Skip to contentSkip to conversation

rsxx: Moving pci_map_page to prevent overflow

PATCH Superseded Philip J. Kelleher 2 participants 2 messages Aug 27, 2013 – Sep 3, 2013 Block layer

v2 of 3 Superseded Latest is v3 →

lore
Share & tools
lore
#0 lore raw

PATCH v2 2/2 rsxx: Moving pci_map_page to prevent overflow

From: Philip J Kelleher <[email protected]>

The pci_map_page function has been moved into our issued workqueue to prevent an us running out of mappable addresses on non-HWWD PCIe x8 slots. The maximum amount that can possible be mapped at one time now is: 255 dmas X 4 dma channels X 4096 Bytes.

-------------------------------------------------------------------------------

drivers/block/rsxx/core.c +0 -5
diff -uprN -X linux-block-vanilla/Documentation/dontdiff linux-block-vanilla/drivers/block/rsxx/core.c linux-block/drivers/block/rsxx/core.c
@@ -749,10 +749,6 @@ static pci_ers_result_t rsxx_slot_reset(
 
 	card->eeh_state = 0;
 
-	st = rsxx_eeh_remap_dmas(card);
-	if (st)
-		goto failed_remap_dmas;
-
 	spin_lock_irqsave(&card->irq_lock, flags);
 	if (card->n_targets & RSXX_MAX_TARGETS)
 		rsxx_enable_ier_and_isr(card, CR_INTR_ALL_G);
@@ -779,7 +775,6 @@ static pci_ers_result_t rsxx_slot_reset(
 	return PCI_ERS_RESULT_RECOVERED;
 
 failed_hw_buffers_init:
-failed_remap_dmas:
 	for (i = 0; i < card->n_targets; i++) {
 		if (card->ctrl[i].status.buf)
 			pci_free_consistent(card->dev,
drivers/block/rsxx/dma.c +27 -42

105 lines not on this page · raw message

#1 lore raw
Quoted 14 lines · Philip J. Kelleher

diff -uprN -X linux-block-vanilla/Documentation/dontdiff linux-block-vanilla/drivers/block/rsxx/dma.c linux-block/drivers/block/rsxx/dma.c

drivers/block/rsxx/dma.c +7 -0
@@ -433,6 +434,27 @@ static void rsxx_issue_dmas(struct rsxx_
 			continue;
 		}
 
+		if (dma->cmd == HW_CMD_BLK_WRITE)
+			dir = PCI_DMA_TODEVICE;
+		else
+			dir = PCI_DMA_FROMDEVICE;
+
+		/*
+		 * The function pci_map_page is placed here because we can

drivers/block/rsxx/dma.c static void rsxx_issue_dmas(struct rsxx_ · line 453

Quoted 10 lines
drivers/block/rsxx/dma.c +10 -0
+		 * only, by design, issue up to 255 commands to the hardware
+		 * at one time per DMA channel. So the maximum amount of mapped
+		 * memory would be 255 * 4 channels * 4096 Bytes which is less
+		 * than 2GB, the limit of a x8 Non-HWWD PCIe slot. This way the
+		 * pci_map_page function should never fail because of a
+		 * lack of mappable memory.
+		 */
+		dma->dma_addr = pci_map_page(ctrl->card->dev, dma->page,
+				     dma->pg_off, dma->sub_page.cnt << 9, dir);
+		if (pci_dma_mapping_error(ctrl->card->dev, dma->dma_addr)) {

Do you also need to do a push_tracker here as well so you don't leak a tag? Why is this error case handled different compared to the dma_fault case earlier in this same function?

drivers/block/rsxx/dma.c static void rsxx_issue_dmas(struct rsxx_ · line 460

Quoted 7 lines
drivers/block/rsxx/dma.c +4 -0
+			kmem_cache_free(rsxx_dma_pool, dma);
+			return;
+		}
+
 		set_tracker_dma(ctrl->trackers, tag, dma);
 		hw_cmd_buf[ctrl->cmd.idx].command  = dma->cmd;
 		hw_cmd_buf[ctrl->cmd.idx].tag      = tag;

-Brian

Signature · 3 lines
Brian King
Power Linux I/O
IBM Linux Technology Center