Message ID | 186656540d3e6225abd98374e791a13d10d86fab.1560868106.git.naveen.n.rao@linux.vnet.ibm.com (mailing list archive) |
---|---|
State | Superseded |
Headers | show |
Series | powerpc/ftrace: Patch out -mprofile-kernel instructions | expand |
Context | Check | Description |
---|---|---|
snowpatch_ozlabs/apply_patch | success | Successfully applied on branch next (e610a466d16a086e321f0bd421e2fc75cff28605) |
snowpatch_ozlabs/checkpatch | warning | total: 0 errors, 0 warnings, 1 checks, 60 lines checked |
On Tue, 18 Jun 2019 20:17:04 +0530 "Naveen N. Rao" <naveen.n.rao@linux.vnet.ibm.com> wrote: > @@ -1551,7 +1551,7 @@ unsigned long ftrace_location_range(unsigned long start, unsigned long end) > key.flags = end; /* overload flags, as it is unsigned long */ > > for (pg = ftrace_pages_start; pg; pg = pg->next) { > - if (end < pg->records[0].ip || > + if (end <= pg->records[0].ip || This breaks the algorithm. "end" is inclusive. That is, if you look for a single byte, where "start" and "end" are the same, and it happens to be the first ip on the pg page, it will be skipped, and not found. -- Steve > start >= (pg->records[pg->index - 1].ip + MCOUNT_INSN_SIZE)) > continue; > rec = bsearch(&key, pg->records, pg->index,
Steven Rostedt wrote: > On Tue, 18 Jun 2019 20:17:04 +0530 > "Naveen N. Rao" <naveen.n.rao@linux.vnet.ibm.com> wrote: > >> @@ -1551,7 +1551,7 @@ unsigned long ftrace_location_range(unsigned long start, unsigned long end) >> key.flags = end; /* overload flags, as it is unsigned long */ >> >> for (pg = ftrace_pages_start; pg; pg = pg->next) { >> - if (end < pg->records[0].ip || >> + if (end <= pg->records[0].ip || > > This breaks the algorithm. "end" is inclusive. That is, if you look for > a single byte, where "start" and "end" are the same, and it happens to > be the first ip on the pg page, it will be skipped, and not found. Thanks. It looks like I should be over-riding ftrace_location() instead. I will update this patch. - Naveen
Naveen N. Rao wrote: > Steven Rostedt wrote: >> On Tue, 18 Jun 2019 20:17:04 +0530 >> "Naveen N. Rao" <naveen.n.rao@linux.vnet.ibm.com> wrote: >> >>> @@ -1551,7 +1551,7 @@ unsigned long ftrace_location_range(unsigned long start, unsigned long end) >>> key.flags = end; /* overload flags, as it is unsigned long */ >>> >>> for (pg = ftrace_pages_start; pg; pg = pg->next) { >>> - if (end < pg->records[0].ip || >>> + if (end <= pg->records[0].ip || >> >> This breaks the algorithm. "end" is inclusive. That is, if you look for >> a single byte, where "start" and "end" are the same, and it happens to >> be the first ip on the pg page, it will be skipped, and not found. > > Thanks. It looks like I should be over-riding ftrace_location() instead. > I will update this patch. I think I will have ftrace own the two instruction range, regardless of whether the preceding instruction is a 'mflr r0' or not. This simplifies things and I don't see an issue with it as of now. I will do more testing to confirm. - Naveen --- a/arch/powerpc/kernel/trace/ftrace.c +++ b/arch/powerpc/kernel/trace/ftrace.c @@ -951,6 +951,16 @@ void arch_ftrace_update_code(int command) } #ifdef CONFIG_MPROFILE_KERNEL +/* + * We consider two instructions -- 'mflr r0', 'bl _mcount' -- to be part + * of ftrace. When checking for the first instruction, we want to include + * the next instruction in the range check. + */ +unsigned long ftrace_location(unsigned long ip) +{ + return ftrace_location_range(ip, ip + MCOUNT_INSN_SIZE); +} + /* Returns 1 if we patched in the mflr */ static int __ftrace_make_call_prep(struct dyn_ftrace *rec) { diff --git a/kernel/trace/ftrace.c b/kernel/trace/ftrace.c index 21d8e201ee80..122e2bb4a739 100644 --- a/kernel/trace/ftrace.c +++ b/kernel/trace/ftrace.c @@ -1573,7 +1573,7 @@ unsigned long ftrace_location_range(unsigned long start, unsigned long end) * the function tracer. It checks the ftrace internal tables to * determine if the address belongs or not. */ -unsigned long ftrace_location(unsigned long ip) +unsigned long __weak ftrace_location(unsigned long ip) { return ftrace_location_range(ip, ip); }
On Tue, 18 Jun 2019 23:53:11 +0530 "Naveen N. Rao" <naveen.n.rao@linux.vnet.ibm.com> wrote: > Naveen N. Rao wrote: > > Steven Rostedt wrote: > >> On Tue, 18 Jun 2019 20:17:04 +0530 > >> "Naveen N. Rao" <naveen.n.rao@linux.vnet.ibm.com> wrote: > >> > >>> @@ -1551,7 +1551,7 @@ unsigned long ftrace_location_range(unsigned long start, unsigned long end) > >>> key.flags = end; /* overload flags, as it is unsigned long */ > >>> > >>> for (pg = ftrace_pages_start; pg; pg = pg->next) { > >>> - if (end < pg->records[0].ip || > >>> + if (end <= pg->records[0].ip || > >> > >> This breaks the algorithm. "end" is inclusive. That is, if you look for > >> a single byte, where "start" and "end" are the same, and it happens to > >> be the first ip on the pg page, it will be skipped, and not found. > > > > Thanks. It looks like I should be over-riding ftrace_location() instead. > > I will update this patch. > > I think I will have ftrace own the two instruction range, regardless of > whether the preceding instruction is a 'mflr r0' or not. This simplifies > things and I don't see an issue with it as of now. I will do more > testing to confirm. > > - Naveen > > > --- a/arch/powerpc/kernel/trace/ftrace.c > +++ b/arch/powerpc/kernel/trace/ftrace.c > @@ -951,6 +951,16 @@ void arch_ftrace_update_code(int command) > } > > #ifdef CONFIG_MPROFILE_KERNEL > +/* > + * We consider two instructions -- 'mflr r0', 'bl _mcount' -- to be part > + * of ftrace. When checking for the first instruction, we want to include > + * the next instruction in the range check. > + */ > +unsigned long ftrace_location(unsigned long ip) > +{ > + return ftrace_location_range(ip, ip + MCOUNT_INSN_SIZE); > +} > + > /* Returns 1 if we patched in the mflr */ > static int __ftrace_make_call_prep(struct dyn_ftrace *rec) > { > diff --git a/kernel/trace/ftrace.c b/kernel/trace/ftrace.c > index 21d8e201ee80..122e2bb4a739 100644 > --- a/kernel/trace/ftrace.c > +++ b/kernel/trace/ftrace.c > @@ -1573,7 +1573,7 @@ unsigned long ftrace_location_range(unsigned long start, unsigned long end) > * the function tracer. It checks the ftrace internal tables to > * determine if the address belongs or not. > */ > -unsigned long ftrace_location(unsigned long ip) > +unsigned long __weak ftrace_location(unsigned long ip) > { > return ftrace_location_range(ip, ip); > } Actually, instead of making this a weak function, let's do this: In include/ftrace.h: #ifndef FTRACE_IP_EXTENSION # define FTRACE_IP_EXTENSION 0 #endif In arch/powerpc/include/asm/ftrace.h #define FTRACE_IP_EXTENSION MCOUNT_INSN_SIZE Then we can just have: unsigned long ftrace_location(unsigned long ip) { return ftrace_location_range(ip, ip + FTRACE_IP_EXTENSION); } -- Steve
Steven Rostedt wrote: > On Tue, 18 Jun 2019 23:53:11 +0530 > "Naveen N. Rao" <naveen.n.rao@linux.vnet.ibm.com> wrote: > >> Naveen N. Rao wrote: >> > Steven Rostedt wrote: >> >> On Tue, 18 Jun 2019 20:17:04 +0530 >> >> "Naveen N. Rao" <naveen.n.rao@linux.vnet.ibm.com> wrote: >> >> >> >>> @@ -1551,7 +1551,7 @@ unsigned long ftrace_location_range(unsigned long start, unsigned long end) >> >>> key.flags = end; /* overload flags, as it is unsigned long */ >> >>> >> >>> for (pg = ftrace_pages_start; pg; pg = pg->next) { >> >>> - if (end < pg->records[0].ip || >> >>> + if (end <= pg->records[0].ip || >> >> >> >> This breaks the algorithm. "end" is inclusive. That is, if you look for >> >> a single byte, where "start" and "end" are the same, and it happens to >> >> be the first ip on the pg page, it will be skipped, and not found. >> > >> > Thanks. It looks like I should be over-riding ftrace_location() instead. >> > I will update this patch. >> >> I think I will have ftrace own the two instruction range, regardless of >> whether the preceding instruction is a 'mflr r0' or not. This simplifies >> things and I don't see an issue with it as of now. I will do more >> testing to confirm. >> >> - Naveen >> >> >> --- a/arch/powerpc/kernel/trace/ftrace.c >> +++ b/arch/powerpc/kernel/trace/ftrace.c >> @@ -951,6 +951,16 @@ void arch_ftrace_update_code(int command) >> } >> >> #ifdef CONFIG_MPROFILE_KERNEL >> +/* >> + * We consider two instructions -- 'mflr r0', 'bl _mcount' -- to be part >> + * of ftrace. When checking for the first instruction, we want to include >> + * the next instruction in the range check. >> + */ >> +unsigned long ftrace_location(unsigned long ip) >> +{ >> + return ftrace_location_range(ip, ip + MCOUNT_INSN_SIZE); >> +} >> + >> /* Returns 1 if we patched in the mflr */ >> static int __ftrace_make_call_prep(struct dyn_ftrace *rec) >> { >> diff --git a/kernel/trace/ftrace.c b/kernel/trace/ftrace.c >> index 21d8e201ee80..122e2bb4a739 100644 >> --- a/kernel/trace/ftrace.c >> +++ b/kernel/trace/ftrace.c >> @@ -1573,7 +1573,7 @@ unsigned long ftrace_location_range(unsigned long start, unsigned long end) >> * the function tracer. It checks the ftrace internal tables to >> * determine if the address belongs or not. >> */ >> -unsigned long ftrace_location(unsigned long ip) >> +unsigned long __weak ftrace_location(unsigned long ip) >> { >> return ftrace_location_range(ip, ip); >> } > > Actually, instead of making this a weak function, let's do this: > > > In include/ftrace.h: > > #ifndef FTRACE_IP_EXTENSION > # define FTRACE_IP_EXTENSION 0 > #endif > > > In arch/powerpc/include/asm/ftrace.h > > #define FTRACE_IP_EXTENSION MCOUNT_INSN_SIZE > > > Then we can just have: > > unsigned long ftrace_location(unsigned long ip) > { > return ftrace_location_range(ip, ip + FTRACE_IP_EXTENSION); > } Thanks, that's indeed nice. I hope you don't mind me adding your SOB for that. - Naveen
On Wed, 19 Jun 2019 13:26:37 +0530 "Naveen N. Rao" <naveen.n.rao@linux.vnet.ibm.com> wrote: > > In include/ftrace.h: > > > > #ifndef FTRACE_IP_EXTENSION > > # define FTRACE_IP_EXTENSION 0 > > #endif > > > > > > In arch/powerpc/include/asm/ftrace.h > > > > #define FTRACE_IP_EXTENSION MCOUNT_INSN_SIZE > > > > > > Then we can just have: > > > > unsigned long ftrace_location(unsigned long ip) > > { > > return ftrace_location_range(ip, ip + FTRACE_IP_EXTENSION); > > } > > Thanks, that's indeed nice. I hope you don't mind me adding your SOB for > that. Actually, it's best not to put a SOB by anyone other than yourself. It actually has legal meaning. In this case, please add: Suggested-by: Steven Rostedt (VMware) <rostedt@goodmis.org> Thanks! -- Steve
diff --git a/arch/powerpc/kernel/trace/ftrace.c b/arch/powerpc/kernel/trace/ftrace.c index 5e2b29808af1..b84046e43207 100644 --- a/arch/powerpc/kernel/trace/ftrace.c +++ b/arch/powerpc/kernel/trace/ftrace.c @@ -951,6 +951,37 @@ void arch_ftrace_update_code(int command) } #ifdef CONFIG_MPROFILE_KERNEL +/* + * We need to check if the previous instruction is a 'nop' or 'mflr r0'. + * If so, we will patch those subsequently and that instruction must be + * considered as part of ftrace. + */ +int ftrace_cmp_recs(const void *a, const void *b) +{ + const struct dyn_ftrace *key = a; + const struct dyn_ftrace *rec = b; + unsigned int op; + + if (key->flags < rec->ip - MCOUNT_INSN_SIZE) + return -1; + if (key->ip >= rec->ip + MCOUNT_INSN_SIZE) + return 1; + + if (key->flags > rec->ip) + return 0; + + /* check the previous instruction */ + if (probe_kernel_read(&op, (void *)rec->ip - MCOUNT_INSN_SIZE, + sizeof(op))) + /* assume we own it */ + return 0; + + if (op != PPC_INST_NOP && op != PPC_INST_MFLR) + return -1; + + return 0; +} + /* Returns 1 if we patched in the mflr */ static int __ftrace_make_call_prep(struct dyn_ftrace *rec) { diff --git a/include/linux/ftrace.h b/include/linux/ftrace.h index fa653a561da5..9941987bf510 100644 --- a/include/linux/ftrace.h +++ b/include/linux/ftrace.h @@ -435,6 +435,7 @@ struct dyn_ftrace *ftrace_rec_iter_record(struct ftrace_rec_iter *iter); int ftrace_update_record(struct dyn_ftrace *rec, int enable); int ftrace_test_record(struct dyn_ftrace *rec, int enable); void ftrace_run_stop_machine(int command); +int ftrace_cmp_recs(const void *a, const void *b); unsigned long ftrace_location(unsigned long ip); unsigned long ftrace_location_range(unsigned long start, unsigned long end); unsigned long ftrace_get_addr_new(struct dyn_ftrace *rec); diff --git a/kernel/trace/ftrace.c b/kernel/trace/ftrace.c index 21d8e201ee80..b5c61db0b452 100644 --- a/kernel/trace/ftrace.c +++ b/kernel/trace/ftrace.c @@ -1517,7 +1517,7 @@ ftrace_ops_test(struct ftrace_ops *ops, unsigned long ip, void *regs) } -static int ftrace_cmp_recs(const void *a, const void *b) +int __weak ftrace_cmp_recs(const void *a, const void *b) { const struct dyn_ftrace *key = a; const struct dyn_ftrace *rec = b; @@ -1551,7 +1551,7 @@ unsigned long ftrace_location_range(unsigned long start, unsigned long end) key.flags = end; /* overload flags, as it is unsigned long */ for (pg = ftrace_pages_start; pg; pg = pg->next) { - if (end < pg->records[0].ip || + if (end <= pg->records[0].ip || start >= (pg->records[pg->index - 1].ip + MCOUNT_INSN_SIZE)) continue; rec = bsearch(&key, pg->records, pg->index,
Now that we are patching the preceding 'mflr r0' instruction with -mprofile-kernel, we need to update ftrace_location[_range]() to recognise that as being part of ftrace. To do this, we make a small change to ftrace_location_range() and convert ftrace_cmp_recs() into a weak function. We implement a custom version of ftrace_cmp_recs() which looks at the instruction preceding the branch to _mcount() and marks that instruction as belonging to ftrace if it is a 'nop' or 'mflr r0'. Signed-off-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com> --- arch/powerpc/kernel/trace/ftrace.c | 31 ++++++++++++++++++++++++++++++ include/linux/ftrace.h | 1 + kernel/trace/ftrace.c | 4 ++-- 3 files changed, 34 insertions(+), 2 deletions(-)