Description of problem: [HVM][SVM] Do not delay ExtInt event injection if RFLAGS.IF==0. AMD-V does not require the delaying of interrupt injection if the guest IF_FLAG disallows interrupts. Remove the code in svm_intr_assist() that was checking the guest IF in eflags. This patch fixes the problem with HVM Windows guests, with observing a very slow timer countdown on the initial boot menu, when there is more than one boot option. Kbd response in this same Windows boot menu is also acceptably responsive with this patch. Version-Release number of selected component (if applicable): xen-3.0.3-testing the patch is from xen-unstable.hg repository changeset 11679 the patch is also attached in the below info field. How reproducible: always. Steps to Reproduce: 1. start up HVM windows XP SP2 guest with >1 boot menu options 2. observe the timer on the boot menu Actual results: 1. boot menu timer will display approximately 1 sec per 10 realtime seconds 2. kdb will be not very responsive Expected results: 1. timer will display approximately 1 sec per 1 realtime seconds 2. kdb will be responsive Additional info: This patch is a safe patch to apply. The removed code was tested on 32bit, 64bit and 32bit-pae hypervisors. Signed-off-by: Tom Woller <thomas.woller>=20 diff -r 9977b8785570 -r a949bd6ceb85 xen/arch/x86/hvm/svm/intr.c --- a/xen/arch/x86/hvm/svm/intr.c Fri Sep 29 18:12:15 2006 +++ b/xen/arch/x86/hvm/svm/intr.c Sat Sep 30 09:49:08 2006 @@ -74,7 +74,6 @@ int intr_type = APIC_DM_EXTINT; int intr_vector = -1; int re_injecting = 0; - unsigned long rflags; ASSERT(vmcb); @@ -87,14 +86,6 @@ re_injecting = 1; } - /* Guest's interrputs masked? */ - rflags = vmcb->rflags; - if (irq_masked(rflags)) { - HVM_DBG_LOG(DBG_LEVEL_1, "Guest IRQs masked: rflags: %lx", rflags); - /* bail out, we won't be injecting an interrupt this time */ - return; - } - /* Previous interrupt still pending? */ if (vmcb->vintr.fields.irq) { // printk("Re-injecting IRQ from Vintr\n");
Created attachment 137601 [details] export of c/s 11679 from xen-unstable.hg repository
*** Bug 209002 has been marked as a duplicate of this bug. ***
Please hold of on applying this patch as the Xen team is seeing some problems on Linux. There is a workaround whihc is being testing and the bugzilla will updated in a few days.
Created attachment 138706 [details] first patch to resolve issue, applies to 3.0.3 base.
Created attachment 138708 [details] 2/2 patch to resolve issue, applies to 3.0.3.
The 2 attached patches have been tested at XenSources and internally at AMD. These patches have been incorporated into the xen enterprise code base, but have not been pushed into xen-unstable.hg to-date. the patches resolve the windows timer issue(s) but have not seen any negative affects for other HVM guests.
Please test tomorrow's milestone beta and let us know if the issue persists.
AMD has the latest test build (2.6.18-1.2736) but would like to know exactly what is fixed that makes Red Hat feel that the issue might go away on it's own. That's statement by itself is frankly confusing. Please commit the two patches that are attched to this bugzilla so that we can test milestone 7 "with" these patches.
Are these patches merged into -unstable yet?
Yes, both VINTR intercept signal and Delay ExtInt Injection patches are in xen-unstable.
This request was evaluated by Red Hat Product Management for inclusion in a Red Hat Enterprise Linux release. Product Management has requested further review of this request by Red Hat Engineering. This request is not yet committed for inclusion in release.
Bhavana will be posting the patches to RHKL shortly.
QE ack for RHEL5B2
in kernel-2.6.18-1.2737.el5