[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <7467f313-761a-42ed-ae2d-cdadfd93db90@intel.com>
Date: Tue, 17 Sep 2024 08:08:50 -0700
From: Reinette Chatre <reinette.chatre@...el.com>
To: Martin Kletzander <nert.pinx@...il.com>
CC: Fenghua Yu <fenghua.yu@...el.com>, Thomas Gleixner <tglx@...utronix.de>,
Ingo Molnar <mingo@...hat.com>, Borislav Petkov <bp@...en8.de>, Dave Hansen
<dave.hansen@...ux.intel.com>, <x86@...nel.org>, "H. Peter Anvin"
<hpa@...or.com>, <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH v2] x86/resctrl: Avoid overflow in MB settings in
bw_validate()
Hi Martin,
On 9/17/24 1:12 AM, Martin Kletzander wrote:
> On Mon, Sep 16, 2024 at 09:56:01AM -0700, Reinette Chatre wrote:
>> On 9/16/24 6:07 AM, Martin Kletzander wrote:
>>> When resctrl is mounted with the "mba_MBps" option the default (maximum)
>>> bandwidth is the maximum unsigned value for the type. However when
>>> using the same value that already exists in the schemata file it is then
>>> rounded up to the bandwidth granularity and overflows to a small number
>>> instead, making it difficult to reset memory bandwidth allocation value
>>> back to its default.
>>>
>>> Since the granularity and minimum bandwidth are not used when the
>>> software controller is used (resctrl is mounted with the "mba_MBps"),
>>> skip the rounding up as well and return early from bw_validate().
>>
>> Thank you very much for finding the issue and proposing a fix.
>>
>>>
>>> Signed-off-by: Martin Kletzander <nert.pinx@...il.com>
>>> ---
>>> Changes in v2:
>>> - actually save the value in the output parameter @data
>>>
>>> arch/x86/kernel/cpu/resctrl/ctrlmondata.c | 9 +++++++--
>>> 1 file changed, 7 insertions(+), 2 deletions(-)
>>>
>>> diff --git a/arch/x86/kernel/cpu/resctrl/ctrlmondata.c b/arch/x86/kernel/cpu/resctrl/ctrlmondata.c
>>> index 50fa1fe9a073..702b1a372e9c 100644
>>> --- a/arch/x86/kernel/cpu/resctrl/ctrlmondata.c
>>> +++ b/arch/x86/kernel/cpu/resctrl/ctrlmondata.c
>>> @@ -48,8 +48,13 @@ static bool bw_validate(char *buf, unsigned long *data, struct rdt_resource *r)
>>> return false;
>>> }
>>>
>>> - if ((bw < r->membw.min_bw || bw > r->default_ctrl) &&
>>> - !is_mba_sc(r)) {
>>> + /* Nothing else to do if software controller is enabled */
>>> + if (is_mba_sc(r)) {
>>> + *data = bw;
>>> + return true;
>>
>> While this would fix the scenario tested I do not believe this fully addresses the
>> overflow issue. As I understand the test wrote U32_MAX to the schemata file,
>> which triggered the overflow because of the rounding and is fixed by this patch. Looks like,
>> after this patch, writing "U32_MAX + 1" will still trigger the overflow.
>>
>> The overflow appears to result from some inconsistent type use and not using
>> appropriate parsing API that is able to detect overflow.
>>
>> How about something like below:
>>
>
> That makes much more sense, I have not considered changing the data type
> as I wanted to keep the changes at minimum, but your solution is even
> better. Should I leave the fix up to you or do you want me to send a v3?
Could you please try it out to ensure it works for you and then send a v3?
Thank you very much.
Reinette
Powered by blists - more mailing lists