Re: [PATCH V2 1/1] net: cdc_ncm: Reduce memory use when kernel memory low

From: BjÃrn Mork
Date: Fri Jun 30 2017 - 13:39:04 EST


Jim Baxter <jim_baxter@xxxxxxxxxx> writes:

> The CDC-NCM driver can require large amounts of memory to create
> skb's and this can be a problem when the memory becomes fragmented.
>
> This especially affects embedded systems that have constrained
> resources but wish to maximise the throughput of CDC-NCM with 16KiB
> NTB's.
>
> The issue is after running for a while the kernel memory can become
> fragmented and it needs compacting.
> If the NTB allocation is needed before the memory has been compacted
> the atomic allocation can fail which can cause increased latency,
> large re-transmissions or disconnections depending upon the data
> being transmitted at the time.
> This situation occurs for less than a second until the kernel has
> compacted the memory but the failed devices can take a lot longer to
> recover from the failed TX packets.
>
> To ease this temporary situation I modified the CDC-NCM TX path to
> temporarily switch into a reduced memory mode which allocates an NTB
> that will fit into a USB_CDC_NCM_NTB_MIN_OUT_SIZE (default 2048 Bytes)
> sized memory block and only transmit NTB's with a single network frame
> until the memory situation is resolved.
> Each time this issue occurs we wait for an increasing number of
> reduced size allocations before requesting a full size one to not
> put additional pressure on a low memory system.
>
> Once the memory is compacted the CDC-NCM data can resume transmitting
> at the normal tx_max rate once again.
>
> Signed-off-by: Jim Baxter <jim_baxter@xxxxxxxxxx>

This looks good to me.

I would still be happier if we didn't need something like this, but I
understand that we do. And this patch looks as clean as it can get. I
haven't tested the patch under any sort of memory pressure, but I did a
basic runtime test on a single MBIM device. As expected, I did not
notice any changes with this patch applied.

But regarding noticable effects: The patch adds no printks, counters or
sysfs attributes which could tell the user that the initial buffer
allocation has failed. Maybe add some sort of debug helper(s) in a
followup patch? How did you verify the patch operation while testing it?

Anyway, that's no show stopper of course. So FWIW:

Reviewed-by: BjÃrn Mork <bjorn@xxxxxxx>