Skip to contentSkip to conversation

media: bt878: check register mapping in probe

PATCH New Ruoyu Wang 2 participants 2 messages Jun 7 Media (V4L/DVB)

lore
Share & tools
lore

Ruoyu Wang

author
#0 lore raw

PATCH 1/1 media: bt878: check register mapping in probe

bt878_probe() maps the device register window and then immediately clears the interrupt mask through btwrite(). If ioremap() fails, the register write dereferences a NULL mapping.

Check the mapping before touching the registers, and unmap it again on later probe failures after the IRQ or DMA buffer allocation paths fail.

---
 drivers/media/pci/bt8xx/bt878.c | 10 ++++++++++
 1 file changed, 10 insertions(+)
drivers/media/pci/bt8xx/bt878.c +10 -0
index 62a6c4a80..245a5f0ef 100644
@@ -445,6 +445,10 @@ static int bt878_probe(struct pci_dev *dev, const struct pci_device_id *pci_id)
 	bt->bt878_mem = (unsigned char *) bt->bt878_adr;
 #else
 	bt->bt878_mem = ioremap(bt->bt878_adr, 0x1000);
+	if (!bt->bt878_mem) {
+		result = -ENOMEM;
+		goto fail1;
+	}
 #endif
 
 	/* clear interrupt mask */
@@ -486,6 +490,12 @@ static int bt878_probe(struct pci_dev *dev, const struct pci_device_id *pci_id)
       fail2:
 	free_irq(bt->irq, bt);
       fail1:
+#ifndef __sparc__
+	if (bt->bt878_mem) {
+		iounmap(bt->bt878_mem);
+		bt->bt878_mem = NULL;
+	}
+#endif
 	release_mem_region(pci_resource_start(bt->dev, 0),
 			   pci_resource_len(bt->dev, 0));
       fail0:
Signature · 1 line
2.51.0
#1 lore raw
Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:

Pre-existing issues:
- [High] Race condition in bt878_probe due to unprotected global variable bt878_num --

commit 5c524e09ca816b9e46cc6d10e5292d06fb75ca23
Author: Ruoyu Wang <[email protected]>

media: bt878: check register mapping in probe

bt878_probe() maps the device register window and then immediately clears the interrupt mask through btwrite(). If ioremap() fails, the register write dereferences a NULL mapping. This patch adds a check for the mapping before touching the registers, and unmaps it on later probe failures.

drivers/media/pci/bt8xx/bt878.c static int bt878_probe(struct pci_dev *dev, const struct pci_device_id *pci_id) · line 447

Quoted 8 lines
drivers/media/pci/bt8xx/bt878.c +0 -0
index 62a6c4a80bed2..245a5f0ef4fec 100644
@@ -445,6 +445,10 @@ static int bt878_probe(struct pci_dev *dev, const struct pci_device_id *pci_id)
 	bt->bt878_mem = (unsigned char *) bt->bt878_adr;
 #else
 	bt->bt878_mem = ioremap(bt->bt878_adr, 0x1000);

[Severity: High]
This isn't a bug introduced by this patch, but looking at how bt is assigned earlier in bt878_probe():

	bt = &bt878[bt878_num];
	bt->dev = dev;
	bt->nr = bt878_num;

Could this cause a race condition since the global bt878_num counter and bt878 array are modified without any locking?

If multiple bt878 devices are bound simultaneously (e.g. via concurrent userspace sysfs bind writes), multiple threads could read the same bt878_num index and concurrently overwrite the same bt878[bt878_num] element.

Could this lead to resource leaks (like losing the ioremap pointer assigned here) or memory corruption when devices process interrupts using clobbered state?

Quoted 7 lines
drivers/media/pci/bt8xx/bt878.c +4 -0
+	if (!bt->bt878_mem) {
+		result = -ENOMEM;
+		goto fail1;
+	}
 #endif
 
 	/* clear interrupt mask */
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=1