ioat(4): Use memory completion rather than device register

The CHANSTS register is a split 64-bit register on CBDMA units before
hardware v3.3.  If a torn read happens during ioat_process_events(),
software cannot know when to stop completing descriptors correctly.

So, just use the device-pushed main memory channel status instead.

Remove the ioat_get_active() seatbelt as well.  It does nothing if the
completion address is valid.

Sponsored by:	Dell EMC Isilon
This commit is contained in:
Conrad Meyer 2016-10-28 23:53:37 +00:00
parent 76305bb865
commit 0d0f264099
2 changed files with 11 additions and 2 deletions

View File

@ -677,7 +677,7 @@ ioat_process_events(struct ioat_softc *ioat)
}
completed = 0;
comp_update = ioat_get_chansts(ioat);
comp_update = *ioat->comp_update;
status = comp_update & IOAT_CHANSTS_COMPLETED_DESCRIPTOR_MASK;
if (status == ioat->last_seen) {
@ -691,7 +691,7 @@ ioat_process_events(struct ioat_softc *ioat)
__func__, ioat->chan_idx, comp_update, ioat->last_seen);
desc = ioat_get_ring_entry(ioat, ioat->tail - 1);
while (desc->hw_desc_bus_addr != status && ioat_get_active(ioat) > 0) {
while (desc->hw_desc_bus_addr != status) {
desc = ioat_get_ring_entry(ioat, ioat->tail);
dmadesc = &desc->bus_dmadesc;
CTR5(KTR_IOAT, "channel=%u completing desc idx %u (%p) ok cb %p(%p)",

View File

@ -523,6 +523,15 @@ struct ioat_softc {
void ioat_test_attach(void);
void ioat_test_detach(void);
/*
* XXX DO NOT USE this routine for obtaining the current completed descriptor.
*
* The double_4 read on ioat<3.3 appears to result in torn reads. And v3.2
* hardware is still commonplace (Broadwell Xeon has it). Instead, use the
* device-pushed *comp_update.
*
* It is safe to use ioat_get_chansts() for the low status bits.
*/
static inline uint64_t
ioat_get_chansts(struct ioat_softc *ioat)
{