Attachment 303657 Details for Bug 442949 – clear %fs when loading new TLS descriptors (2/2)

[patch] clear %fs when loading new TLS descriptors (2/2)

fs-descriptor-clear-2.patch (text/plain), 2.03 KB, created by Eduardo Habkost on 2008-04-24 16:10:37 UTC

(hide)

Description:

Filename:

MIME Type:

Creator: Eduardo Habkost

Created: 2008-04-24 16:10:37 UTC

Size: 2.03 KB

patch

obsolete

>commit 5ccfa2809516aa50beb9bd54c2e2d4a367c7ea6e
>Author: Eduardo Habkost <ehabkost@redhat.com>
>Date:   Mon Apr 7 15:48:00 2008 -0300
>
>    xen-64: Clear %fs on xen_load_tls()
>    
>    We need to do this, otherwise we can get a GPF on hypercall return
>    after TLS descriptor is cleared but %fs is still pointing to it.
>    
>    Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
>
>diff --git a/arch/x86/xen/enlighten.c b/arch/x86/xen/enlighten.c
>index 5151b0f..bbee2f0 100644
>--- a/arch/x86/xen/enlighten.c
>+++ b/arch/x86/xen/enlighten.c
>@@ -385,7 +385,6 @@ static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
> 
> 	xen_mc_issue(PARAVIRT_LAZY_CPU);
> 
>-#ifdef CONFIG_X86_32
> 	/*
> 	 * XXX sleazy hack: If we're being called in a lazy-cpu zone,
> 	 * it means we're in a context switch, and %gs has just been
>@@ -395,20 +394,22 @@ static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
> 	 * loaded properly.  This will go away as soon as Xen has been
> 	 * modified to not save/restore %gs for normal hypercalls.
> 	 *
>-	 * This hack is not used on x86_64, because:
>-	 * - For %gs: x86_64 gs points to KERNEL_GS_BASE (and uses it for PDA
>-	 *   references), so we must not zero %gs on x86_64
>-	 * - For %fs: x86_64 may have %fs pointing to TLS descriptors, but
>-	 *   __switch_to() reads %fs after load_TLS(). We can't clear it
>+	 * On x86_64, this hack is not used for %gs, because gs points
>+	 * to KERNEL_GS_BASE (and uses it for PDA references), so we
>+	 * must not zero %gs on x86_64
> 	 *
>-	 * FIXME: we need to be sure we don't _need_ the hack on x86_64.
>-	 * Can we get a #GP when switch back from the hypervisor if
>-	 * TLS descriptor is cleared but %fs still points to it?
>+	 * For x86_64, we need to zero %fs, otherwise we may
>+	 * get an exception between the new %fs descriptor being
>+	 * loaded and %fs being effectively cleared at __switch_to().
> 	 */
> 	if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) {
>+#ifdef CONFIG_X86_64
>+		loadsegment(fs, 0);
>+#endif
>+#ifdef CONFIG_X86_32
> 		loadsegment(gs, 0);
>-	}
> #endif
>+	}
> }
> 
> #ifdef CONFIG_X86_64

Actions: View | Diff

Attachments on bug 442949: 302798 | 303656 | 303657