| Message ID | 161599468311.1752480.3413807875011702040.stgit@bahia.lan |
|---|---|
| State | New |
| Headers | show |
| Series | target/ppc/kvm: Cache timebase frequency | expand |
On 3/17/21 4:24 PM, Greg Kurz wrote: > Each vCPU core exposes its timebase frequency in the DT. When running > under KVM, this means parsing /proc/cpuinfo in order to get the timebase > frequency of the host CPU. > > The parsing appears to slow down the boot quite a bit with higher number > of cores: > > # of cores seconds spent in spapr_dt_cpus() > 8 0.550122 > 16 1.342375 > 32 2.850316 > 64 5.922505 > 96 9.109224 > 128 12.245504 > 256 24.957236 > 384 37.389113 > > The timebase frequency of the host CPU is identical for all > cores and it is an invariant for the VM lifetime. Cache it > instead of doing the same expensive parsing again and again. > > With this patch applied: > > 384 0.518382 > > Signed-off-by: Greg Kurz <groug@kaod.org> > --- > target/ppc/kvm.c | 11 +++++++++-- > 1 file changed, 9 insertions(+), 2 deletions(-) > > diff --git a/target/ppc/kvm.c b/target/ppc/kvm.c > index 298c1f882c67..9ad3dae29132 100644 > --- a/target/ppc/kvm.c > +++ b/target/ppc/kvm.c > @@ -1819,7 +1819,13 @@ uint32_t kvmppc_get_tbfreq(void) > { > char line[512]; > char *ns; > - uint32_t retval = NANOSECONDS_PER_SECOND; > + static uint32_t retval = -1; Please document why in the code ... > + > + if (retval != -1) { > + return retval; > + } > + > + retval = NANOSECONDS_PER_SECOND; > > if (read_cpuinfo("timebase", line, sizeof(line))) { > return retval; > @@ -1832,7 +1838,8 @@ uint32_t kvmppc_get_tbfreq(void) > > ns++; > > - return atoi(ns); > + retval = atoi(ns); > + return retval; > } ... or alternatively use self-documented code: -- >8 -- diff --git a/target/ppc/kvm.c b/target/ppc/kvm.c index 298c1f882c6..2b2fe5d8148 100644 --- a/target/ppc/kvm.c +++ b/target/ppc/kvm.c @@ -1815,7 +1815,7 @@ static int read_cpuinfo(const char *field, char *value, int len) return ret; } -uint32_t kvmppc_get_tbfreq(void) +static uint32_t kvmppc_get_tbfreq_procfs(void) { char line[512]; char *ns; @@ -1835,6 +1835,17 @@ uint32_t kvmppc_get_tbfreq(void) return atoi(ns); } +uint32_t kvmppc_get_tbfreq(void) +{ + static uint32_t cached_tbfreq; + + if (!cached_tbfreq) { + cached_tbfreq = kvmppc_get_tbfreq_procfs(); + } + + return cached_tbfreq; +} + bool kvmppc_get_host_serial(char **value) { return g_file_get_contents("/proc/device-tree/system-id", value, NULL, ---
On Wed, 17 Mar 2021 16:39:04 +0100 Philippe Mathieu-Daudé <f4bug@amsat.org> wrote: > On 3/17/21 4:24 PM, Greg Kurz wrote: > > Each vCPU core exposes its timebase frequency in the DT. When running > > under KVM, this means parsing /proc/cpuinfo in order to get the timebase > > frequency of the host CPU. > > > > The parsing appears to slow down the boot quite a bit with higher number > > of cores: > > > > # of cores seconds spent in spapr_dt_cpus() > > 8 0.550122 > > 16 1.342375 > > 32 2.850316 > > 64 5.922505 > > 96 9.109224 > > 128 12.245504 > > 256 24.957236 > > 384 37.389113 > > > > The timebase frequency of the host CPU is identical for all > > cores and it is an invariant for the VM lifetime. Cache it > > instead of doing the same expensive parsing again and again. > > > > With this patch applied: > > > > 384 0.518382 > > > > Signed-off-by: Greg Kurz <groug@kaod.org> > > --- > > target/ppc/kvm.c | 11 +++++++++-- > > 1 file changed, 9 insertions(+), 2 deletions(-) > > > > diff --git a/target/ppc/kvm.c b/target/ppc/kvm.c > > index 298c1f882c67..9ad3dae29132 100644 > > --- a/target/ppc/kvm.c > > +++ b/target/ppc/kvm.c > > @@ -1819,7 +1819,13 @@ uint32_t kvmppc_get_tbfreq(void) > > { > > char line[512]; > > char *ns; > > - uint32_t retval = NANOSECONDS_PER_SECOND; > > + static uint32_t retval = -1; > > Please document why in the code ... > Yeah, I've been lazy :) > > + > > + if (retval != -1) { > > + return retval; > > + } > > + > > + retval = NANOSECONDS_PER_SECOND; > > > > if (read_cpuinfo("timebase", line, sizeof(line))) { > > return retval; > > @@ -1832,7 +1838,8 @@ uint32_t kvmppc_get_tbfreq(void) > > > > ns++; > > > > - return atoi(ns); > > + retval = atoi(ns); > > + return retval; > > } > > ... or alternatively use self-documented code: > > -- >8 -- > diff --git a/target/ppc/kvm.c b/target/ppc/kvm.c > index 298c1f882c6..2b2fe5d8148 100644 > --- a/target/ppc/kvm.c > +++ b/target/ppc/kvm.c > @@ -1815,7 +1815,7 @@ static int read_cpuinfo(const char *field, char > *value, int len) > return ret; > } > > -uint32_t kvmppc_get_tbfreq(void) > +static uint32_t kvmppc_get_tbfreq_procfs(void) > { > char line[512]; > char *ns; > @@ -1835,6 +1835,17 @@ uint32_t kvmppc_get_tbfreq(void) > return atoi(ns); > } > > +uint32_t kvmppc_get_tbfreq(void) > +{ > + static uint32_t cached_tbfreq; > + > + if (!cached_tbfreq) { > + cached_tbfreq = kvmppc_get_tbfreq_procfs(); > + } > + > + return cached_tbfreq; > +} > + This is much nicer indeed. I'll do just that in v2. Thanks ! :) > bool kvmppc_get_host_serial(char **value) > { > return g_file_get_contents("/proc/device-tree/system-id", value, NULL, > ---
diff --git a/target/ppc/kvm.c b/target/ppc/kvm.c index 298c1f882c67..9ad3dae29132 100644 --- a/target/ppc/kvm.c +++ b/target/ppc/kvm.c @@ -1819,7 +1819,13 @@ uint32_t kvmppc_get_tbfreq(void) { char line[512]; char *ns; - uint32_t retval = NANOSECONDS_PER_SECOND; + static uint32_t retval = -1; + + if (retval != -1) { + return retval; + } + + retval = NANOSECONDS_PER_SECOND; if (read_cpuinfo("timebase", line, sizeof(line))) { return retval; @@ -1832,7 +1838,8 @@ uint32_t kvmppc_get_tbfreq(void) ns++; - return atoi(ns); + retval = atoi(ns); + return retval; } bool kvmppc_get_host_serial(char **value)
Each vCPU core exposes its timebase frequency in the DT. When running under KVM, this means parsing /proc/cpuinfo in order to get the timebase frequency of the host CPU. The parsing appears to slow down the boot quite a bit with higher number of cores: # of cores seconds spent in spapr_dt_cpus() 8 0.550122 16 1.342375 32 2.850316 64 5.922505 96 9.109224 128 12.245504 256 24.957236 384 37.389113 The timebase frequency of the host CPU is identical for all cores and it is an invariant for the VM lifetime. Cache it instead of doing the same expensive parsing again and again. With this patch applied: 384 0.518382 Signed-off-by: Greg Kurz <groug@kaod.org> --- target/ppc/kvm.c | 11 +++++++++-- 1 file changed, 9 insertions(+), 2 deletions(-)