linux-kernel - Re: [PATCH] Typo in x86 apic.c with DIVISOR setup

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:	Fri, 10 Oct 2008 09:39:15 +0400
From:	"Cyrill Gorcunov" <gorcunov@...il.com>
To:	"Pallipadi, Venkatesh" <venkatesh.pallipadi@...el.com>
Cc:	"Ingo Molnar" <mingo@...e.hu>,
	"Thomas Gleixner" <tglx@...utronix.de>,
	"H. Peter Anvin" <hpa@...or.com>,
	linux-kernel <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH] Typo in x86 apic.c with DIVISOR setup

On Fri, Oct 10, 2008 at 9:10 AM, Pallipadi, Venkatesh
<venkatesh.pallipadi@...el.com> wrote:
>
>
>>-----Original Message-----
>>From: Cyrill Gorcunov [mailto:gorcunov@...il.com]
>>Sent: Thursday, October 09, 2008 9:38 PM
>>To: Pallipadi, Venkatesh
>>Cc: Ingo Molnar; Thomas Gleixner; H. Peter Anvin; linux-kernel
>>Subject: Re: [PATCH] Typo in x86 apic.c with DIVISOR setup
>>
>>On Fri, Oct 10, 2008 at 4:22 AM, Venki Pallipadi
>><venkatesh.pallipadi@...el.com> wrote:
>>>
>>> Fix a typo in x86 apic.c CONFG_X86_64 -> CONFIG_X86_64.
>>> The intention looks like was to have divisor of 1 on x86_64 kernel.
>>>
>>> Also, for DIVISOR value of 1 to work, APIC_TDR_DIV should also change
>>> depending on divide by 1 or divide by 16 is used.
>>>
>>> Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@...el.com>
>>>
>>> ---
>>>  arch/x86/kernel/apic.c |    6 ++++--
>>>  1 file changed, 4 insertions(+), 2 deletions(-)
>>>
>>> Index: tip/arch/x86/kernel/apic.c
>>> ===================================================================
>>> --- tip.orig/arch/x86/kernel/apic.c     2008-10-09
>>17:12:39.000000000 -0700
>>> +++ tip/arch/x86/kernel/apic.c  2008-10-09 17:13:29.000000000 -0700
>>> @@ -332,10 +332,12 @@ int lapic_get_maxlvt(void)
>>>  */
>>>
>>>  /* Clock divisor */
>>> -#ifdef CONFG_X86_64
>>> +#ifdef CONFIG_X86_64
>>>  #define APIC_DIVISOR 1
>>> +#define APIC_TDR_DIV APIC_TDR_DIV_1
>>>  #else
>>>  #define APIC_DIVISOR 16
>>> +#define APIC_TDR_DIV APIC_TDR_DIV_16
>>>  #endif
>>>
>>>  /*
>>> @@ -369,7 +371,7 @@ static void __setup_APIC_LVTT(unsigned i
>>>        tmp_value = apic_read(APIC_TDCR);
>>>        apic_write(APIC_TDCR,
>>>                (tmp_value & ~(APIC_TDR_DIV_1 |
>>APIC_TDR_DIV_TMBASE)) |
>>> -               APIC_TDR_DIV_16);
>>> +               APIC_TDR_DIV);
>>>
>>>        if (!oneshot)
>>>                apic_write(APIC_TMICT, clocks / APIC_DIVISOR);
>>>
>>
>>Hi Venki,
>>
>>you know I don't have code under my hands now to take a look on
>>(i'm in office now) -- but if I recall correctly it's not a typo :) By
>>writting (1) instead of (16) you make counter decrease faster by
>>16 times (if to compare with what we have now). So we could get
>>counter underflow before we've finish calibration. That is was the
>>original intension of this 16 divisor being used. Did you test this
>>patch?
>>
>
> Typo part I mentioned was this
>
>>> -#ifdef CONFG_X86_64
>>> +#ifdef CONFIG_X86_64
>
> Looks like intention was to use DIVISOR of 1 for x86_64 and divisor of
> 16 for 32 bit kernel. Once I changed that, there was still one place in
> 64 bit code that was assuming divisor of 16. The rest of the patch
> changes that part.
>
> Yes. Tested this patch on couple of systems here and worked fine.
>
> Thanks,
> Venki

Ah, it was CONFG who is guilty :) This is true indeed, thanks!

But here is not the same

       apic_write(APIC_TDCR,
               (tmp_value & ~(APIC_TDR_DIV_1 | APIC_TDR_DIV_TMBASE)) |
-               APIC_TDR_DIV_16);
+               APIC_TDR_DIV);

On x86_64 it will be 1 now but we've used 16 for a long time in purpose
to slowdown processor's bus CLKs from APIC point of view. So I don't think
it's good idea to change it now. If we start using divisor 1 today -- it would
work for some time 'till processor bus raised to the some speed when
we'll get counter underflow before calibration finished.

So could you please split the patch into two:
1) Plain CONFG typo fix (which is completely Ack'ed)
2) APIC divisor patch (which I'm not sure if we've to touch)

Thanks,
Cyrill
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/