From: Alistair Popple <alistair@popple.id.au>
To: Hari Bathini <hbathini@linux.vnet.ibm.com>
Cc: Ankit Kumar <ankit@linux.vnet.ibm.com>,
Kexec-ml <kexec@lists.infradead.org>,
Ananth N Mavinakayanahalli <ananth@linux.vnet.ibm.com>,
Simon Horman <horms@verge.net.au>,
Anshuman Khandual <khandual@linux.vnet.ibm.com>
Subject: Re: [PATCH] kexec-tools: ppc64: avoid adding coherent memory regions to crash memory ranges
Date: Fri, 18 Aug 2017 14:56:30 +1000 [thread overview]
Message-ID: <2943090.QOlPB5ZzCH@new-mexico> (raw)
In-Reply-To: <150297311110.25328.11468130779639120510.stgit@hbathini.in.ibm.com>
Hi,
I haven't reviewed the code in depth but skipping memory regions marked
ibm,coherent-device-memory is the right thing to do as the links to this memory
will usually go down prior during a kexec.
Also device-drivers typically put this memory into ZONE_MOVABLE which means
there will be no kernel allocations present in this memory.
Regards,
Alistair
On Thu, 17 Aug 2017 06:01:51 PM Hari Bathini wrote:
> Accelerator devices like GPU and FPGA cards contain onboard memory. This
> onboard memory is represented as a memory only NUMA node, integrating it
> with core memory subsystem. Since, the link through which these devices
> are integrated to core memory goes down after a system crash and they are
> meant for user workloads, avoid adding coherent device memory regions to
> crash memory ranges. Without this change, makedumpfile tool tries to save
> unaccessible coherent device memory regions, crashing the system.
>
> Signed-off-by: Hari Bathini <hbathini@linux.vnet.ibm.com>
> ---
> kexec/arch/ppc64/crashdump-ppc64.c | 64 +++++++++++++++++++++++++++++++++++-
> kexec/arch/ppc64/kexec-ppc64.h | 1 +
> 2 files changed, 63 insertions(+), 2 deletions(-)
>
> diff --git a/kexec/arch/ppc64/crashdump-ppc64.c b/kexec/arch/ppc64/crashdump-ppc64.c
> index 13995bf..7ea3983 100644
> --- a/kexec/arch/ppc64/crashdump-ppc64.c
> +++ b/kexec/arch/ppc64/crashdump-ppc64.c
> @@ -181,6 +181,53 @@ static int get_dyn_reconf_crash_memory_ranges(void)
> return 0;
> }
>
> +/*
> + * For a given memory node, check if it is mapped to system RAM or
> + * to onboard memory on accelerator device like GPU card or such.
> + */
> +static int is_coherent_device_mem(const char *fname)
> +{
> + char fpath[PATH_LEN];
> + char buf[32];
> + DIR *dmem;
> + FILE *file;
> + struct dirent *mentry;
> + int cnt, ret = 0;
> +
> + strcpy(fpath, fname);
> + if ((dmem = opendir(fpath)) == NULL) {
> + perror(fpath);
> + return -1;
> + }
> +
> + while ((mentry = readdir(dmem)) != NULL) {
> + if (strcmp(mentry->d_name, "compatible"))
> + continue;
> +
> + strcat(fpath, "/compatible");
> + if ((file = fopen(fpath, "r")) == NULL) {
> + perror(fpath);
> + ret = -1;
> + break;
> + }
> + if ((cnt = fread(buf, 1, 32, file)) < 0) {
> + perror(fpath);
> + fclose(file);
> + ret = -1;
> + break;
> + }
> + if (!strncmp(buf, "ibm,coherent-device-memory", 26)) {
> + ret = 1;
> + break;
> + }
> + fclose(file);
> + }
> +
> + closedir(dmem);
> + return ret;
> +}
> +
> +
> /* Reads the appropriate file and retrieves the SYSTEM RAM regions for whom to
> * create Elf headers. Keeping it separate from get_memory_ranges() as
> * requirements are different in the case of normal kexec and crashdumps.
> @@ -196,12 +243,12 @@ static int get_crash_memory_ranges(struct memory_range **range, int *ranges)
> {
>
> char device_tree[256] = "/proc/device-tree/";
> - char fname[256];
> + char fname[PATH_LEN];
> char buf[MAXBYTES];
> DIR *dir, *dmem;
> FILE *file;
> struct dirent *dentry, *mentry;
> - int n, crash_rng_len = 0;
> + int n, ret, crash_rng_len = 0;
> unsigned long long start, end;
> int page_size;
>
> @@ -240,6 +287,19 @@ static int get_crash_memory_ranges(struct memory_range **range, int *ranges)
> continue;
> strcpy(fname, device_tree);
> strcat(fname, dentry->d_name);
> +
> + ret = is_coherent_device_mem(fname);
> + if (ret == -1) {
> + closedir(dir);
> + goto err;
> + } else if (ret == 1) {
> + /*
> + * Avoid adding this memory region as it is not
> + * mapped to system RAM.
> + */
> + continue;
> + }
> +
> if ((dmem = opendir(fname)) == NULL) {
> perror(fname);
> closedir(dir);
> diff --git a/kexec/arch/ppc64/kexec-ppc64.h b/kexec/arch/ppc64/kexec-ppc64.h
> index 633ae77..434b4bf 100644
> --- a/kexec/arch/ppc64/kexec-ppc64.h
> +++ b/kexec/arch/ppc64/kexec-ppc64.h
> @@ -1,6 +1,7 @@
> #ifndef KEXEC_PPC64_H
> #define KEXEC_PPC64_H
>
> +#define PATH_LEN 256
> #define MAXBYTES 128
> #define MAX_LINE 160
> #define CORE_TYPE_ELF32 1
>
_______________________________________________
kexec mailing list
kexec@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/kexec
next prev parent reply other threads:[~2017-08-18 4:57 UTC|newest]
Thread overview: 5+ messages / expand[flat|nested] mbox.gz Atom feed top
2017-08-17 12:31 [PATCH] kexec-tools: ppc64: avoid adding coherent memory regions to crash memory ranges Hari Bathini
2017-08-18 4:56 ` Alistair Popple [this message]
2017-08-28 3:12 ` Pingfan Liu
2017-08-28 13:37 ` Simon Horman
2017-08-29 17:47 ` Hari Bathini
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=2943090.QOlPB5ZzCH@new-mexico \
--to=alistair@popple.id.au \
--cc=ananth@linux.vnet.ibm.com \
--cc=ankit@linux.vnet.ibm.com \
--cc=hbathini@linux.vnet.ibm.com \
--cc=horms@verge.net.au \
--cc=kexec@lists.infradead.org \
--cc=khandual@linux.vnet.ibm.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox