Skip to contentSkip to conversation

x86/pvh: Call C code via the kernel virtual mapping

PATCH In review Jason Andryuk 2 participants 3 messages Apr 11, 2025 – Apr 19, 2025

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 x86/pvh: Call C code via the kernel virtual mapping

From: Ard Biesheuvel <[email protected]>

[ Upstream commit e8fbc0d9cab6c1ee6403f42c0991b0c1d5dbc092 ]

Calling C code via a different mapping than it was linked at is problematic, because the compiler assumes that RIP-relative and absolute symbol references are interchangeable. GCC in particular may use RIP-relative per-CPU variable references even when not using -fpic.

So call xen_prepare_pvh() via its kernel virtual mapping on x86_64, so that those RIP-relative references produce the correct values. This matches the pre-existing behavior for i386, which also invokes xen_prepare_pvh() via the kernel virtual mapping before invoking startup_32 with paging disabled again.

Fixes 7243b93345f7 ("xen/pvh: Bootstrap PVH guest")

Message-ID: <[email protected]>

[ Stable context update ]

---
Stable backport for 6.6 .. 5.10.

Direct cherry-pick needed context fixups, which are made here. This upstream commit was previously included in stable, but with the pre-req of b464b461d27d ("x86/pvh: Set phys_base when calling xen_prepare_pvh()"). Both were subsequently reverted as b464b461d27d caused regressions. This backport, e8fbc0d9cab6, in isolation is correct.

This fixes a regression introduced by the backport of upstream commit b4845bb6383821a9516ce30af3a27dc873e37fd4 ("x86/xen: add central hypercall functions")

b4845bb63838 adds a comparison between rip-relative xen_hypercall_amd() and kernel virtual address of xen_hypercall_amd() to determine whether to use the AMD or Intel variant. When running from the identity mapped address, the comparison always fail. The leads to calling xen_hypercall_intel(), even on AMD processors, which faults and halts boot. This affects PVH dom0 - domU doesn't seem to be affected.

This patch performs the rip-relative mapping from the kernel virtual mapping, so the values can be properly compared.

---
 arch/x86/platform/pvh/head.S | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)
arch/x86/platform/pvh/head.S +6 -1
index c4365a05ab83..fc46b4dfbd74 100644
@@ -100,7 +100,12 @@ SYM_CODE_START_LOCAL(pvh_start_xen)
 	xor %edx, %edx
 	wrmsr
 
-	call xen_prepare_pvh
+	/* Call xen_prepare_pvh() via the kernel virtual mapping */
+	leaq xen_prepare_pvh(%rip), %rax
+	subq phys_base(%rip), %rax
+	addq $__START_KERNEL_map, %rax
+	ANNOTATE_RETPOLINE_SAFE
+	call *%rax
 
 	/* startup_64 expects boot_params in %rsi. */
 	mov $_pa(pvh_bootparams), %rsi
Signature · 1 line
2.49.0
#1 lore raw

Hi Greg and Sasha,

It's been a week, so I'm following up on the status of this.

Should I use a different tag than "PATCH" when sending a manual fixup?

Thanks,
Jason

Quoted 68 lines · Jason Andryuk

From: Ard Biesheuvel <[email protected]>

[ Upstream commit e8fbc0d9cab6c1ee6403f42c0991b0c1d5dbc092 ]

Calling C code via a different mapping than it was linked at is problematic, because the compiler assumes that RIP-relative and absolute symbol references are interchangeable. GCC in particular may use RIP-relative per-CPU variable references even when not using -fpic.

So call xen_prepare_pvh() via its kernel virtual mapping on x86_64, so that those RIP-relative references produce the correct values. This matches the pre-existing behavior for i386, which also invokes xen_prepare_pvh() via the kernel virtual mapping before invoking startup_32 with paging disabled again.

Fixes 7243b93345f7 ("xen/pvh: Bootstrap PVH guest")

Message-ID: <[email protected]>

[ Stable context update ]

---
Stable backport for 6.6 .. 5.10.

Direct cherry-pick needed context fixups, which are made here. This upstream commit was previously included in stable, but with the pre-req of b464b461d27d ("x86/pvh: Set phys_base when calling xen_prepare_pvh()"). Both were subsequently reverted as b464b461d27d caused regressions. This backport, e8fbc0d9cab6, in isolation is correct.

This fixes a regression introduced by the backport of upstream commit b4845bb6383821a9516ce30af3a27dc873e37fd4 ("x86/xen: add central hypercall functions")

b4845bb63838 adds a comparison between rip-relative xen_hypercall_amd() and kernel virtual address of xen_hypercall_amd() to determine whether to use the AMD or Intel variant. When running from the identity mapped address, the comparison always fail. The leads to calling xen_hypercall_intel(), even on AMD processors, which faults and halts boot. This affects PVH dom0 - domU doesn't seem to be affected.

This patch performs the rip-relative mapping from the kernel virtual mapping, so the values can be properly compared.

---
  arch/x86/platform/pvh/head.S | 7 ++++++-
  1 file changed, 6 insertions(+), 1 deletion(-)
arch/x86/platform/pvh/head.S +6 -1
index c4365a05ab83..fc46b4dfbd74 100644
@@ -100,7 +100,12 @@ SYM_CODE_START_LOCAL(pvh_start_xen)
  	xor %edx, %edx
  	wrmsr
 
-	call xen_prepare_pvh
+	/* Call xen_prepare_pvh() via the kernel virtual mapping */
+	leaq xen_prepare_pvh(%rip), %rax
+	subq phys_base(%rip), %rax
+	addq $__START_KERNEL_map, %rax
+	ANNOTATE_RETPOLINE_SAFE
+	call *%rax
 
  	/* startup_64 expects boot_params in %rsi. */
  	mov $_pa(pvh_bootparams), %rsi
#2 lore raw
Quoted 3 lines · Jason Andryuk

Hi Greg and Sasha,

It's been a week, so I'm following up on the status of this.

Please relax, right now I have 477 "backports" to dig through for stable tree inclusion, including your 4 emails. They will be processed eventually, when we have the time to do so. Newest kernel trees are always handled first, and then we move down to older ones.

thanks,

greg k-h