lkml.org 
[lkml]   [2021]   [May]   [29]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
From
Subject[RFC PATCH v6 31/34] dyndbg: detect repeated site recs in add_module
Date
Since the 1st patch in this series, which partitioned into structs
_ddebug & _ddebug_site while leaving lineno in _ddebug, we have had
chunks of repeating site records; every pr_debug in a function
has identical site data.

To avoid reliance on the redundant data, we need:
0. ._map field (added HEAD~1)
1. detection of repeating sites.

In ddebug_add_module(), alter the index initialization loop to recognize
the 1st record in each block of repeats, remember it in "j", and use it
to ._map all the pr_debugs in that block. To find the block starts,
we can just look for a new .function in each row.

This gives us, for each function; 1st used row, 2nd-Nth unused rows.
To actually reclaim useful memory chunks, we will have to pack the
site records up towards the header.

Heres what booting with dynamic_debug.verbose=3 will show:

1st 3 builtin modules:
- cols 3,4 row 1 - shows +1 due to previous head64
- cols 2,4 - shows repeat detection working

rcu: Hierarchical SRCU implementation.
dyndbg: add-module: head64.1 sites
dyndbg: skip header head64
dyndbg: add-module: main.6 sites
Callback from call_rcu_tasks_trace() invoked.
dyndbg: 0 0 1 1 main.run_init_process.1358
dyndbg: 1 0 2 1 main.run_init_process.1356
dyndbg: 2 0 3 1 main.run_init_process.1355
dyndbg: 3 0 4 1 main.run_init_process.1353
dyndbg: 4 4 5 5 main.initcall_blacklisted.1142
dyndbg: 5 5 6 6 main.initcall_blacklist.1103
dyndbg: 6 debug prints in module main (in 3 functions)
dyndbg: add-module: initramfs.1 sites
dyndbg: 0 0 7 7 initramfs.unpack_to_rootfs.496
dyndbg: 1 debug prints in module initramfs (in 1 functions)
dyndbg: add-module: ibs.3 sites
dyndbg: 0 0 8 8 ibs.force_ibs_eilvt_setup.926
dyndbg: 1 1 9 9 ibs.setup_ibs_ctl.897
dyndbg: 2 1 10 9 ibs.setup_ibs_ctl.890
dyndbg: 3 debug prints in module ibs (in 2 functions)

for a loaded module:
- row 0000 - is the header (function==module) - lineno is garbage
- cols 2,4 - show repeat detect
- 9 functions - over by 1, due to counting the header

dyndbg: add-module: intel_rapl_common.13 sites
dyndbg: 0 0 0 0 intel_rapl_common.intel_rapl_common.16928
dyndbg: 1 1 1 1 intel_rapl_common.rapl_remove_package.1279
dyndbg: 2 2 2 2 intel_rapl_common.rapl_detect_domains.1245
dyndbg: 3 2 3 2 intel_rapl_common.rapl_detect_domains.1242
dyndbg: 4 4 4 4 intel_rapl_common.rapl_package_register_powercap.1159
dyndbg: 5 4 5 4 intel_rapl_common.rapl_package_register_powercap.1145
dyndbg: 6 4 6 4 intel_rapl_common.rapl_package_register_powercap.1114
dyndbg: 7 4 7 4 intel_rapl_common.rapl_package_register_powercap.1108
dyndbg: 8 8 8 8 intel_rapl_common.rapl_update_domain_data.1083
dyndbg: 9 9 9 9 intel_rapl_common.rapl_check_unit_atom.824
dyndbg: 10 10 10 10 intel_rapl_common.rapl_check_unit_core.796
dyndbg: 11 11 11 11 intel_rapl_common.rapl_read_data_raw.722
dyndbg: 12 12 12 12 intel_rapl_common.contraint_to_pl.303
dyndbg: 13 debug prints in module intel_rapl_common (in 9 functions)
Signed-off-by: Jim Cromie <jim.cromie@gmail.com>
---
lib/dynamic_debug.c | 26 ++++++++++++++++++++------
1 file changed, 20 insertions(+), 6 deletions(-)

diff --git a/lib/dynamic_debug.c b/lib/dynamic_debug.c
index 524303a04462..e12538640b51 100644
--- a/lib/dynamic_debug.c
+++ b/lib/dynamic_debug.c
@@ -1073,7 +1073,7 @@ static int __ddebug_add_module(struct _ddebug *tab, struct _ddebug_site *sites,
{
struct ddebug_table *dt;
union _ddebug_header *dh = (union _ddebug_header *) &tab[0];
- int i;
+ int i, j, ct;

v3pr_info("add-module: %s.%d sites\n", modname, numdbgs);

@@ -1107,10 +1107,22 @@ static int __ddebug_add_module(struct _ddebug *tab, struct _ddebug_site *sites,
dt->ddebugs = tab;
dt->sites = sites;

- for (i = 0; i < numdbgs; i++) {
- tab[i]._back = base + i;
- tab[i]._map = base + i;
- v3pr_info(" %d %d %d %s.%s.%d\n", i, tab[i]._back, tab[i]._map,
+ /*
+ * Since sites don't have lineno (thats in struct _ddebug),
+ * all pr_debugs in a function have identical site data. So
+ * we watch for repeats on site.function, and reuse the index
+ * of the 1st site record in a repeating block.
+ */
+ for (ct = i = j = 0; i < numdbgs; i++) {
+ tab[i]._back = base + i; /* index back to header */
+
+ if (sites[i].function != sites[j].function) {
+ j = i; /* remember 1st row with new fn */
+ ct++;
+ }
+ tab[i]._map = base + j;
+
+ v3pr_info(" %d %d %d %d %s.%s.%d\n", i, j, tab[i]._back, tab[i]._map,
modname, sites[i].function, tab[i].lineno);
}

@@ -1118,7 +1130,9 @@ static int __ddebug_add_module(struct _ddebug *tab, struct _ddebug_site *sites,
list_add(&dt->link, &ddebug_tables);
mutex_unlock(&ddebug_lock);

- v2pr_info("%3u debug prints in module %s\n", numdbgs, modname);
+ v2pr_info("%3u debug prints in module %s (in %d functions)\n",
+ numdbgs, modname, ++ct);
+
return 0;
}

--
2.31.1
\
 
 \ /
  Last update: 2021-05-29 22:04    [W:2.068 / U:0.012 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site