|
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] Re: [Xen-devel] [PATCH v3 3/3] xentrace: Implement cpu mask range parsing of human values (-c).
On Fri, Apr 3, 2015 at 8:34 PM, Konrad Rzeszutek Wilk <konrad@xxxxxxxxxx> wrote:
>> > + if ( !map )
>> > + {
>> > + fprintf(stderr, "Out of memory!\n");
>> > + return EXIT_FAILURE;
>> > + }
>> > + c = c_old = totaldigits = 0;
>> > + s = arg;
>> > + do {
>> > + in_range = 0;
>> > + a = b = 0;
>> > + /* The buflen can become zero in the '} while(..) below. */
>> > + while ( buflen )
>> > + {
>> > + c_old = c;
>> > + c = *s++;
>> > + buflen--;
>> > +
>> > + if ( isspace(c) )
>> > + continue;
>> > +
>> > + if ( totaldigits && c && isspace(c_old) )
>> > + {
>> > + fprintf(stderr, "No embedded whitespaces allowed in:
>> > %s\n", arg);
>> > + goto err_out;
>> > + }
>>
>> Why are we even bothering to handle spaces here? Who would put
>> xentrace -c " 0-3,5" /tmp/foo
>> and why would we want to accept that input?
>
> This checks that and exits out. We would _not_ accept this input.
It doesn't look like it to me -- it looks like this whole "c_old"
stuff is meant to tolerate (i.e., skip) spaces at the beginning of the
cpumask string. AFAIK the only way to actually do that practically
would be to add quotes with spaces, like above.
> Are you saying that before going in this function I should check
> that opts.cpu_mask_start starts with an space? But what if the string is:
> xentrace -c "0- 3,5".
>
> This check will catch it.
>
> Or should I have an check before this function that does:
>
> for ( i = 0; i < strlen(opts.cpu_mask_str); i++ )
> if (isspace(opts.cpu_mask_str[i]))
> goto err_out;
>
> ?
No, that's the opposite of what I'm saying. I'm saying we should
accept only the characters [0-9-,] in a cpumask string; the existence
of any other character -- even at the beginning -- should return an
error. If we take out the whole "c_old" and "isspace()" checking
entirely, this will fall out naturally when it hits the "!is_digit(c)"
below.
>>
>> Not handling initial whitespace lets us get rid of totaldigits and c_old
>> as well.
>>
>> > +
>> > + /* A '\0' or a ',' signal the end of a cpu# or range */
>> > + if ( c == '\0' || c == ',' )
>> > + break;
>>
>> Can c=='\0' ever be true here? Isn't that why we do all the accounting
>
> Yes.
>> w/ buflen?
>
> At the start of the loop we decrease buflen by one which means it can
> get to zero - so on next iteration we will exit. However, if we do not
> have this check, then:
>
> 993 if ( !isdigit(c) )
> 994 {
> 995 fprintf(stderr, "Only digits allowed in: %s\n", arg);
> 996 goto err_out;
> 997 }
>
> hit that and error out. We could move this check higher up, but would
> have to special case the ",", "-". And we would still have the
> conditionals on those characters.
Right. Well I think all that is unnecessarily hard to understand,
primarily because it's organized in a confusing way.
Rather than try to nitpick you into doing it the way that seems more
sensible to me, I've rewritten the patch a bit below. This time I've
actually done some basic testing with it. What do you think?
-George
xentrace: Implement cpu mask range parsing of human values (-c).
Instead of just using -c 0x<some hex value> we can
also use: -c <starting cpu>-<end cpu>, -c <cpu1>,<cpu2>, or a
combination of them, or 'all' for all cpus.
This new format can include just singular CPUs: -c <cpu1>,
or ranges without an start or end (and xentrace will figure out
the values), such as: -c -<cpu2> (which will include cpu0, cpu1,
and cpu2) or -c <cpu2>- (which will include cpu2 and up to MAX_CPUS).
That should make it easier to trace the right CPU if
using this along with 'xl vcpu-list'.
The code has been lifted from the Linux kernel, see file
lib/bitmap.c, function __bitmap_parselist.
To make the old behavior and the new function work, we check
to see if the arguments have '0x' in them. If they do
we use the old style parsing (limited to 32 CPUs). If that
does not exist we use the new parsing.
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@xxxxxxxxxx>
Signed-off-by: George Dunlap <george.dunlap@xxxxxxxxxxxxx>
---
v5:
- Simplify parse_cpu_range
- Rename "parse_cpu_mask" to parse_cpu_mask
diff --git a/tools/xentrace/xentrace.8 b/tools/xentrace/xentrace.8
index ac18e9f..7b3172b 100644
--- a/tools/xentrace/xentrace.8
+++ b/tools/xentrace/xentrace.8
@@ -36,8 +36,38 @@ all new records to the output
set the time, p, (in milliseconds) to sleep between polling the buffers
for new data.
.TP
-.B -c, --cpu-mask=c
-set bitmask of CPUs to trace. It is limited to 32-bits.
+.B -c, --cpu-mask=[\fIc\fP|\fICPU-LIST\fP|\fIall\fP]
+This can be: a hex value (of the form 0xNNNN...), or a set of cpu
+ranges as described below, or the string \fIall\fP. Hex values are limited
+to 32 bits. If not specified, the cpu-mask as set during bootup will be
+constructed. If using the \fICPU-LIST\fP it expects decimal numbers, which
+may be specified as follows:
+
+.RS 4
+.ie n .IP """0-3""" 4
+.el .IP "``0-3''" 4
+.IX Item "0-3"
+Trace only on CPUs 0 through 3
+.ie n .IP """0,2,5-7""" 4
+.el .IP "``0,2,5-7''" 4
+.IX Item "0,2,5-7"
+Trace only on CPUs 0, 2, and 5 through 7.
+.ie n .IP """-3""" 4
+.el .IP "``-3''" 4
+.IX Item "-3"
+Trace only on CPUs 0 through 3
+.ie n .IP """-3,7""" 4
+.el .IP "``-3,7''" 4
+.IX Item "-3,7"
+Trace only on CPUs 0 through 3 and 7
+.ie n .IP """3-""" 4
+.el .IP "``3-''" 4
+.IX Item "-3-"
+Trace only on CPUs 3 up to maximum numbers of CPUs the host has.
+.RE
+.Sp
+
+If using \fIall\fP it will use all of the CPUs the host has.
.TP
.B -e, --evt-mask=e
set event capture mask. If not specified the TRC_ALL will be used.
diff --git a/tools/xentrace/xentrace.c b/tools/xentrace/xentrace.c
index 8b40e88..1558496 100644
--- a/tools/xentrace/xentrace.c
+++ b/tools/xentrace/xentrace.c
@@ -23,6 +23,7 @@
#include <string.h>
#include <getopt.h>
#include <assert.h>
+#include <ctype.h>
#include <sys/poll.h>
#include <sys/statvfs.h>
@@ -52,7 +53,7 @@ typedef struct settings_st {
char *outfile;
unsigned long poll_sleep; /* milliseconds to sleep between polls */
uint32_t evt_mask;
- uint32_t cpu_mask;
+ char *cpu_mask_str;
unsigned long tbuf_size;
unsigned long disk_rsvd;
unsigned long timeout;
@@ -542,33 +543,17 @@ void print_cpu_mask(xc_cpumap_t map)
fprintf(stderr, "\n");
}
-static void set_cpu_mask(uint32_t mask)
+static int set_cpu_mask(xc_cpumap_t map)
{
- int i, ret = 0;
- xc_cpumap_t map;
-
- map = xc_cpumap_alloc(xc_handle);
- if ( !map )
- goto out;
+ int ret = xc_tbuf_set_cpu_mask(xc_handle, map);
- /*
- * If mask is set, copy the bits out of it. This still works for
- * systems with more than 32 cpus, as the shift will just shift
- * mask down to zero.
- */
- for ( i = 0; i < xc_get_cpumap_size(xc_handle); i++ )
- map[i] = (mask >> (i * 8)) & 0xff;
-
- ret = xc_tbuf_set_cpu_mask(xc_handle, map);
- if ( ret != 0 )
- goto out;
-
- print_cpu_mask(map);
- free(map);
- return;
-out:
+ if ( ret == 0 )
+ {
+ print_cpu_mask(map);
+ return 0;
+ }
PERROR("Failure to get trace buffer pointer from Xen and set the
new mask");
- exit(EXIT_FAILURE);
+ return EXIT_FAILURE;
}
/**
@@ -819,7 +804,8 @@ static void usage(void)
"Usage: xentrace [OPTION...] [output file]\n" \
"Tool to capture Xen trace buffer data\n" \
"\n" \
-" -c, --cpu-mask=c Set cpu-mask\n" \
+" -c, --cpu-mask=c Set cpu-mask, using either hex, CPU ranges, or\n" \
+" for all CPUs\n" \
" -e, --evt-mask=e Set evt-mask\n" \
" -s, --poll-sleep=p Set sleep time, p, in milliseconds between\n" \
" polling the trace buffer for new data\n" \
@@ -951,6 +937,134 @@ static int parse_evtmask(char *arg)
return 0;
}
+#define ZERO_DIGIT '0'
+
+#define is_terminator(c) ((c)=='\0' || (c)==',')
+
+static int parse_cpumask_range(const char *mask_str, xc_cpumap_t map)
+{
+ unsigned int a, b;
+ int nmaskbits;
+ char c;
+ int in_range;
+ const char *s;
+
+ nmaskbits = xc_get_max_cpus(xc_handle);
+ if ( nmaskbits <= 0 )
+ {
+ fprintf(stderr, "Failed to get max number of CPUs! rc: %d\n",
nmaskbits);
+ return EXIT_FAILURE;
+ }
+
+ c = 0;
+ s = mask_str;
+ do {
+ in_range = 0;
+ a = b = 0;
+
+ /* Process until we find a range terminator */
+ for(c=*s++; !is_terminator(c); c=*s++)
+ {
+ if ( c == '-' )
+ {
+ if ( in_range )
+ goto err_out;
+ b = 0;
+ in_range = 1;
+ continue;
+ }
+
+ if ( !isdigit(c) )
+ {
+ fprintf(stderr, "Invalid character in cpumask: %s\n",
mask_str);
+ goto err_out;
+ }
+
+ b = b * 10 + (c - ZERO_DIGIT);
+ if ( !in_range )
+ a = b;
+ }
+
+ /* Syntax: <digit>-[,] - expand to number of CPUs. */
+ if ( b == 0 && in_range )
+ b = nmaskbits-1;
+
+ if ( a > b )
+ {
+ fprintf(stderr, "Wrong order of %d and %d\n", a, b);
+ goto err_out;
+ }
+
+ if ( b >= nmaskbits )
+ {
+ fprintf(stderr, "Specified higher value then there are CPUS!\n");
+ goto err_out;
+ }
+
+ while ( a <= b )
+ {
+ xc_cpumap_setcpu(a, map);
+ a++;
+ }
+ } while ( c );
+
+ return 0;
+ err_out:
+ errno = EINVAL;
+ return EXIT_FAILURE;
+}
+
+/**
+ * Figure out which of the CPU types the user has provided - either the hex
+ * variant, the cpu-list, or 'all'. Once done set the CPU mask.
+ */
+static int parse_cpu_mask(void)
+{
+ int i, ret = EXIT_FAILURE;
+ xc_cpumap_t map;
+
+ map = xc_cpumap_alloc(xc_handle);
+ if ( !map )
+ goto out;
+
+ if ( strlen(opts.cpu_mask_str) < 1 )
+ {
+ errno = ENOSPC;
+ goto out;
+ }
+
+ ret = 0;
+ if ( strncmp("0x", opts.cpu_mask_str, 2) == 0 )
+ {
+ uint32_t v;
+
+ v = argtol(opts.cpu_mask_str, 0);
+ /*
+ * If mask is set, copy the bits out of it. This still works for
+ * systems with more than 32 cpus, as the shift will just shift
+ * mask down to zero.
+ */
+ for ( i = 0; i < sizeof(uint32_t); i++ )
+ map[i] = (v >> (i * 8)) & 0xff;
+ }
+ else if ( strcmp("all", opts.cpu_mask_str) == 0 )
+ {
+ for ( i = 0; i < xc_get_cpumap_size(xc_handle); i++ )
+ map[i] = 0xff;
+ }
+ else
+ ret = parse_cpumask_range(opts.cpu_mask_str, map);
+
+ if ( !ret )
+ ret = set_cpu_mask(map);
+ out:
+ /* We don't use them pass this point. */
+ free(map);
+ free(opts.cpu_mask_str);
+ opts.cpu_mask_str = NULL;
+ return ret;
+}
+
/* parse command line arguments */
static void parse_args(int argc, char **argv)
{
@@ -981,10 +1095,9 @@ static void parse_args(int argc, char **argv)
opts.poll_sleep = argtol(optarg, 0);
break;
- case 'c': /* set new cpu mask for filtering*/
- opts.cpu_mask = argtol(optarg, 0);
+ case 'c': /* set new cpu mask for filtering (when xch is set). */
+ opts.cpu_mask_str = strdup(optarg);
break;
-
case 'e': /* set new event mask for filtering*/
parse_evtmask(optarg);
break;
@@ -1047,7 +1160,7 @@ int main(int argc, char **argv)
opts.outfile = 0;
opts.poll_sleep = POLL_SLEEP_MILLIS;
opts.evt_mask = 0;
- opts.cpu_mask = 0;
+ opts.cpu_mask_str = NULL;
opts.disk_rsvd = 0;
opts.disable_tracing = 1;
opts.start_disabled = 0;
@@ -1065,9 +1178,11 @@ int main(int argc, char **argv)
if ( opts.evt_mask != 0 )
set_evt_mask(opts.evt_mask);
-
- if ( opts.cpu_mask != 0 )
- set_cpu_mask(opts.cpu_mask);
+ if ( opts.cpu_mask_str )
+ {
+ if ( parse_cpu_mask() )
+ exit(EXIT_FAILURE);
+ }
if ( opts.timeout != 0 )
alarm(opts.timeout);
Attachment:
xentrace-cpu-mask.diff _______________________________________________ Xen-devel mailing list Xen-devel@xxxxxxxxxxxxx http://lists.xen.org/xen-devel
|
 |
Lists.xenproject.org is hosted with RackSpace, monitoring our |