Re: [PATCH -next] i3c: master: svc: drop free_irq of devm_request_irq allocated irq

From: Yang Yingliang
Date: Thu May 27 2021 - 21:44:03 EST



On 2021/5/27 22:40, Miquel Raynal wrote:
Hi Yang,

Yang Yingliang <yangyingliang@xxxxxxxxxx> wrote on Thu, 27 May 2021
21:49:53 +0800:

Hi,

On 2021/5/27 18:01, Miquel Raynal wrote:
Hi Yang,

Yang Yingliang <yangyingliang@xxxxxxxxxx> wrote on Tue, 18 May 2021
21:11:27 +0800:
irq allocated with devm_request_irq should not be freed using
free_irq, because doing so causes a dangling pointer, and a
subsequent double free.

Reported-by: Hulk Robot <hulkci@xxxxxxxxxx>
Signed-off-by: Yang Yingliang <yangyingliang@xxxxxxxxxx>
---
drivers/i3c/master/svc-i3c-master.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/i3c/master/svc-i3c-master.c b/drivers/i3c/master/svc-i3c-master.c
index 1f6ba4221817..761c9c468357 100644
--- a/drivers/i3c/master/svc-i3c-master.c
+++ b/drivers/i3c/master/svc-i3c-master.c
@@ -1448,7 +1448,7 @@ static int svc_i3c_master_remove(struct platform_device *pdev)
if (ret)
return ret;
>> - free_irq(master->irq, master);
+ devm_free_irq(&pdev->dev, master->irq, master);
Wouldn't removing this call the right solution? If it's a device
managed resource, it won't probably be needed to free it explicitly in
the remove path.
Some drivers would expect to free irq itself,
I don't get it. Drivers do not expect anything, they should just comply
with the API. If robots complain because a device managed resource is
being freed without the device managed helper, this does not mean that
the resource should explicitly be freed, it just means that *if* it
must be explicitly freed, the wrong helper is being used.

I am not sure if it's ok to remove the free_irq() in i3c,
What is the link with I3C? Sorry I might be missing something but
master->irq is a driver variable, I don't get the link with the I3C
framework and why it would interfere.

I just keep the original logic here and avoid double free.
I don't think it is sane. Calling devm_free_irq() maybe is the right
solution - I don't feel like it is - but your certainly can't hide
behind a 'I just want the robots to be happy' justification. Hiding
bugs on purpose is not something that I personally appreciate much.
Freeing irq in ->remove() is earlier than in device manage framework, if
just remove the free_irq() in svc_i3c_master_remove() and free the irq by
device manage framework, I am not sure if it breaks the resource free
sequence in Silvaco I3C master driver. If it's OK, I can resend a patch with
removing the free_irq().

Thanks,
Yang

Thanks,
Miquèl
.