Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Message-ID: <dd5815fa-ec61-b7d1-c0b3-70cd0f33bbee@intel.com>
Date:   Thu, 13 Oct 2022 10:44:22 -0700
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:91.0) Gecko/20100101
 Thunderbird/91.11.0
Subject: Re: [PATCH] x86/fpu: Remove dynamic features from xcomp_bv for
 init_fpstate
Content-Language: en-US
To:     "Chang S. Bae" <chang.seok.bae@intel.com>,
        "Yao, Yuan" <yuan.yao@intel.com>,
        Dave Hansen <dave.hansen@linux.intel.com>,
        "linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>
Cc:     "x86@kernel.org" <x86@kernel.org>,
        Thomas Gleixner <tglx@linutronix.de>
References: <20221011222425.866137-1-dave.hansen@linux.intel.com>
 <BYAPR11MB3717EDEF2351C958F2C86EED95259@BYAPR11MB3717.namprd11.prod.outlook.com>
 <9b198ed3-4b2d-c857-710b-3f7115bbcf74@intel.com>
 <78f2021e-339f-9dfb-2aee-51f58ea77ec9@intel.com>
 <9be1413b-ce30-acf8-86fb-3e302dc98396@intel.com>
From:   Dave Hansen <dave.hansen@intel.com>
In-Reply-To: <9be1413b-ce30-acf8-86fb-3e302dc98396@intel.com>
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
Precedence: bulk

On 10/13/22 10:33, Chang S. Bae wrote:
>>>
>>> -       mask = fpstate->user_xfeatures;
>>> +       mask = fpstate->user_xfeatures & fpstate->xfeatures;
>>
>> I'm not sure this is quite right either.
>>
>> Doesn't kvm expect that all of the ->user_xfeatures will end up being
>> copied out?  We surely can't copy them from 'fpstate' if the feature
>> isn't there, but we can't skip them entirely, can we?
> 
> No, we can't skip them. IIUC, the code will zero out:
> 
>     /*
>      * ... The gap tracking will zero these states.
>      */
>     mask = fpstate->user_xfeatures;
> 
>     for_each_extended_xfeature(i, mask) {
>         /*
>          * If there was a feature or alignment gap, zero the space
>          * in the destination buffer.
>          */
>         if (zerofrom < xstate_offsets[i])
>             membuf_zero(&to, xstate_offsets[i] - zerofrom);
> 
>         <snip>

Ahh, good point.  A better comment for that would be:

	 * Some user_xfeatures may not be present in the fpstate.
	 * Remove those from 'mask' to zero those features in the
	 * user buffer instead of retrieving them from fpstate.