From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from ozlabs.org (ozlabs.org [103.22.144.67]) (using TLSv1.2 with cipher AECDH-AES256-SHA (256/256 bits)) (No client certificate requested) by lists.ozlabs.org (Postfix) with ESMTPS id F3BE51A0D8C for ; Mon, 7 Dec 2015 21:30:47 +1100 (AEDT) In-Reply-To: <20151112054442.GD22330@fergus.ozlabs.ibm.com> To: Paul Mackerras , linuxppc-dev@ozlabs.org From: Michael Ellerman Cc: Michael Neuling , kvm-ppc@vger.kernel.org Subject: Re: powerpc/64: Include KVM guest test in all interrupt vectors Message-Id: <20151207103047.BF8561402DE@ozlabs.org> Date: Mon, 7 Dec 2015 21:30:47 +1100 (AEDT) List-Id: Linux on PowerPC Developers Mail List List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , On Thu, 2015-12-11 at 05:44:42 UTC, Paul Mackerras wrote: > Currently, if HV KVM is configured but PR KVM isn't, we don't include > a test to see whether we were interrupted in KVM guest context for the > set of interrupts which get delivered directly to the guest by hardware > if they occur in the guest. This includes things like program > interrupts. > > However, the recent bug where userspace could set the MSR for a VCPU > to have an illegal value in the TS field, and thus cause a TM Bad Thing > type of program interrupt on the hrfid that enters the guest, showed that > we can never be completely sure that these interrupts can never occur > in the guest entry/exit code. If one of these interrupts does happen > and we have HV KVM configured but not PR KVM, then we end up trying to > run the handler in the host with the MMU set to the guest MMU context, > which generally ends badly. > > Thus, for robustness it is better to have the test in every interrupt > vector, so that if some way is found to trigger some interrupt in the > guest entry/exit path, we can handle it without immediately crashing > the host. > > This means that the distinction between KVMTEST and KVMTEST_PR goes > away. Thus we delete KVMTEST_PR and associated macros and use KVMTEST > everywhere that we previously used either KVMTEST_PR or KVMTEST. It > also means that SOFTEN_TEST_HV_201 becomes the same as SOFTEN_TEST_PR, > so we deleted SOFTEN_TEST_HV_201 and use SOFTEN_TEST_PR instead. > > Signed-off-by: Paul Mackerras Applied to powerpc next, thanks. https://git.kernel.org/powerpc/c/31a40e2b052c0f2b80df7b56 cheers