2020-04-12 13:30:37

by Ivan Safonov

[permalink] [raw]
Subject: Re: [PATCH 5/8] staging: rtl8188eu: Remove function rtw_modular64()

> Remove function rtw_modular64 as all it does is call do_div.

This is wrong. Macro do_div(x, y) change first argument x, but
rtw_modular64(x, y) preserve it.

> + tsf = pmlmeext->TSFValue - do_div(pmlmeext->TSFValue, (pmlmeinfo->bcn_interval*1024)) - 1024; /* us */

rounddown(pmlmeext->TSFValue, pmlmeinfo->bcn_interval * 1024) - 1024
is a better replacement for

> - tsf = pmlmeext->TSFValue - rtw_modular64(pmlmeext->TSFValue, (pmlmeinfo->bcn_interval*1024)) - 1024; /* us */

Patch '[PATCH 01/10] staging: rtl8723bs: Remove function
rtw_modular64()' have same bug.

Ivan Safonov.


2020-04-14 15:25:52

by Dan Carpenter

[permalink] [raw]
Subject: Re: [PATCH 5/8] staging: rtl8188eu: Remove function rtw_modular64()

On Sun, Apr 12, 2020 at 04:34:08PM +0300, Ivan Safonov wrote:
> > Remove function rtw_modular64 as all it does is call do_div.
>
> This is wrong. Macro do_div(x, y) change first argument x, but
> rtw_modular64(x, y) preserve it.
>
> > + tsf = pmlmeext->TSFValue - do_div(pmlmeext->TSFValue, (pmlmeinfo->bcn_interval*1024)) - 1024; /* us */
>
> rounddown(pmlmeext->TSFValue, pmlmeinfo->bcn_interval * 1024) - 1024
> is a better replacement for

You're absolutely correct that the patch is buggy, but I'm not sure that
rounddown() is what we want.

rtw_modular64() took the MOD of x. So it should be something like:

tsf = pmlmeext->TSFValue - (pmlmeext->TSFValue % (pmlmeinfo->bcn_interval * 1024)) - 1024; /* us */

But what the heck is that even??? If pmlmeinfo->bcn_interval is zero
or one then the subtraction ends up giving us a negative.

regards,
dan carpenter

2020-04-14 15:29:54

by Ivan Safonov

[permalink] [raw]
Subject: Re: [PATCH 5/8] staging: rtl8188eu: Remove function rtw_modular64()

On 4/14/20 2:56 PM, Dan Carpenter wrote:
> On Sun, Apr 12, 2020 at 04:34:08PM +0300, Ivan Safonov wrote:
>>> Remove function rtw_modular64 as all it does is call do_div.
>>
>> This is wrong. Macro do_div(x, y) change first argument x, but
>> rtw_modular64(x, y) preserve it.
>>
>>> + tsf = pmlmeext->TSFValue - do_div(pmlmeext->TSFValue, (pmlmeinfo->bcn_interval*1024)) - 1024; /* us */
>>
>> rounddown(pmlmeext->TSFValue, pmlmeinfo->bcn_interval * 1024) - 1024
>> is a better replacement for
>
> You're absolutely correct that the patch is buggy, but I'm not sure that
> rounddown() is what we want.
>
> rtw_modular64() took the MOD of x. So it should be something like:
>
> tsf = pmlmeext->TSFValue - (pmlmeext->TSFValue % (pmlmeinfo->bcn_interval * 1024)) - 1024; /* us */
>
> But what the heck is that even??? If pmlmeinfo->bcn_interval is zero
> or one then the subtraction ends up giving us a negative.
>
> regards,
> dan carpenter
>

1. pmlmeext->TSFValue can not be negative, because it is uint64_t;
2. pmlmeext->TSFValue is cyclic value:
https://en.wikipedia.org/wiki/Timing_synchronization_function ;
3. (rounddown(a, b)) is equal to (a - a % b) by definition.

Ivan Safonov.

2020-04-14 15:50:44

by Dan Carpenter

[permalink] [raw]
Subject: Re: [PATCH 5/8] staging: rtl8188eu: Remove function rtw_modular64()

On Tue, Apr 14, 2020 at 03:22:59PM +0300, Ivan Safonov wrote:
> On 4/14/20 2:56 PM, Dan Carpenter wrote:
> > On Sun, Apr 12, 2020 at 04:34:08PM +0300, Ivan Safonov wrote:
> > > > Remove function rtw_modular64 as all it does is call do_div.
> > >
> > > This is wrong. Macro do_div(x, y) change first argument x, but
> > > rtw_modular64(x, y) preserve it.
> > >
> > > > + tsf = pmlmeext->TSFValue - do_div(pmlmeext->TSFValue, (pmlmeinfo->bcn_interval*1024)) - 1024; /* us */
> > >
> > > rounddown(pmlmeext->TSFValue, pmlmeinfo->bcn_interval * 1024) - 1024
> > > is a better replacement for
> >
> > You're absolutely correct that the patch is buggy, but I'm not sure that
> > rounddown() is what we want.
> >
> > rtw_modular64() took the MOD of x. So it should be something like:
> >
> > tsf = pmlmeext->TSFValue - (pmlmeext->TSFValue % (pmlmeinfo->bcn_interval * 1024)) - 1024; /* us */
> >
> > But what the heck is that even??? If pmlmeinfo->bcn_interval is zero
> > or one then the subtraction ends up giving us a negative.
> >
> > regards,
> > dan carpenter
> >
>
> 1. pmlmeext->TSFValue can not be negative, because it is uint64_t;
> 2. pmlmeext->TSFValue is cyclic value:
> https://en.wikipedia.org/wiki/Timing_synchronization_function ;
> 3. (rounddown(a, b)) is equal to (a - a % b) by definition.

Yeah. You're right. I got mixed up and I misread what you were
suggesting.

tsf = rounddown(pmlmeext->TSFValue, pmlmeinfo->bcn_interval * 1024) - 1024;

regards,
dan carpenter