Re: [PATCH] memory: mtk-smi: Fix a null dereference for the ostd

From: Yong Wu
Date: Mon Nov 01 2021 - 06:41:02 EST


On Mon, 2021-11-01 at 09:45 +0100, Krzysztof Kozlowski wrote:
> On 01/11/2021 07:09, Yong Wu wrote:
> > On Fri, 2021-10-29 at 19:35 +0200, Krzysztof Kozlowski wrote:
> > > On 28/10/2021 07:50, Yong Wu wrote:
> > > > We add the ostd setting for mt8195. It introduces a abort for
> > > > the
> > > > previous SoC which doesn't have ostd setting. This is the log:
> > > >
> > > > Unable to handle kernel NULL pointer dereference at virtual
> > > > address
> > > > 0000000000000080
> > > > ...
> > > > pc : mtk_smi_larb_config_port_gen2_general+0x64/0x130
> > > > lr : mtk_smi_larb_resume+0x54/0x98
> > > > ...
> > > > Call trace:
> > > > mtk_smi_larb_config_port_gen2_general+0x64/0x130
> > > > pm_generic_runtime_resume+0x2c/0x48
> > > > __genpd_runtime_resume+0x30/0xa8
> > > > genpd_runtime_resume+0x94/0x2c8
> > > > __rpm_callback+0x44/0x150
> > > > rpm_callback+0x6c/0x78
> > > > rpm_resume+0x310/0x558
> > > > __pm_runtime_resume+0x3c/0x88
> > > >
> > > > In the code: larbostd = larb->larb_gen->ostd[larb->larbid],
> > > > if "larb->larb_gen->ostd" is null, the "larbostd" is the
> > > > offset, it
> > > > is
> > > > also a valid value, thus, use the larb->larb_gen->ostd as the
> > > > condition
> > > > inside the "for" loop.
> > >
> > > You need to write more clearly, what you are fixing here.
> > >
> > > >
> > > > Signed-off-by: Yong Wu <yong.wu@xxxxxxxxxxxx>
> > > > ---
> > > > Hi Krzysztof,
> > > > Could you help review and conside this as a fix for the mt8195
> > > > patchset?
> > > > The mt8195 patchset are not in mainline, thus, I don't know its
> > > > sha-id,
> > > > and don't add Fixes tag.
> > > > Thanks
> > > > ---
> > > > drivers/memory/mtk-smi.c | 2 +-
> > > > 1 file changed, 1 insertion(+), 1 deletion(-)
> > > >
> > > > diff --git a/drivers/memory/mtk-smi.c b/drivers/memory/mtk-
> > > > smi.c
> > > > index b883dcc0bbfa..0262a59a2d6e 100644
> > > > --- a/drivers/memory/mtk-smi.c
> > > > +++ b/drivers/memory/mtk-smi.c
> > > > @@ -257,7 +257,7 @@ static void
> > > > mtk_smi_larb_config_port_gen2_general(struct device *dev)
> > > > if (MTK_SMI_CAPS(flags_general, MTK_SMI_FLAG_SW_FLAG))
> > > > writel_relaxed(SMI_LARB_SW_FLAG_1, larb->base +
> > > > SMI_LARB_SW_FLAG);
> > > >
> > > > - for (i = 0; i < SMI_LARB_PORT_NR_MAX && larbostd &&
> > > > !!larbostd[i]; i++)
> > > > + for (i = 0; i < SMI_LARB_PORT_NR_MAX && larb->larb_gen-
> > > > >ostd &&
> > > > !!larbostd[i]; i++)
> > > > writel_relaxed(larbostd[i], larb->base +
> > > > SMI_LARB_OSTDL_PORTx(i));
> > >
> > > The code does not look good. You have already a dereference at
> > > line
> > > 244:
> > >
> > > const u8 *larbostd = larb->larb_gen->ostd[larb->larbid];
> >
> > if larb->larb_gen->ostd is null, larbostd is the offset, e.g. 0x80
> > in
> > the log above. thus, we can not use "larbostd[i]" in the "for"
> > loop.
> >
> > sorry for the unreadable. In this case, is the change ok?
>
> No, it's ok, I did not check the type of ostd and it's confusing a
> bit
> that it is defined as a pointer to an array but you actually use it
> as
> array of pointers to 32-elemenet arrays... Anyway I was mistaken and
> there will be indeed no dereference at the assignment, but for code
> clarity I would still prefer to do the check earlier, so:
>
> >
> > or like this:
> >
> > -const u8 *larbostd = larb->larb_gen->ostd[larb->larbid];
> > +const u8 *larbostd = larb->larb_gen->ostd ? larb->larb_gen-
> > ostd[larb-
> > > larbid] : NULL;
>
> Although I think now the proper type should be explicit.
> mtk_smi_larb_mt8195_ostd is an 28-element array of
> SMI_LARB_PORT_NR_MAX-element u8 arrays, therefore struct
> mtk_smi_larb_gen should be:
> const u8 (*ostd)[][SMI_LARB_PORT_NR_MAX];

mt8195_ostd only is a 2 dimensional array,
static const u8 mtk_smi_larb_mt8195_ostd[][SMI_LARB_PORT_NR_MAX];

If using this, it will cause build fail:

.../drivers/memory/mtk-smi.c:359:23: error: initialization from
incompatible pointer type [-Werror=incompatible-pointer-types]
.ostd = mtk_smi_larb_mt8195_ostd,
^~~~~~~~~~~~~~~~~~~~~~~~

More background about this 2 dimensional array:
The platform have many larbs(up to MTK_LARB_NR_MAX), each a larb have
many ports(up to SMI_LARB_PORT_NR_MAX). In the
function(*_config_port_gen2_general), we will update the value for each
a port inside a larb, thus we need a pointer point to the line of that
larb.

If I miss something, please tell me. or how to manage this will be
better.

Thanks.

>
> Right?
>
>
> Best regards,
> Krzysztof