| Message ID | 20230123140233.790103-2-tim.gardner@canonical.com |
|---|---|
| State | New |
| Headers | show |
| Series | Azure: TDX enabled hyper-visors cause segfault | expand |
LGTM Acked-by: Ian May <ian.may@canonical.com> On 2023-01-23 07:02:33 , Tim Gardner wrote: > From: Dexuan Cui <decui@microsoft.com> > > BugLink: https://bugs.launchpad.net/bugs/2003714 > > glibc 2.34/2.35 (and 2.36?) had a bug (2.32 is good): > See https://sourceware.org/bugzilla/show_bug.cgi?id=28784 > > The bug has been fixed in upstream glibc: > https://sourceware.org/git/?p=glibc.git;a=commitdiff;h=c242fcce06e3102ca663b2f992611d0bda4f2668 > > However, it looks like a lot of distros haven't picked up the fix yet, > e.g. Ubuntu 22.04/22.10/23.04's glibc need pick up the glibc fix (c242fcce06e3102ca663b2f992611d0bda4f2668). > RHEL 9's glibc needs the glibc fix as well. > > Before the glibc packages in the distros are fixed, we can use this > kernel side workaround patch for now. The workaround is from Intel. > See the below for the rationale: > > x86/tdx: Virtualize CPUID leaf 0x2 > CPUID leaf 0x2 provides cache and TLB information. In TDX guest access > to the leaf causes #VE. > > Current implementation returns all zero, but it confuses some users: > some recent versions of GLIBC hit segfaults. It is a GLIBC bug, but it is > also a user-visible regression comparing to non-TDX environment. > > Kernel can generate a sensible response to the #VE to work around the > glibc segfault for now. > > The leaf is obsolete. There are leafs that provides the same > information in a structured form. See leaf 0x4 on cache info and > leaf 0x18 on TLB info. > > Generate a response that indicates that CPUID leaf 0x4 and 0x18 have to > be used instead. > > (cherry picked from commit 16218cf73491e867fd39c16c9e4b8aa926cbda68 https://github.com/dcui/tdx) > Signed-off-by: Tim Gardner <tim.gardner@canonical.com> > --- > arch/x86/coco/tdx/tdx.c | 12 ++++++++++++ > 1 file changed, 12 insertions(+) > > diff --git a/arch/x86/coco/tdx/tdx.c b/arch/x86/coco/tdx/tdx.c > index c32c7ef55249..928ca748bb26 100644 > --- a/arch/x86/coco/tdx/tdx.c > +++ b/arch/x86/coco/tdx/tdx.c > @@ -329,6 +329,18 @@ static int handle_cpuid(struct pt_regs *regs, struct ve_info *ve) > .r13 = regs->cx, > }; > > + /* > + * Work around the segfault issue in glibc 2.35 in Ubuntu 22.04. > + * See https://sourceware.org/bugzilla/show_bug.cgi?id=28784 > + * Ubuntu 22.04/22.10/23.04's glibc should pick up this glibc fix: > + * https://sourceware.org/git/?p=glibc.git;a=commitdiff;h=c242fcce06e3102ca663b2f992611d0bda4f2668 > + */ > + if (regs->ax == 2) { > + regs->ax = 0xf1ff01; > + regs->bx = regs->cx = regs->dx = 0; > + return ve_instr_len(ve); > + } > + > /* > * Only allow VMM to control range reserved for hypervisor > * communication. > -- > 2.34.1 > > > -- > kernel-team mailing list > kernel-team@lists.ubuntu.com > https://lists.ubuntu.com/mailman/listinfo/kernel-team
diff --git a/arch/x86/coco/tdx/tdx.c b/arch/x86/coco/tdx/tdx.c index c32c7ef55249..928ca748bb26 100644 --- a/arch/x86/coco/tdx/tdx.c +++ b/arch/x86/coco/tdx/tdx.c @@ -329,6 +329,18 @@ static int handle_cpuid(struct pt_regs *regs, struct ve_info *ve) .r13 = regs->cx, }; + /* + * Work around the segfault issue in glibc 2.35 in Ubuntu 22.04. + * See https://sourceware.org/bugzilla/show_bug.cgi?id=28784 + * Ubuntu 22.04/22.10/23.04's glibc should pick up this glibc fix: + * https://sourceware.org/git/?p=glibc.git;a=commitdiff;h=c242fcce06e3102ca663b2f992611d0bda4f2668 + */ + if (regs->ax == 2) { + regs->ax = 0xf1ff01; + regs->bx = regs->cx = regs->dx = 0; + return ve_instr_len(ve); + } + /* * Only allow VMM to control range reserved for hypervisor * communication.