On 24/11/2022 19:04, Tony Lu wrote:
On Thu, Nov 24, 2022 at 03:06:50PM +0100, Alexandra Winter wrote:
On 24.11.22 14:00, Alexandra Winter wrote:
[ ... ]
On Wed, Nov 23, 2022 at 11:58:30AM +0100, Jan Karcher wrote:
In the past SMC used the values of tcp_{w|r}mem to create the send
buffer and RMB. We now have our own sysctl knobs to tune them without
influencing the TCP default.
This patch removes the dependency on the TCP control by providing our
own initial values which aim for a low memory footprint
[ ... ]>>> Iiuc you are changing the default values in this a patch and
your other patch:
Default values for real_buf for send and receive:
before 0227f058aa29 ("net/smc: Unbind r/w buffer size from clcsock and make them tunable")
real_buf=net.ipv4.tcp_{w|r}mem[1]/2 send: 8k recv: 64k
Jan, you explained to me: "the minima is 16Kib. This is enforced in smc_compress_bufsize
which would move any value <= 16Kib into bucket 0 - which is 16KiB "
so send 16k recv: 64k
after 0227f058aa29 ("net/smc: Unbind r/w buffer size from clcsock and make them tunable")
real_buf=net.ipv4.tcp_{w|r}mem[1] send: 16k (16*1024) recv: 128k (131072)
after net/smc: Fix expected buffersizes and sync logic
real_buf=net.ipv4.tcp_{w|r}mem[1] send: 16k (16*1024) recv: 128k (131072)
after net/smc: Unbind smc control from tcp control
real_buf=SMC_*BUF_INIT_SIZE send: 16k (16384) recv: 64k (65536)
If my understanding is correct, then I nack this.
Defaults should be restored to the values before 0227f058aa29.
Otherwise users will notice a change in memory usage that needs to
be avoided or announced more explicitely. (and don't change them twice)
See above, I stand corrected. This does restore the default receive buffer size.
It should got to net with a Fixes: 0227f058aa29 tag.
And the descrition should clarify that this restores the default recv buffer size
(and uncouples from tcp control)
Maybe the users and the distros would thank you, if you merge them into 1 patch? >
AFAIK, updating kernel to the version of upstream will take a long time
for distros and users. Since we have sysctls, maybe we could let
applications and distros use sysctl to set their desired and default
values with documentation. It's a more common way to tweak kernel. >
To whomever it may interest. The second patch in discussion is:
https://lore.kernel.org/netdev/20221123104907.14624-1-jaka@xxxxxxxxxxxxx/
(Just in case someone stumbles on this thread without being included prior).
I decided to split those two separate in favor of the user. While the
fix corrects specific behavior this patch does provides a new
functionality and could be used without the fixed patch.
AFAIK the distros do not care how many patches they apply. We just need
to tell them to backport one more patch. Which IMO is a good trade-off
if it gives other members of the community more freedom.
Cheers,
Tony Lu
Signed-off-by: Jan Karcher <jaka@xxxxxxxxxxxxx>
Reviewed-by: Wenjia Zhang <wenjia@xxxxxxxxxxxxx>
---
Documentation/networking/smc-sysctl.rst | 4 ++--
net/smc/smc_core.h | 6 ++++--
net/smc/smc_sysctl.c | 10 ++++++----
3 files changed, 12 insertions(+), 8 deletions(-)
diff --git a/Documentation/networking/smc-sysctl.rst b/Documentation/networking/smc-sysctl.rst
index 6d8acdbe9be1..a1c634d3690a 100644
--- a/Documentation/networking/smc-sysctl.rst
+++ b/Documentation/networking/smc-sysctl.rst
@@ -44,7 +44,7 @@ smcr_testlink_time - INTEGER
wmem - INTEGER
Initial size of send buffer used by SMC sockets.
- The default value inherits from net.ipv4.tcp_wmem[1].
+ The default value aims for a small memory footprint and is set to 16KiB.
The minimum value is 16KiB and there is no hard limit for max value, but
only allowed 512KiB for SMC-R and 1MiB for SMC-D.
@@ -53,7 +53,7 @@ wmem - INTEGER
rmem - INTEGER
Initial size of receive buffer (RMB) used by SMC sockets.
- The default value inherits from net.ipv4.tcp_rmem[1].
+ The default value aims for a small memory footprint and is set to 64KiB.
The minimum value is 16KiB and there is no hard limit for max value, but
only allowed 512KiB for SMC-R and 1MiB for SMC-D.
diff --git a/net/smc/smc_core.h b/net/smc/smc_core.h
index 285f9bd8e232..67c3937f341d 100644
--- a/net/smc/smc_core.h
+++ b/net/smc/smc_core.h
@@ -206,8 +206,10 @@ struct smc_rtoken { /* address/key of remote RMB */
u32 rkey;
};
-#define SMC_BUF_MIN_SIZE 16384 /* minimum size of an RMB */
-#define SMC_RMBE_SIZES 16 /* number of distinct RMBE sizes */
+#define SMC_SNDBUF_INIT_SIZE 16384 /* initial size of send buffer */
+#define SMC_RCVBUF_INIT_SIZE 65536 /* initial size of receive buffer */
+#define SMC_BUF_MIN_SIZE 16384 /* minimum size of an RMB */
+#define SMC_RMBE_SIZES 16 /* number of distinct RMBE sizes */
There is one blank added to these lines? Why? They still don't align.
Argh good catch. I think my new workstation had a different tab spacing...
/* theoretically, the RFC states that largest size would be 512K,
* i.e. compressed 5 and thus 6 sizes (0..5), despite
* struct smc_clc_msg_accept_confirm.rmbe_size being a 4 bit value (0..15)
diff --git a/net/smc/smc_sysctl.c b/net/smc/smc_sysctl.c
index b6f79fabb9d3..a63aa79d4856 100644
--- a/net/smc/smc_sysctl.c
+++ b/net/smc/smc_sysctl.c
@@ -19,8 +19,10 @@
#include "smc_llc.h"
#include "smc_sysctl.h"
-static int min_sndbuf = SMC_BUF_MIN_SIZE;
-static int min_rcvbuf = SMC_BUF_MIN_SIZE;
+static int initial_sndbuf = SMC_SNDBUF_INIT_SIZE;
+static int initial_rcvbuf = SMC_RCVBUF_INIT_SIZE;
+static int min_sndbuf = SMC_BUF_MIN_SIZE;
+static int min_rcvbuf = SMC_BUF_MIN_SIZE;
Broken formatting
static struct ctl_table smc_table[] = {
{
@@ -88,8 +90,8 @@ int __net_init smc_sysctl_net_init(struct net *net)
net->smc.sysctl_autocorking_size = SMC_AUTOCORKING_DEFAULT_SIZE;
net->smc.sysctl_smcr_buf_type = SMCR_PHYS_CONT_BUFS;
net->smc.sysctl_smcr_testlink_time = SMC_LLC_TESTLINK_DEFAULT_TIME;
- WRITE_ONCE(net->smc.sysctl_wmem, READ_ONCE(net->ipv4.sysctl_tcp_wmem[1]));
- WRITE_ONCE(net->smc.sysctl_rmem, READ_ONCE(net->ipv4.sysctl_tcp_rmem[1]));
+ WRITE_ONCE(net->smc.sysctl_wmem, initial_sndbuf);
+ WRITE_ONCE(net->smc.sysctl_rmem, initial_rcvbuf);
Maybe we can use SMC_{SND|RCV}BUF_INIT_SIZE macro directly, instead of
new variables.
The reason i created the new variables is that min_{snd|rcv}buf also have
their own variables. I know it is not needed but thought it was cleaner.
If you have a strong opinion on using the value directly i can change it.
Please let me know if you want it changed.
Yep, it's okay for me to use variables or macros. Just let it be.
I think it's better coding style to use the macros instead of unneccessary variables.
At least the variables could be defined as const.
I just thought about that. I'm going to fix the formatting any way in a
version 2 so i try to clean up here as well.
Thank you for your comments
- Jan
Reviewed-by: Tony Lu <tonylu@xxxxxxxxxxxxxxxxx>
Cheers,
Tony Lu
- Jan
Cheers,
Tony Lu
return 0;
--
2.34.1