Halil Pasic points out [1] that the full revert of that commit (revert in bddac7c1e02b), and that a partial revert that only reverts the problematic case, but still keeps some of the cleanups is probably better.  And that partial revert [2] had already been verified by Oleksandr Natalenko to also fix the issue, I had just missed that in the long discussion. So let's reinstate the cleanups from commit aa6f8dcbab47 ("swiotlb: rework "fix info leak with DMA_FROM_DEVICE""), and effectively only revert the part that caused problems.
Unified · Split
| @@ -130,11 +130,3 @@ accesses to DMA buffers in both privileged "supervisor" and unprivileged · 129 unmodified lines | |||
| 130 | subsystem that the buffer is fully accessible at the elevated privilege | 130 | subsystem that the buffer is fully accessible at the elevated privilege |
| 131 | level (and ideally inaccessible or at least read-only at the | 131 | level (and ideally inaccessible or at least read-only at the |
| 132 | lesser-privileged levels). | 132 | lesser-privileged levels). |
| 133 | |||
| 134 | DMA_ATTR_OVERWRITE | ||
| 135 | ------------------ | ||
| 136 | |||
| 137 | This is a hint to the DMA-mapping subsystem that the device is expected to | ||
| 138 | overwrite the entire mapped size, thus the caller does not require any of the | ||
| 139 | previous buffer contents to be preserved. This allows bounce-buffering | ||
| 140 | implementations to optimise DMA_FROM_DEVICE transfers. | ||
| @@ -61,14 +61,6 @@ · 60 unmodified lines | |||
| 61 | */ | 61 | */ |
| 62 | #define DMA_ATTR_PRIVILEGED (1UL << 9) | 62 | #define DMA_ATTR_PRIVILEGED (1UL << 9) |
| 63 | 63 | ||
| 64 | /* | ||
| 65 | * This is a hint to the DMA-mapping subsystem that the device is expected | ||
| 66 | * to overwrite the entire mapped size, thus the caller does not require any | ||
| 67 | * of the previous buffer contents to be preserved. This allows | ||
| 68 | * bounce-buffering implementations to optimise DMA_FROM_DEVICE transfers. | ||
| 69 | */ | ||
| 70 | #define DMA_ATTR_OVERWRITE (1UL << 10) | ||
| 71 | |||
| 72 | /* | 64 | /* |
| 73 | * A dma_addr_t can hold any valid DMA or bus address for the platform. It can | 65 | * A dma_addr_t can hold any valid DMA or bus address for the platform. It can |
| 74 | * be given to a device to use as a DMA source or target. It is specific to a | 66 | * be given to a device to use as a DMA source or target. It is specific to a |
| @@ -627,10 +627,14 @@ phys_addr_t swiotlb_tbl_map_single(struct device *dev, phys_addr_t orig_addr, · 626 unmodified lines | |||
| 627 | for (i = 0; i < nr_slots(alloc_size + offset); i++) | 627 | for (i = 0; i < nr_slots(alloc_size + offset); i++) |
| 628 | mem->slots[index + i].orig_addr = slot_addr(orig_addr, i); | 628 | mem->slots[index + i].orig_addr = slot_addr(orig_addr, i); |
| 629 | tlb_addr = slot_addr(mem->start, index) + offset; | 629 | tlb_addr = slot_addr(mem->start, index) + offset; |
| 630 | if (!(attrs & DMA_ATTR_SKIP_CPU_SYNC) && | 630 | /* |
| 631 | (!(attrs & DMA_ATTR_OVERWRITE) || dir == DMA_TO_DEVICE || | 631 | * When dir == DMA_FROM_DEVICE we could omit the copy from the orig |
| 632 | dir == DMA_BIDIRECTIONAL)) | 632 | * to the tlb buffer, if we knew for sure the device will |
| 633 | swiotlb_bounce(dev, tlb_addr, mapping_size, DMA_TO_DEVICE); | 633 | * overwirte the entire current content. But we don't. Thus |
| 634 | * unconditional bounce may prevent leaking swiotlb content (i.e. | ||
| 635 | * kernel memory) to user-space. | ||
| 636 | */ | ||
| 637 | swiotlb_bounce(dev, tlb_addr, mapping_size, DMA_TO_DEVICE); | ||
| 634 | return tlb_addr; | 638 | return tlb_addr; |
| 635 | } | 639 | } |
| 636 | 640 | ||