RE: [PATCH] usb: dwc3: gadget: Correct the logic for finding last SG entry

From: Anurag Kumar Vulisha
Date: Thu Jun 06 2019 - 11:36:42 EST

Hi Felipe,

>-----Original Message-----
>From: Felipe Balbi [mailto:balbi@xxxxxxxxxx]
>Sent: Wednesday, June 5, 2019 1:34 PM
>To: Anurag Kumar Vulisha <anuragku@xxxxxxxxxx>; Greg Kroah-Hartman
>Cc: linux-usb@xxxxxxxxxxxxxxx; linux-kernel@xxxxxxxxxxxxxxx;
>v.anuragkumar@xxxxxxxxx; Anurag Kumar Vulisha <anuragku@xxxxxxxxxx>
>Subject: Re: [PATCH] usb: dwc3: gadget: Correct the logic for finding last SG
>Anurag Kumar Vulisha <anurag.kumar.vulisha@xxxxxxxxxx> writes:
>> As a process of preparing TRBs usb_gadget_map_request_by_dev() is
>> called from dwc3_prepare_trbs() for mapping the request. This will
>> call dma_map_sg() if req->num_sgs are greater than 0. dma_map_sg()
>> will map the sg entries in sglist and return the number of mapped SGs.
>> As a part of mapping, some sg entries having contigous memory may be
>> merged together into a single sg (when IOMMU used). So, the number of
>> mapped sg entries may not be equal to the number of orginal sg entries
>> in the request (req->num_sgs).
>> As a part of preparing the TRBs, dwc3_prepare_one_trb_sg() iterates
>> over the sg entries present in the sglist and calls sg_is_last() to
>> identify whether the sg entry is last and set IOC bit for the last sg
>> entry. The
>> sg_is_last() determines last sg if SG_END is set in sg->page_link.
>> When IOMMU used, dma_map_sg() merges 2 or more sgs into a single sg
>> and it doesn't retain the page_link properties. Because of this reason
>> the
>> sg_is_last() may not find SG_END and thus resulting in IOC bit never
>> getting set.
>> For example:
>> Consider a request having 8 sg entries with each entry having a length
>> of
>> 4096 bytes. Assume that sg1 & sg2, sg3 & sg4, sg5 & sg6, sg7 & sg8 are
>> having contigous memory regions.
>> Before calling dma_map_sg():
>> sg1-->sg2-->sg3-->sg4-->sg6-->sg7-->sg8
>> dma_length: 4K 4K 4K 4K 4K 4K 4K
>> SG_END: False False False False False False True
>> num_sgs = 8
>> num_mapped_sgs = 0
>> The dma_map_sg() merges sg1 & sg2 memory regions into sg1-
>> Similarly sg3 & sg4 into sg2->dma_address, sg5 & sg6 into the
>> sg3->dma_address and sg6 & sg8 into sg4->dma_address. Here the
>> regions are merged but the page_link properties like SG_END are not
>> retained into the merged sgs.
>isn't this a bug in the scatterlist mapping code? Why doesn't it keep

Thanks for providing your comment.
I don't think it is a bug, instead I feel some enhancement needs to be done in
dma-mapping code. SG_END represents the last sg entry in the sglist and it is
correctly getting set to the last sg entry. The issue happens only when 2 or more
sg entry pages are merged into contiguous dma-able address and sg_is_last()
is used to find the last sg entry with valid dma address. I think that along with
sg_is_last() a new flag (SG_DMA_END) and function (something like sg_dma_is_last() )
needs to be added into dma-mapping code for identifying the last valid sg entry
with valid dma address. So that we can make use of that function instead of sg_is_last().

Anurag Kumar Vulisha