Re: [PATCH v2] tools/power turbostat: Fix RAPL summary collection on AMD processors

[Chen Yu <yu.c.chen@intel.com>]

On Tue, Apr 20, 2021 at 10:42:09AM -0400, Calvin Walton wrote:
> On Tue, 2021-04-20 at 22:37 +0800, Chen Yu wrote:
> > On Tue, Apr 20, 2021 at 09:28:06AM -0400, Calvin Walton wrote:
> > > This patch has the same issue I noticed with the initial revision
> > > of
> > > Terry's patch - the idx_to_offset function returns type int (32-bit
> > > signed), but MSR_PKG_ENERGY_STAT is greater than INT_MAX (or
> > > rather,
> > > would be interpreted as a negative number)
> > > 
> > > The end result is, as far as I can tell, that it hits the if
> > > (offset <
> > > 0) check in update_msr_sum() resulting in the timer callback for
> > > updating the stat in the background when long durations are used to
> > > not
> > > happen.
> > > 
> > > For short durations it still works fine since the background update
> > > isn't used.
> > > 
> > Ah, got it, nice catch. How about an incremental patch based on Bas'
> > one
> > to fix this 'overflow' issue? Would converting offset_to_idx(),
> > idx_to_offset() and
> > update_msr_sum() to use off_t instead of int be enough? Do you or
> > Terry have interest
> > to cook that patch? For Terry's version, I'm not sure if spliting
> > the code into different CPU vendor would benefit in the future,
> > except
> > that we would have plenty of new MSRs to be introduced in the future.
> 
> Yes, I believe updating the offset_to_idx(), idx_to_offset(), and
> update_msr_sum() functions is sufficient. I can do the incremental
> patch for that this evening if nobody beats me to it :)
>
Calvin, could you please take a look at the following version if it is suitible?

From b2e63fe4f02e17289414b4f61237da822df115fb Mon Sep 17 00:00:00 2001
From: Calvin Walton <calvin.walton@kepstin.ca>
Date: Fri, 23 Apr 2021 17:32:13 +0800
Subject: [PATCH 3/5] tools/power turbostat: Fix offset overflow issue in index
 converting

The idx_to_offset() function returns type int (32-bit signed), but
MSR_PKG_ENERGY_STAT is greater than INT_MAX (or rather, would be
interpreted as a negative number). The end result is that it hits
the if (offset < 0) check in update_msr_sum() resulting in the timer
callback for updating the stat in the background when long durations
are used to not happen. The similar issue exists in offset_to_idx()
and update_msr_sum().

This patch fixes this issue by converting the 'int' type to 'off_t'
accordingly.

Fixes: 9972d5d84d76 ("tools/power turbostat: Enable accumulate RAPL display")
Signed-off-by: Chen Yu <yu.c.chen@intel.com>
---
 tools/power/x86/turbostat/turbostat.c | 10 +++++-----
 1 file changed, 5 insertions(+), 5 deletions(-)

diff --git a/tools/power/x86/turbostat/turbostat.c b/tools/power/x86/turbostat/turbostat.c
index a211264b57fd..77557122b292 100644
--- a/tools/power/x86/turbostat/turbostat.c
+++ b/tools/power/x86/turbostat/turbostat.c
@@ -296,9 +296,9 @@ struct msr_sum_array {
 /* The percpu MSR sum array.*/
 struct msr_sum_array *per_cpu_msr_sum;
 
-int idx_to_offset(int idx)
+off_t idx_to_offset(int idx)
 {
-	int offset;
+	off_t offset;
 
 	switch (idx) {
 	case IDX_PKG_ENERGY:
@@ -328,7 +328,7 @@ int idx_to_offset(int idx)
 	return offset;
 }
 
-int offset_to_idx(int offset)
+int offset_to_idx(off_t offset)
 {
 	int idx;
 
@@ -3338,7 +3338,7 @@ static int update_msr_sum(struct thread_data *t, struct core_data *c, struct pkg
 
 	for (i = IDX_PKG_ENERGY; i < IDX_COUNT; i++) {
 		unsigned long long msr_cur, msr_last;
-		int offset;
+		off_t offset;
 
 		if (!idx_valid(i))
 			continue;
@@ -3347,7 +3347,7 @@ static int update_msr_sum(struct thread_data *t, struct core_data *c, struct pkg
 			continue;
 		ret = get_msr(cpu, offset, &msr_cur);
 		if (ret) {
-			fprintf(outf, "Can not update msr(0x%x)\n", offset);
+			fprintf(outf, "Can not update msr(0x%llx)\n", (long long int)offset);
 			continue;
 		}

thanks,
Chenyu 
-- 
2.25.1

 
> -- 
> Calvin Walton <calvin.walton@kepstin.ca>
>