diff options
| author | Brendan Cunningham <bcunningham@cornelisnetworks.com> | 2023-05-19 12:32:16 -0400 |
|---|---|---|
| committer | Jason Gunthorpe <jgg@nvidia.com> | 2023-06-01 14:38:00 -0300 |
| commit | c9358de193ecfb360c3ce75f27ce839ca0b0bc8c (patch) | |
| tree | c7f4c09ca5a4b606f355e6a1409f13e5e425a11c /drivers/infiniband/hw/hfi1/sdma.c | |
| parent | 1cc625cecce9cbbe7898c88cc9e6c06021d7d8f0 (diff) | |
IB/hfi1: Fix wrong mmu_node used for user SDMA packet after invalidate
The hfi1 user SDMA pinned-page cache will leave a stale cache entry when
the cache-entry's virtual address range is invalidated but that cache
entry is in-use by an outstanding SDMA request.
Subsequent user SDMA requests with buffers in or spanning the virtual
address range of the stale cache entry will result in packets constructed
from the wrong memory, the physical pages pointed to by the stale cache
entry.
To fix this, remove mmu_rb_node cache entries from the mmu_rb_handler
cache independent of the cache entry's refcount. Add 'struct kref
refcount' to struct mmu_rb_node and manage mmu_rb_node lifetime with
kref_get() and kref_put().
mmu_rb_node.refcount makes sdma_mmu_node.refcount redundant. Remove
'atomic_t refcount' from struct sdma_mmu_node and change sdma_mmu_node
code to use mmu_rb_node.refcount.
Move the mmu_rb_handler destructor call after a
wait-for-SDMA-request-completion call so mmu_rb_nodes that need
mmu_rb_handler's workqueue to queue themselves up for destruction from an
interrupt context may do so.
Fixes: f48ad614c100 ("IB/hfi1: Move driver out of staging")
Fixes: 00cbce5cbf88 ("IB/hfi1: Fix bugs with non-PAGE_SIZE-end multi-iovec user SDMA requests")
Link: https://lore.kernel.org/r/168451393605.3700681.13493776139032178861.stgit@awfm-02.cornelisnetworks.com
Reviewed-by: Dean Luick <dean.luick@cornelisnetworks.com>
Signed-off-by: Brendan Cunningham <bcunningham@cornelisnetworks.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@cornelisnetworks.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Diffstat (limited to 'drivers/infiniband/hw/hfi1/sdma.c')
| -rw-r--r-- | drivers/infiniband/hw/hfi1/sdma.c | 23 |
1 files changed, 18 insertions, 5 deletions
diff --git a/drivers/infiniband/hw/hfi1/sdma.c b/drivers/infiniband/hw/hfi1/sdma.c index bb2552dd29c1..26c62162759b 100644 --- a/drivers/infiniband/hw/hfi1/sdma.c +++ b/drivers/infiniband/hw/hfi1/sdma.c @@ -1593,7 +1593,20 @@ static inline void sdma_unmap_desc( struct hfi1_devdata *dd, struct sdma_desc *descp) { - system_descriptor_complete(dd, descp); + switch (sdma_mapping_type(descp)) { + case SDMA_MAP_SINGLE: + dma_unmap_single(&dd->pcidev->dev, sdma_mapping_addr(descp), + sdma_mapping_len(descp), DMA_TO_DEVICE); + break; + case SDMA_MAP_PAGE: + dma_unmap_page(&dd->pcidev->dev, sdma_mapping_addr(descp), + sdma_mapping_len(descp), DMA_TO_DEVICE); + break; + } + + if (descp->pinning_ctx && descp->ctx_put) + descp->ctx_put(descp->pinning_ctx); + descp->pinning_ctx = NULL; } /* @@ -3113,8 +3126,8 @@ int ext_coal_sdma_tx_descs(struct hfi1_devdata *dd, struct sdma_txreq *tx, /* Add descriptor for coalesce buffer */ tx->desc_limit = MAX_DESC; - return _sdma_txadd_daddr(dd, SDMA_MAP_SINGLE, NULL, tx, - addr, tx->tlen); + return _sdma_txadd_daddr(dd, SDMA_MAP_SINGLE, tx, + addr, tx->tlen, NULL, NULL, NULL); } return 1; @@ -3157,9 +3170,9 @@ int _pad_sdma_tx_descs(struct hfi1_devdata *dd, struct sdma_txreq *tx) make_tx_sdma_desc( tx, SDMA_MAP_NONE, - NULL, dd->sdma_pad_phys, - sizeof(u32) - (tx->packet_len & (sizeof(u32) - 1))); + sizeof(u32) - (tx->packet_len & (sizeof(u32) - 1)), + NULL, NULL, NULL); tx->num_desc++; _sdma_close_tx(dd, tx); return rval; |