diff mbox series

[1/3] usb: cdns3: gadget: suspicious implicit sign extension

Message ID 20201016101659.29482-2-peter.chen@nxp.com (mailing list archive)
State Superseded
Headers show
Series usb: cdns3: three bug fixes for v5.10 | expand

Commit Message

Peter Chen Oct. 16, 2020, 10:16 a.m. UTC
For code:
trb->length = cpu_to_le32(TRB_BURST_LEN(priv_ep->trb_burst_size)
	       	| TRB_LEN(length));

TRB_BURST_LEN(priv_ep->trb_burst_size) may be overflow for int 32 if
priv_ep->trb_burst_size is equal or larger than 0x80;

Below is the Coverity warning:
sign_extension: Suspicious implicit sign extension: priv_ep->trb_burst_size
with type u8 (8 bits, unsigned) is promoted in priv_ep->trb_burst_size << 24
to type int (32 bits, signed), then sign-extended to type unsigned long
(64 bits, unsigned). If priv_ep->trb_burst_size << 24 is greater than 0x7FFFFFFF,
the upper bits of the result will all be 1.

Cc: <stable@vger.kernel.org> #v5.8+
Fixes: 7733f6c32e36 ("usb: cdns3: Add Cadence USB3 DRD Driver")
Signed-off-by: Peter Chen <peter.chen@nxp.com>
---
 drivers/usb/cdns3/gadget.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

Comments

Felipe Balbi Oct. 27, 2020, 9:03 a.m. UTC | #1
Hi,

Peter Chen <peter.chen@nxp.com> writes:
> For code:
> trb->length = cpu_to_le32(TRB_BURST_LEN(priv_ep->trb_burst_size)
> 	       	| TRB_LEN(length));
>
> TRB_BURST_LEN(priv_ep->trb_burst_size) may be overflow for int 32 if
> priv_ep->trb_burst_size is equal or larger than 0x80;
>
> Below is the Coverity warning:
> sign_extension: Suspicious implicit sign extension: priv_ep->trb_burst_size
> with type u8 (8 bits, unsigned) is promoted in priv_ep->trb_burst_size << 24
> to type int (32 bits, signed), then sign-extended to type unsigned long
> (64 bits, unsigned). If priv_ep->trb_burst_size << 24 is greater than 0x7FFFFFFF,
> the upper bits of the result will all be 1.

looks like a false positive...

> Cc: <stable@vger.kernel.org> #v5.8+
> Fixes: 7733f6c32e36 ("usb: cdns3: Add Cadence USB3 DRD Driver")
> Signed-off-by: Peter Chen <peter.chen@nxp.com>
> ---
>  drivers/usb/cdns3/gadget.h | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/drivers/usb/cdns3/gadget.h b/drivers/usb/cdns3/gadget.h
> index 1ccecd237530..020936cb9897 100644
> --- a/drivers/usb/cdns3/gadget.h
> +++ b/drivers/usb/cdns3/gadget.h
> @@ -1072,7 +1072,7 @@ struct cdns3_trb {
>  #define TRB_TDL_SS_SIZE_GET(p)	(((p) & GENMASK(23, 17)) >> 17)
>  
>  /* transfer_len bitmasks - bits 31:24 */
> -#define TRB_BURST_LEN(p)	(((p) << 24) & GENMASK(31, 24))
> +#define TRB_BURST_LEN(p)	(unsigned int)(((p) << 24) & GENMASK(31, 24))

... because TRB_BURST_LEN() is used to intialize a __le32 type. Even if
it ends up being sign extended, the top 32-bits will be ignored.

I'll apply, but it looks like a pointless fix. We shouldn't need it for stable
Felipe Balbi Oct. 27, 2020, 9:03 a.m. UTC | #2
Hi,

Peter Chen <peter.chen@nxp.com> writes:
> For code:
> trb->length = cpu_to_le32(TRB_BURST_LEN(priv_ep->trb_burst_size)
> 	       	| TRB_LEN(length));
>
> TRB_BURST_LEN(priv_ep->trb_burst_size) may be overflow for int 32 if
> priv_ep->trb_burst_size is equal or larger than 0x80;
>
> Below is the Coverity warning:
> sign_extension: Suspicious implicit sign extension: priv_ep->trb_burst_size
> with type u8 (8 bits, unsigned) is promoted in priv_ep->trb_burst_size << 24
> to type int (32 bits, signed), then sign-extended to type unsigned long
> (64 bits, unsigned). If priv_ep->trb_burst_size << 24 is greater than 0x7FFFFFFF,
> the upper bits of the result will all be 1.

looks like a false positive...

> Cc: <stable@vger.kernel.org> #v5.8+
> Fixes: 7733f6c32e36 ("usb: cdns3: Add Cadence USB3 DRD Driver")
> Signed-off-by: Peter Chen <peter.chen@nxp.com>
> ---
>  drivers/usb/cdns3/gadget.h | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/drivers/usb/cdns3/gadget.h b/drivers/usb/cdns3/gadget.h
> index 1ccecd237530..020936cb9897 100644
> --- a/drivers/usb/cdns3/gadget.h
> +++ b/drivers/usb/cdns3/gadget.h
> @@ -1072,7 +1072,7 @@ struct cdns3_trb {
>  #define TRB_TDL_SS_SIZE_GET(p)	(((p) & GENMASK(23, 17)) >> 17)
>  
>  /* transfer_len bitmasks - bits 31:24 */
> -#define TRB_BURST_LEN(p)	(((p) << 24) & GENMASK(31, 24))
> +#define TRB_BURST_LEN(p)	(unsigned int)(((p) << 24) & GENMASK(31, 24))

... because TRB_BURST_LEN() is used to intialize a __le32 type. Even if
it ends up being sign extended, the top 32-bits will be ignored.

I'll apply, but it looks like a pointless fix. We shouldn't need it for
stable.
Peter Chen Oct. 27, 2020, 9:48 a.m. UTC | #3
On 20-10-27 11:03:29, Felipe Balbi wrote:
> 
> Hi,
> 
> Peter Chen <peter.chen@nxp.com> writes:
> > For code:
> > trb->length = cpu_to_le32(TRB_BURST_LEN(priv_ep->trb_burst_size)
> > 	       	| TRB_LEN(length));
> >
> > TRB_BURST_LEN(priv_ep->trb_burst_size) may be overflow for int 32 if
> > priv_ep->trb_burst_size is equal or larger than 0x80;
> >
> > Below is the Coverity warning:
> > sign_extension: Suspicious implicit sign extension: priv_ep->trb_burst_size
> > with type u8 (8 bits, unsigned) is promoted in priv_ep->trb_burst_size << 24
> > to type int (32 bits, signed), then sign-extended to type unsigned long
> > (64 bits, unsigned). If priv_ep->trb_burst_size << 24 is greater than 0x7FFFFFFF,
> > the upper bits of the result will all be 1.
> 
> looks like a false positive...
> 
> > Cc: <stable@vger.kernel.org> #v5.8+
> > Fixes: 7733f6c32e36 ("usb: cdns3: Add Cadence USB3 DRD Driver")
> > Signed-off-by: Peter Chen <peter.chen@nxp.com>
> > ---
> >  drivers/usb/cdns3/gadget.h | 2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
> >
> > diff --git a/drivers/usb/cdns3/gadget.h b/drivers/usb/cdns3/gadget.h
> > index 1ccecd237530..020936cb9897 100644
> > --- a/drivers/usb/cdns3/gadget.h
> > +++ b/drivers/usb/cdns3/gadget.h
> > @@ -1072,7 +1072,7 @@ struct cdns3_trb {
> >  #define TRB_TDL_SS_SIZE_GET(p)	(((p) & GENMASK(23, 17)) >> 17)
> >  
> >  /* transfer_len bitmasks - bits 31:24 */
> > -#define TRB_BURST_LEN(p)	(((p) << 24) & GENMASK(31, 24))
> > +#define TRB_BURST_LEN(p)	(unsigned int)(((p) << 24) & GENMASK(31, 24))
> 
> ... because TRB_BURST_LEN() is used to intialize a __le32 type. Even if
> it ends up being sign extended, the top 32-bits will be ignored.
> 
> I'll apply, but it looks like a pointless fix. We shouldn't need it for stable
> 
At my v2:

It is:
#define TRB_BURST_LEN(p)	((unsigned int)((p) << 24) & GENMASK(31, 24))

It is not related to high 32-bits issue, it is sign extended issue for
32 bits. If p is a unsigned char data, the compiler will consider
(p) << 24 is int, but not unsigned int. So, if the p is larger than
0x80, the (p) << 24 will be overflow. 

If you compile the code:

unsigned int k = 0x80 << 24 + 0x81;

It will report build warning:
warning: left shift count >= width of type [-Wshift-count-overflow]
David Laight Oct. 27, 2020, 10:08 a.m. UTC | #4
From: Peter Chen
> Sent: 27 October 2020 09:49
> 
> On 20-10-27 11:03:29, Felipe Balbi wrote:
> >
> > Hi,
> >
> > Peter Chen <peter.chen@nxp.com> writes:
> > > For code:
> > > trb->length = cpu_to_le32(TRB_BURST_LEN(priv_ep->trb_burst_size)
> > > 	       	| TRB_LEN(length));
> > >
> > > TRB_BURST_LEN(priv_ep->trb_burst_size) may be overflow for int 32 if
> > > priv_ep->trb_burst_size is equal or larger than 0x80;
> > >
> > > Below is the Coverity warning:
> > > sign_extension: Suspicious implicit sign extension: priv_ep->trb_burst_size
> > > with type u8 (8 bits, unsigned) is promoted in priv_ep->trb_burst_size << 24
> > > to type int (32 bits, signed), then sign-extended to type unsigned long
> > > (64 bits, unsigned). If priv_ep->trb_burst_size << 24 is greater than 0x7FFFFFFF,
> > > the upper bits of the result will all be 1.
> >
> > looks like a false positive...
> >
> > > Cc: <stable@vger.kernel.org> #v5.8+
> > > Fixes: 7733f6c32e36 ("usb: cdns3: Add Cadence USB3 DRD Driver")
> > > Signed-off-by: Peter Chen <peter.chen@nxp.com>
> > > ---
> > >  drivers/usb/cdns3/gadget.h | 2 +-
> > >  1 file changed, 1 insertion(+), 1 deletion(-)
> > >
> > > diff --git a/drivers/usb/cdns3/gadget.h b/drivers/usb/cdns3/gadget.h
> > > index 1ccecd237530..020936cb9897 100644
> > > --- a/drivers/usb/cdns3/gadget.h
> > > +++ b/drivers/usb/cdns3/gadget.h
> > > @@ -1072,7 +1072,7 @@ struct cdns3_trb {
> > >  #define TRB_TDL_SS_SIZE_GET(p)	(((p) & GENMASK(23, 17)) >> 17)
> > >
> > >  /* transfer_len bitmasks - bits 31:24 */
> > > -#define TRB_BURST_LEN(p)	(((p) << 24) & GENMASK(31, 24))
> > > +#define TRB_BURST_LEN(p)	(unsigned int)(((p) << 24) & GENMASK(31, 24))
> >
> > ... because TRB_BURST_LEN() is used to intialize a __le32 type. Even if
> > it ends up being sign extended, the top 32-bits will be ignored.
> >
> > I'll apply, but it looks like a pointless fix. We shouldn't need it for stable
> >
> At my v2:
> 
> It is:
> #define TRB_BURST_LEN(p)	((unsigned int)((p) << 24) & GENMASK(31, 24))
> 
> It is not related to high 32-bits issue, it is sign extended issue for
> 32 bits. If p is a unsigned char data, the compiler will consider
> (p) << 24 is int, but not unsigned int. So, if the p is larger than
> 0x80, the (p) << 24 will be overflow.
> 
> If you compile the code:
> 
> unsigned int k = 0x80 << 24 + 0x81;
> 
> It will report build warning:
> warning: left shift count >= width of type [-Wshift-count-overflow]

Something like:
#define TRB_BURST_LEN(p)  (((p) + 0u) << 24)
will remove the warning.

The GENMASK() is pretty pointless - the compiler may fail to optimise
it away.
If p is 64bit the high bits should get discarded pretty quickly.

	David

-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK
Registration No: 1397386 (Wales)
Alan Stern Oct. 27, 2020, 2:21 p.m. UTC | #5
On Tue, Oct 27, 2020 at 09:48:54AM +0000, Peter Chen wrote:
> If you compile the code:
> 
> unsigned int k = 0x80 << 24 + 0x81;
> 
> It will report build warning:
> warning: left shift count >= width of type [-Wshift-count-overflow]

That's a separate issue.  I believe (but haven't checked) that the << 
operator has lower precedence than +, so the compiler interprets the 
expression as:

unsigned int k = 0x80 << (24 + 0x81);

and it's pretty obvious why this causes an error.  Instead, try 
compiling:

unsigned int k = (0x80 << 24) + 0x81;

You may get an error message about signed-integer overflow, but not 
about shift-count overflow.

Alan Stern
Peter Chen Oct. 28, 2020, 6:41 a.m. UTC | #6
> That's a separate issue.  I believe (but haven't checked) that the << operator
> has lower precedence than +, so the compiler interprets the expression as:
> 
> unsigned int k = 0x80 << (24 + 0x81);
> 
> and it's pretty obvious why this causes an error.  Instead, try
> compiling:
> 
> unsigned int k = (0x80 << 24) + 0x81;
> 
> You may get an error message about signed-integer overflow, but not about
> shift-count overflow.
> 

Hi Alan,

Your analysis is correct, I did not check the warning message correctly.

Peter
diff mbox series

Patch

diff --git a/drivers/usb/cdns3/gadget.h b/drivers/usb/cdns3/gadget.h
index 1ccecd237530..020936cb9897 100644
--- a/drivers/usb/cdns3/gadget.h
+++ b/drivers/usb/cdns3/gadget.h
@@ -1072,7 +1072,7 @@  struct cdns3_trb {
 #define TRB_TDL_SS_SIZE_GET(p)	(((p) & GENMASK(23, 17)) >> 17)
 
 /* transfer_len bitmasks - bits 31:24 */
-#define TRB_BURST_LEN(p)	(((p) << 24) & GENMASK(31, 24))
+#define TRB_BURST_LEN(p)	(unsigned int)(((p) << 24) & GENMASK(31, 24))
 #define TRB_BURST_LEN_GET(p)	(((p) & GENMASK(31, 24)) >> 24)
 
 /* Data buffer pointer bitmasks*/