 insertions(+), 8 deletions(-)
diff --git a/drivers/misc/habanalabs/common/device.c b/drivers/misc/habanalabs/common/device.c
index 97c7c86580e6..8f96f2bb14ec 100644
--- a/drivers/misc/habanalabs/common/device.c
+++ b/drivers/misc/habanalabs/common/device.c
@@ -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;
}
diff --git a/drivers/misc/habanalabs/common/habanalabs.h b/drivers/misc/habanalabs/common/habanalabs.h
index f8e23ca18a57..99de80915eed 100644
--- a/drivers/misc/habanalabs/common/habanalabs.h
+++ b/drivers/misc/habanalabs/common/habanalabs.h
@@ -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;
diff --git a/drivers/misc/habanalabs/common/sysfs.c b/drivers/misc/habanalabs/common/sysfs.c
index 34f9f2779962..42c1769ad25d 100644
--- a/drivers/misc/habanalabs/common/sysfs.c
+++ b/drivers/misc/habanalabs/common/sysfs.c
@@ -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);
diff --git a/drivers/misc/habanalabs/goya/goya.c b/drivers/misc/habanalabs/goya/goya.c
index 031c1849da14..ef67e2586ede 100644
--- a/drivers/misc/habanalabs/goya/goya.c
+++ b/drivers/misc/habanalabs/goya/goya.c
@@ -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);
--
2.17.1
next prev parent reply other threads:[~2021-10-04 10:08 UTC|newest]
Thread overview: 6+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-10-04 10:08 [PATCH 1/6] habanalabs: add support for a long interrupt target value Oded Gabbay
2021-10-04 10:08 ` [PATCH 2/6] habanalabs: fix debugfs device memory MMU VA translation Oded Gabbay
2021-10-04 10:08 ` Oded Gabbay [this message]
2021-10-04 10:08 ` [PATCH 4/6] habanalabs: refactor reset log message Oded Gabbay
2021-10-04 10:08 ` [PATCH 5/6] habanalabs: prevent race between fd close/open Oded Gabbay
2021-10-04 10:08 ` [PATCH 6/6] habanalabs: take timestamp on wait for interrupt 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=20211004100830.27573-3-ogabbay@kernel.org \
--to=ogabbay@kernel.org \
--cc=linux-kernel@vger.kernel.org \
/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 an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.