linux-kernel - Re: [PATCH v19 007/130] x86/virt/tdx: Export SEAMCALL functions

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: <228473c4-fcda-448e-b859-008687fb052d@intel.com>
Date: Fri, 15 Mar 2024 16:53:34 -0700
From: Dave Hansen <dave.hansen@...el.com>
To: Sean Christopherson <seanjc@...gle.com>
Cc: Rick P Edgecombe <rick.p.edgecombe@...el.com>,
 "kvm@...r.kernel.org" <kvm@...r.kernel.org>,
 "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
 Kai Huang <kai.huang@...el.com>, Isaku Yamahata <isaku.yamahata@...el.com>,
 "x86@...nel.org" <x86@...nel.org>, Tina Zhang <tina.zhang@...el.com>,
 Hang Yuan <hang.yuan@...el.com>, Bo2 Chen <chen.bo@...el.com>,
 "sagis@...gle.com" <sagis@...gle.com>,
 "isaku.yamahata@...il.com" <isaku.yamahata@...il.com>,
 Erdem Aktas <erdemaktas@...gle.com>,
 "pbonzini@...hat.com" <pbonzini@...hat.com>
Subject: Re: [PATCH v19 007/130] x86/virt/tdx: Export SEAMCALL functions

On 3/15/24 12:38, Sean Christopherson wrote:
>> tdh_mem_page_remove() _should_ just be logically:
>>
>> 	* initialize tdx_module_args.  Move a few things into place on
>> 	  the stack and zero the rest.
> The "zero the rest" is what generates the fugly code.  The underlying problem is
> that the SEAMCALL assembly functions unpack _all_ registers from tdx_module_args.
> As a result, tdx_module_args needs to be zeroed to avoid loading registers with
> unitialized stack data.

It's the "zero the rest" and also the copy:

> +	if (out) {
> +		*out = *in;
> +		ret = seamcall_ret(op, out);
> +	} else
> +		ret = seamcall(op, in);

Things get a wee bit nicer if you do an out-of-line mempcy() instead of
the structure copy.  But the really fun part is that 'out' is NULL and
the compiler *SHOULD* know it.  I'm not actually sure what trips it up.

In any case, I think it ends up generating code for both sides of the
if/else including the entirely superfluous copy.

The two nested while loops (one for TDX_RND_NO_ENTROPY and the other for
TDX_ERROR_SEPT_BUSY) also don't make for great code generation.

So, sure, the generated code here could be a better.  But there's a lot
more going on here than just shuffling gunk in and out of the 'struct
tdx_module_args', and there's a _lot_ more work to do for one of these
than for a plain old kvm_hypercall*().

It might make sense to separate out the "out" functionality into and
maybe to uninline _some_ of the helper levels.  But after that, there's
not a lot of low hanging fruit.