Skip to contentSkip to conversation

tpci: fix NULL pointer dereference on wrong test invocation

PATCH New Krzysztof Kozlowski 2 participants 6 messages Jun 18, 2021 – Jul 26, 2021

lore
Share & tools
lore
#0 lore raw

PATCH 1/2 tpci: fix NULL pointer dereference on wrong test invocation

Fix NULL pointer dereference when a ltp_tpci test case is started before choosing the device:

    ltp_tpci: test-case 12
    ltp_tpci: assign resources
    ltp_tpci: assign resource #0
    BUG: kernel NULL pointer dereference, address: 00000000000003b8
    ...
     Call Trace:
      dev_attr_store+0x17/0x30
      sysfs_kf_write+0x3e/0x50
      kernfs_fop_write_iter+0x13c/0x1d0
      new_sync_write+0x113/0x1a0
      vfs_write+0x1c5/0x200
      ksys_write+0x67/0xe0
      __x64_sys_write+0x1a/0x20
      do_syscall_64+0x49/0xc0
      entry_SYSCALL_64_after_hwframe+0x44/0xa9
---
 testcases/kernel/device-drivers/pci/tpci_kernel/ltp_tpci.c | 5 +++++
 1 file changed, 5 insertions(+)
testcases/kernel/device-drivers/pci/tpci_kernel/ltp_tpci.c +5 -0
index 5b48aa0c7ece..f2d4a4ba497c 100644
@@ -556,6 +556,11 @@ static int test_case(unsigned int cmd)
 {
 	int rc = TSKIP;
 
+	if (!ltp_pci.dev || !ltp_pci.bus) {
+		prk_err("device or bus not selected for test");
+		return TFAIL;
+	}
+
 	switch (cmd) {
 	case PCI_ENABLE:
 		rc = pci_enable();
Signature · 1 line
2.27.0
#1 lore raw

PATCH 1/2 tpci: fix NULL pointer dereference on wrong test invocation

Quoted 19 lines · Krzysztof Kozlowski

Fix NULL pointer dereference when a ltp_tpci test case is started before choosing the device:

    ltp_tpci: test-case 12
    ltp_tpci: assign resources
    ltp_tpci: assign resource #0
    BUG: kernel NULL pointer dereference, address: 00000000000003b8
    ...
     Call Trace:
      dev_attr_store+0x17/0x30
      sysfs_kf_write+0x3e/0x50
      kernfs_fop_write_iter+0x13c/0x1d0
      new_sync_write+0x113/0x1a0
      vfs_write+0x1c5/0x200
      ksys_write+0x67/0xe0
      __x64_sys_write+0x1a/0x20
      do_syscall_64+0x49/0xc0
      entry_SYSCALL_64_after_hwframe+0x44/0xa9
Quoted 5 lines
---
 testcases/kernel/device-drivers/pci/tpci_kernel/ltp_tpci.c | 5 +++++
 1 file changed, 5 insertions(+)

Hi everyone,

Any comments on these two tpci patches? Do they look reasonable?

Best regards,
Krzysztof

#2 lore raw

PATCH 1/2 tpci: fix NULL pointer dereference on wrong test invocation

Hi Krzysztof,

Quoted 19 lines · Krzysztof Kozlowski
Quoted 18 lines

Fix NULL pointer dereference when a ltp_tpci test case is started before choosing the device:

    ltp_tpci: test-case 12
    ltp_tpci: assign resources
    ltp_tpci: assign resource #0
    BUG: kernel NULL pointer dereference, address: 00000000000003b8
    ...
     Call Trace:
      dev_attr_store+0x17/0x30
      sysfs_kf_write+0x3e/0x50
      kernfs_fop_write_iter+0x13c/0x1d0
      new_sync_write+0x113/0x1a0
      vfs_write+0x1c5/0x200
      ksys_write+0x67/0xe0
      __x64_sys_write+0x1a/0x20
      do_syscall_64+0x49/0xc0
      entry_SYSCALL_64_after_hwframe+0x44/0xa9

Merged this one, thanks!

Kind regards,
Petr

#3 lore raw

PATCH 2/2 tpci: accept ENOMEM resource failure with virtio-pci

In reply to Krzysztof Kozlowski

Assigning the memory prefetch resource to virtio-pci might fail on VM guests. For example on Oracle cloud instance (describing itself as "Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.4.1 12/03/2020"):

    test_pci   76  TFAIL  :  tpci.c:73: PCI bus 00 slot 20 : Test-case '12'

In dmesg:

ltp_tpci: test-case 12
ltp_tpci: assign resources
ltp_tpci: assign resource #0
ltp_tpci: name = 0000:00:04.0, flags = 262401, start 0xc000, end 0xc03f
ltp_tpci: assign resource #1
ltp_tpci: name = 0000:00:04.0, flags = 262656, start 0xc1010000, end 0xc1010fff
ltp_tpci: assign resource #2
ltp_tpci: name = 0000:00:04.0, flags = 0, start 0x0, end 0x0
ltp_tpci: assign resource #3
ltp_tpci: name = 0000:00:04.0, flags = 0, start 0x0, end 0x0
ltp_tpci: assign resource #4
ltp_tpci: name = 0000:00:04.0, flags = 538190348, start 0x800004000, end 0x800007fff virtio-pci 0000:00:04.0: BAR 4: no space for [mem size 0x00004000 64bit pref] virtio-pci 0000:00:04.0: BAR 4: failed to assign [mem size 0x00004000 64bit pref]

---

Found the failure on multiple different kernels in different cloud providers (Oracle, AWS, Azure):
https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1878389

---
 .../device-drivers/pci/tpci_kernel/ltp_tpci.c     | 15 ++++++++++++---
 1 file changed, 12 insertions(+), 3 deletions(-)
testcases/kernel/device-drivers/pci/tpci_kernel/ltp_tpci.c +12 -3
index f2d4a4ba497c..41462d4ead6d 100644
@@ -108,8 +108,9 @@ static int probe_pci_dev(unsigned int bus, unsigned int slot)
 	if (!dev || !dev->driver)
 		return -ENODEV;
 
-	prk_info("found pci_dev '%s', bus %u, devfn %u",
-		pci_name(dev), bus, slot);
+	prk_info("found pci_dev '%s', driver '%s', bus %u, devfn %u",
+		pci_name(dev), (dev->driver) ? dev->driver->name : "",
+		bus, slot);
 
 	ltp_pci.dev = dev;
 	ltp_pci.bus = dev->bus;
@@ -444,7 +445,15 @@ static int test_assign_resources(void)
 			r->flags & IORESOURCE_PREFETCH) {
 			ret = pci_assign_resource(dev, i);
 			prk_info("assign resource to '%d', ret '%d'", i, ret);
-			rc |= (ret < 0 && ret != -EBUSY) ? TFAIL : TPASS;
+			if (dev->driver && !strncmp(dev->driver->name, "virtio-pci",
+						    strlen("virtio-pci"))) {
+				if (ret < 0 && ret != -EBUSY && ret != -ENOMEM)
+					rc |= TFAIL;
+				else
+					rc |= TPASS;
+			} else {
+				rc |= (ret < 0 && ret != -EBUSY) ? TFAIL : TPASS;
+			}
 		}
 	}
 
Signature · 1 line
2.27.0
#4 lore raw

PATCH 2/2 tpci: accept ENOMEM resource failure with virtio-pci

Hi Krzysztof,

Quoted 6 lines

Assigning the memory prefetch resource to virtio-pci might fail on VM guests. For example on Oracle cloud instance (describing itself as "Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.4.1 12/03/2020"):

    test_pci   76  TFAIL  :  tpci.c:73: PCI bus 00 slot 20 : Test-case '12'

I'm not sure if we don't cover actual bug. Because I haven't found any issues on virtio_pci used on VM boxes. But probably failing to assign memory is something else than kernel bug.

Quoted 45 lines

In dmesg:

ltp_tpci: test-case 12
ltp_tpci: assign resources
ltp_tpci: assign resource #0
ltp_tpci: name = 0000:00:04.0, flags = 262401, start 0xc000, end 0xc03f
ltp_tpci: assign resource #1
ltp_tpci: name = 0000:00:04.0, flags = 262656, start 0xc1010000, end 0xc1010fff
ltp_tpci: assign resource #2
ltp_tpci: name = 0000:00:04.0, flags = 0, start 0x0, end 0x0
ltp_tpci: assign resource #3
ltp_tpci: name = 0000:00:04.0, flags = 0, start 0x0, end 0x0
ltp_tpci: assign resource #4
ltp_tpci: name = 0000:00:04.0, flags = 538190348, start 0x800004000, end 0x800007fff virtio-pci 0000:00:04.0: BAR 4: no space for [mem size 0x00004000 64bit pref] virtio-pci 0000:00:04.0: BAR 4: failed to assign [mem size 0x00004000 64bit pref]

---

Found the failure on multiple different kernels in different cloud providers (Oracle, AWS, Azure):
https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1878389

---
 .../device-drivers/pci/tpci_kernel/ltp_tpci.c     | 15 ++++++++++++---
 1 file changed, 12 insertions(+), 3 deletions(-)
testcases/kernel/device-drivers/pci/tpci_kernel/ltp_tpci.c +3 -2
index f2d4a4ba497c..41462d4ead6d 100644
@@ -108,8 +108,9 @@ static int probe_pci_dev(unsigned int bus, unsigned int slot)
 	if (!dev || !dev->driver)
 		return -ENODEV;
 
-	prk_info("found pci_dev '%s', bus %u, devfn %u",
-		pci_name(dev), bus, slot);
+	prk_info("found pci_dev '%s', driver '%s', bus %u, devfn %u",
+		pci_name(dev), (dev->driver) ? dev->driver->name : "",
+		bus, slot);
 
 	ltp_pci.dev = dev;
 	ltp_pci.bus = dev->bus;

testcases/kernel/device-drivers/pci/tpci_kernel/ltp_tpci.c static int test_assign_resources(void) · line 452

Quoted 9 lines
testcases/kernel/device-drivers/pci/tpci_kernel/ltp_tpci.c +5 -1
@@ -444,7 +445,15 @@ static int test_assign_resources(void)
 			r->flags & IORESOURCE_PREFETCH) {
 			ret = pci_assign_resource(dev, i);
 			prk_info("assign resource to '%d', ret '%d'", i, ret);
-			rc |= (ret < 0 && ret != -EBUSY) ? TFAIL : TPASS;
+			if (dev->driver && !strncmp(dev->driver->name, "virtio-pci",
+						    strlen("virtio-pci"))) {
+				if (ret < 0 && ret != -EBUSY && ret != -ENOMEM)
+					rc |= TFAIL;
+				else

Shouldn't ENOMEM/EBUSY result to TCONF or TBROK?

https://github.com/linux-test-project/ltp/wiki/C-Test-API#12-basic-test-interface

Kind regards,
Petr

Quoted 6 lines
testcases/kernel/device-drivers/pci/tpci_kernel/ltp_tpci.c +4 -0
+					rc |= TPASS;
+			} else {
+				rc |= (ret < 0 && ret != -EBUSY) ? TFAIL : TPASS;
+			}
 		}
 	}