Re: [PATCH] RDMA/cma: Make CM response timeout and # CM retries configurable

[HÃkon Bugge]

> On 23 Feb 2019, at 09:49, Leon Romanovsky <leon@xxxxxxxxxx> wrote:
> 
> On Fri, Feb 22, 2019 at 12:51:55PM -0500, Doug Ledford wrote:
>> 
>> 
>>> On Feb 22, 2019, at 12:14 PM, Steve Wise <swise@xxxxxxxxxxxxxxxxxxxxx> wrote:
>>> 
>>> 
>>> On 2/22/2019 10:36 AM, Jason Gunthorpe wrote:
>>>> On Sun, Feb 17, 2019 at 06:09:09PM +0100, HÃkon Bugge wrote:
>>>>> During certain workloads, the default CM response timeout is too
>>>>> short, leading to excessive retries. Hence, make it configurable
>>>>> through sysctl. While at it, also make number of CM retries
>>>>> configurable.
>>>>> 
>>>>> The defaults are not changed.
>>>>> 
>>>>> Signed-off-by: HÃkon Bugge <haakon.bugge@xxxxxxxxxx>
>>>>> drivers/infiniband/core/cma.c | 51 ++++++++++++++++++++++++++++++-----
>>>>> 1 file changed, 44 insertions(+), 7 deletions(-)
>>>>> 
>>>>> diff --git a/drivers/infiniband/core/cma.c b/drivers/infiniband/core/cma.c
>>>>> index c43512752b8a..ce99e1cd1029 100644
>>>>> +++ b/drivers/infiniband/core/cma.c
>>>>> @@ -43,6 +43,7 @@
>>>>> #include <linux/inetdevice.h>
>>>>> #include <linux/slab.h>
>>>>> #include <linux/module.h>
>>>>> +#include <linux/sysctl.h>
>>>>> #include <net/route.h>
>>>>> 
>>>>> #include <net/net_namespace.h>
>>>>> @@ -68,13 +69,46 @@ MODULE_AUTHOR("Sean Hefty");
>>>>> MODULE_DESCRIPTION("Generic RDMA CM Agent");
>>>>> MODULE_LICENSE("Dual BSD/GPL");
>>>>> 
>>>>> -#define CMA_CM_RESPONSE_TIMEOUT 20
>>>>> #define CMA_QUERY_CLASSPORT_INFO_TIMEOUT 3000
>>>>> -#define CMA_MAX_CM_RETRIES 15
>>>>> #define CMA_CM_MRA_SETTING (IB_CM_MRA_FLAG_DELAY | 24)
>>>>> #define CMA_IBOE_PACKET_LIFETIME 18
>>>>> #define CMA_PREFERRED_ROCE_GID_TYPE IB_GID_TYPE_ROCE_UDP_ENCAP
>>>>> 
>>>>> +#define CMA_DFLT_CM_RESPONSE_TIMEOUT 20
>>>>> +static int cma_cm_response_timeout = CMA_DFLT_CM_RESPONSE_TIMEOUT;
>>>>> +static int cma_cm_response_timeout_min = 8;
>>>>> +static int cma_cm_response_timeout_max = 31;
>>>>> +#undef CMA_DFLT_CM_RESPONSE_TIMEOUT
>>>>> +
>>>>> +#define CMA_DFLT_MAX_CM_RETRIES 15
>>>>> +static int cma_max_cm_retries = CMA_DFLT_MAX_CM_RETRIES;
>>>>> +static int cma_max_cm_retries_min = 1;
>>>>> +static int cma_max_cm_retries_max = 100;
>>>>> +#undef CMA_DFLT_MAX_CM_RETRIES
>>>>> +
>>>>> +static struct ctl_table_header *cma_ctl_table_hdr;
>>>>> +static struct ctl_table cma_ctl_table[] = {
>>>>> +	{
>>>>> +		.procname	= "cma_cm_response_timeout",
>>>>> +		.data		= &cma_cm_response_timeout,
>>>>> +		.maxlen		= sizeof(cma_cm_response_timeout),
>>>>> +		.mode		= 0644,
>>>>> +		.proc_handler	= proc_dointvec_minmax,
>>>>> +		.extra1		= &cma_cm_response_timeout_min,
>>>>> +		.extra2		= &cma_cm_response_timeout_max,
>>>>> +	},
>>>>> +	{
>>>>> +		.procname	= "cma_max_cm_retries",
>>>>> +		.data		= &cma_max_cm_retries,
>>>>> +		.maxlen		= sizeof(cma_max_cm_retries),
>>>>> +		.mode		= 0644,
>>>>> +		.proc_handler	= proc_dointvec_minmax,
>>>>> +		.extra1		= &cma_max_cm_retries_min,
>>>>> +		.extra2		= &cma_max_cm_retries_max,
>>>>> +	},
>>>>> +	{ }
>>>>> +};
>>>> Is sysctl the right approach here? Should it be rdma tool instead?
>>>> 
>>>> Jason
>>> 
>>> There are other rdma sysctls currently:  net.rdma_ucm.max_backlog and
>>> net.iw_cm.default_backlog.  The core network stack seems to use sysctl
>>> and not ip tool to set basically globals.
>>> 
>>> To use rdma tool, we'd have to have some concept of a "module" object, I
>>> guess.  IE there's dev, link, and resource rdma tool objects currently.
>>> But these cma timeout settings are really not per dev, link, nor a
>>> resource.   Maybe we have just a "core" object:  rdma core set
>>> cma_max_cm_retries min 8 max 30.
>> 
>> I donât know, I think you make a fairly good argument for leaving it as a sysctl.  We have infrastructure in place for admins to set persistent sysctl settings.  The per device/link settings need something different because link names and such can change.  Since these are globals, Iâd leave them where they are.
>> 
> 
> I have patches from Parav which extend rdmatool to set global to whole
> stack parameters, something like "rdma system ...", so the option to
> set through rdmatool global parameters for modules e.g. "rdma system cma set ..."
> exists. But I'm not sure if it is right thing to do.

If decided to for the sysctl flavour, I must just say: ahem, ahem, I forgot to put in a unregister_net_sysctl_table() in cma_cleanup(). Will send a v2.


Thxs, HÃkon

> 
>> --
>> Doug Ledford <dledford@xxxxxxxxxx>
>>    GPG KeyID: B826A3330E572FDD
>>    Key fingerprint = AE6B 1BDA 122B 23B4 265B  1274 B826 A333 0E57 2FDD