| Message ID | 20211209182624.2316453-3-aaronlewis@google.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | Add additional testing for routing L2 exceptions | expand |
On Thu, Dec 09, 2021, Aaron Lewis wrote: > Setting the stack to PAGE_SIZE - 1 sets the stack to being 1-byte LOL, nice. It's also pointless because any access larger than a byte that occurs without first adjusting the stack will explode. > aligned, which fails in usermode with alignment checks enabled (ie: with > flags cr0.am set and eflags.ac set). This was causing an #AC in > usermode.c when preparing to call the callback in run_in_user(). > Aligning the stack fixes the issue. > > For the purposes of fixing the #AC in usermode.c the stack has to be > aligned to at least an 8-byte boundary. Setting it to a page aligned > boundary ensures any stack alignment requirements are met as x86_64 > stacks generally want to be 16-byte aligned. > > Signed-off-by: Aaron Lewis <aaronlewis@google.com> > --- > x86/vmx.c | 4 ++-- > 1 file changed, 2 insertions(+), 2 deletions(-) > > diff --git a/x86/vmx.c b/x86/vmx.c > index 6dc9a55..44f4861 100644 > --- a/x86/vmx.c > +++ b/x86/vmx.c > @@ -1242,7 +1242,7 @@ static void init_vmcs_guest(void) > vmcs_write(GUEST_CR4, guest_cr4); > vmcs_write(GUEST_SYSENTER_CS, KERNEL_CS); > vmcs_write(GUEST_SYSENTER_ESP, > - (u64)(guest_syscall_stack + PAGE_SIZE - 1)); > + (u64)(guest_syscall_stack + PAGE_SIZE)); > vmcs_write(GUEST_SYSENTER_EIP, (u64)(&entry_sysenter)); > vmcs_write(GUEST_DR7, 0); > vmcs_write(GUEST_EFER, rdmsr(MSR_EFER)); > @@ -1292,7 +1292,7 @@ static void init_vmcs_guest(void) > > /* 26.3.1.4 */ > vmcs_write(GUEST_RIP, (u64)(&guest_entry)); > - vmcs_write(GUEST_RSP, (u64)(guest_stack + PAGE_SIZE - 1)); > + vmcs_write(GUEST_RSP, (u64)(guest_stack + PAGE_SIZE)); Rather than do the math here, which looks arbitrary at first blance, what about adjusting on allocation and renaming the variables accordingly? E.g. diff --git a/x86/vmx.c b/x86/vmx.c index 6dc9a55..bc8be77 100644 --- a/x86/vmx.c +++ b/x86/vmx.c @@ -1388,8 +1388,8 @@ void init_vmx(u64 *vmxon_region) static void alloc_bsp_vmx_pages(void) { bsp_vmxon_region = alloc_page(); - guest_stack = alloc_page(); - guest_syscall_stack = alloc_page(); + guest_stack_top = alloc_page() + PAGE_SIZE; + guest_syscall_stack_top = alloc_page() + PAGE_SIZE; vmcs_root = alloc_page(); } > vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED); > > /* 26.3.1.5 */ > -- > 2.34.1.173.g76aa8bc2d0-goog >
diff --git a/x86/vmx.c b/x86/vmx.c index 6dc9a55..44f4861 100644 --- a/x86/vmx.c +++ b/x86/vmx.c @@ -1242,7 +1242,7 @@ static void init_vmcs_guest(void) vmcs_write(GUEST_CR4, guest_cr4); vmcs_write(GUEST_SYSENTER_CS, KERNEL_CS); vmcs_write(GUEST_SYSENTER_ESP, - (u64)(guest_syscall_stack + PAGE_SIZE - 1)); + (u64)(guest_syscall_stack + PAGE_SIZE)); vmcs_write(GUEST_SYSENTER_EIP, (u64)(&entry_sysenter)); vmcs_write(GUEST_DR7, 0); vmcs_write(GUEST_EFER, rdmsr(MSR_EFER)); @@ -1292,7 +1292,7 @@ static void init_vmcs_guest(void) /* 26.3.1.4 */ vmcs_write(GUEST_RIP, (u64)(&guest_entry)); - vmcs_write(GUEST_RSP, (u64)(guest_stack + PAGE_SIZE - 1)); + vmcs_write(GUEST_RSP, (u64)(guest_stack + PAGE_SIZE)); vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED); /* 26.3.1.5 */
Setting the stack to PAGE_SIZE - 1 sets the stack to being 1-byte aligned, which fails in usermode with alignment checks enabled (ie: with flags cr0.am set and eflags.ac set). This was causing an #AC in usermode.c when preparing to call the callback in run_in_user(). Aligning the stack fixes the issue. For the purposes of fixing the #AC in usermode.c the stack has to be aligned to at least an 8-byte boundary. Setting it to a page aligned boundary ensures any stack alignment requirements are met as x86_64 stacks generally want to be 16-byte aligned. Signed-off-by: Aaron Lewis <aaronlewis@google.com> --- x86/vmx.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)