linux-kernel - Re: [PATCH v1 1/3] x86/traps: Move DR7_RESET

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]

Message-ID: <00358cf3-e59a-4a5f-8cfd-06a174da72b4@zytor.com>
Date: Fri, 13 Jun 2025 10:58:34 -0700
From: Xin Li <xin@...or.com>
To: Sean Christopherson <seanjc@...gle.com>
Cc: linux-kernel@...r.kernel.org, kvm@...r.kernel.org, tglx@...utronix.de,
        mingo@...hat.com, bp@...en8.de, dave.hansen@...ux.intel.com,
        x86@...nel.org, hpa@...or.com, pbonzini@...hat.com,
        peterz@...radead.org, brgerst@...il.com, tony.luck@...el.com,
        fenghuay@...dia.com
Subject: Re: [PATCH v1 1/3] x86/traps: Move DR7_RESET_VALUE to
 <uapi/asm/debugreg.h>

On 6/13/2025 7:18 AM, Sean Christopherson wrote:
> On Fri, Jun 13, 2025, Xin Li (Intel) wrote:
>> Move DR7_RESET_VALUE to <uapi/asm/debugreg.h> to prepare to write DR7
>> with DR7_RESET_VALUE at boot time.
> 
> Alternatively, what about dropping DR7_RESET_VALUE,  moving KVM's DR6 and DR7
> #defines out of arch/x86/include/asm/kvm_host.h, and then using DR7_FIXED_1?

We definitely should do it, I see quite a few architectural definitions
are in KVM only headers (the native FRED patches needed to reuse the 
event types that were previously VMX-specific and moved them out of KVM
headers).

Because there is an UAPI header, we probably don't want to remove
definitions from it?  Ofc there is a non-UAPI header we can move into.

> 
> Arguably, that'd be an improvement for 2 of the 3 uses of DR7_RESET_VALUE in SEV
> code:
> 
> 	/* Early non-zero writes to DR7 are not supported */
> 	if (!data && (val & ~DR7_RESET_VALUE))
> 		return ES_UNSUPPORTED;
> 
> vs.
> 
> 	/* Early non-zero writes to DR7 are not supported */
> 	if (!data && (val & ~DR7_FIXED_1))
> 		return ES_UNSUPPORTED;
> 
> And in vc_handle_dr7_read():
> 
> 	if (data)
> 		*reg = data->dr7;
> 	else
> 		*reg = DR7_RESET_VALUE;
> 
> vs.
> 
> 	if (data)
> 		*reg = data->dr7;
> 	else
> 		*reg = DR7_FIXED_1;
> 
> In both of those cases, it isn't the RESET value that's interesting, it's that
> architecturally bit 10 is fixed to '1'.
> 
> I haven't looked at the kernel code, but I suspect DR6_ACTIVE_LOW, DR6_VOLATILE,
> and/or DR6_FIXED_1 could also come in handy.

I can find time to take a look after the bug-fixing patches.

Thanks!
     Xin