d/freebsd-nq
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Partial workaround for Intel PCI adapters reading past the end of the

Browse Source 
host-programmed DMA regions.  This change seemingly fixes the
descriptor fetches, but the packet memory accesses are left
problematic.

Reviewed by:	emaste, erj, sbruno
Sponsored by:	The FreeBSD Foundation
MFC after:	2 weeks
Differential revision:	https://reviews.freebsd.org/D8282
This commit is contained in:
Konstantin Belousov 2016-10-19 18:15:44 +00:00
parent 801b39194a
commit cb31871867
1 changed files with 15 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

19
sys/dev/e1000/if_lem.c
View File

@ -543,8 +543,16 @@ lem_attach(device_t dev)
*/
adapter->hw.mac.report_tx_early = 1;
tsize = roundup2(adapter->num_tx_desc * sizeof(struct e1000_tx_desc),
EM_DBA_ALIGN);
/*
* It seems that the descriptor DMA engine on some PCI cards
* fetches memory past the end of the last descriptor in the
* ring. These reads are problematic when VT-d (DMAR) busdma
* is used. Allocate the scratch space to avoid getting
* faults from DMAR, by requesting scratch memory for one more
* descriptor.
*/
tsize = roundup2((adapter->num_tx_desc + 1) *
sizeof(struct e1000_tx_desc), EM_DBA_ALIGN);
/* Allocate Transmit Descriptor ring */
if (lem_dma_malloc(adapter, tsize, &adapter->txdma, BUS_DMA_NOWAIT)) {
@ -555,8 +563,11 @@ lem_attach(device_t dev)
adapter->tx_desc_base =
(struct e1000_tx_desc *)adapter->txdma.dma_vaddr;
rsize = roundup2(adapter->num_rx_desc * sizeof(struct e1000_rx_desc),
EM_DBA_ALIGN);
/*
* See comment above txdma allocation for rationale behind +1.
*/
rsize = roundup2((adapter->num_rx_desc + 1) *
sizeof(struct e1000_rx_desc), EM_DBA_ALIGN);
/* Allocate Receive Descriptor ring */
if (lem_dma_malloc(adapter, rsize, &adapter->rxdma, BUS_DMA_NOWAIT)) {