[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <f755b775-b87a-4a3f-b7b9-f37cdadf21e3@intel.com>
Date: Wed, 3 Sep 2025 08:26:21 -0700
From: Dave Jiang <dave.jiang@...el.com>
To: yuanli fu <fuyuanli0722@...il.com>
Cc: jdmason@...zu.us, allenbh@...il.com, ntb@...ts.linux.dev,
linux-kernel@...r.kernel.org
Subject: Re: [PATCH v2] ntb: Add mutex to make link_event_callback executed
linearly.
On 9/2/25 7:20 PM, yuanli fu wrote:
> Dave Jiang <dave.jiang@...el.com> 于2025年8月25日周一 23:06写道:
>>
>>
>>
>> On 8/25/25 2:15 AM, fuyuanli wrote:
>>> Since the CPU selected by schedule_work is uncertain, multiple link_event
>>> callbacks may be executed at same time. For example, after peer's link is
>>> up, it is down quickly before local link_work completed. If link_cleanup
>>> is added to the workqueue of another CPU, then link_work and link_cleanup
>>> may be executed at the same time. So add a mutex to prevent them from being
>>> executed concurrently.
>>>
>>> Signed-off-by: fuyuanli <fuyuanli@...iglobal.com>
>>
>> Reviewed-by: Dave Jiang <dave.jiang@...el.com>
>
> Hi Dave,
>
> Hope you are doing well.
>
> Just wanted to gently follow up on this patch which you had acked
> before. Is there
> anything else I can do to help get this merged? Perhaps it needs a rebase on a
> different tree?
Jon will merge it when he has a chance.
>
> Thanks for your time and all your work!
>
> Best regards,
> Yuanli Fu
>
>
>>
>>> ---
>>> v2:
>>> 1) use guard() instead of lock & unlock functions.
>>>
>>> v1:
>>> Link: https://lore.kernel.org/all/aKiBi4ZDlbgzed%2Fz@didi-ThinkCentre-M930t-N000/
>>> ---
>>> drivers/ntb/ntb_transport.c | 7 +++++++
>>> 1 file changed, 7 insertions(+)
>>>
>>> diff --git a/drivers/ntb/ntb_transport.c b/drivers/ntb/ntb_transport.c
>>> index 4f775c3e218f..eb875e3db2e3 100644
>>> --- a/drivers/ntb/ntb_transport.c
>>> +++ b/drivers/ntb/ntb_transport.c
>>> @@ -59,6 +59,7 @@
>>> #include <linux/slab.h>
>>> #include <linux/types.h>
>>> #include <linux/uaccess.h>
>>> +#include <linux/mutex.h>
>>> #include "linux/ntb.h"
>>> #include "linux/ntb_transport.h"
>>>
>>> @@ -241,6 +242,9 @@ struct ntb_transport_ctx {
>>> struct work_struct link_cleanup;
>>>
>>> struct dentry *debugfs_node_dir;
>>> +
>>> + /* Make sure workq of link event be executed serially */
>>> + struct mutex link_event_lock;
>>> };
>>>
>>> enum {
>>> @@ -1024,6 +1028,7 @@ static void ntb_transport_link_cleanup_work(struct work_struct *work)
>>> struct ntb_transport_ctx *nt =
>>> container_of(work, struct ntb_transport_ctx, link_cleanup);
>>>
>>> + guard(mutex)(&nt->link_event_lock);
>>> ntb_transport_link_cleanup(nt);
>>> }
>>>
>>> @@ -1047,6 +1052,8 @@ static void ntb_transport_link_work(struct work_struct *work)
>>> u32 val;
>>> int rc = 0, i, spad;
>>>
>>> + guard(mutex)(&nt->link_event_lock);
>>> +
>>> /* send the local info, in the opposite order of the way we read it */
>>>
>>> if (nt->use_msi) {
>>
Powered by blists - more mailing lists