Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1752782AbYGVQF1 (ORCPT ); Tue, 22 Jul 2008 12:05:27 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S1751329AbYGVQFN (ORCPT ); Tue, 22 Jul 2008 12:05:13 -0400 Received: from lizzard.sbs.de ([194.138.37.39]:21033 "EHLO lizzard.sbs.de" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751061AbYGVQFL (ORCPT ); Tue, 22 Jul 2008 12:05:11 -0400 Message-ID: <488604F8.1040008@siemens.com> Date: Tue, 22 Jul 2008 18:04:08 +0200 From: Jan Kiszka User-Agent: Mozilla/5.0 (X11; U; Linux i686 (x86_64); de; rv:1.8.1.12) Gecko/20080226 SUSE/2.0.0.12-1.1 Thunderbird/2.0.0.12 Mnenhy/0.7.5.666 MIME-Version: 1.0 To: Mathieu Desnoyers CC: akpm@linux-foundation.org, Ingo Molnar , linux-kernel@vger.kernel.org, Peter Zijlstra , Avi Kivity , kvm@vger.kernel.org, "Feng(Eric) Liu" Subject: Re: [patch 4/4] KVM-trace port to tracepoints References: <20080717155724.897537670@polymtl.ca> <20080717160003.359557938@polymtl.ca> <487F7800.4010502@siemens.com> <20080717172853.GB29855@Krystal> In-Reply-To: <20080717172853.GB29855@Krystal> Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 5448 Lines: 134 Mathieu Desnoyers wrote: > * Jan Kiszka (jan.kiszka@siemens.com) wrote: >> Mathieu Desnoyers wrote: >>> Port/cleanup of KVM-trace to tracepoints. >>> >>> Tracepoints allow dormat instrumentation, like the kernel markers, but also >>> allows to describe the trace points in global headers so they can be easily >>> managed. They also do not use format strings. >>> >>> Anything that would involve an action (dereference a pointer, vmcs read, ...) >>> only required when tracing is placed in the probes created in kvm_trace.c >>> >>> This patch depends on the "Tracepoints" patch. >>> >>> Signed-off-by: Mathieu Desnoyers >>> CC: 'Peter Zijlstra' >>> CC: 'Feng(Eric) Liu' >>> CC: Avi Kivity >>> CC: kvm@vger.kernel.org >>> --- >>> arch/x86/kvm/vmx.c | 38 ++--- >>> arch/x86/kvm/x86.c | 43 ++---- >>> include/trace/kvm.h | 83 ++++++++++++ >>> virt/kvm/kvm_trace.c | 336 +++++++++++++++++++++++++++++++++++++++++++-------- >>> 4 files changed, 398 insertions(+), 102 deletions(-) >> Is it a specific property of KVM-trace that causes this LOC blow-up? Or >> is this a generic side-effect of tracepoints? >> >> [ Hmm, hope I didn't missed too much of the tracepoint discussion... ] >> > > This LOC blow-up is caused by the creation of one probe per > instrumentation site. So instead of placing the argument setup of > everything that goes in the trace (0 to 5 u32 arguments) in the kvm > code, it can be placed separately in a probe object, which could > eventually be a dynamically loadable module. > > The primary objective of tracepoints is to make sure the kernel code > does not become harder to read because of added instrumentation and to > provide type-checking at compile-time without needing to put format > strings into the kernel code, which, to some, looks like debugging code. > The other aspect it try to address is maintainability of trace points : > it's much easier to look at all the prototype definitions in > include/trace/*.h and to manage them (and external tracers which would > like to connect on those points) than to try to figure out in which C > files tracing statements has been hidden. We can think of it as a > standard tracing API providing a more or less stable list of kernel > tracepoints. > > So, while KVMTRACE_?D() statements suits closely kvm-trace > specificities, it's useless to impose constraints such as splitting > unsigned longs into two u32 for tracers which can support a wider > variety of data types. > > After refactoring the patch to put the probes in arch/x86/kvm, the > result is : > > arch/x86/kvm/Makefile | 1 > arch/x86/kvm/kvm_trace_probes.c | 251 ++++++++++++++++++++++++++++++++++++++++ > arch/x86/kvm/vmx.c | 38 ++---- > arch/x86/kvm/x86.c | 43 ++---- > include/asm-x86/kvm_host.h | 8 + > include/trace/kvm.h | 83 +++++++++++++ > virt/kvm/kvm_trace.c | 93 ++++++-------- > 7 files changed, 414 insertions(+), 103 deletions(-) > > So actually, is it better to have less LOC which looks like this : > > KVMTRACE_5D(CPUID, vcpu, function, > (u32)kvm_register_read(vcpu, VCPU_REGS_RAX), > (u32)kvm_register_read(vcpu, VCPU_REGS_RBX), > (u32)kvm_register_read(vcpu, VCPU_REGS_RCX), > (u32)kvm_register_read(vcpu, VCPU_REGS_RDX), handler); > > > or more LOC looking like this : > > include/trace/kvm.h: > DEFINE_TRACE(kvm_cpuid, > TPPROTO(struct kvm_vcpu *vcpu, u32 function), > TPARGS(vcpu, function)); > > arch/x86/kvm/x86.c: > trace_kvm_cpuid(vcpu, function); > > arch/x86/kvm/kvm_trace_probes.c: > static void probe_kvm_cpuid(struct kvm_vcpu *vcpu, u32 function) > { > kvm_add_trace(KVM_TRC_CPUID, vcpu, 5, > (u32 []){ function, > (u32)kvm_register_read(vcpu, VCPU_REGS_RAX), > (u32)kvm_register_read(vcpu, VCPU_REGS_RBX), > (u32)kvm_register_read(vcpu, VCPU_REGS_RCX), > (u32)kvm_register_read(vcpu, VCPU_REGS_RDX) }); > } > > int register_kvm_tracepoints(void) > { > ... > ret = register_trace_kvm_cpuid(probe_kvm_cpuid); > WARN_ON(ret); > ... > } > > void unregister_kvm_tracepoints(void) > { > ... > unregister_trace_kvm_cpuid(probe_kvm_cpuid); > ... > } > > ? > > Notice that only a single line of code is inserted to the kernel code, > while all the rest sits outsite in a separated probe module. So I think > it's very important to distinguish between LOC which impair kernel code > readability and LOC which sit in their own sandbox. That's true - as long as you don't have to add/remove/modify tracepoints. I had to do this job in the past (not for KVM). Having 1 spot in 1 file (based on generic probes) would be handier in that case than 5 spots in 3 files. But if the KVM tracepoints are considered stable in their number and structure, that shouldn't be an issue here. Jan -- Siemens AG, Corporate Technology, CT SE 2 Corporate Competence Center Embedded Linux -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/