scsi: mpt2sas: try 64 bit DMA when 32 bit DMA fails

The MULDIV macro has been designed for small
numbers. It emits an overflow warning on 64 bit
systems. This patch places type casts in the
parameters to fix the compiler warning.

Signed-off-by Sinan Kaya <[email protected]>

---
 drivers/scsi/sg.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

drivers/scsi/sg.c +0 -0

index 9d7b7db..eb2739d 100644

drivers/scsi/sg.c +0 -1

 #define SG_DEFAULT_TIMEOUT MULDIV(SG_DEFAULT_TIMEOUT_USER, HZ, USER_HZ)
 
--

Qualcomm Technologies, Inc. on behalf of Qualcomm Innovation Center, Inc.
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, a Linux Foundation Collaborative Project

--
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html

Quoted 15 lines · Sinan Kaya

The MULDIV macro has been designed for small
numbers. It emits an overflow warning on 64 bit
systems. This patch places type casts in the
parameters to fix the compiler warning.

Signed-off-by Sinan Kaya <[email protected]>

---
  drivers/scsi/sg.c | 5 ++++-
  1 file changed, 4 insertions(+), 1 deletion(-)

drivers/scsi/sg.c +0 -0

index 9d7b7db..eb2739d 100644

drivers/scsi/sg.c static void sg_proc_cleanup(void); · line 94

Quoted 8 lines

drivers/scsi/sg.c +4 -1

@@ -88,7 +88,10 @@ static void sg_proc_cleanup(void);
   * Of course an overflow is inavoidable if the result of muldiv doesn't fit
   * in 32 bits.
   */
-#define MULDIV(X,MUL,DIV) ((((X % DIV) * MUL) / DIV) + ((X / DIV) * MUL))
+static inline u64 MULDIV(u64 X, u32 MUL, u32 DIV)
+{
+       return ((((X % DIV) * MUL) / DIV) + ((X / DIV) * MUL));
+}

Like kbuild bot already told you it would be nice to think of 32-bit architectures.

Moreover we have mult_frac() macro already for 32-bit numbers.

For 64 bit numbers you need to do do_div().

Like:

static inline u64 mult_frac64(u64 x, u32 m, u32 n) {
u64 ret;

ret = do_div(x, n);
return ret * m;
}

Quoted 12 lines

drivers/scsi/sg.c +0 -1

  #define SG_DEFAULT_TIMEOUT MULDIV(SG_DEFAULT_TIMEOUT_USER, HZ, USER_HZ)
 
--

Qualcomm Technologies, Inc. on behalf of Qualcomm Innovation Center, Inc.
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, a Linux Foundation Collaborative Project

--
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html

#define MULDIV(X,MUL,DIV) mult_frac64(X, MUL, DIV)

Quoted 39 lines · Andy Shevchenko

Quoted 22 lines · Sinan Kaya

The MULDIV macro has been designed for small
numbers. It emits an overflow warning on 64 bit
systems. This patch places type casts in the
parameters to fix the compiler warning.

Signed-off-by Sinan Kaya <[email protected]>

---
  drivers/scsi/sg.c | 5 ++++-
  1 file changed, 4 insertions(+), 1 deletion(-)

drivers/scsi/sg.c +0 -0

index 9d7b7db..eb2739d 100644
@@ -88,7 +88,10 @@ static void sg_proc_cleanup(void);
   * Of course an overflow is inavoidable if the result of muldiv doesn't

fit
* in 32 bits.
*/
-#define MULDIV(X,MUL,DIV) ((((X % DIV) * MUL) / DIV) + ((X / DIV) * MUL))

drivers/scsi/sg.c static void sg_proc_cleanup(void); · line 94

Quoted 4 lines

drivers/scsi/sg.c +4 -0

+static inline u64 MULDIV(u64 X, u32 MUL, u32 DIV)
+{
+       return ((((X % DIV) * MUL) / DIV) + ((X / DIV) * MUL));
+}

Like kbuild bot already told you it would be nice to think of 32-bit architectures.

Moreover we have mult_frac() macro already for 32-bit numbers.

For 64 bit numbers you need to do do_div().

Like:

Quoted 8 lines

static inline u64 mult_frac64(u64 x, u32 m, u32 n) {
u64 ret;

ret = do_div(x, n);
return ret * m;
}

OK, I didn't know that we had such a macro. To make this look like the other macro, I can do this.

static inline u64 mult_frac64(u64 x, u32 numer, u32 denom)

{
        u64 quot;
        u64 rem  = x % denom;
        u64 rem2;

        quot = x;
        do_div(quot, denom);

        rem2 = rem * numer;
        do_div(rem2, denom);

        return (quot * numer) + rem2;
}

#define MULDIV(X,MUL,DIV) mult_frac64(X, MUL, DIV)

Quoted 9 lines

Quoted 9 lines

  #define SG_DEFAULT_TIMEOUT MULDIV(SG_DEFAULT_TIMEOUT_USER, HZ, USER_HZ)

--
Qualcomm Technologies, Inc. on behalf of Qualcomm Innovation Center, Inc. Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, a Linux Foundation Collaborative Project

Quoted 8 lines

Quoted 4 lines

--
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html

--
Sinan Kaya
Qualcomm Technologies, Inc. on behalf of Qualcomm Innovation Center, Inc. Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, a Linux Foundation Collaborative Project

Quoted 12 lines

OK, I didn't know that we had such a macro. To make this look like the other

Quoted 11 lines

macro, I can do this.

static inline u64 mult_frac64(u64 x, u32 numer, u32 denom)

{
        u64 quot;
        u64 rem  = x % denom;
        u64 rem2;

        quot = x;
        do_div(quot, denom);

Quoted 5 lines

Quoted 5 lines

        rem2 = rem * numer;
        do_div(rem2, denom);

        return (quot * numer) + rem2;
}

Might be I did a wrong smaple, but do_div() returns two values actually. You perhaps overlooked it and thus wrote something redundant above.

On 11/5/2015 1:07 PM, Andy Shevchenko wrote:

Let's try again.

static inline u64 mult_frac64(u64 x, u32 numer, u32 denom) {
        u64 rem  = x % denom;
        u64 quot = do_div(x, denom);
        u64 mul = rem * numer;

        return (quot * numer) + do_div(mul, denom);
}

Quoted 11 lines · Sinan Kaya

On 11/5/2015 1:07 PM, Andy Shevchenko wrote:

Let's try again.

static inline u64 mult_frac64(u64 x, u32 numer, u32 denom) {
        u64 rem  = x % denom;
        u64 quot = do_div(x, denom);
        u64 mul = rem * numer;

        return (quot * numer) + do_div(mul, denom);
}

First of all why not to put this to generic header? We have math64.h and kernel.h.
Might be a good idea (needs to check current users) to move mult_frac to math64.h.

Then, x % y is already a problem. After all, you seems messed quot and remainder.

What about something like

#if BITS_PER_LONG == 64

#define mult_frac64(x,n,d) mult_frac(x,n,d)

#else

static inline u64 mult_frac64(u64 x, u32 numer, u32 denom) {
        u64 r1 = do_div(x, denom);
        u64 r2 = r1 * numer;

        do_div(r2, denom);
        return (x * numer) + r2;
}

#endif

?

One more look to the users of MULDIV.

They all seems 32 bit for x.
It means you don't need two do_div()s at all.

Just do something like:

u64 d = x * numer;
do_div(d, denom);
return d;

Quoted 29 lines · Andy Shevchenko

Quoted 11 lines · Sinan Kaya

On 11/5/2015 1:07 PM, Andy Shevchenko wrote:

Let's try again.

static inline u64 mult_frac64(u64 x, u32 numer, u32 denom) {
         u64 rem  = x % denom;
         u64 quot = do_div(x, denom);
         u64 mul = rem * numer;

         return (quot * numer) + do_div(mul, denom);
}

First of all why not to put this to generic header? We have math64.h and kernel.h.
Might be a good idea (needs to check current users) to move mult_frac to math64.h.

Then, x % y is already a problem. After all, you seems messed quot and remainder.

What about something like

#if BITS_PER_LONG == 64

#define mult_frac64(x,n,d) mult_frac(x,n,d)

#else

Quoted 7 lines

static inline u64 mult_frac64(u64 x, u32 numer, u32 denom) {
         u64 r1 = do_div(x, denom);
         u64 r2 = r1 * numer;

         do_div(r2, denom);
         return (x * numer) + r2;
}

Quoted 4 lines

#endif

?

One more look to the users of MULDIV.

They all seems 32 bit for x.
It means you don't need two do_div()s at all.

Just do something like:

u64 d = x * numer;
do_div(d, denom);
return d;