Re: [PATCH v2] iser: explicitly set shost max_segment_size if non virtual boundary devices

Re: [PATCH v2] iser: explicitly set shost max_segment_size if non virtual boundary devices

[Jason Gunthorpe]

On Thu, Jun 06, 2019 at 06:29:14PM -0700, Sagi Grimberg wrote:
> if the rdma device supports sg gaps, we don't need to set a virtual
> boundary but we then need to explicitly set the max_segment_size, otherwise
> scsi takes BLK_MAX_SEGMENT_SIZE and sets it using dma_set_max_seg_size()
> and this affects all the rdma device consumers.
> 
> Fix it by setting shost max_segment_size according to the device
> capability if SG_GAPS are not supported.
> 
> Reported-by: Jason Gunthorpe <jgg@ziepe.ca>
> Suggested-by: Christoph Hellwig <hch@lst.de>
> Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
> ---
> Changes from v1:
> - set max_segment_size only for non virtual boundary devices
> 
>  drivers/infiniband/ulp/iser/iscsi_iser.c | 4 +++-
>  1 file changed, 3 insertions(+), 1 deletion(-)

Sagi, are you respinning this or ??

https://patchwork.kernel.org/patch/10980657/

Jason

Re: [PATCH v2] iser: explicitly set shost max_segment_size if non virtual boundary devices

[Sagi Grimberg]

>> if the rdma device supports sg gaps, we don't need to set a virtual
>> boundary but we then need to explicitly set the max_segment_size, otherwise
>> scsi takes BLK_MAX_SEGMENT_SIZE and sets it using dma_set_max_seg_size()
>> and this affects all the rdma device consumers.
>>
>> Fix it by setting shost max_segment_size according to the device
>> capability if SG_GAPS are not supported.
>>
>> Reported-by: Jason Gunthorpe <jgg@ziepe.ca>
>> Suggested-by: Christoph Hellwig <hch@lst.de>
>> Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
>> ---
>> Changes from v1:
>> - set max_segment_size only for non virtual boundary devices
>>
>>   drivers/infiniband/ulp/iser/iscsi_iser.c | 4 +++-
>>   1 file changed, 3 insertions(+), 1 deletion(-)
> 
> Sagi, are you respinning this or ??

I can take the change-log message and paste it in the code,
but is that something we want? Usually people look in the
change-log history...

Re: [PATCH v2] iser: explicitly set shost max_segment_size if non virtual boundary devices

[Sagi Grimberg]

>>> if the rdma device supports sg gaps, we don't need to set a virtual
>>> boundary but we then need to explicitly set the max_segment_size, 
>>> otherwise
>>> scsi takes BLK_MAX_SEGMENT_SIZE and sets it using dma_set_max_seg_size()
>>> and this affects all the rdma device consumers.
>>>
>>> Fix it by setting shost max_segment_size according to the device
>>> capability if SG_GAPS are not supported.
>>>
>>> Reported-by: Jason Gunthorpe <jgg@ziepe.ca>
>>> Suggested-by: Christoph Hellwig <hch@lst.de>
>>> Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
>>> ---
>>> Changes from v1:
>>> - set max_segment_size only for non virtual boundary devices
>>>
>>>   drivers/infiniband/ulp/iser/iscsi_iser.c | 4 +++-
>>>   1 file changed, 3 insertions(+), 1 deletion(-)
>>
>> Sagi, are you respinning this or ??
> 
> I can take the change-log message and paste it in the code,
> but is that something we want? Usually people look in the
> change-log history...

I actually thought it was already in...