Skip site navigation (1)Skip section navigation (2) 
Date:      Sun, 1 Sep 2019 02:03:48 +0530
From:      shreyank amartya <shreyankfbsd@gmail.com>
To:        freebsd-drivers@freebsd.org
Cc:        freebsd-hackers@freebsd.org, avg@freebsd.org, mmacy@freebsd.org
Subject:   hwpmc NMI/cpuxx ... going to debugger
Message-ID:  <CAD9jf8Ddzyiu8cTup790UGaM4Y9d17Cu%2BTPwXnJFXFWisz7Eew@mail.gmail.com>
index | next in thread | raw e-mail 

Hi,

I'm trying to profile a java application using hwpmc. While running the
profiler I run into these errors:

NMI ISA 30, EISA ff
NMI/cpu123 ... going to debugger
NMI ISA 30, EISA ff
NMI/cpu58 ... going to debugger
NMI ISA 20, EISA ff
NMI/cpu59 ... going to debugger

Due to this, pmclog contains very few samples and sometimes none at all. I
enabled hwpmc kernel traces and with that it seems that cpu which is unable
to catch the pmc interrupt hits true for this condition in
amd_intr(hwpmc_amd.c):

                if ((pm = pac->pc_amdpmcs[i].phw_pmc) == NULL ||
                    !PMC_IS_SAMPLING_MODE(PMC_TO_MODE(pm))) {
                        isnull++;
                        continue;
                }

                if (!AMD_PMC_HAS_OVERFLOWED(i)) {
                        ovrflw++;
                        continue;
                }

These are some logs

Aug 31 20:01:11 amd kernel: cpu245 MDP:INT:1: cpu=245 tf=0xfffffe00015cbf30
um=0
Aug 31 20:01:11 amd kernel: cpu245 MDP:INT:2: retval=1 isnull=0 ovrflw=0
Aug 31 20:01:11 amd kernel: cpu245 MDP:INT:1: cpu=245 tf=0xfffffe00015cbf30
um=1
Aug 31 20:01:11 amd kernel: cpu245 MDP:INT:2: retval=1 isnull=0 ovrflw=0
Aug 31 20:01:11 amd kernel: cpu245 MDP:INT:1: cpu=245 tf=0xfffffe00015cbf30
um=0
Aug 31 20:01:11 amd kernel: cpu245 MDP:INT:2: retval=0 isnull=16 ovrflw=0
Aug 31 20:01:11 amd kernel: NMI ISA 3c, EISA ff
Aug 31 20:01:11 amd kernel: NMI/cpu245 ... going to debugger
Aug 31 20:01:11 amd kernel: cpu236 MDP:INT:1: cpu=236 tf=0xfffffe0001595f30
um=0
Aug 31 20:01:11 amd kernel: cpu236 MDP:INT:2: retval=1 isnull=0 ovrflw=0
Aug 31 20:01:11 amd kernel: cpu252 MDP:INT:1: cpu=252 tf=0xfffffe00015f5f30
um=0
Aug 31 20:01:11 amd kernel: cpu252 MDP:INT:2: retval=1 isnull=0 ovrflw=0

Other observations:
If I enable all the md=* flags, I do not see this issue. When i run the
profiler for a C++ application, I do not encounter this problem.
There is no panic observed.
Any idea on what could be causing this? or pointers on how to debug this
further?

Thanks
Shreyank
home | help

Want to link to this message? Use this
URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?CAD9jf8Ddzyiu8cTup790UGaM4Y9d17Cu%2BTPwXnJFXFWisz7Eew>