[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [PATCH 1/3] x86/EFI: Fix detection of buildid


  • To: Jan Beulich <jbeulich@xxxxxxxx>
  • From: Andrew Cooper <andrew.cooper3@xxxxxxxxxx>
  • Date: Thu, 5 Jun 2025 18:01:02 +0100
  • Autocrypt: addr=andrew.cooper3@xxxxxxxxxx; keydata= xsFNBFLhNn8BEADVhE+Hb8i0GV6mihnnr/uiQQdPF8kUoFzCOPXkf7jQ5sLYeJa0cQi6Penp VtiFYznTairnVsN5J+ujSTIb+OlMSJUWV4opS7WVNnxHbFTPYZVQ3erv7NKc2iVizCRZ2Kxn srM1oPXWRic8BIAdYOKOloF2300SL/bIpeD+x7h3w9B/qez7nOin5NzkxgFoaUeIal12pXSR Q354FKFoy6Vh96gc4VRqte3jw8mPuJQpfws+Pb+swvSf/i1q1+1I4jsRQQh2m6OTADHIqg2E ofTYAEh7R5HfPx0EXoEDMdRjOeKn8+vvkAwhviWXTHlG3R1QkbE5M/oywnZ83udJmi+lxjJ5 YhQ5IzomvJ16H0Bq+TLyVLO/VRksp1VR9HxCzItLNCS8PdpYYz5TC204ViycobYU65WMpzWe LFAGn8jSS25XIpqv0Y9k87dLbctKKA14Ifw2kq5OIVu2FuX+3i446JOa2vpCI9GcjCzi3oHV e00bzYiHMIl0FICrNJU0Kjho8pdo0m2uxkn6SYEpogAy9pnatUlO+erL4LqFUO7GXSdBRbw5 gNt25XTLdSFuZtMxkY3tq8MFss5QnjhehCVPEpE6y9ZjI4XB8ad1G4oBHVGK5LMsvg22PfMJ ISWFSHoF/B5+lHkCKWkFxZ0gZn33ju5n6/FOdEx4B8cMJt+cWwARAQABzSlBbmRyZXcgQ29v cGVyIDxhbmRyZXcuY29vcGVyM0BjaXRyaXguY29tPsLBegQTAQgAJAIbAwULCQgHAwUVCgkI CwUWAgMBAAIeAQIXgAUCWKD95wIZAQAKCRBlw/kGpdefoHbdD/9AIoR3k6fKl+RFiFpyAhvO 59ttDFI7nIAnlYngev2XUR3acFElJATHSDO0ju+hqWqAb8kVijXLops0gOfqt3VPZq9cuHlh IMDquatGLzAadfFx2eQYIYT+FYuMoPZy/aTUazmJIDVxP7L383grjIkn+7tAv+qeDfE+txL4 SAm1UHNvmdfgL2/lcmL3xRh7sub3nJilM93RWX1Pe5LBSDXO45uzCGEdst6uSlzYR/MEr+5Z JQQ32JV64zwvf/aKaagSQSQMYNX9JFgfZ3TKWC1KJQbX5ssoX/5hNLqxMcZV3TN7kU8I3kjK mPec9+1nECOjjJSO/h4P0sBZyIUGfguwzhEeGf4sMCuSEM4xjCnwiBwftR17sr0spYcOpqET ZGcAmyYcNjy6CYadNCnfR40vhhWuCfNCBzWnUW0lFoo12wb0YnzoOLjvfD6OL3JjIUJNOmJy RCsJ5IA/Iz33RhSVRmROu+TztwuThClw63g7+hoyewv7BemKyuU6FTVhjjW+XUWmS/FzknSi dAG+insr0746cTPpSkGl3KAXeWDGJzve7/SBBfyznWCMGaf8E2P1oOdIZRxHgWj0zNr1+ooF /PzgLPiCI4OMUttTlEKChgbUTQ+5o0P080JojqfXwbPAyumbaYcQNiH1/xYbJdOFSiBv9rpt TQTBLzDKXok86M7BTQRS4TZ/ARAAkgqudHsp+hd82UVkvgnlqZjzz2vyrYfz7bkPtXaGb9H4 Rfo7mQsEQavEBdWWjbga6eMnDqtu+FC+qeTGYebToxEyp2lKDSoAsvt8w82tIlP/EbmRbDVn 7bhjBlfRcFjVYw8uVDPptT0TV47vpoCVkTwcyb6OltJrvg/QzV9f07DJswuda1JH3/qvYu0p vjPnYvCq4NsqY2XSdAJ02HrdYPFtNyPEntu1n1KK+gJrstjtw7KsZ4ygXYrsm/oCBiVW/OgU g/XIlGErkrxe4vQvJyVwg6YH653YTX5hLLUEL1NS4TCo47RP+wi6y+TnuAL36UtK/uFyEuPy wwrDVcC4cIFhYSfsO0BumEI65yu7a8aHbGfq2lW251UcoU48Z27ZUUZd2Dr6O/n8poQHbaTd 6bJJSjzGGHZVbRP9UQ3lkmkmc0+XCHmj5WhwNNYjgbbmML7y0fsJT5RgvefAIFfHBg7fTY/i kBEimoUsTEQz+N4hbKwo1hULfVxDJStE4sbPhjbsPCrlXf6W9CxSyQ0qmZ2bXsLQYRj2xqd1 bpA+1o1j2N4/au1R/uSiUFjewJdT/LX1EklKDcQwpk06Af/N7VZtSfEJeRV04unbsKVXWZAk uAJyDDKN99ziC0Wz5kcPyVD1HNf8bgaqGDzrv3TfYjwqayRFcMf7xJaL9xXedMcAEQEAAcLB XwQYAQgACQUCUuE2fwIbDAAKCRBlw/kGpdefoG4XEACD1Qf/er8EA7g23HMxYWd3FXHThrVQ HgiGdk5Yh632vjOm9L4sd/GCEACVQKjsu98e8o3ysitFlznEns5EAAXEbITrgKWXDDUWGYxd pnjj2u+GkVdsOAGk0kxczX6s+VRBhpbBI2PWnOsRJgU2n10PZ3mZD4Xu9kU2IXYmuW+e5KCA vTArRUdCrAtIa1k01sPipPPw6dfxx2e5asy21YOytzxuWFfJTGnVxZZSCyLUO83sh6OZhJkk b9rxL9wPmpN/t2IPaEKoAc0FTQZS36wAMOXkBh24PQ9gaLJvfPKpNzGD8XWR5HHF0NLIJhgg 4ZlEXQ2fVp3XrtocHqhu4UZR4koCijgB8sB7Tb0GCpwK+C4UePdFLfhKyRdSXuvY3AHJd4CP 4JzW0Bzq/WXY3XMOzUTYApGQpnUpdOmuQSfpV9MQO+/jo7r6yPbxT7CwRS5dcQPzUiuHLK9i nvjREdh84qycnx0/6dDroYhp0DFv4udxuAvt1h4wGwTPRQZerSm4xaYegEFusyhbZrI0U9tJ B8WrhBLXDiYlyJT6zOV2yZFuW47VrLsjYnHwn27hmxTC/7tvG3euCklmkn9Sl9IAKFu29RSo d5bD8kMSCYsTqtTfT6W4A3qHGvIDta3ptLYpIAOD2sY3GYq2nf3Bbzx81wZK14JdDDHUX2Rs 6+ahAA==
  • Cc: Ross Lagerwall <ross.lagerwall@xxxxxxxxxx>, Roger Pau Monné <roger.pau@xxxxxxxxxx>, Marek Marczykowski-Górecki <marmarek@xxxxxxxxxxxxxxxxxxxxxx>, "Daniel P . Smith" <dpsmith@xxxxxxxxxxxxxxxxxxxx>, Xen-devel <xen-devel@xxxxxxxxxxxxxxxxxxxx>
  • Delivery-date: Thu, 05 Jun 2025 17:01:07 +0000
  • List-id: Xen developer discussion <xen-devel.lists.xenproject.org>

On 05/06/2025 2:24 pm, Jan Beulich wrote:
> On 05.06.2025 14:14, Andrew Cooper wrote:
>> On 05/06/2025 1:02 pm, Jan Beulich wrote:
>>> On 05.06.2025 13:16, Andrew Cooper wrote:
>>>> The format of the buildid is a property of the binary, not a property of 
>>>> how
>>>> it was loaded.  This fixes buildid recognition when starting xen.efi from 
>>>> it's
>>>> MB2 entrypoint.
>>>>
>>>> Suggested-by: Ross Lagerwall <ross.lagerwall@xxxxxxxxxx>
>>>> Signed-off-by: Andrew Cooper <andrew.cooper3@xxxxxxxxxx>
>>> I'll pick this up without a Fixes: tag, but I think it ought to have one. (I
>>> didn't check whether MB2 or build-id support came later, hence introducing 
>>> the
>>> issue.)
>> MB2+EFI came long before any buildid support.  If you want a fixes tag,
>> it's eee5909e9d1
> That commit talks of an earlier hack, though. And no, MB2 work came later,
> albeit still in the 4.9 dev window (commit 9180f53655245).

The "previous hack" was embedding note.o (from the livepatch test
infrastructure) back in the main xen binary.

That's still present.  For xen.gz you get an elf note.  For xen.efi, it
may be an elf note or a CodeView region, depending on the toolchain.

>
>>>> --- a/xen/common/version.c
>>>> +++ b/xen/common/version.c
>>>> @@ -203,8 +203,11 @@ void __init xen_build_init(void)
>>>>      rc = xen_build_id_check(n, sz, &build_id_p, &build_id_len);
>>>>  
>>>>  #ifdef CONFIG_X86
>>>> -    /* Alternatively we may have a CodeView record from an EFI build. */
>>>> -    if ( rc && efi_enabled(EFI_LOADER) )
>>>> +    /*
>>>> +     * xen.efi built with a new enough toolchain will have a CodeView 
>>>> record,
>>>> +     * not an ELF note.
>>>> +     */
>>>> +    if ( rc )
>>> Instead of dropping the efi_enabled(), I think you want to replace 
>>> EFI_LOADER
>>> by EFI_BOOT.
>> No - that's differently buggy.  I suppose the commit message wasn't
>> clear enough?
>>
>> We'd still have a CodeView record if we booted xen.efi using it's MB2
>> entrypoint without the EFI extensions.
> Hmm, if that's a possible mode of operation (as said in reply to Marek, I
> wasn't aware of that) - yes.

It's how I found the bug.

There's an awful lot of hackery in the patch-queue holding it together,
but in this case it's not really about MB2 or anything else; it's
xen_build_init() being incorrect in how it determines whether there's a
CodeView region or not.

>
>> This really is a property of being a PE32+ binary, and nothing to do
>> with EFI.
> Which still can be checked for without having this code path being taken
> for xen.gz, too: You could e.g. check for &efi > &_end. That's firmly an
> image property (yet I expect you're going to sigh about yet another hack).

It's all hacks, but no.

I'm amazed MISRA hasn't spotted that we've got a global `struct efi
efi;` and a label named efi, creating an alias for the object with it
out of bounds in the compiled image.  But even then, it's based on
XEN_BUILD_EFI not XEN_BUILD_PE and does not distinguish the property
that matters.

But the argument I'm going to make this this:  Why do you want a check,
even if you can find a correct one (and as said before, I cannot)?

This function is run exactly once.  We've excluded "nothing given by the
toolchain", and excluded "what the toolchain gave us was not the
expected ELF note".  The only thing left (modulo toolchain bugs) is the
CodeView region, and if it's not a valid CodeView region then we've
wasted a handful of cycles.

~Andrew



 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.