Re: [PATCH] tuners: tda8290: reduce stack usage with kasan

[Joe Perches]

On Tue, 2017-12-12 at 15:21 +0100, Arnd Bergmann wrote:
> On Tue, Dec 12, 2017 at 1:45 PM, Mauro Carvalho Chehab
>  wrote:
> > Em Tue, 12 Dec 2017 03:42:32 -0800
> > Joe Perches  escreveu:
> > 
> > > > I actually thought about marking them 'const' here before sending
> > > > (without noticing the changelog text) and then ran into what must
> > > > have led me to drop the 'const' originally: tuner_i2c_xfer_send()
> > > > takes a non-const pointer. This can be fixed but it requires
> > > > an ugly cast:
> > > 
> > > Casting away const is always a horrible hack.
> > > 
> > > Until it could be changed, my preference would
> > > be to update the changelog and perhaps add to
> > > the changelog the reason why it can not be const
> > > as detailed below.
> > > 
> > > ie: xfer_send and xfer_xend_recv both take a
> > > non-const unsigned char *
> 
> Ok.
> 
> > Perhaps, on a separate changeset, we could change I2C routines to
> > accept const unsigned char pointers. This is unrelated to tda8290
> > KASAN fixes. So, it should go via I2C tree, and, once accepted
> > there, we can change V4L2 drivers (and other drivers) accordingly.
> 
> I don't see how that would work unfortunately. i2c_msg contains
> a pointer to the data, and that is used for both input and output,
> including arrays like
> 
> struct i2c_msg msgs[] = {
> {
> .addr = dvo->slave_addr,
> .flags = 0,
> .len = 1,
> .buf = ,
> },
> {
> .addr = dvo->slave_addr,
> .flags = I2C_M_RD,
> .len = 1,
> .buf = val,
> }
> };
> 
> that have one constant output pointer and one non-constant
> input pointer. We could add an anonymous union for 'buf'
> to make that two separate pointers, but that's barely any
> better than the cast, and it would break the named initializers
> in the example above, at least on older compilers. Adding
> a second pointer to i2c_msg would add a bit of bloat and
> also require tree-wide changes or ugly hacks.

Perhaps add something like

struct i2c_msg_set {
__u16 addr; /* slave address*/
__u16 flags;
__u16 len;  /* msg length   */
const __u8 *buf;/* pointer to read-only msg data*/
};

struct i2c_msg_get {
__u16 addr; /* slave address*/
__u16 flags;
__u16 len;  /* msg length   */
__u8 *buf;  /* pointer to writeable msg data*/
};

to the uapi include and use that where appropriate
but where a write then read is done via a single
i2c_msg array, it's not really feasible either.

Probably better to avoid any churn and just mark
all these as static rather than static const.

Re: [PATCH] tuners: tda8290: reduce stack usage with kasan

[Arnd Bergmann]

On Tue, Dec 12, 2017 at 1:45 PM, Mauro Carvalho Chehab
 wrote:
> Em Tue, 12 Dec 2017 03:42:32 -0800
> Joe Perches  escreveu:
>
>> > I actually thought about marking them 'const' here before sending
>> > (without noticing the changelog text) and then ran into what must
>> > have led me to drop the 'const' originally: tuner_i2c_xfer_send()
>> > takes a non-const pointer. This can be fixed but it requires
>> > an ugly cast:
>>
>> Casting away const is always a horrible hack.
>>
>> Until it could be changed, my preference would
>> be to update the changelog and perhaps add to
>> the changelog the reason why it can not be const
>> as detailed below.
>>
>> ie: xfer_send and xfer_xend_recv both take a
>> non-const unsigned char *

Ok.

> Perhaps, on a separate changeset, we could change I2C routines to
> accept const unsigned char pointers. This is unrelated to tda8290
> KASAN fixes. So, it should go via I2C tree, and, once accepted
> there, we can change V4L2 drivers (and other drivers) accordingly.

I don't see how that would work unfortunately. i2c_msg contains
a pointer to the data, and that is used for both input and output,
including arrays like

struct i2c_msg msgs[] = {
{
.addr = dvo->slave_addr,
.flags = 0,
.len = 1,
.buf = ,
},
{
.addr = dvo->slave_addr,
.flags = I2C_M_RD,
.len = 1,
.buf = val,
}
};

that have one constant output pointer and one non-constant
input pointer. We could add an anonymous union for 'buf'
to make that two separate pointers, but that's barely any
better than the cast, and it would break the named initializers
in the example above, at least on older compilers. Adding
a second pointer to i2c_msg would add a bit of bloat and
also require tree-wide changes or ugly hacks.

   Arnd

Re: [PATCH] tuners: tda8290: reduce stack usage with kasan

[Mauro Carvalho Chehab]

Em Tue, 12 Dec 2017 03:42:32 -0800
Joe Perches  escreveu:

> > I actually thought about marking them 'const' here before sending
> > (without noticing the changelog text) and then ran into what must
> > have led me to drop the 'const' originally: tuner_i2c_xfer_send()
> > takes a non-const pointer. This can be fixed but it requires
> > an ugly cast:  
> 
> Casting away const is always a horrible hack.
> 
> Until it could be changed, my preference would
> be to update the changelog and perhaps add to
> the changelog the reason why it can not be const
> as detailed below.
> 
> ie: xfer_send and xfer_xend_recv both take a
> non-const unsigned char *

Perhaps, on a separate changeset, we could change I2C routines to
accept const unsigned char pointers. This is unrelated to tda8290
KASAN fixes. So, it should go via I2C tree, and, once accepted
there, we can change V4L2 drivers (and other drivers) accordingly.


Thanks,
Mauro

Re: [PATCH] tuners: tda8290: reduce stack usage with kasan

[Joe Perches]

On Tue, 2017-12-12 at 11:24 +0100, Arnd Bergmann wrote:
> On Mon, Dec 11, 2017 at 10:17 PM, Michael Ira Krufky
>  wrote:
> > On Mon, Dec 11, 2017 at 2:34 PM, Joe Perches  wrote:
> > > On Mon, 2017-12-11 at 13:06 +0100, Arnd Bergmann wrote:
> > > > With CONFIG_KASAN enabled, we get a relatively large stack frame in one 
> > > > function
> > > > 
> > > > drivers/media/tuners/tda8290.c: In function 'tda8290_set_params':
> > > > drivers/media/tuners/tda8290.c:310:1: warning: the frame size of 1520 
> > > > bytes is larger than 1024 bytes [-Wframe-larger-than=]
> > > > 
> > > > With CONFIG_KASAN_EXTRA this goes up to
> > > > 
> > > > drivers/media/tuners/tda8290.c: In function 'tda8290_set_params':
> > > > drivers/media/tuners/tda8290.c:310:1: error: the frame size of 3200 
> > > > bytes is larger than 3072 bytes [-Werror=frame-larger-than=]
> > > > 
> > > > We can significantly reduce this by marking local arrays as 'static 
> > > > const', and
> > > > this should result in better compiled code for everyone.
> > > 
> > > []
> > > > diff --git a/drivers/media/tuners/tda8290.c 
> > > > b/drivers/media/tuners/tda8290.c
> > > 
> > > []
> > > > @@ -63,8 +63,8 @@ static int tda8290_i2c_bridge(struct dvb_frontend 
> > > > *fe, int close)
> > > >  {
> > > >   struct tda8290_priv *priv = fe->analog_demod_priv;
> > > > 
> > > > - unsigned char  enable[2] = { 0x21, 0xC0 };
> > > > - unsigned char disable[2] = { 0x21, 0x00 };
> > > > + static unsigned char  enable[2] = { 0x21, 0xC0 };
> > > > + static unsigned char disable[2] = { 0x21, 0x00 };
> > > 
> > > Doesn't match commit message.
> > > 
> > > static const or just static?
> > > 
> > > > @@ -84,9 +84,9 @@ static int tda8295_i2c_bridge(struct dvb_frontend 
> > > > *fe, int close)
> > > >  {
> > > >   struct tda8290_priv *priv = fe->analog_demod_priv;
> > > > 
> > > > - unsigned char  enable[2] = { 0x45, 0xc1 };
> > > > - unsigned char disable[2] = { 0x46, 0x00 };
> > > > - unsigned char buf[3] = { 0x45, 0x01, 0x00 };
> > > > + static unsigned char  enable[2] = { 0x45, 0xc1 };
> > > > + static unsigned char disable[2] = { 0x46, 0x00 };
> > > 
> > > etc.
> > > 
> > > 
> > 
> > 
> > Joe is correct - they can be CONSTified. My bad -- a lot of the code I
> > wrote many years ago has this problem -- I wasn't so stack-conscious
> > back then.
> > 
> > The bytes in `enable` / `disable` don't get changed, but they may be
> > copied to another byte array that does get changed.  If would be best
> > to make these `static const`
> 
> Right. This was an older patch of mine that I picked up again
> after running into a warning that I had been ignoring for a while,
> and I didn't double-check the message.
> 
> I actually thought about marking them 'const' here before sending
> (without noticing the changelog text) and then ran into what must
> have led me to drop the 'const' originally: tuner_i2c_xfer_send()
> takes a non-const pointer. This can be fixed but it requires
> an ugly cast:

Casting away const is always a horrible hack.

Until it could be changed, my preference would
be to update the changelog and perhaps add to
the changelog the reason why it can not be const
as detailed below.

ie: xfer_send and xfer_xend_recv both take a
non-const unsigned char *

> diff --git a/drivers/media/tuners/tuner-i2c.h 
> b/drivers/media/tuners/tuner-i2c.h
> index bda67a5a76f2..809466eec780 100644
> --- a/drivers/media/tuners/tuner-i2c.h
> +++ b/drivers/media/tuners/tuner-i2c.h
> @@ -34,10 +34,10 @@ struct tuner_i2c_props {
>  };
> 
>  static inline int tuner_i2c_xfer_send(struct tuner_i2c_props *props,
> - unsigned char *buf, int len)
> + const unsigned char *buf, int len)
>  {
> struct i2c_msg msg = { .addr = props->addr, .flags = 0,
> -  .buf = buf, .len = len };
> +  .buf = (unsigned char *)buf, .len = len };
> int ret = i2c_transfer(props->adap, , 1);
> 
> return (ret == 1) ? len : ret;
> @@ -54,11 +54,11 @@ static inline int tuner_i2c_xfer_recv(struct
> tuner_i2c_props *props,
>  }
> 
>  static inline int tuner_i2c_xfer_send_recv(struct tuner_i2c_props *props,
> -  unsigned char *obuf, int olen,
> +  const unsigned char *obuf, int 
> olen,
>unsigned char *ibuf, int ilen)
>  {
> struct i2c_msg msg[2] = { { .addr = props->addr, .flags = 0,
> -   .buf = obuf, .len = olen },
> +   .buf = (unsigned char *)obuf, .len = olen 
> },
>   { .addr = props->addr, .flags = I2C_M_RD,
> .buf = ibuf, .len = ilen } };
> int ret = i2c_transfer(props->adap, msg, 2);
> 
> Should I submit it as a two-patch 

Re: [PATCH] tuners: tda8290: reduce stack usage with kasan

[Arnd Bergmann]

On Mon, Dec 11, 2017 at 10:17 PM, Michael Ira Krufky
 wrote:
> On Mon, Dec 11, 2017 at 2:34 PM, Joe Perches  wrote:
>> On Mon, 2017-12-11 at 13:06 +0100, Arnd Bergmann wrote:
>>> With CONFIG_KASAN enabled, we get a relatively large stack frame in one 
>>> function
>>>
>>> drivers/media/tuners/tda8290.c: In function 'tda8290_set_params':
>>> drivers/media/tuners/tda8290.c:310:1: warning: the frame size of 1520 bytes 
>>> is larger than 1024 bytes [-Wframe-larger-than=]
>>>
>>> With CONFIG_KASAN_EXTRA this goes up to
>>>
>>> drivers/media/tuners/tda8290.c: In function 'tda8290_set_params':
>>> drivers/media/tuners/tda8290.c:310:1: error: the frame size of 3200 bytes 
>>> is larger than 3072 bytes [-Werror=frame-larger-than=]
>>>
>>> We can significantly reduce this by marking local arrays as 'static const', 
>>> and
>>> this should result in better compiled code for everyone.
>> []
>>> diff --git a/drivers/media/tuners/tda8290.c b/drivers/media/tuners/tda8290.c
>> []
>>> @@ -63,8 +63,8 @@ static int tda8290_i2c_bridge(struct dvb_frontend *fe, 
>>> int close)
>>>  {
>>>   struct tda8290_priv *priv = fe->analog_demod_priv;
>>>
>>> - unsigned char  enable[2] = { 0x21, 0xC0 };
>>> - unsigned char disable[2] = { 0x21, 0x00 };
>>> + static unsigned char  enable[2] = { 0x21, 0xC0 };
>>> + static unsigned char disable[2] = { 0x21, 0x00 };
>>
>> Doesn't match commit message.
>>
>> static const or just static?
>>
>>> @@ -84,9 +84,9 @@ static int tda8295_i2c_bridge(struct dvb_frontend *fe, 
>>> int close)
>>>  {
>>>   struct tda8290_priv *priv = fe->analog_demod_priv;
>>>
>>> - unsigned char  enable[2] = { 0x45, 0xc1 };
>>> - unsigned char disable[2] = { 0x46, 0x00 };
>>> - unsigned char buf[3] = { 0x45, 0x01, 0x00 };
>>> + static unsigned char  enable[2] = { 0x45, 0xc1 };
>>> + static unsigned char disable[2] = { 0x46, 0x00 };
>>
>> etc.
>>
>>
>
>
> Joe is correct - they can be CONSTified. My bad -- a lot of the code I
> wrote many years ago has this problem -- I wasn't so stack-conscious
> back then.
>
> The bytes in `enable` / `disable` don't get changed, but they may be
> copied to another byte array that does get changed.  If would be best
> to make these `static const`

Right. This was an older patch of mine that I picked up again
after running into a warning that I had been ignoring for a while,
and I didn't double-check the message.

I actually thought about marking them 'const' here before sending
(without noticing the changelog text) and then ran into what must
have led me to drop the 'const' originally: tuner_i2c_xfer_send()
takes a non-const pointer. This can be fixed but it requires
an ugly cast:

diff --git a/drivers/media/tuners/tuner-i2c.h b/drivers/media/tuners/tuner-i2c.h
index bda67a5a76f2..809466eec780 100644
--- a/drivers/media/tuners/tuner-i2c.h
+++ b/drivers/media/tuners/tuner-i2c.h
@@ -34,10 +34,10 @@ struct tuner_i2c_props {
 };

 static inline int tuner_i2c_xfer_send(struct tuner_i2c_props *props,
- unsigned char *buf, int len)
+ const unsigned char *buf, int len)
 {
struct i2c_msg msg = { .addr = props->addr, .flags = 0,
-  .buf = buf, .len = len };
+  .buf = (unsigned char *)buf, .len = len };
int ret = i2c_transfer(props->adap, , 1);

return (ret == 1) ? len : ret;
@@ -54,11 +54,11 @@ static inline int tuner_i2c_xfer_recv(struct
tuner_i2c_props *props,
 }

 static inline int tuner_i2c_xfer_send_recv(struct tuner_i2c_props *props,
-  unsigned char *obuf, int olen,
+  const unsigned char *obuf, int olen,
   unsigned char *ibuf, int ilen)
 {
struct i2c_msg msg[2] = { { .addr = props->addr, .flags = 0,
-   .buf = obuf, .len = olen },
+   .buf = (unsigned char *)obuf, .len = olen },
  { .addr = props->addr, .flags = I2C_M_RD,
.buf = ibuf, .len = ilen } };
int ret = i2c_transfer(props->adap, msg, 2);

Should I submit it as a two-patch series with that added in, or update
the changelog to not mention 'const' instead?

   Arnd

Re: [PATCH] tuners: tda8290: reduce stack usage with kasan

[Michael Ira Krufky]

On Mon, Dec 11, 2017 at 2:34 PM, Joe Perches  wrote:
> On Mon, 2017-12-11 at 13:06 +0100, Arnd Bergmann wrote:
>> With CONFIG_KASAN enabled, we get a relatively large stack frame in one 
>> function
>>
>> drivers/media/tuners/tda8290.c: In function 'tda8290_set_params':
>> drivers/media/tuners/tda8290.c:310:1: warning: the frame size of 1520 bytes 
>> is larger than 1024 bytes [-Wframe-larger-than=]
>>
>> With CONFIG_KASAN_EXTRA this goes up to
>>
>> drivers/media/tuners/tda8290.c: In function 'tda8290_set_params':
>> drivers/media/tuners/tda8290.c:310:1: error: the frame size of 3200 bytes is 
>> larger than 3072 bytes [-Werror=frame-larger-than=]
>>
>> We can significantly reduce this by marking local arrays as 'static const', 
>> and
>> this should result in better compiled code for everyone.
> []
>> diff --git a/drivers/media/tuners/tda8290.c b/drivers/media/tuners/tda8290.c
> []
>> @@ -63,8 +63,8 @@ static int tda8290_i2c_bridge(struct dvb_frontend *fe, int 
>> close)
>>  {
>>   struct tda8290_priv *priv = fe->analog_demod_priv;
>>
>> - unsigned char  enable[2] = { 0x21, 0xC0 };
>> - unsigned char disable[2] = { 0x21, 0x00 };
>> + static unsigned char  enable[2] = { 0x21, 0xC0 };
>> + static unsigned char disable[2] = { 0x21, 0x00 };
>
> Doesn't match commit message.
>
> static const or just static?
>
>> @@ -84,9 +84,9 @@ static int tda8295_i2c_bridge(struct dvb_frontend *fe, int 
>> close)
>>  {
>>   struct tda8290_priv *priv = fe->analog_demod_priv;
>>
>> - unsigned char  enable[2] = { 0x45, 0xc1 };
>> - unsigned char disable[2] = { 0x46, 0x00 };
>> - unsigned char buf[3] = { 0x45, 0x01, 0x00 };
>> + static unsigned char  enable[2] = { 0x45, 0xc1 };
>> + static unsigned char disable[2] = { 0x46, 0x00 };
>
> etc.
>
>


Joe is correct - they can be CONSTified. My bad -- a lot of the code I
wrote many years ago has this problem -- I wasn't so stack-conscious
back then.

The bytes in `enable` / `disable` don't get changed, but they may be
copied to another byte array that does get changed.  If would be best
to make these `static const`

Best regards,

Michael Ira Krufky

Re: [PATCH] tuners: tda8290: reduce stack usage with kasan

[Joe Perches]

On Mon, 2017-12-11 at 13:06 +0100, Arnd Bergmann wrote:
> With CONFIG_KASAN enabled, we get a relatively large stack frame in one 
> function
> 
> drivers/media/tuners/tda8290.c: In function 'tda8290_set_params':
> drivers/media/tuners/tda8290.c:310:1: warning: the frame size of 1520 bytes 
> is larger than 1024 bytes [-Wframe-larger-than=]
> 
> With CONFIG_KASAN_EXTRA this goes up to
> 
> drivers/media/tuners/tda8290.c: In function 'tda8290_set_params':
> drivers/media/tuners/tda8290.c:310:1: error: the frame size of 3200 bytes is 
> larger than 3072 bytes [-Werror=frame-larger-than=]
> 
> We can significantly reduce this by marking local arrays as 'static const', 
> and
> this should result in better compiled code for everyone.
[]
> diff --git a/drivers/media/tuners/tda8290.c b/drivers/media/tuners/tda8290.c
[]
> @@ -63,8 +63,8 @@ static int tda8290_i2c_bridge(struct dvb_frontend *fe, int 
> close)
>  {
>   struct tda8290_priv *priv = fe->analog_demod_priv;
>  
> - unsigned char  enable[2] = { 0x21, 0xC0 };
> - unsigned char disable[2] = { 0x21, 0x00 };
> + static unsigned char  enable[2] = { 0x21, 0xC0 };
> + static unsigned char disable[2] = { 0x21, 0x00 };

Doesn't match commit message.

static const or just static?

> @@ -84,9 +84,9 @@ static int tda8295_i2c_bridge(struct dvb_frontend *fe, int 
> close)
>  {
>   struct tda8290_priv *priv = fe->analog_demod_priv;
>  
> - unsigned char  enable[2] = { 0x45, 0xc1 };
> - unsigned char disable[2] = { 0x46, 0x00 };
> - unsigned char buf[3] = { 0x45, 0x01, 0x00 };
> + static unsigned char  enable[2] = { 0x45, 0xc1 };
> + static unsigned char disable[2] = { 0x46, 0x00 };

etc.

Re: [PATCH] tuners: tda8290: reduce stack usage with kasan

[Michael Ira Krufky]

On Mon, Dec 11, 2017 at 7:06 AM, Arnd Bergmann  wrote:
> With CONFIG_KASAN enabled, we get a relatively large stack frame in one 
> function
>
> drivers/media/tuners/tda8290.c: In function 'tda8290_set_params':
> drivers/media/tuners/tda8290.c:310:1: warning: the frame size of 1520 bytes 
> is larger than 1024 bytes [-Wframe-larger-than=]
>
> With CONFIG_KASAN_EXTRA this goes up to
>
> drivers/media/tuners/tda8290.c: In function 'tda8290_set_params':
> drivers/media/tuners/tda8290.c:310:1: error: the frame size of 3200 bytes is 
> larger than 3072 bytes [-Werror=frame-larger-than=]
>
> We can significantly reduce this by marking local arrays as 'static const', 
> and
> this should result in better compiled code for everyone.
>
> I have another patch for the same symptom to patch tuner_i2c_xfer_*, and we
> actually want both of them.
>
> Signed-off-by: Arnd Bergmann 

Thank you, Arnd.

Reviewed-by: Michael Ira Krufky 


> ---
>  drivers/media/tuners/tda8290.c | 76 
> ++
>  1 file changed, 39 insertions(+), 37 deletions(-)
>
> diff --git a/drivers/media/tuners/tda8290.c b/drivers/media/tuners/tda8290.c
> index a59c567c55d6..19854221b72d 100644
> --- a/drivers/media/tuners/tda8290.c
> +++ b/drivers/media/tuners/tda8290.c
> @@ -63,8 +63,8 @@ static int tda8290_i2c_bridge(struct dvb_frontend *fe, int 
> close)
>  {
> struct tda8290_priv *priv = fe->analog_demod_priv;
>
> -   unsigned char  enable[2] = { 0x21, 0xC0 };
> -   unsigned char disable[2] = { 0x21, 0x00 };
> +   static unsigned char  enable[2] = { 0x21, 0xC0 };
> +   static unsigned char disable[2] = { 0x21, 0x00 };
> unsigned char *msg;
>
> if (close) {
> @@ -84,9 +84,9 @@ static int tda8295_i2c_bridge(struct dvb_frontend *fe, int 
> close)
>  {
> struct tda8290_priv *priv = fe->analog_demod_priv;
>
> -   unsigned char  enable[2] = { 0x45, 0xc1 };
> -   unsigned char disable[2] = { 0x46, 0x00 };
> -   unsigned char buf[3] = { 0x45, 0x01, 0x00 };
> +   static unsigned char  enable[2] = { 0x45, 0xc1 };
> +   static unsigned char disable[2] = { 0x46, 0x00 };
> +   static unsigned char buf[3] = { 0x45, 0x01, 0x00 };
> unsigned char *msg;
>
> if (close) {
> @@ -178,24 +178,24 @@ static void tda8290_set_params(struct dvb_frontend *fe,
>  {
> struct tda8290_priv *priv = fe->analog_demod_priv;
>
> -   unsigned char soft_reset[]  = { 0x00, 0x00 };
> +   static unsigned char soft_reset[]  = { 0x00, 0x00 };
> unsigned char easy_mode[]   = { 0x01, priv->tda8290_easy_mode };
> -   unsigned char expert_mode[] = { 0x01, 0x80 };
> -   unsigned char agc_out_on[]  = { 0x02, 0x00 };
> -   unsigned char gainset_off[] = { 0x28, 0x14 };
> -   unsigned char if_agc_spd[]  = { 0x0f, 0x88 };
> -   unsigned char adc_head_6[]  = { 0x05, 0x04 };
> -   unsigned char adc_head_9[]  = { 0x05, 0x02 };
> -   unsigned char adc_head_12[] = { 0x05, 0x01 };
> -   unsigned char pll_bw_nom[]  = { 0x0d, 0x47 };
> -   unsigned char pll_bw_low[]  = { 0x0d, 0x27 };
> -   unsigned char gainset_2[]   = { 0x28, 0x64 };
> -   unsigned char agc_rst_on[]  = { 0x0e, 0x0b };
> -   unsigned char agc_rst_off[] = { 0x0e, 0x09 };
> -   unsigned char if_agc_set[]  = { 0x0f, 0x81 };
> -   unsigned char addr_adc_sat  = 0x1a;
> -   unsigned char addr_agc_stat = 0x1d;
> -   unsigned char addr_pll_stat = 0x1b;
> +   static unsigned char expert_mode[] = { 0x01, 0x80 };
> +   static unsigned char agc_out_on[]  = { 0x02, 0x00 };
> +   static unsigned char gainset_off[] = { 0x28, 0x14 };
> +   static unsigned char if_agc_spd[]  = { 0x0f, 0x88 };
> +   static unsigned char adc_head_6[]  = { 0x05, 0x04 };
> +   static unsigned char adc_head_9[]  = { 0x05, 0x02 };
> +   static unsigned char adc_head_12[] = { 0x05, 0x01 };
> +   static unsigned char pll_bw_nom[]  = { 0x0d, 0x47 };
> +   static unsigned char pll_bw_low[]  = { 0x0d, 0x27 };
> +   static unsigned char gainset_2[]   = { 0x28, 0x64 };
> +   static unsigned char agc_rst_on[]  = { 0x0e, 0x0b };
> +   static unsigned char agc_rst_off[] = { 0x0e, 0x09 };
> +   static unsigned char if_agc_set[]  = { 0x0f, 0x81 };
> +   static unsigned char addr_adc_sat  = 0x1a;
> +   static unsigned char addr_agc_stat = 0x1d;
> +   static unsigned char addr_pll_stat = 0x1b;
> unsigned char adc_sat, agc_stat,
>   pll_stat;
> int i;
> @@ -468,9 +468,9 @@ static void tda8290_standby(struct dvb_frontend *fe)
>  {
> struct tda8290_priv *priv = fe->analog_demod_priv;
>
> -   unsigned char cb1[] = { 0x30, 0xD0 };
> -   unsigned char tda8290_standby[] = { 0x00, 0x02 };
> -   unsigned char tda8290_agc_tri[] = { 0x02, 0x20 };
> +   static unsigned char cb1[] = { 0x30, 0xD0 };
> +   static unsigned char 

[PATCH] tuners: tda8290: reduce stack usage with kasan

[Arnd Bergmann]

With CONFIG_KASAN enabled, we get a relatively large stack frame in one function

drivers/media/tuners/tda8290.c: In function 'tda8290_set_params':
drivers/media/tuners/tda8290.c:310:1: warning: the frame size of 1520 bytes is 
larger than 1024 bytes [-Wframe-larger-than=]

With CONFIG_KASAN_EXTRA this goes up to

drivers/media/tuners/tda8290.c: In function 'tda8290_set_params':
drivers/media/tuners/tda8290.c:310:1: error: the frame size of 3200 bytes is 
larger than 3072 bytes [-Werror=frame-larger-than=]

We can significantly reduce this by marking local arrays as 'static const', and
this should result in better compiled code for everyone.

I have another patch for the same symptom to patch tuner_i2c_xfer_*, and we
actually want both of them.

Signed-off-by: Arnd Bergmann 
---
 drivers/media/tuners/tda8290.c | 76 ++
 1 file changed, 39 insertions(+), 37 deletions(-)

diff --git a/drivers/media/tuners/tda8290.c b/drivers/media/tuners/tda8290.c
index a59c567c55d6..19854221b72d 100644
--- a/drivers/media/tuners/tda8290.c
+++ b/drivers/media/tuners/tda8290.c
@@ -63,8 +63,8 @@ static int tda8290_i2c_bridge(struct dvb_frontend *fe, int 
close)
 {
struct tda8290_priv *priv = fe->analog_demod_priv;
 
-   unsigned char  enable[2] = { 0x21, 0xC0 };
-   unsigned char disable[2] = { 0x21, 0x00 };
+   static unsigned char  enable[2] = { 0x21, 0xC0 };
+   static unsigned char disable[2] = { 0x21, 0x00 };
unsigned char *msg;
 
if (close) {
@@ -84,9 +84,9 @@ static int tda8295_i2c_bridge(struct dvb_frontend *fe, int 
close)
 {
struct tda8290_priv *priv = fe->analog_demod_priv;
 
-   unsigned char  enable[2] = { 0x45, 0xc1 };
-   unsigned char disable[2] = { 0x46, 0x00 };
-   unsigned char buf[3] = { 0x45, 0x01, 0x00 };
+   static unsigned char  enable[2] = { 0x45, 0xc1 };
+   static unsigned char disable[2] = { 0x46, 0x00 };
+   static unsigned char buf[3] = { 0x45, 0x01, 0x00 };
unsigned char *msg;
 
if (close) {
@@ -178,24 +178,24 @@ static void tda8290_set_params(struct dvb_frontend *fe,
 {
struct tda8290_priv *priv = fe->analog_demod_priv;
 
-   unsigned char soft_reset[]  = { 0x00, 0x00 };
+   static unsigned char soft_reset[]  = { 0x00, 0x00 };
unsigned char easy_mode[]   = { 0x01, priv->tda8290_easy_mode };
-   unsigned char expert_mode[] = { 0x01, 0x80 };
-   unsigned char agc_out_on[]  = { 0x02, 0x00 };
-   unsigned char gainset_off[] = { 0x28, 0x14 };
-   unsigned char if_agc_spd[]  = { 0x0f, 0x88 };
-   unsigned char adc_head_6[]  = { 0x05, 0x04 };
-   unsigned char adc_head_9[]  = { 0x05, 0x02 };
-   unsigned char adc_head_12[] = { 0x05, 0x01 };
-   unsigned char pll_bw_nom[]  = { 0x0d, 0x47 };
-   unsigned char pll_bw_low[]  = { 0x0d, 0x27 };
-   unsigned char gainset_2[]   = { 0x28, 0x64 };
-   unsigned char agc_rst_on[]  = { 0x0e, 0x0b };
-   unsigned char agc_rst_off[] = { 0x0e, 0x09 };
-   unsigned char if_agc_set[]  = { 0x0f, 0x81 };
-   unsigned char addr_adc_sat  = 0x1a;
-   unsigned char addr_agc_stat = 0x1d;
-   unsigned char addr_pll_stat = 0x1b;
+   static unsigned char expert_mode[] = { 0x01, 0x80 };
+   static unsigned char agc_out_on[]  = { 0x02, 0x00 };
+   static unsigned char gainset_off[] = { 0x28, 0x14 };
+   static unsigned char if_agc_spd[]  = { 0x0f, 0x88 };
+   static unsigned char adc_head_6[]  = { 0x05, 0x04 };
+   static unsigned char adc_head_9[]  = { 0x05, 0x02 };
+   static unsigned char adc_head_12[] = { 0x05, 0x01 };
+   static unsigned char pll_bw_nom[]  = { 0x0d, 0x47 };
+   static unsigned char pll_bw_low[]  = { 0x0d, 0x27 };
+   static unsigned char gainset_2[]   = { 0x28, 0x64 };
+   static unsigned char agc_rst_on[]  = { 0x0e, 0x0b };
+   static unsigned char agc_rst_off[] = { 0x0e, 0x09 };
+   static unsigned char if_agc_set[]  = { 0x0f, 0x81 };
+   static unsigned char addr_adc_sat  = 0x1a;
+   static unsigned char addr_agc_stat = 0x1d;
+   static unsigned char addr_pll_stat = 0x1b;
unsigned char adc_sat, agc_stat,
  pll_stat;
int i;
@@ -468,9 +468,9 @@ static void tda8290_standby(struct dvb_frontend *fe)
 {
struct tda8290_priv *priv = fe->analog_demod_priv;
 
-   unsigned char cb1[] = { 0x30, 0xD0 };
-   unsigned char tda8290_standby[] = { 0x00, 0x02 };
-   unsigned char tda8290_agc_tri[] = { 0x02, 0x20 };
+   static unsigned char cb1[] = { 0x30, 0xD0 };
+   static unsigned char tda8290_standby[] = { 0x00, 0x02 };
+   static unsigned char tda8290_agc_tri[] = { 0x02, 0x20 };
struct i2c_msg msg = {.addr = priv->tda827x_addr, .flags=0, .buf=cb1, 
.len = 2};
 
if (fe->ops.analog_ops.i2c_gate_ctrl)
@@ -495,9 +495,9 @@ static void tda8290_init_if(struct dvb_frontend *fe)
 {
struct tda8290_priv