| Message ID | 20230809155438.22470-8-rf@opensource.cirrus.com (mailing list archive) |
|---|---|
| State | New |
| Delegated to: | Brendan Higgins |
| Headers | show |
| Series | kunit: Add dynamically-extending log | expand |
On Wed, Aug 9, 2023 at 10:54 AM Richard Fitzgerald <rf@opensource.cirrus.com> wrote: > > It's wasteful to call vsnprintf() only to figure out the length of the > string. The string might fit in the available buffer space but we have to > vsnprintf() again to do that. > > Instead, attempt to format the string to the available buffer at the same > time as finding the string length. Only if the string didn't fit the > available space is it necessary to extend the log and format the string > again to a new fragment buffer. > > Signed-off-by: Richard Fitzgerald <rf@opensource.cirrus.com> Hello! I find this version slightly more confusing but I realize that the extra vsnprintf() call is unnecessary. So I am ok with this version except for one question below. I am curious what David Gow thinks and it would also be good to have another set of eyes here. My testing of this showed no problems. Thanks! -Rae > --- > lib/kunit/test.c | 33 +++++++++++++++++---------------- > 1 file changed, 17 insertions(+), 16 deletions(-) > > diff --git a/lib/kunit/test.c b/lib/kunit/test.c > index 28d0048d6171..230ec5e9824f 100644 > --- a/lib/kunit/test.c > +++ b/lib/kunit/test.c > @@ -196,11 +196,21 @@ void kunit_log_append(struct list_head *log, const char *fmt, ...) > if (!log) > return; > > - /* Evaluate length of line to add to log */ > + frag = list_last_entry(log, struct kunit_log_frag, list); > + log_len = strlen(frag->buf); > + len_left = sizeof(frag->buf) - log_len - 1; > + > + /* Attempt to print formatted line to current fragment */ > va_start(args, fmt); > - len = vsnprintf(NULL, 0, fmt, args) + 1; > + len = vsnprintf(frag->buf + log_len, len_left, fmt, args) + 1; > va_end(args); > > + if (len <= len_left) > + goto out_newline; > + > + /* Abandon the truncated result of vsnprintf */ > + frag->buf[log_len] = '\0'; > + > if (len > sizeof(frag->buf) - 1) { > va_start(args, fmt); > tmp = kvasprintf(GFP_KERNEL, fmt, args); > @@ -212,24 +222,15 @@ void kunit_log_append(struct list_head *log, const char *fmt, ...) > return; > } > > - frag = list_last_entry(log, struct kunit_log_frag, list); > - log_len = strlen(frag->buf); > - len_left = sizeof(frag->buf) - log_len - 1; > - > - if (len > len_left) { > - frag = kunit_log_extend(log); > - if (!frag) > - return; > - > - len_left = sizeof(frag->buf) - 1; > - log_len = 0; > - } > + frag = kunit_log_extend(log); Are we meant to be using the remainder of the last fragment to its capacity or just start again with a new fragment and leave some of the last fragment empty? I worry we abandoned using the last fragment or is that the intention? Let me know if I understand this correctly. > + if (!frag) > + return; > > /* Print formatted line to the log */ > va_start(args, fmt); > - vsnprintf(frag->buf + log_len, min(len, len_left), fmt, args); > + vsnprintf(frag->buf, sizeof(frag->buf) - 1, fmt, args); > va_end(args); > - > +out_newline: > /* Add newline to end of log if not already present. */ > kunit_log_newline(frag); > } > -- > 2.30.2 >
On Wed, 9 Aug 2023 at 23:54, Richard Fitzgerald <rf@opensource.cirrus.com> wrote: > > It's wasteful to call vsnprintf() only to figure out the length of the > string. The string might fit in the available buffer space but we have to > vsnprintf() again to do that. > > Instead, attempt to format the string to the available buffer at the same > time as finding the string length. Only if the string didn't fit the > available space is it necessary to extend the log and format the string > again to a new fragment buffer. > > Signed-off-by: Richard Fitzgerald <rf@opensource.cirrus.com> > --- This looks good. The only case I'm not totally convinced about is the last one, where the message doesn't fit in the current log fragment, but is not a whole fragment itself. I'm not sure if the added efficiency of not doing a second vsprintf() is worth the memory fragmentation of always starting a new fragment: the worst case where a log message is just over half the length of frag->buf would result in every message being in its own fragment (which would not necessarily have a consistent size), and memory use would be ~doubled. But I could accept it either way: until there's a real-world example which strongly benefits from either implementation, let's go with whatever we have working. Reviewed-by: David Gow <davidgow@google.com> Cheers, -- David > lib/kunit/test.c | 33 +++++++++++++++++---------------- > 1 file changed, 17 insertions(+), 16 deletions(-) > > diff --git a/lib/kunit/test.c b/lib/kunit/test.c > index 28d0048d6171..230ec5e9824f 100644 > --- a/lib/kunit/test.c > +++ b/lib/kunit/test.c > @@ -196,11 +196,21 @@ void kunit_log_append(struct list_head *log, const char *fmt, ...) > if (!log) > return; > > - /* Evaluate length of line to add to log */ > + frag = list_last_entry(log, struct kunit_log_frag, list); > + log_len = strlen(frag->buf); > + len_left = sizeof(frag->buf) - log_len - 1; > + > + /* Attempt to print formatted line to current fragment */ > va_start(args, fmt); > - len = vsnprintf(NULL, 0, fmt, args) + 1; > + len = vsnprintf(frag->buf + log_len, len_left, fmt, args) + 1; > va_end(args); > > + if (len <= len_left) > + goto out_newline; > + > + /* Abandon the truncated result of vsnprintf */ > + frag->buf[log_len] = '\0'; > + > if (len > sizeof(frag->buf) - 1) { > va_start(args, fmt); > tmp = kvasprintf(GFP_KERNEL, fmt, args); > @@ -212,24 +222,15 @@ void kunit_log_append(struct list_head *log, const char *fmt, ...) > return; > } > > - frag = list_last_entry(log, struct kunit_log_frag, list); > - log_len = strlen(frag->buf); > - len_left = sizeof(frag->buf) - log_len - 1; > - > - if (len > len_left) { > - frag = kunit_log_extend(log); > - if (!frag) > - return; > - > - len_left = sizeof(frag->buf) - 1; > - log_len = 0; > - } > + frag = kunit_log_extend(log); > + if (!frag) > + return; > > /* Print formatted line to the log */ > va_start(args, fmt); > - vsnprintf(frag->buf + log_len, min(len, len_left), fmt, args); > + vsnprintf(frag->buf, sizeof(frag->buf) - 1, fmt, args); > va_end(args); > - > +out_newline: > /* Add newline to end of log if not already present. */ > kunit_log_newline(frag); > } > -- > 2.30.2 >
diff --git a/lib/kunit/test.c b/lib/kunit/test.c index 28d0048d6171..230ec5e9824f 100644 --- a/lib/kunit/test.c +++ b/lib/kunit/test.c @@ -196,11 +196,21 @@ void kunit_log_append(struct list_head *log, const char *fmt, ...) if (!log) return; - /* Evaluate length of line to add to log */ + frag = list_last_entry(log, struct kunit_log_frag, list); + log_len = strlen(frag->buf); + len_left = sizeof(frag->buf) - log_len - 1; + + /* Attempt to print formatted line to current fragment */ va_start(args, fmt); - len = vsnprintf(NULL, 0, fmt, args) + 1; + len = vsnprintf(frag->buf + log_len, len_left, fmt, args) + 1; va_end(args); + if (len <= len_left) + goto out_newline; + + /* Abandon the truncated result of vsnprintf */ + frag->buf[log_len] = '\0'; + if (len > sizeof(frag->buf) - 1) { va_start(args, fmt); tmp = kvasprintf(GFP_KERNEL, fmt, args); @@ -212,24 +222,15 @@ void kunit_log_append(struct list_head *log, const char *fmt, ...) return; } - frag = list_last_entry(log, struct kunit_log_frag, list); - log_len = strlen(frag->buf); - len_left = sizeof(frag->buf) - log_len - 1; - - if (len > len_left) { - frag = kunit_log_extend(log); - if (!frag) - return; - - len_left = sizeof(frag->buf) - 1; - log_len = 0; - } + frag = kunit_log_extend(log); + if (!frag) + return; /* Print formatted line to the log */ va_start(args, fmt); - vsnprintf(frag->buf + log_len, min(len, len_left), fmt, args); + vsnprintf(frag->buf, sizeof(frag->buf) - 1, fmt, args); va_end(args); - +out_newline: /* Add newline to end of log if not already present. */ kunit_log_newline(frag); }
It's wasteful to call vsnprintf() only to figure out the length of the string. The string might fit in the available buffer space but we have to vsnprintf() again to do that. Instead, attempt to format the string to the available buffer at the same time as finding the string length. Only if the string didn't fit the available space is it necessary to extend the log and format the string again to a new fragment buffer. Signed-off-by: Richard Fitzgerald <rf@opensource.cirrus.com> --- lib/kunit/test.c | 33 +++++++++++++++++---------------- 1 file changed, 17 insertions(+), 16 deletions(-)