From: Mario Limonciello <mario.limonciello@amd.com>
To: Lizhi Hou <lizhi.hou@amd.com>,
ogabbay@kernel.org, quic_jhugo@quicinc.com,
dri-devel@lists.freedesktop.org,
maciej.falkowski@linux.intel.com
Cc: linux-kernel@vger.kernel.org, max.zhen@amd.com, sonal.santan@amd.com
Subject: Re: [PATCH V2] accel/amdxdna: Fix dead lock for suspend and resume
Date: Wed, 11 Feb 2026 16:05:50 -0600 [thread overview]
Message-ID: <20cdfe29-403c-4093-9aad-55639a53a0c5@amd.com> (raw)
In-Reply-To: <ade9f2b8-3bec-77e2-d5ab-6850a7ee3f9b@amd.com>
On 2/11/26 3:59 PM, Lizhi Hou wrote:
>
> On 2/11/26 13:54, Mario Limonciello wrote:
>> On 2/11/26 2:46 PM, Lizhi Hou wrote:
>>> When an application issues a query IOCTL while auto suspend is running,
>>> a deadlock can occur. The query path holds dev_lock and then calls
>>> pm_runtime_resume_and_get(), which waits for the ongoing suspend to
>>> complete. Meanwhile, the suspend callback attempts to acquire dev_lock
>>> and blocks, resulting in a deadlock.
>>>
>>> Fix this by releasing dev_lock before calling
>>> pm_runtime_resume_and_get()
>>> and reacquiring it after the call completes. Also acquire dev_lock in
>>> the
>>> resume callback to keep the locking consistent.
>>>
>>> Fixes: 063db451832b ("accel/amdxdna: Enhance runtime power management")
>>> Signed-off-by: Lizhi Hou <lizhi.hou@amd.com>
>>> ---
>>> drivers/accel/amdxdna/aie2_ctx.c | 4 ++--
>>> drivers/accel/amdxdna/aie2_pci.c | 7 +++----
>>> drivers/accel/amdxdna/aie2_pm.c | 2 +-
>>> drivers/accel/amdxdna/amdxdna_ctx.c | 19 +++++++------------
>>> drivers/accel/amdxdna/amdxdna_pm.c | 2 ++
>>> drivers/accel/amdxdna/amdxdna_pm.h | 11 +++++++++++
>>> 6 files changed, 26 insertions(+), 19 deletions(-)
>>>
>>> diff --git a/drivers/accel/amdxdna/aie2_ctx.c b/drivers/accel/
>>> amdxdna/aie2_ctx.c
>>> index 37d05f2e986f..58e146172b61 100644
>>> --- a/drivers/accel/amdxdna/aie2_ctx.c
>>> +++ b/drivers/accel/amdxdna/aie2_ctx.c
>>> @@ -629,7 +629,7 @@ int aie2_hwctx_init(struct amdxdna_hwctx *hwctx)
>>> goto free_entity;
>>> }
>>> - ret = amdxdna_pm_resume_get(xdna);
>>> + ret = amdxdna_pm_resume_get_locked(xdna);
>>> if (ret)
>>> goto free_col_list;
>>> @@ -760,7 +760,7 @@ static int aie2_hwctx_cu_config(struct
>>> amdxdna_hwctx *hwctx, void *buf, u32 size
>>> if (!hwctx->cus)
>>> return -ENOMEM;
>>> - ret = amdxdna_pm_resume_get(xdna);
>>> + ret = amdxdna_pm_resume_get_locked(xdna);
>>> if (ret)
>>> goto free_cus;
>>> diff --git a/drivers/accel/amdxdna/aie2_pci.c b/drivers/accel/
>>> amdxdna/aie2_pci.c
>>> index f70ccf0f3c01..5b326e4610e6 100644
>>> --- a/drivers/accel/amdxdna/aie2_pci.c
>>> +++ b/drivers/accel/amdxdna/aie2_pci.c
>>> @@ -451,7 +451,6 @@ static int aie2_hw_suspend(struct amdxdna_dev *xdna)
>>> {
>>> struct amdxdna_client *client;
>>> - guard(mutex)(&xdna->dev_lock);
>>> list_for_each_entry(client, &xdna->client_list, node)
>>> aie2_hwctx_suspend(client);
>>> @@ -951,7 +950,7 @@ static int aie2_get_info(struct amdxdna_client
>>> *client, struct amdxdna_drm_get_i
>>> if (!drm_dev_enter(&xdna->ddev, &idx))
>>> return -ENODEV;
>>> - ret = amdxdna_pm_resume_get(xdna);
>>> + ret = amdxdna_pm_resume_get_locked(xdna);
>>> if (ret)
>>> goto dev_exit;
>>> @@ -1044,7 +1043,7 @@ static int aie2_get_array(struct
>>> amdxdna_client *client,
>>> if (!drm_dev_enter(&xdna->ddev, &idx))
>>> return -ENODEV;
>>> - ret = amdxdna_pm_resume_get(xdna);
>>> + ret = amdxdna_pm_resume_get_locked(xdna);
>>> if (ret)
>>> goto dev_exit;
>>> @@ -1134,7 +1133,7 @@ static int aie2_set_state(struct
>>> amdxdna_client *client,
>>> if (!drm_dev_enter(&xdna->ddev, &idx))
>>> return -ENODEV;
>>> - ret = amdxdna_pm_resume_get(xdna);
>>> + ret = amdxdna_pm_resume_get_locked(xdna);
>>> if (ret)
>>> goto dev_exit;
>>> diff --git a/drivers/accel/amdxdna/aie2_pm.c b/drivers/accel/
>>> amdxdna/aie2_pm.c
>>> index 579b8be13b18..29bd4403a94d 100644
>>> --- a/drivers/accel/amdxdna/aie2_pm.c
>>> +++ b/drivers/accel/amdxdna/aie2_pm.c
>>> @@ -31,7 +31,7 @@ int aie2_pm_set_dpm(struct amdxdna_dev_hdl *ndev,
>>> u32 dpm_level)
>>> {
>>> int ret;
>>> - ret = amdxdna_pm_resume_get(ndev->xdna);
>>> + ret = amdxdna_pm_resume_get_locked(ndev->xdna);
>>> if (ret)
>>> return ret;
>>> diff --git a/drivers/accel/amdxdna/amdxdna_ctx.c b/drivers/accel/
>>> amdxdna/amdxdna_ctx.c
>>> index d17aef89a0ad..db3aa26fb55f 100644
>>> --- a/drivers/accel/amdxdna/amdxdna_ctx.c
>>> +++ b/drivers/accel/amdxdna/amdxdna_ctx.c
>>> @@ -266,9 +266,9 @@ int amdxdna_drm_config_hwctx_ioctl(struct
>>> drm_device *dev, void *data, struct dr
>>> struct amdxdna_drm_config_hwctx *args = data;
>>> struct amdxdna_dev *xdna = to_xdna_dev(dev);
>>> struct amdxdna_hwctx *hwctx;
>>> - int ret, idx;
>>> u32 buf_size;
>>> void *buf;
>>> + int ret;
>>> u64 val;
>>> if (XDNA_MBZ_DBG(xdna, &args->pad, sizeof(args->pad)))
>>> @@ -310,20 +310,17 @@ int amdxdna_drm_config_hwctx_ioctl(struct
>>> drm_device *dev, void *data, struct dr
>>> return -EINVAL;
>>> }
>>> - mutex_lock(&xdna->dev_lock);
>>> - idx = srcu_read_lock(&client->hwctx_srcu);
>>> + guard(mutex)(&xdna->dev_lock);
>>> hwctx = xa_load(&client->hwctx_xa, args->handle);
>>> if (!hwctx) {
>>> XDNA_DBG(xdna, "PID %d failed to get hwctx %d", client-
>>> >pid, args->handle);
>>> ret = -EINVAL;
>>> - goto unlock_srcu;
>>> + goto free_buf;
>>> }
>>> ret = xdna->dev_info->ops->hwctx_config(hwctx, args-
>>> >param_type, val, buf, buf_size);
>>> -unlock_srcu:
>>> - srcu_read_unlock(&client->hwctx_srcu, idx);
>>> - mutex_unlock(&xdna->dev_lock);
>>> +free_buf:
>>> kfree(buf);
>>> return ret;
>>> }
>>> @@ -334,7 +331,7 @@ int amdxdna_hwctx_sync_debug_bo(struct
>>> amdxdna_client *client, u32 debug_bo_hdl)
>>> struct amdxdna_hwctx *hwctx;
>>> struct amdxdna_gem_obj *abo;
>>> struct drm_gem_object *gobj;
>>> - int ret, idx;
>>> + int ret;
>>> if (!xdna->dev_info->ops->hwctx_sync_debug_bo)
>>> return -EOPNOTSUPP;
>>> @@ -345,17 +342,15 @@ int amdxdna_hwctx_sync_debug_bo(struct
>>> amdxdna_client *client, u32 debug_bo_hdl)
>>> abo = to_xdna_obj(gobj);
>>> guard(mutex)(&xdna->dev_lock);
>>> - idx = srcu_read_lock(&client->hwctx_srcu);
>>> hwctx = xa_load(&client->hwctx_xa, abo->assigned_hwctx);
>>> if (!hwctx) {
>>> ret = -EINVAL;
>>> - goto unlock_srcu;
>>> + goto put_obj;
>>> }
>>> ret = xdna->dev_info->ops->hwctx_sync_debug_bo(hwctx,
>>> debug_bo_hdl);
>>> -unlock_srcu:
>>> - srcu_read_unlock(&client->hwctx_srcu, idx);
>>> +put_obj:
>>> drm_gem_object_put(gobj);
>>> return ret;
>>> }
>>> diff --git a/drivers/accel/amdxdna/amdxdna_pm.c b/drivers/accel/
>>> amdxdna/amdxdna_pm.c
>>> index d024d480521c..b1fafddd7ad5 100644
>>> --- a/drivers/accel/amdxdna/amdxdna_pm.c
>>> +++ b/drivers/accel/amdxdna/amdxdna_pm.c
>>> @@ -16,6 +16,7 @@ int amdxdna_pm_suspend(struct device *dev)
>>> struct amdxdna_dev *xdna = to_xdna_dev(dev_get_drvdata(dev));
>>> int ret = -EOPNOTSUPP;
>>> + guard(mutex)(&xdna->dev_lock);
>>> if (xdna->dev_info->ops->suspend)
>>> ret = xdna->dev_info->ops->suspend(xdna);
>>> @@ -28,6 +29,7 @@ int amdxdna_pm_resume(struct device *dev)
>>> struct amdxdna_dev *xdna = to_xdna_dev(dev_get_drvdata(dev));
>>> int ret = -EOPNOTSUPP;
>>> + guard(mutex)(&xdna->dev_lock);
>>> if (xdna->dev_info->ops->resume)
>>> ret = xdna->dev_info->ops->resume(xdna);
>>> diff --git a/drivers/accel/amdxdna/amdxdna_pm.h b/drivers/accel/
>>> amdxdna/amdxdna_pm.h
>>> index 77b2d6e45570..3d26b973e0e3 100644
>>> --- a/drivers/accel/amdxdna/amdxdna_pm.h
>>> +++ b/drivers/accel/amdxdna/amdxdna_pm.h
>>> @@ -15,4 +15,15 @@ void amdxdna_pm_suspend_put(struct amdxdna_dev
>>> *xdna);
>>> void amdxdna_pm_init(struct amdxdna_dev *xdna);
>>> void amdxdna_pm_fini(struct amdxdna_dev *xdna);
>>> +static inline int amdxdna_pm_resume_get_locked(struct amdxdna_dev
>>> *xdna)
>>> +{
>>> + int ret;
>>> +
>>> + mutex_unlock(&xdna->dev_lock);
>>> + ret = amdxdna_pm_resume_get(xdna);
>>> + mutex_lock(&xdna->dev_lock);
>>> +
>>> + return ret;
>>
>> Like the other code you added in this patch, couldn't this be
>> simplified with a guard(mutex)?
>
> This is unlock before resume_get, then lock after. guard(mutex) might
> not help here.
>
Oh right; it's reversed. Agreed!
Reviewed-by: Mario Limonciello (AMD) <superm1@kernel.org>
> Lizhi
>
>>
>>> +}
>>> +
>>> #endif /* _AMDXDNA_PM_H_ */
>>
prev parent reply other threads:[~2026-02-11 22:05 UTC|newest]
Thread overview: 4+ messages / expand[flat|nested] mbox.gz Atom feed top
2026-02-11 20:46 Lizhi Hou
2026-02-11 21:54 ` Mario Limonciello
2026-02-11 21:59 ` Lizhi Hou
2026-02-11 22:05 ` Mario Limonciello [this message]
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=20cdfe29-403c-4093-9aad-55639a53a0c5@amd.com \
--to=mario.limonciello@amd.com \
--cc=dri-devel@lists.freedesktop.org \
--cc=linux-kernel@vger.kernel.org \
--cc=lizhi.hou@amd.com \
--cc=maciej.falkowski@linux.intel.com \
--cc=max.zhen@amd.com \
--cc=ogabbay@kernel.org \
--cc=quic_jhugo@quicinc.com \
--cc=sonal.santan@amd.com \
/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 a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox
all inboxes | Powered by JetHome®