Re: [PATCH] powerpc: 8xx: fix a return value error in mpc8xx_pic_init

From: Michael Ellerman <mpe@ellerman.id.au>
To: Christophe Leroy <christophe.leroy@csgroup.eu>,
	Hangyu Hua <hbh25y@gmail.com>,
	"benh@kernel.crashing.org" <benh@kernel.crashing.org>,
	"paulus@samba.org" <paulus@samba.org>,
	"peng.hao2@zte.com.cn" <peng.hao2@zte.com.cn>,
	"wen.yang99@zte.com.cn" <wen.yang99@zte.com.cn>
Cc: "linuxppc-dev@lists.ozlabs.org" <linuxppc-dev@lists.ozlabs.org>,
	"linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>
Subject: Re: [PATCH] powerpc: 8xx: fix a return value error in mpc8xx_pic_init
Date: Thu, 10 Mar 2022 10:16:18 +1100	[thread overview]
Message-ID: <87fsnqn2d9.fsf@mpe.ellerman.id.au> (raw)
In-Reply-To: <867e5b54-5990-578e-bfae-b638efce2a8a@csgroup.eu>

Christophe Leroy <christophe.leroy@csgroup.eu> writes:
> Le 09/03/2022 à 11:46, Michael Ellerman a écrit :
>> Christophe Leroy <christophe.leroy@csgroup.eu> writes:
>>> Le 09/03/2022 à 04:24, Michael Ellerman a écrit :
>>>> Hangyu Hua <hbh25y@gmail.com> writes:
>>>>> mpc8xx_pic_init() should return -ENOMEM instead of 0 when
>>>>> irq_domain_add_linear() return NULL. This cause mpc8xx_pics_init to continue
>>>>> executing even if mpc8xx_pic_host is NULL.
>>>>>
>>>>> Fixes: cc76404feaed ("powerpc/8xx: Fix possible device node reference leak")
>>>>> Signed-off-by: Hangyu Hua <hbh25y@gmail.com>
>>>>> ---
>>>>>    arch/powerpc/platforms/8xx/pic.c | 1 +
>>>>>    1 file changed, 1 insertion(+)
>>>>>
>>>>> diff --git a/arch/powerpc/platforms/8xx/pic.c b/arch/powerpc/platforms/8xx/pic.c
>>>>> index f2ba837249d6..04a6abf14c29 100644
>>>>> --- a/arch/powerpc/platforms/8xx/pic.c
>>>>> +++ b/arch/powerpc/platforms/8xx/pic.c
>>>>> @@ -153,6 +153,7 @@ int __init mpc8xx_pic_init(void)
>>>>
>>>> Expanding the context:
>>>>
>>>> 	siu_reg = ioremap(res.start, resource_size(&res));
>>>> 	if (siu_reg == NULL) {
>>>> 		ret = -EINVAL;
>>>> 		goto out;
>>>> 	}
>>>>
>>>> 	mpc8xx_pic_host = irq_domain_add_linear(np, 64, &mpc8xx_pic_host_ops, NULL);
>>>>>    	if (mpc8xx_pic_host == NULL) {
>>>>>    		printk(KERN_ERR "MPC8xx PIC: failed to allocate irq host!\n");
>>>>>    		ret = -ENOMEM;
>>>>> +		goto out;
>>>>>    	}
>>>>>    
>>>>>    	ret = 0;
>>>>>    	
>>>> out:
>>>> 	of_node_put(np);
>>>> 	return ret;
>>>> }
>>>>
>>>> Proper error cleanup should also undo the ioremap() if
>>>> irq_domain_add_linear() fails.
>>>
>>> Uh ...
>>>
>>> If siu_reg is NULL, you get a serious problem when __do_irq() calls
>>> mpc8xx_get_irq()
>> 
>> Arguably it shouldn't be assigned to ppc_md.get_irq unless
>> mpc8xx_pic_init() succeeds. See eg. xics_init().
>
> I agree with that, but it's a huge work I guess. Most platforms set 
> .get_irq in ppc_md() at buildtime. See the generic mpic_get_irq() which 
> has a BUG_ON() in call mpic_primary is NULL. There are 50 platforms with 
> buildtime assignment.

Yes I agree. And __do_irq() will just oops if ppc_md.get_irq() is NULL,
so it's a bit of a mess.

> That would however be a good opportunity to switch get_irq() to a static 
> call. I'll open a github issue to follow it.

Thanks.

cheers