Page MenuHomeFreeBSD
Differential D30036 Diff 88373 sys/netinet/tcp_ratelimit.h

Changeset View

Standalone View

View Options

sys/netinet/tcp_ratelimit.h

Context not available.
struct tcp_hwrate_limit_table { struct tcp_hwrate_limit_table {
const struct tcp_rate_set *ptbl; /* Pointer to parent table */ const struct tcp_rate_set *ptbl; /* Pointer to parent table */
struct m_snd_tag *tag; /* Send tag if needed (chelsio) */ struct m_snd_tag *tag; /* Send tag if needed (chelsio) */
uint64_t rate; /* Rate we get in Bytes per second (Bps) */ long rate; /* Rate we get in Bytes per second (Bps) */
long using; /* Temporary -- rrs remove */
kbowlingUnsubmitted
Not Done
Inline Actions

should it be here?

kbowling: should it be here?
rrsAuthorUnsubmitted
Done
Inline Actions

I am not understanding your comment. Maybe you are referring to the comment "Temporary".. and yeah
that should probably not be there i.e. the comment should say " Total number of connections using this rate".
I will fix that.

The thing I thought, when I put it in, that it was just a nice temp for debugging. But I have found when
working with Drew on hardware pacing it is really important to have this bit of information while you
are testing and having a sysctl to get it is really good.

rrs: I am not understanding your comment. Maybe you are referring to the comment "Temporary".. and…
long rs_num_enobufs;
tuexenUnsubmitted
Not Done
Inline Actions

Why do you change it from uint64_t to long?

tuexen: Why do you change it from uint64_t to long?
rrsAuthorUnsubmitted
Done
Inline Actions

uint64_t breaks with 32 bit platforms. This code was yanked after the
ratelimit changes went in by Hans to make all platforms compile. Using
a long is a compromise, it makes it so 32 bit platforms "work" but they
will be limited to say 16Gbps (0x7fffffff x 8) since this is in bytes per second.

The other limit is in using, but 2 billion connections using a rate is a bit
more than any 32 bit platform would handle. And for the counter of
enobufs, well that should not wrap too often, and its only used to indicate
we need more pacing time.

So I think its an ok compromise though it will be better when the 32 bit
platforms all go away (and I am saying that as someone who has a stack
of 1U 32bit i386 machines) ;-)

rrs: uint64_t breaks with 32 bit platforms. This code was yanked after the ratelimit changes went in…
tuexenUnsubmitted
Not Done
Inline Actions

Then I'm fine with the changes. Go ahead and commit them.

tuexen: Then I'm fine with the changes. Go ahead and commit them.
uint32_t time_between; /* Time-Gap between packets at this rate */ uint32_t time_between; /* Time-Gap between packets at this rate */
uint32_t flags; uint32_t flags;
}; };
Context not available.