On Tip of Tree (47cf1b422e60) I'm seeing the following IOMMU error when enumerating my eMMC device: [ 13.097314] mmc0: SDHCI controller on PCI [0000:02:00.0] using ADMA 64-bit [ 13.104550] mmc1: SDHCI controller on ACPI [AMDI0040:00] using ADMA 64-bit [ 13.136359] mmc1: Allocated DMA: align_addr: 0x000000041ab55000, adma_addr: 0x000000041ab55200, dma_size: e0c. <- I added this to sdhci_setup_host [ 13.340852] mmc1: ADMA error: 0x02000000 [ 13.344780] mmc1: sdhci: ============ SDHCI REGISTER DUMP =========== [ 13.351222] mmc1: sdhci: Sys addr: 0x00000000 | Version: 0x00001002 [ 13.357669] mmc1: sdhci: Blk size: 0x00007200 | Blk cnt: 0x00000001 [ 13.364118] mmc1: sdhci: Argument: 0x00000000 | Trn mode: 0x00000013 [ 13.370567] mmc1: sdhci: Present: 0xf1ff0000 | Host ctl: 0x00000019 [ 13.377014] mmc1: sdhci: Power: 0x0000000f | Blk gap: 0x00000000 [ 13.383463] mmc1: sdhci: Wake-up: 0x00000000 | Clock: 0x0000fa07 [ 13.389911] mmc1: sdhci: Timeout: 0x0000000a | Int stat: 0x00000000 [ 13.396360] mmc1: sdhci: Int enab: 0x03ff000b | Sig enab: 0x03ff000b [ 13.402808] mmc1: sdhci: ACmd stat: 0x00000000 | Slot int: 0x00000000 [ 13.409257] mmc1: sdhci: Caps: 0x71fec8b2 | Caps_1: 0x00000577 [ 13.415706] mmc1: sdhci: Cmd: 0x0000083a | Max curr: 0x00c80064 [ 13.422154] mmc1: sdhci: Resp[0]: 0x00000700 | Resp[1]: 0xffffffff [ 13.428603] mmc1: sdhci: Resp[2]: 0x328f5903 | Resp[3]: 0x00d00f00 [ 13.435048] mmc1: sdhci: Host ctl2: 0x00000000 [ 13.439506] mmc1: sdhci: ADMA Err: 0x00000001 | ADMA Ptr: 0x000000041ab55200 [ 13.446641] mmc1: sdhci: ============================================ [ 13.453077] mmc1: sdhci: 41ab55200: DMA 0x000000041a462400, LEN 0x0200, Attr=0x21 [ 13.460551] mmc1: sdhci: 41ab5520c: DMA 0x0000000000000000, LEN 0x0000, Attr=0x03 [ 13.468070] AMD-Vi: Event logged [IO_PAGE_FAULT device=00:13.1 domain=0x0000 address=0x41ab55200 flags=0x0050] [ 13.478109] mmc1: error -5 whilst initialising MMC card If I read the error correctly (not sure), it looks like the ADMA descriptor was at 0x41ab55200. The fault was also for 0x41ab55200. Decoding the flags: 2.5.3 IO_PAGE_FAULT Event * PE (peripheral did not have the permissions required to perform the transaction) * PR (transaction was to a page marked as present (including V=1b in DTE) or interrupt marked as remapped (RemapEn=1)). So it looks like the eMMC controller had a page fault trying to read the DMA descriptor. Is this a known issue? Does anyone have any suggestions? I little background: We recently cherry-picked `be62dbf554c5 iommu/amd: Convert AMD iommu driver to the dma-iommu api` into the chromiumos 5.4 tree: https://chromium-review.googlesource.com/c/chromiumos/third_party/kernel/+/2169153 This caused the eMMC page faults to start happening on boot. When we reverted the patch from our tree the problems went away. When I tried on ToT, I saw the same errors. So I reverted be62dbf554c5 from my ToT branch expecting the errors to go away, but they remained. So this makes me wonder if we are missing an mmc patch or the iommu wasn't working correctly on 5.4 before we landed be62dbf554c5. I looked through the recent mmc patches, but didn't see anything that stood out. This is the driver we are using: https://source.chromium.org/chromiumos/chromiumos/codesearch/+/master:src/third_party/kernel/v5.4/drivers/mmc/host/sdhci-acpi.c;l=607 Any suggestions would be appreciated.
