Sean Christopherson <[email protected]> writes: > On Wed, Jun 03, 2026, Ackerley Tng via B4 Relay wrote: >> From: Ackerley Tng <[email protected]> >> >> The original implementation of guest_code_xsave makes a jmp to >> guest_sev_es_code in inline assembly. When code that uses guest_sev_es_code >> is removed, guest_sev_es_code will be optimized out, leading to a linking >> error since guest_code_xsave still tries to jmp to guest_sev_es_code. > > So, don't do that? >
I was commenting out function calls while working on another part of this selftest and ran into linking issues. >> Rewrite guest_code_xsave() to instead make a call, in C, to >> guest_sev_es_code(), so that usage of guest_sev_es_code() is made known to >> the compiler. >> >> This rewriting also gives a name to the xsave inline assembly, improving >> readability. >> >> Signed-off-by: Ackerley Tng <[email protected]> >> --- >> tools/testing/selftests/kvm/x86/sev_smoke_test.c | 24 >> +++++++++++++++++------- >> 1 file changed, 17 insertions(+), 7 deletions(-) >> >> diff --git a/tools/testing/selftests/kvm/x86/sev_smoke_test.c >> b/tools/testing/selftests/kvm/x86/sev_smoke_test.c >> index 1a49ee3915864..8b859adf4cf6f 100644 >> --- a/tools/testing/selftests/kvm/x86/sev_smoke_test.c >> +++ b/tools/testing/selftests/kvm/x86/sev_smoke_test.c >> @@ -80,13 +80,23 @@ static void guest_sev_code(void) >> GUEST_DONE(); >> } >> >> -/* Stash state passed via VMSA before any compiled code runs. */ > > Uh, so as the comment says, the goal is to stash state before _any_ compiled > code runs. Shoving the code into inline asm breaks that. The compiler > *probably* > won't shove anything before the first inline assembly, but there are > absolutely > no guarantees. E.g. you're subtly relying on a tail-call optimization to > avoid > any stack operations. If I force guest_sev_es_code() to be inlined, then the > prologue becomes: > > 0000000000402a10 <guest_code_xsave>: > 402a10: 48 83 ec 08 sub > $0x8,%rsp > 402a14: b8 07 00 00 00 mov > $0x7,%eax > 402a19: 31 d2 xor > %edx,%edx > 402a1b: 0f ae 27 xsave (%rdi) > 402a1e: b9 31 01 01 c0 mov > $0xc0010131,%ecx > > and we're hosed. > And omg, I thought I was running the tests!! >> -extern void guest_code_xsave(void); >> -asm("guest_code_xsave:\n" >> - "mov $" __stringify(XFEATURE_MASK_X87_AVX) ", %eax\n" >> - "xor %edx, %edx\n" >> - "xsave (%rdi)\n" >> - "jmp guest_sev_es_code"); >> +static void xsave_all_registers(void *addr) >> +{ >> + __asm__ __volatile__( >> + "mov $" __stringify(XFEATURE_MASK_X87_AVX) ", %eax\n" >> + "xor %edx, %edx\n" > > This doesn't even build. When using input and/or output params, named > registers > like eax and edx need an extra '%' to escape them, e.g. > > asm volatile("mov $" __stringify(XFEATURE_MASK_X87_AVX) ", %%eax\n\t" > "xor %%edx, %%edx\n\t" > "xsave (%0)" > : > : "r"(addr) > : "eax", "edx", "memory" > ); > My bad, I was compiling this while having the caller of this function commented out. Didn't realise that even within the same file, the inline assembly won't even be compiled if there's no caller. >> + "xsave (%0)" >> + : >> + : "r"(addr) >> + : "eax", "edx", "memory" >> + ); >> +} >> + >> >> [...snip...] >> My bad, this patch is such a goof! I still feel that it'd be nice to allow commenting out functions parts of selftests while developing other parts, but let's shelve this for now. I think to clean this up I would do something like 1. Enter the guest, GUEST_SYNC(STAGE_READY) 2. In the host, write VMSA 3. vcpu_run(), guest will perform that __asm__ block, then GUEST_DONE() 4. Host performs compare_vmsa() to verify But this relies on ucall support for SNP.
