Hi all: Several days ago I sent out a series of patches to support Realtek SD/MMC reader. Boris helped to give it a test using his Thinkpad x121e, with a built-in Realtek card reader named rts5209. (Hi, Thank you, Boris .. ) And he reported that some error messages, as below shows, would arise when formating a 16G SD UHS-I card using mkfs.ext3. [ 286.244866] mmc0: new ultra high speed SDR50 SDHC card at address e624 [ 286.277199] mmcblk0: mmc0:e624 SU16G 14.8 GiB [ 286.318163] mmcblk0: [ 363.754647] mmcblk0: p1 [ 379.545510] DMA: Out of SW-IOMMU space for 438272 bytes at device 0000:03:00.0 [ 389.578514] DMA: Out of SW-IOMMU space for 380928 bytes at device 0000:03:00.0 [ 399.590953] DMA: Out of SW-IOMMU space for 524288 bytes at device 0000:03:00.0 [ 409.590698] DMA: Out of SW-IOMMU space for 380928 bytes at device 0000:03:00.0 [ 419.589054] DMA: Out of SW-IOMMU space for 380928 bytes at device 0000:03:00.0 That's to say, in some situation dma_map_sg would fail when transferring massive data. By adding more debug message I found some facts: 1. In Thinkpad x121e, the transfer size in one transfer is often 512KB. (But in my notebook, the largest transfer size I found is 64KB); 2. When dma_map_sg succeeds, the largest sg size in the sglist will not be larger than 256KB; 3. When dma_map_sg fails, the sg number of the sglist is always 2, and the total transfer size is circa 512KB. So I guess that the largest sg size may be larger than 256KB. Based on the above facts and guess, I asked Boris to help to change mmc->max_seg_size from 512KB to 64KB in Realtek SD/MMC host driver, and everything is OK. (Hi Boris, you have also tested 256KB later, and the result is still OK, right ?) So my question comes out. >From our experiment, it seems that mmc host driver must set a proper value to max_seg_size at first. But I really don't think it's the host driver's responsibility to guess/inquiry the max DMA segment size before declaring it. The host driver is all along passive to response the request from the upper layer. If the segment is too large to handle, that buffer should be divided into more smaller pieces at first in the upper layer and then pass on to the lower layer, which is our host driver in this case. And I didn't find any host driver in drivers/mmc/host to inquiry the max DMA segment size that this platform can afford before declaring their max_seg_size. Maybe I missed something , but I still can't find the answer for now. So please help :-) BR, wwang -- To unsubscribe from this list: send the line "unsubscribe linux-mmc" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
