Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Message-ID: <3a72852c-2f28-2916-f02c-b52cb06efd7d@gmx.de>
Date:   Wed, 14 Sep 2022 11:04:33 +0200
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:102.0) Gecko/20100101
 Thunderbird/102.2.1
Subject: Re: [PATCH] parisc: Increase the usage check of kmalloc allocated
 object a
Content-Language: en-US
To:     Rolf Eike Beer <eike-kernel@sf-tec.de>,
        Li zeming <zeming@nfschina.com>,
        James.Bottomley@hansenpartnership.com
Cc:     linux-parisc@vger.kernel.org, linux-kernel@vger.kernel.org
References: <20220914060428.4039-1-zeming@nfschina.com>
 <f346579c-e6a3-d2bb-1c21-d914aeb9a77d@gmx.de>
 <1668528.WaZXzWfjLs@eto.sf-tec.de>
From:   Helge Deller <deller@gmx.de>
In-Reply-To: <1668528.WaZXzWfjLs@eto.sf-tec.de>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: quoted-printable
Precedence: bulk

On 9/14/22 08:43, Rolf Eike Beer wrote:
> Am Mittwoch, 14. September 2022, 08:18:19 CEST schrieb Helge Deller:
>> On 9/14/22 08:04, Li zeming wrote:
>>> In the case of memory allocation failure, no alignment operation is
>>> required.
>>>
>>> Signed-off-by: Li zeming <zeming@nfschina.com>
>>> ---
>>>
>>>    drivers/parisc/iosapic.c | 4 +++-
>>>    1 file changed, 3 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/drivers/parisc/iosapic.c b/drivers/parisc/iosapic.c
>>> index 3a8c98615634..33de438916d3 100644
>>> --- a/drivers/parisc/iosapic.c
>>> +++ b/drivers/parisc/iosapic.c
>>> @@ -229,7 +229,9 @@ static struct irt_entry *iosapic_alloc_irt(int
>>> num_entries)>
>>>    	 * 4-byte alignment on 32-bit kernels
>>>    	 */
>>>
>>>    	a =3D (unsigned long)kmalloc(sizeof(struct irt_entry) * num_entrie=
s
> + 8,
>>>    	GFP_KERNEL);>
>>> -	a =3D (a + 7UL) & ~7UL;
>>> +	if (a)
>>> +		a =3D (a + 7UL) & ~7UL;
>>> +
>>
>> As you said, the adjustment isn't required, but it's still ok.
>> So I think the additional "if" isn't necessary and so I'm not
>> applying your patch.
>>
>> Anyway, thanks for your help to try to improve the code!
>
> I was about to say the same, but from looking at the code I don't think =
what
> is in there is correct either. The comment seems outdated, because
> __assume_kmalloc_alignment, which is __alignof__(unsigned long long). Th=
is
> code is untouched for the entire git history, so maybe we can just chang=
e the
> whole thing to
>
>    return kcalloc(num_entries, sizeof(struct irt_entry))
>
> now?

Yes, your proposal is good.
Anyone want to send a patch (with a small comment that kcalloc() will retu=
rn
at least the required 8-byte alignment)?

> And these functions end up propagating an allocation error in this file =
and it
> will never reach kernel/setup.c, which seems bad.

That part I don't understand.
The return value of iosapic_alloc_irt() is checked afterwards, but you pro=
bably
meant something else?

> But I guess the only point where this really can go wrong if the PDC
> returns an absurdly large number of entries.
Helge