[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] [Xen-devel] [PATCH] Fix OOS typecounting [was: Test report for Xen-3.3.0-rc4 (#18314)]
Hello, Keir Fraser wrote: We still need to decide whether to fix the second issue or disable OOS. The attached patch should fix this issue. It was an all-my-fault breakage of set_l1e atomicity. We're not decided on that just yet. If it reproed more reliably for us then I'd be more optimistic about fixing it. Perhaps I will switch to 32pae as so far I've been running 32e host. A very easy way to reproduce this bug is to set SHADOW_OOS_FIXUPS to 1 in xen/include/asm-x86/mm.h. This will reproduce very quickly the typecount corruption. Gianluca diff -r d3947223dfae xen/arch/x86/mm/shadow/multi.c --- a/xen/arch/x86/mm/shadow/multi.c Thu Aug 14 13:46:48 2008 +0100 +++ b/xen/arch/x86/mm/shadow/multi.c Thu Aug 14 10:31:12 2008 -0400 @@ -1415,6 +1415,15 @@ static int shadow_set_l1e(struct vcpu *v mfn_t new_gmfn = shadow_l1e_get_mfn(new_sl1e); #endif ASSERT(sl1e != NULL); + +#if SHADOW_OPTIMIZATIONS & SHOPT_OUT_OF_SYNC + if ( mfn_valid(new_gmfn) && mfn_oos_may_write(new_gmfn) + && ((shadow_l1e_get_flags(new_sl1e) & (_PAGE_RW|_PAGE_PRESENT)) + == (_PAGE_RW|_PAGE_PRESENT)) ) + { + oos_fixup_add(v, new_gmfn, sl1mfn, pgentry_ptr_to_slot(sl1e)); + } +#endif old_sl1e = *sl1e; @@ -1434,14 +1443,6 @@ static int shadow_set_l1e(struct vcpu *v else { shadow_vram_get_l1e(new_sl1e, sl1e, sl1mfn, d); -#if SHADOW_OPTIMIZATIONS & SHOPT_OUT_OF_SYNC - if ( mfn_valid(new_gmfn) && mfn_oos_may_write(new_gmfn) - && (shadow_l1e_get_flags(new_sl1e) & _PAGE_RW) ) - { - oos_fixup_add(v, new_gmfn, sl1mfn, pgentry_ptr_to_slot(sl1e)); - } -#endif - } } } _______________________________________________ Xen-devel mailing list Xen-devel@xxxxxxxxxxxxxxxxxxx http://lists.xensource.com/xen-devel
|
Lists.xenproject.org is hosted with RackSpace, monitoring our |