[DTrace-devel] [PATCH v2 5/5] Add -xcpu support for general providers
Kris Van Hees
kvanhees at kvh-deb-bpf.us.oracle.com
Tue Jan 23 17:06:19 UTC 2024
On Thu, Dec 21, 2023 at 09:46:53PM -0500, eugene.loh at oracle.com wrote:
>
> Add code to dt_cg_tramp_prologue() to filter on the desired cpu.
> Then call the new dt_cg_tramp_prologue_cpu() function, which
> assumes a cpu has already been selected, if necessary.
>
> Providers that handle -xcpu specially can bypass that filtering
> by calling the new dt_cg_tramp_prologue_cpu() function.
>
> The new function simply calls
> dt_cg_tramp_prologue_act(pcb, DT_ACTIVITY_ACTIVE). The _act
> function is called directly by only one provider: dtrace.
>
> Signed-off-by: Eugene Loh <eugene.loh at oracle.com>
Reviewed-by: Kris Van Hees <kris.van.hees at oracle.com>
> ---
> libdtrace/dt_cg.c | 28 ++++-
> libdtrace/dt_cg.h | 1 +
> libdtrace/dt_prov_cpc.c | 4 +-
> libdtrace/dt_prov_profile.c | 4 +-
> test/unittest/options/tst.cpu-syscall.sh | 151 +++++++++++++++++++++++
> 5 files changed, 181 insertions(+), 7 deletions(-)
> create mode 100755 test/unittest/options/tst.cpu-syscall.sh
>
> diff --git a/libdtrace/dt_cg.c b/libdtrace/dt_cg.c
> index 7ae16db8..c8e80abb 100644
> --- a/libdtrace/dt_cg.c
> +++ b/libdtrace/dt_cg.c
> @@ -356,10 +356,32 @@ dt_cg_tramp_prologue_act(dt_pcb_t *pcb, dt_activity_t act)
> }
> }
>
> +void
> +dt_cg_tramp_prologue_cpu(dt_pcb_t *pcb) {
> + dt_cg_tramp_prologue_act(pcb, DT_ACTIVITY_ACTIVE);
> +}
> +
> void
> dt_cg_tramp_prologue(dt_pcb_t *pcb)
> {
> - dt_cg_tramp_prologue_act(pcb, DT_ACTIVITY_ACTIVE);
> + dtrace_hdl_t *dtp = pcb->pcb_hdl;
> +
> + /* Check if we are on the specified CPU (if any). */
> + if (dtp->dt_options[DTRACEOPT_CPU] != DTRACEOPT_UNSET) {
> + dt_irlist_t *dlp = &pcb->pcb_ir;
> +
> + emit(dlp, BPF_MOV_REG(BPF_REG_8, BPF_REG_1));
> +
> + emit(dlp, BPF_CALL_HELPER(BPF_FUNC_get_smp_processor_id));
> + emit(dlp, BPF_BRANCH_IMM(BPF_JNE, BPF_REG_0,
> + dtp->dt_options[DTRACEOPT_CPU],
> + pcb->pcb_exitlbl));
> +
> + emit(dlp, BPF_MOV_REG(BPF_REG_1, BPF_REG_8));
> + }
> +
> + /* Call the rest of the prologue code generation. */
> + dt_cg_tramp_prologue_cpu(pcb);
> }
>
> /*
> @@ -506,8 +528,8 @@ dt_cg_tramp_copy_args_from_regs(dt_pcb_t *pcb, int called)
> * So put the PC in both arg0 and arg1, test the PC, and then zero out
> * either arg0 or arg1, as apropriate.
> *
> - * The caller must ensure that %r7 and %r8 contain the values set by
> - * the dt_cg_tramp_prologue*() functions.
> + * The caller must ensure that %r7 and %r8 contain the values set by the
> + * dt_cg_tramp_prologue*() functions.
> */
> void
> dt_cg_tramp_copy_pc_from_regs(dt_pcb_t *pcb)
> diff --git a/libdtrace/dt_cg.h b/libdtrace/dt_cg.h
> index 95bf507f..50dd256b 100644
> --- a/libdtrace/dt_cg.h
> +++ b/libdtrace/dt_cg.h
> @@ -21,6 +21,7 @@ extern void dt_cg(dt_pcb_t *, dt_node_t *);
> extern void dt_cg_xsetx(dt_irlist_t *, dt_ident_t *, uint_t, int, uint64_t);
> extern dt_irnode_t *dt_cg_node_alloc(uint_t, struct bpf_insn);
> extern void dt_cg_tramp_prologue_act(dt_pcb_t *pcb, dt_activity_t act);
> +extern void dt_cg_tramp_prologue_cpu(dt_pcb_t *pcb);
> extern void dt_cg_tramp_prologue(dt_pcb_t *pcb);
> extern void dt_cg_tramp_clear_regs(dt_pcb_t *pcb);
> extern void dt_cg_tramp_copy_regs(dt_pcb_t *pcb);
> diff --git a/libdtrace/dt_prov_cpc.c b/libdtrace/dt_prov_cpc.c
> index 9e0f5542..93968950 100644
> --- a/libdtrace/dt_prov_cpc.c
> +++ b/libdtrace/dt_prov_cpc.c
> @@ -395,10 +395,10 @@ static int trampoline(dt_pcb_t *pcb, uint_t exitlbl)
> int i;
> dt_irlist_t *dlp = &pcb->pcb_ir;
>
> - dt_cg_tramp_prologue(pcb);
> + dt_cg_tramp_prologue_cpu(pcb);
>
> /*
> - * After the dt_cg_tramp_prologue() call, we have:
> + * After the dt_cg_tramp_prologue_cpu() call, we have:
> * // (%r7 = dctx->mst)
> * // (%r8 = dctx->ctx)
> */
> diff --git a/libdtrace/dt_prov_profile.c b/libdtrace/dt_prov_profile.c
> index f3f3bf23..967ca8c2 100644
> --- a/libdtrace/dt_prov_profile.c
> +++ b/libdtrace/dt_prov_profile.c
> @@ -220,10 +220,10 @@ static int trampoline(dt_pcb_t *pcb, uint_t exitlbl)
> int i;
> dt_irlist_t *dlp = &pcb->pcb_ir;
>
> - dt_cg_tramp_prologue(pcb);
> + dt_cg_tramp_prologue_cpu(pcb);
>
> /*
> - * After the dt_cg_tramp_prologue() call, we have:
> + * After the dt_cg_tramp_prologue_cpu() call, we have:
> * // (%r7 = dctx->mst)
> * // (%r8 = dctx->ctx)
> */
> diff --git a/test/unittest/options/tst.cpu-syscall.sh b/test/unittest/options/tst.cpu-syscall.sh
> new file mode 100755
> index 00000000..8713e9d4
> --- /dev/null
> +++ b/test/unittest/options/tst.cpu-syscall.sh
> @@ -0,0 +1,151 @@
> +#!/bin/bash
> +#
> +# Oracle Linux DTrace.
> +# Copyright (c) 2023, Oracle and/or its affiliates. All rights reserved.
> +# Licensed under the Universal Permissive License v 1.0 as shown at
> +# http://oss.oracle.com/licenses/upl.
> +#
> +
> +dtrace=$1
> +
> +DIRNAME="$tmpdir/cpu-syscall.$$.$RANDOM"
> +mkdir -p $DIRNAME
> +cd $DIRNAME
> +
> +#
> +# Make trigger code and compile it.
> +#
> +
> +cat << EOF > main.c
> +#define _GNU_SOURCE
> +#include <sched.h>
> +
> +#include <stdio.h>
> +#include <stdlib.h>
> +#include <unistd.h>
> +
> +#include <sys/types.h>
> +#include <sys/stat.h>
> +#include <fcntl.h>
> +
> +#include <errno.h>
> +
> +int ncpus;
> +cpu_set_t *mask;
> +size_t size;
> +
> +/*
> + * Grow the CPU mask as needed.
> + *
> + * Other ways of determining the number of CPUs available on the system:
> + * - inspecting the contents of /proc/cpuinfo
> + * - using sysconf(3) to obtain _SC_NPROCESSORS_CONF and _SC_NPROCESSORS_ONLN
> + * - inspecting the list of CPU directories under /sys/devices/system/cpu/
> + */
> +int grow_mask() {
> + ncpus = 1024;
> + while ((mask = CPU_ALLOC(ncpus)) != NULL &&
> + sched_getaffinity(0, CPU_ALLOC_SIZE(ncpus), mask) != 0 &&
> + errno == EINVAL) {
> + CPU_FREE(mask);
> + errno = 0;
> + ncpus *= 2;
> + }
> + if (mask == NULL || (errno != 0 && errno != EINVAL))
> + return 1;
> + size = CPU_ALLOC_SIZE(ncpus);
> + return 0;
> +}
> +
> +int bind_to_cpu(size_t cpu) {
> + CPU_ZERO_S(size, mask);
> + CPU_SET_S(cpu, size, mask);
> + if (sched_setaffinity(0, size, mask) != 0)
> + return 1;
> + return 0;
> +}
> +
> +int main(int c, char **v) {
> + int i, fd = open("/dev/null", O_WRONLY);
> +
> + if (fd == -1)
> + return 1;
> +
> + if (grow_mask() == -1)
> + return 1;
> +
> + /* Loop over CPUs (in argv[]). */
> + for (i = 1; i < c; i++) {
> + /* Bind to CPU. */
> + if (bind_to_cpu(atol(v[i])) == -1)
> + return 1;
> +
> + /* Call write() from CPU. */
> + write(fd, &i, sizeof(i));
> + }
> +
> + close(fd);
> + CPU_FREE(mask);
> +
> + return 0;
> +}
> +EOF
> +
> +gcc main.c
> +if [ $? -ne 0 ]; then
> + echo ERROR compilation failed
> + exit 1
> +fi
> +
> +#
> +# Get CPU list and form expected-results file.
> +#
> +
> +cpulist=`awk '/^processor[ ]: [0-9]*$/ { print $3 }' /proc/cpuinfo`
> +echo $cpulist
> +
> +echo > expect.txt
> +for cpu in $cpulist; do
> + echo $cpu >> expect.txt
> +done
> +
> +#
> +# Run DTrace (without -xcpu). Check that all CPUs appear.
> +#
> +
> +$dtrace $dt_flags \
> + -qn 'syscall::write:entry { printf("%d\n", cpu); }' \
> + -c "./a.out $cpulist" | sort | uniq > actual.txt
> +if [ $? -ne 0 ]; then
> + echo ERROR dtrace failed
> + exit 1
> +fi
> +
> +if ! diff -q expect.txt actual.txt > /dev/null; then
> + echo ERROR did not see expected CPUs in baseline case
> + echo expect $cpulist
> + echo actual:
> + cat actual.txt
> + exit 1
> +fi
> +
> +#
> +# Run DTrace (with -xcpu). Check that only the specified CPU appears.
> +#
> +
> +nerr=0
> +for cpu0 in $cpulist; do
> + $dtrace $dt_flags -xcpu=$cpu0 \
> + -qn 'syscall::write:entry { printf("%d\n", cpu); }' \
> + -c "./a.out $cpulist" | sort | uniq > actual.txt
> + echo > expect.txt
> + echo $cpu0 >> expect.txt
> +
> + if ! diff -q expect.txt actual.txt > /dev/null; then
> + echo ERROR did not see expected CPUs in cpu $cpu0 case, saw:
> + cat actual.txt
> + nerr=$(($nerr + 1))
> + fi
> +done
> +
> +exit $nerr
> --
> 2.18.4
>
>
More information about the DTrace-devel
mailing list