2014-01-08 07:32:38

by Hannes Reinecke

[permalink] [raw]
Subject: Re: [PATCH] drivers: target: target_core_mod: use div64_u64_rem() instead of operator '%' for u64

On 12/24/2013 04:35 AM, Chen Gang wrote:
> On 12/23/2013 02:51 PM, Nicholas A. Bellinger wrote:
>> On Sun, 2013-12-22 at 17:17 +0800, Chen Gang wrote:
>>> On 12/22/2013 10:56 AM, Nicholas A. Bellinger wrote:
>>>> Hi Chen,
>>>>
>>>> On Sat, 2013-12-21 at 10:08 +0800, Chen Gang wrote:
>>>>> In kernel, need use div64_u64_rem() instead of operator '%' for u64, or
>>>>> can not pass compiling (with allmodconfig under metag):
>>>>>
>>>>> MODPOST 2909 modules
>>>>> ERROR: "__umoddi3" [drivers/target/target_core_mod.ko] undefined!
>>>>>
>>>>> Also need u64 type cast for u32 variable multiply u32 variable, or will
>>>>> cause type overflow issue.
>>>>>
>>>>>
>>>>> Signed-off-by: Chen Gang <[email protected]>
>>>>> ---
>>>>> drivers/target/target_core_alua.c | 3 ++-
>>>>> 1 file changed, 2 insertions(+), 1 deletion(-)
>>>>>
>>>>
>>>> FYI, this unsigned long long division in core_alua_state_lba_dependent()
>>>> was fixed for 32-bit in linux-next >= 12192013 code.
>>>>
>>>
>>> OK, thanks.
>>>
>>> The related fix patch changed "start_lba = lba % ..." to "start_lba =
>>> lba / ...", and also assumed "segment_size * segment_mult" is still
>>> within u32 (can not cause type over flow).
>>>
>>> I guess the original author already knew about them, and intended to do
>>> like that (if not, please let me know, thanks).
>>>
>>
>> Sorry, your correct that the original code is using modulo division to
>> calculate start_lba.
>>
>
> Oh, that's all right, (in fact, don't need sorry), I am not quite
> familiar with the details, so need related member help check it. :-)
>
>> Hannes, can you please double check this below..?
>>
>
> Please help check when have time, thanks.
>
I would even convert segment_size and segment_mult to u64,
to ensure no overflows occur:

diff --git a/drivers/target/target_core_alua.c
b/drivers/target/target_core_alua
.c
index 9b1856d..54b1e52 100644
--- a/drivers/target/target_core_alua.c
+++ b/drivers/target/target_core_alua.c
@@ -477,8 +477,7 @@ static inline int core_alua_state_lba_dependent(
u8 *alua_ascq)
{
struct se_device *dev = cmd->se_dev;
- u32 segment_size, segment_mult, sectors;
- u64 lba;
+ u64 segment_size, segment_mult, sectors, lba;

/* Only need to check for cdb actually containing LBAs */
if (!(cmd->se_cmd_flags & SCF_SCSI_DATA_CDB))


Other than that the sector_div() patch is correct.

Cheers,

Hannes
--
Dr. Hannes Reinecke zSeries & Storage
[email protected] +49 911 74053 688
SUSE LINUX Products GmbH, Maxfeldstr. 5, 90409 Nürnberg
GF: J. Hawn, J. Guild, F. Imendörffer, HRB 16746 (AG Nürnberg)


2014-01-08 14:40:29

by Chen Gang

[permalink] [raw]
Subject: Re: [PATCH] drivers: target: target_core_mod: use div64_u64_rem() instead of operator '%' for u64

On 01/08/2014 03:32 PM, Hannes Reinecke wrote:
> On 12/24/2013 04:35 AM, Chen Gang wrote:
>> On 12/23/2013 02:51 PM, Nicholas A. Bellinger wrote:
>>> On Sun, 2013-12-22 at 17:17 +0800, Chen Gang wrote:
>>>> On 12/22/2013 10:56 AM, Nicholas A. Bellinger wrote:
>>>>> Hi Chen,
>>>>>
>>>>> On Sat, 2013-12-21 at 10:08 +0800, Chen Gang wrote:
>>>>>> In kernel, need use div64_u64_rem() instead of operator '%' for u64, or
>>>>>> can not pass compiling (with allmodconfig under metag):
>>>>>>
>>>>>> MODPOST 2909 modules
>>>>>> ERROR: "__umoddi3" [drivers/target/target_core_mod.ko] undefined!
>>>>>>
>>>>>> Also need u64 type cast for u32 variable multiply u32 variable, or will
>>>>>> cause type overflow issue.
>>>>>>
>>>>>>
>>>>>> Signed-off-by: Chen Gang <[email protected]>
>>>>>> ---
>>>>>> drivers/target/target_core_alua.c | 3 ++-
>>>>>> 1 file changed, 2 insertions(+), 1 deletion(-)
>>>>>>
>>>>>
>>>>> FYI, this unsigned long long division in core_alua_state_lba_dependent()
>>>>> was fixed for 32-bit in linux-next >= 12192013 code.
>>>>>
>>>>
>>>> OK, thanks.
>>>>
>>>> The related fix patch changed "start_lba = lba % ..." to "start_lba =
>>>> lba / ...", and also assumed "segment_size * segment_mult" is still
>>>> within u32 (can not cause type over flow).
>>>>
>>>> I guess the original author already knew about them, and intended to do
>>>> like that (if not, please let me know, thanks).
>>>>
>>>
>>> Sorry, your correct that the original code is using modulo division to
>>> calculate start_lba.
>>>
>>
>> Oh, that's all right, (in fact, don't need sorry), I am not quite
>> familiar with the details, so need related member help check it. :-)
>>
>>> Hannes, can you please double check this below..?
>>>
>>
>> Please help check when have time, thanks.
>>
> I would even convert segment_size and segment_mult to u64,
> to ensure no overflows occur:
>
> diff --git a/drivers/target/target_core_alua.c
> b/drivers/target/target_core_alua
> .c
> index 9b1856d..54b1e52 100644
> --- a/drivers/target/target_core_alua.c
> +++ b/drivers/target/target_core_alua.c
> @@ -477,8 +477,7 @@ static inline int core_alua_state_lba_dependent(
> u8 *alua_ascq)
> {
> struct se_device *dev = cmd->se_dev;
> - u32 segment_size, segment_mult, sectors;
> - u64 lba;
> + u64 segment_size, segment_mult, sectors, lba;
>
> /* Only need to check for cdb actually containing LBAs */
> if (!(cmd->se_cmd_flags & SCF_SCSI_DATA_CDB))
>
>

OK, thanks.

> Other than that the sector_div() patch is correct.
>

So we really need use '/' instead of original '%'?


Thanks
--
Chen Gang

Open, share and attitude like air, water and life which God blessed

2014-01-08 23:16:59

by Nicholas A. Bellinger

[permalink] [raw]
Subject: Re: [PATCH] drivers: target: target_core_mod: use div64_u64_rem() instead of operator '%' for u64

On Wed, 2014-01-08 at 08:32 +0100, Hannes Reinecke wrote:
> On 12/24/2013 04:35 AM, Chen Gang wrote:
> > On 12/23/2013 02:51 PM, Nicholas A. Bellinger wrote:
> >> On Sun, 2013-12-22 at 17:17 +0800, Chen Gang wrote:

<SNIP>

> >>> The related fix patch changed "start_lba = lba % ..." to "start_lba =
> >>> lba / ...", and also assumed "segment_size * segment_mult" is still
> >>> within u32 (can not cause type over flow).
> >>>
> >>> I guess the original author already knew about them, and intended to do
> >>> like that (if not, please let me know, thanks).
> >>>
> >>
> >> Sorry, your correct that the original code is using modulo division to
> >> calculate start_lba.
> >>
> >
> > Oh, that's all right, (in fact, don't need sorry), I am not quite
> > familiar with the details, so need related member help check it. :-)
> >
> >> Hannes, can you please double check this below..?
> >>
> >
> > Please help check when have time, thanks.
> >
> I would even convert segment_size and segment_mult to u64,
> to ensure no overflows occur:
>
> diff --git a/drivers/target/target_core_alua.c
> b/drivers/target/target_core_alua
> .c
> index 9b1856d..54b1e52 100644
> --- a/drivers/target/target_core_alua.c
> +++ b/drivers/target/target_core_alua.c
> @@ -477,8 +477,7 @@ static inline int core_alua_state_lba_dependent(
> u8 *alua_ascq)
> {
> struct se_device *dev = cmd->se_dev;
> - u32 segment_size, segment_mult, sectors;
> - u64 lba;
> + u64 segment_size, segment_mult, sectors, lba;
>
> /* Only need to check for cdb actually containing LBAs */
> if (!(cmd->se_cmd_flags & SCF_SCSI_DATA_CDB))
>
>

Will squash the above into the original patch shortly in for-next..

> Other than that the sector_div() patch is correct.
>

<nod> Thanks for confirming that sector_div() is correct here vs. the
original code using modulo that Chen had pointed out.

Thanks Hannes!

--nab

2014-01-09 08:14:41

by Hannes Reinecke

[permalink] [raw]
Subject: Re: [PATCH] drivers: target: target_core_mod: use div64_u64_rem() instead of operator '%' for u64

On 01/09/2014 12:18 AM, Nicholas A. Bellinger wrote:
> On Wed, 2014-01-08 at 08:32 +0100, Hannes Reinecke wrote:
>> On 12/24/2013 04:35 AM, Chen Gang wrote:
>>> On 12/23/2013 02:51 PM, Nicholas A. Bellinger wrote:
>>>> On Sun, 2013-12-22 at 17:17 +0800, Chen Gang wrote:
>
> <SNIP>
>
>>>>> The related fix patch changed "start_lba = lba % ..." to "start_lba =
>>>>> lba / ...", and also assumed "segment_size * segment_mult" is still
>>>>> within u32 (can not cause type over flow).
>>>>>
>>>>> I guess the original author already knew about them, and intended to do
>>>>> like that (if not, please let me know, thanks).
>>>>>
>>>>
>>>> Sorry, your correct that the original code is using modulo division to
>>>> calculate start_lba.
>>>>
>>>
>>> Oh, that's all right, (in fact, don't need sorry), I am not quite
>>> familiar with the details, so need related member help check it. :-)
>>>
>>>> Hannes, can you please double check this below..?
>>>>
>>>
>>> Please help check when have time, thanks.
>>>
>> I would even convert segment_size and segment_mult to u64,
>> to ensure no overflows occur:
>>
>> diff --git a/drivers/target/target_core_alua.c
>> b/drivers/target/target_core_alua
>> .c
>> index 9b1856d..54b1e52 100644
>> --- a/drivers/target/target_core_alua.c
>> +++ b/drivers/target/target_core_alua.c
>> @@ -477,8 +477,7 @@ static inline int core_alua_state_lba_dependent(
>> u8 *alua_ascq)
>> {
>> struct se_device *dev = cmd->se_dev;
>> - u32 segment_size, segment_mult, sectors;
>> - u64 lba;
>> + u64 segment_size, segment_mult, sectors, lba;
>>
>> /* Only need to check for cdb actually containing LBAs */
>> if (!(cmd->se_cmd_flags & SCF_SCSI_DATA_CDB))
>>
>>
>
> Will squash the above into the original patch shortly in for-next..
>
>> Other than that the sector_div() patch is correct.
>>
>
> <nod> Thanks for confirming that sector_div() is correct here vs. the
> original code using modulo that Chen had pointed out.
>
Ah, _that_ was the issue.
I was wondering why you kept on poking me ...

Well.
No, that's actually _not_ correct.
The correct fix would be

diff --git a/drivers/target/target_core_alua.c
b/drivers/target/target_core_alua.c
index 54b1e52..12da9b3 100644
--- a/drivers/target/target_core_alua.c
+++ b/drivers/target/target_core_alua.c
@@ -500,8 +500,7 @@ static inline int core_alua_state_lba_dependent(

if (segment_mult) {
u64 tmp = lba;
- sector_div(tmp, segment_size *
segment_mult);
- start_lba = tmp;
+ start_lba = sector_div(tmp, segment_size
* segment_mult);

last_lba = first_lba + segment_size - 1;
if (start_lba >= first_lba &&
(beware of line breaks ...)
Thing is, we need to calculate the offset into the segment to figure out
which map entry to use.
The actual number of the segment (as had been calculated with the
original fix) is immaterial here.

Sorry for this. The email thread just flew past me during Xmas
with me not paying real attention.

Cheers,

Hannes
--
Dr. Hannes Reinecke zSeries & Storage
[email protected] +49 911 74053 688
SUSE LINUX Products GmbH, Maxfeldstr. 5, 90409 Nürnberg
GF: J. Hawn, J. Guild, F. Imendörffer, HRB 16746 (AG Nürnberg)

2014-01-10 05:46:11

by Nicholas A. Bellinger

[permalink] [raw]
Subject: Re: [PATCH] drivers: target: target_core_mod: use div64_u64_rem() instead of operator '%' for u64

On Thu, 2014-01-09 at 11:17 +0100, Hannes Reinecke wrote:
> On 01/09/2014 12:18 AM, Nicholas A. Bellinger wrote:
> > On Wed, 2014-01-08 at 08:32 +0100, Hannes Reinecke wrote:

<SNIP>

> >> Other than that the sector_div() patch is correct.
> >>
> >
> > <nod> Thanks for confirming that sector_div() is correct here vs. the
> > original code using modulo that Chen had pointed out.
> >
> Ah, _that_ was the issue.
> I was wondering why you kept on poking me ...
>
> Well.
> No, that's actually _not_ correct.
> The correct fix would be
>
> diff --git a/drivers/target/target_core_alua.c
> b/drivers/target/target_core_alua.c
> index 54b1e52..12da9b3 100644
> --- a/drivers/target/target_core_alua.c
> +++ b/drivers/target/target_core_alua.c
> @@ -500,8 +500,7 @@ static inline int core_alua_state_lba_dependent(
>
> if (segment_mult) {
> u64 tmp = lba;
> - sector_div(tmp, segment_size * segment_mult);
> - start_lba = tmp;
> + start_lba = sector_div(tmp, segment_size * segment_mult);
>
> last_lba = first_lba + segment_size - 1;
> if (start_lba >= first_lba &&
> (beware of line breaks ...)
> Thing is, we need to calculate the offset into the segment to figure out
> which map entry to use.
> The actual number of the segment (as had been calculated with the
> original fix) is immaterial here.
>
> Sorry for this. The email thread just flew past me during Xmas
> with me not paying real attention.
>

Applied + squashed. Apologies for the initial pre-holiday BUG..

Thanks Hannes!

--nab

2014-01-10 16:06:06

by Chen Gang

[permalink] [raw]
Subject: Re: [PATCH] drivers: target: target_core_mod: use div64_u64_rem() instead of operator '%' for u64

On 01/10/2014 01:47 PM, Nicholas A. Bellinger wrote:
> On Thu, 2014-01-09 at 11:17 +0100, Hannes Reinecke wrote:
>> On 01/09/2014 12:18 AM, Nicholas A. Bellinger wrote:
>>> On Wed, 2014-01-08 at 08:32 +0100, Hannes Reinecke wrote:
>
> <SNIP>
>
>>>> Other than that the sector_div() patch is correct.
>>>>
>>>
>>> <nod> Thanks for confirming that sector_div() is correct here vs. the
>>> original code using modulo that Chen had pointed out.
>>>
>> Ah, _that_ was the issue.
>> I was wondering why you kept on poking me ...
>>
>> Well.
>> No, that's actually _not_ correct.
>> The correct fix would be
>>
>> diff --git a/drivers/target/target_core_alua.c
>> b/drivers/target/target_core_alua.c
>> index 54b1e52..12da9b3 100644
>> --- a/drivers/target/target_core_alua.c
>> +++ b/drivers/target/target_core_alua.c
>> @@ -500,8 +500,7 @@ static inline int core_alua_state_lba_dependent(
>>
>> if (segment_mult) {
>> u64 tmp = lba;
>> - sector_div(tmp, segment_size * segment_mult);
>> - start_lba = tmp;
>> + start_lba = sector_div(tmp, segment_size * segment_mult);
>>
>> last_lba = first_lba + segment_size - 1;
>> if (start_lba >= first_lba &&
>> (beware of line breaks ...)
>> Thing is, we need to calculate the offset into the segment to figure out
>> which map entry to use.
>> The actual number of the segment (as had been calculated with the
>> original fix) is immaterial here.
>>
>> Sorry for this. The email thread just flew past me during Xmas
>> with me not paying real attention.
>>
>
> Applied + squashed. Apologies for the initial pre-holiday BUG..
>
> Thanks Hannes!
>

Thank all of you.

--
Chen Gang

Open, share and attitude like air, water and life which God blessed