Skip to contentSkip to conversation

ath11k: Change DMA_FROM_DEVICE to DMA_TO_DEVICE when map reinjected packets

PATCH In review Jouni Malinen 3 participants 12 messages Sep 13, 2021 – Oct 11, 2021 Qualcomm Atheros ATH11K wireless driver

lore
Share & tools
lore
#0 lore raw

PATCH 1/5 ath11k: Change DMA_FROM_DEVICE to DMA_TO_DEVICE when map reinjected packets

From: Baochen Qiang <[email protected]>

For fragmented packets, ath11k reassembles each fragment as a normal packet and then reinjects it into HW ring. In this case, the DMA direction should be DMA_TO_DEVICE, not DMA_FROM_DEVICE, otherwise invalid payload will be reinjected to HW and then delivered to host. What is more, since arbitrary memory could be allocated to the frame, we don't know what kind of data is contained in the buffer reinjected. Thus, as a bad result, private info may be leaked.

Note that this issue is only found on Intel platform.

Tested-on: QCA6390 hw2.0 PCI WLAN.HST.1.0.1-01740-QCAHSTSWPLZ_V2_TO_X86-1

---
 drivers/net/wireless/ath/ath11k/dp_rx.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/net/wireless/ath/ath11k/dp_rx.c +1 -1
index 90da56316e7e..0c27eead3e02 100644
@@ -3434,7 +3434,7 @@ static int ath11k_dp_rx_h_defrag_reo_reinject(struct ath11k *ar, struct dp_rx_ti
 
 	paddr = dma_map_single(ab->dev, defrag_skb->data,
 			       defrag_skb->len + skb_tailroom(defrag_skb),
-			       DMA_FROM_DEVICE);
+			       DMA_TO_DEVICE);
 	if (dma_mapping_error(ab->dev, paddr))
 		return -ENOMEM;
 
Signature · 7 lines
2.25.1


-- 
ath11k mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/ath11k
#1 lore raw

PATCH 2/5 ath11k: Drop MSDU with length error in DP rx path

From: Baochen Qiang <[email protected]>

There are MSDUs whose length are invalid. For example, attackers may inject on purpose truncated A-MSDUs with invalid MSDU length.

Such MSDUs are marked with an err bit set in rx attention tlvs, so we can check and drop them.

Tested-on: QCA6390 hw2.0 PCI WLAN.HST.1.0.1-01740-QCAHSTSWPLZ_V2_TO_X86-1
Tested-on: WCN6855 hw2.0 PCI WLAN.HSP.1.1-01720.1-QCAHSPSWPL_V1_V2_SILICONZ_LITE-1

---
 drivers/net/wireless/ath/ath11k/dp_rx.c | 18 ++++++++++++++++++
 1 file changed, 18 insertions(+)
drivers/net/wireless/ath/ath11k/dp_rx.c +18 -0
index 0c27eead3e02..c50f70913583 100644
@@ -142,6 +142,18 @@ static u32 ath11k_dp_rx_h_attn_mpdu_err(struct rx_attention *attn)
 	return errmap;
 }
 
+static bool ath11k_dp_rx_h_attn_msdu_len_err(struct ath11k_base *ab,
+					     struct hal_rx_desc *desc)
+{
+	struct rx_attention *rx_attention;
+	u32 errmap;
+
+	rx_attention = ath11k_dp_rx_get_attention(ab, desc);
+	errmap = ath11k_dp_rx_h_attn_mpdu_err(rx_attention);
+
+	return errmap & DP_RX_MPDU_ERR_MSDU_LEN;
+}
+
 static u16 ath11k_dp_rx_h_msdu_start_msdu_len(struct ath11k_base *ab,
 					      struct hal_rx_desc *desc)
 {
@@ -2525,6 +2537,12 @@ static int ath11k_dp_rx_process_msdu(struct ath11k *ar,
 	}
 
 	rx_desc = (struct hal_rx_desc *)msdu->data;
+	if (ath11k_dp_rx_h_attn_msdu_len_err(ab, rx_desc)) {
+		ath11k_warn(ar->ab, "msdu len not valid\n");
+		ret = -EIO;
+		goto free_out;
+	}
+
 	lrx_desc = (struct hal_rx_desc *)last_buf->data;
 	rx_attention = ath11k_dp_rx_get_attention(ab, lrx_desc);
 	if (!ath11k_dp_rx_h_attn_msdu_done(rx_attention)) {
Signature · 1 line
2.25.1
#2 lore raw

Re: [PATCH 2/5] ath11k: Drop MSDU with length error in DP rx path

Quoted 7 lines · Jouni Malinen

There are MSDUs whose length are invalid. For example, attackers may inject on purpose truncated A-MSDUs with invalid MSDU length.

Such MSDUs are marked with an err bit set in rx attention tlvs, so we can check and drop them.

Quoted 6 lines

Tested-on: QCA6390 hw2.0 PCI WLAN.HST.1.0.1-01740-QCAHSTSWPLZ_V2_TO_X86-1
Tested-on: WCN6855 hw2.0 PCI WLAN.HSP.1.1-01720.1-QCAHSPSWPL_V1_V2_SILICONZ_LITE-1

3 patches applied to ath-next branch of ath.git, thanks.

cd18ed4cf805 ath11k: Drop MSDU with length error in DP rx path 8a0b899f169d ath11k: Fix inaccessible debug registers 72de799aa9e3 ath11k: Fix memory leak in ath11k_qmi_driver_event_work

Signature · 9 lines
https://patchwork.kernel.org/project/linux-wireless/patch/[email protected]/

https://wireless.wiki.kernel.org/en/developers/documentation/submittingpatches


-- 
ath11k mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/ath11k
#3 lore raw

Re: [PATCH 2/5] ath11k: Drop MSDU with length error in DP rx path

In reply to Jouni Malinen
Quoted 7 lines · Jouni Malinen

There are MSDUs whose length are invalid. For example, attackers may inject on purpose truncated A-MSDUs with invalid MSDU length.

Such MSDUs are marked with an err bit set in rx attention tlvs, so we can check and drop them.

Quoted 6 lines

Tested-on: QCA6390 hw2.0 PCI WLAN.HST.1.0.1-01740-QCAHSTSWPLZ_V2_TO_X86-1
Tested-on: WCN6855 hw2.0 PCI WLAN.HSP.1.1-01720.1-QCAHSPSWPL_V1_V2_SILICONZ_LITE-1

3 patches applied to ath-next branch of ath.git, thanks.

cd18ed4cf805 ath11k: Drop MSDU with length error in DP rx path 8a0b899f169d ath11k: Fix inaccessible debug registers 72de799aa9e3 ath11k: Fix memory leak in ath11k_qmi_driver_event_work

Signature · 3 lines
https://patchwork.kernel.org/project/linux-wireless/patch/[email protected]/

https://wireless.wiki.kernel.org/en/developers/documentation/submittingpatches
#4 lore raw

PATCH 3/5 ath11k: Fix inaccessible debug registers

In reply to Jouni Malinen

From: Baochen Qiang <[email protected]>

Current code clears debug registers after SOC global reset performed in ath11k_pci_sw_reset. However at that time those registers are not accessible due to reset, thus they are actually not cleared at all. For WCN6855, it may cause target fail to initialize. This issue can be fixed by moving clear action ahead.

In addition, on some specific platforms, need to add delay to wait those registers to become accessible.

Tested-on: WCN6855 hw2.0 PCI WLAN.HSP.1.1-01720.1-QCAHSPSWPL_V1_V2_SILICONZ_LITE-1

---
 drivers/net/wireless/ath/ath11k/pci.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
drivers/net/wireless/ath/ath11k/pci.c +3 -1
index 5abb38cc3b55..7b3bce0ba76e 100644
@@ -430,6 +430,8 @@ static void ath11k_pci_force_wake(struct ath11k_base *ab)
 
 static void ath11k_pci_sw_reset(struct ath11k_base *ab, bool power_on)
 {
+	mdelay(100);
+
 	if (power_on) {
 		ath11k_pci_enable_ltssm(ab);
 		ath11k_pci_clear_all_intrs(ab);
@@ -439,9 +441,9 @@ static void ath11k_pci_sw_reset(struct ath11k_base *ab, bool power_on)
 	}
 
 	ath11k_mhi_clear_vector(ab);
+	ath11k_pci_clear_dbg_registers(ab);
 	ath11k_pci_soc_global_reset(ab);
 	ath11k_mhi_set_mhictrl_reset(ab);
-	ath11k_pci_clear_dbg_registers(ab);
 }
 
 int ath11k_pci_get_msi_irq(struct device *dev, unsigned int vector)
Signature · 1 line
2.25.1
#5 lore raw

PATCH 4/5 ath11k: Fix memory leak in ath11k_qmi_driver_event_work

In reply to Jouni Malinen

From: Baochen Qiang <[email protected]>

The buffer pointed to by event is not freed in case ATH11K_FLAG_UNREGISTERING bit is set, resulting in memory leak, so fix it.

Tested-on: WCN6855 hw2.0 PCI WLAN.HSP.1.1-01720.1-QCAHSPSWPL_V1_V2_SILICONZ_LITE-1

Fixes d5c65159f289 ("ath11k: driver for Qualcomm IEEE 802.11ax devices")
---
 drivers/net/wireless/ath/ath11k/qmi.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
drivers/net/wireless/ath/ath11k/qmi.c +3 -1
index 7e82d03b0d87..0175e35849bf 100644
@@ -2753,8 +2753,10 @@ static void ath11k_qmi_driver_event_work(struct work_struct *work)
 		list_del(&event->list);
 		spin_unlock(&qmi->event_lock);
 
-		if (test_bit(ATH11K_FLAG_UNREGISTERING, &ab->dev_flags))
+		if (test_bit(ATH11K_FLAG_UNREGISTERING, &ab->dev_flags)) {
+			kfree(event);
 			return;
+		}
 
 		switch (event->type) {
 		case ATH11K_QMI_EVENT_SERVER_ARRIVE:
Signature · 1 line
2.25.1
#6 lore raw

PATCH 5/5 ath11k: Handle MSI enablement during rmmod and SSR

In reply to Jouni Malinen

From: Baochen Qiang <[email protected]>

When doing "rmmod ath11k_pci", ath11k performs global SOC reset and MHI reset, where 0 address access is captured by IOMMU. See log below:

...

[  133.953860] ath11k_pci 0000:02:00.0: setting mhi state: DEINIT(1)
[  133.959714] ath11k_pci 0000:02:00.0: AMD-Vi: Event logged [IO_PAGE_FAULT domain=0x000a address=0x0 flags=0x0020]
[  133.973854] ath11k_pci 0000:02:00.0: MHISTATUS 0xff04
[  133.974095] ath11k_pci 0000:02:00.0: AMD-Vi: Event logged [IO_PAGE_FAULT domain=0x000a address=0x0 flags=0x0020]

...

This issue is also observed in SSR process, cause a similar sequence as above is performed.

Such an invalid access occurs because, during rmmod or SSR, MSI address is cleared but HW MSI functionality not disabled, thus HW target is able to raise an MSI transaction with 0 as MSI address.

So it can be fixed by simply disabling MSI before reset. For SSR, since MSI functionality is still needed after target is brought back, we need to reenable it.

Also change naming of some interfaces related.

Tested-on: QCA6390 hw2.0 PCI WLAN.HST.1.0.1-01740-QCAHSTSWPLZ_V2_TO_X86-1
Tested-on: WCN6855 hw2.0 PCI WLAN.HSP.1.1-01720.1-QCAHSPSWPL_V1_V2_SILICONZ_LITE-1

---
 drivers/net/wireless/ath/ath11k/pci.c | 27 ++++++++++++++++++++++-----
 1 file changed, 22 insertions(+), 5 deletions(-)
drivers/net/wireless/ath/ath11k/pci.c +22 -5
index 7b3bce0ba76e..1094b53465bc 100644
@@ -855,7 +855,18 @@ static void ath11k_pci_ce_irqs_enable(struct ath11k_base *ab)
 	}
 }
 
-static int ath11k_pci_enable_msi(struct ath11k_pci *ab_pci)
+static void ath11k_pci_enable_msi(struct pci_dev *dev, bool enable)
+{
+	u16 control;
+
+	pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &control);
+	control &= ~PCI_MSI_FLAGS_ENABLE;
+	if (enable)
+		control |= PCI_MSI_FLAGS_ENABLE;
+	pci_write_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, control);
+}
+
+static int ath11k_pci_alloc_msi(struct ath11k_pci *ab_pci)
 {
 	struct ath11k_base *ab = ab_pci->ab;
 	const struct ath11k_msi_config *msi_config = ab_pci->msi_config;
@@ -876,6 +887,7 @@ static int ath11k_pci_enable_msi(struct ath11k_pci *ab_pci)
 		else
 			return num_vectors;
 	}
+	ath11k_pci_enable_msi(ab_pci->pdev, false);
 
 	msi_desc = irq_get_msi_desc(ab_pci->pdev->irq);
 	if (!msi_desc) {
@@ -898,7 +910,7 @@ static int ath11k_pci_enable_msi(struct ath11k_pci *ab_pci)
 	return ret;
 }
 
-static void ath11k_pci_disable_msi(struct ath11k_pci *ab_pci)
+static void ath11k_pci_free_msi(struct ath11k_pci *ab_pci)
 {
 	pci_free_irq_vectors(ab_pci->pdev);
 }
@@ -1019,6 +1031,8 @@ static int ath11k_pci_power_up(struct ath11k_base *ab)
 	 */
 	ath11k_pci_aspm_disable(ab_pci);
 
+	ath11k_pci_enable_msi(ab_pci->pdev, true);
+
 	ret = ath11k_mhi_start(ab_pci);
 	if (ret) {
 		ath11k_err(ab, "failed to start mhi: %d\n", ret);
@@ -1039,6 +1053,9 @@ static void ath11k_pci_power_down(struct ath11k_base *ab)
 	ath11k_pci_aspm_restore(ab_pci);
 
 	ath11k_pci_force_wake(ab_pci->ab);
+
+	ath11k_pci_enable_msi(ab_pci->pdev, false);
+
 	ath11k_mhi_stop(ab_pci);
 	clear_bit(ATH11K_PCI_FLAG_INIT_DONE, &ab_pci->flags);
 	ath11k_pci_sw_reset(ab_pci->ab, false);
@@ -1263,7 +1280,7 @@ static int ath11k_pci_probe(struct pci_dev *pdev,
 		goto err_pci_free_region;
 	}
 
-	ret = ath11k_pci_enable_msi(ab_pci);
+	ret = ath11k_pci_alloc_msi(ab_pci);
 	if (ret) {
 		ath11k_err(ab, "failed to enable msi: %d\n", ret);
 		goto err_pci_free_region;
@@ -1317,7 +1334,7 @@ static int ath11k_pci_probe(struct pci_dev *pdev,
 	ath11k_mhi_unregister(ab_pci);
 
 err_pci_disable_msi:
-	ath11k_pci_disable_msi(ab_pci);
+	ath11k_pci_free_msi(ab_pci);
 
 err_pci_free_region:
 	ath11k_pci_free_region(ab_pci);
@@ -1348,7 +1365,7 @@ static void ath11k_pci_remove(struct pci_dev *pdev)
 	ath11k_mhi_unregister(ab_pci);
 
 	ath11k_pci_free_irq(ab);
-	ath11k_pci_disable_msi(ab_pci);
+	ath11k_pci_free_msi(ab_pci);
 	ath11k_pci_free_region(ab_pci);
 
 	ath11k_hal_srng_deinit(ab);
Signature · 7 lines
2.25.1


-- 
ath11k mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/ath11k
#7 lore raw

Re: [PATCH 5/5] ath11k: Handle MSI enablement during rmmod and SSR

Quoted 44 lines · Jouni Malinen

From: Baochen Qiang <[email protected]>

When doing "rmmod ath11k_pci", ath11k performs global SOC reset and MHI reset, where 0 address access is captured by IOMMU. See log below:

...

[  133.953860] ath11k_pci 0000:02:00.0: setting mhi state: DEINIT(1)
[  133.959714] ath11k_pci 0000:02:00.0: AMD-Vi: Event logged [IO_PAGE_FAULT domain=0x000a address=0x0 flags=0x0020]
[  133.973854] ath11k_pci 0000:02:00.0: MHISTATUS 0xff04
[  133.974095] ath11k_pci 0000:02:00.0: AMD-Vi: Event logged [IO_PAGE_FAULT domain=0x000a address=0x0 flags=0x0020]

...

This issue is also observed in SSR process, cause a similar sequence as above is performed.

Such an invalid access occurs because, during rmmod or SSR, MSI address is cleared but HW MSI functionality not disabled, thus HW target is able to raise an MSI transaction with 0 as MSI address.

So it can be fixed by simply disabling MSI before reset. For SSR, since MSI functionality is still needed after target is brought back, we need to reenable it.

Also change naming of some interfaces related.

Tested-on: QCA6390 hw2.0 PCI WLAN.HST.1.0.1-01740-QCAHSTSWPLZ_V2_TO_X86-1
Tested-on: WCN6855 hw2.0 PCI WLAN.HSP.1.1-01720.1-QCAHSPSWPL_V1_V2_SILICONZ_LITE-1

---
 drivers/net/wireless/ath/ath11k/pci.c | 27 ++++++++++++++++++++++-----
 1 file changed, 22 insertions(+), 5 deletions(-)
drivers/net/wireless/ath/ath11k/pci.c +0 -1
index 7b3bce0ba76e..1094b53465bc 100644
@@ -855,7 +855,18 @@ static void ath11k_pci_ce_irqs_enable(struct ath11k_base *ab)
 	}
 }
 
-static int ath11k_pci_enable_msi(struct ath11k_pci *ab_pci)

drivers/net/wireless/ath/ath11k/pci.c static void ath11k_pci_ce_irqs_enable(struct ath11k_base *ab) · line 867

Quoted 10 lines
drivers/net/wireless/ath/ath11k/pci.c +10 -0
+static void ath11k_pci_enable_msi(struct pci_dev *dev, bool enable)
+{
+	u16 control;
+
+	pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &control);
+	control &= ~PCI_MSI_FLAGS_ENABLE;
+	if (enable)
+		control |= PCI_MSI_FLAGS_ENABLE;
+	pci_write_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, control);
+}

To make the function cleaner I renamed this to ath11k_pci_msi_config(), added an else branch and changed it to take structh ath11k_pci. I also added helpers ath11k_pci_msi_enable() and ath11k_pci_msi_disable().

Signature · 3 lines
https://patchwork.kernel.org/project/linux-wireless/list/

https://wireless.wiki.kernel.org/en/developers/documentation/submittingpatches
#8 lore raw

Re: [PATCH 5/5] ath11k: Handle MSI enablement during rmmod and SSR

In reply to Jouni Malinen
Quoted 22 lines · Jouni Malinen

When doing "rmmod ath11k_pci", ath11k performs global SOC reset and MHI reset, where 0 address access is captured by IOMMU. See log below:

...

[  133.953860] ath11k_pci 0000:02:00.0: setting mhi state: DEINIT(1)
[  133.959714] ath11k_pci 0000:02:00.0: AMD-Vi: Event logged [IO_PAGE_FAULT domain=0x000a address=0x0 flags=0x0020]
[  133.973854] ath11k_pci 0000:02:00.0: MHISTATUS 0xff04
[  133.974095] ath11k_pci 0000:02:00.0: AMD-Vi: Event logged [IO_PAGE_FAULT domain=0x000a address=0x0 flags=0x0020]

...

This issue is also observed in SSR process, cause a similar sequence as above is performed.

Such an invalid access occurs because, during rmmod or SSR, MSI address is cleared but HW MSI functionality not disabled, thus HW target is able to raise an MSI transaction with 0 as MSI address.

So it can be fixed by simply disabling MSI before reset. For SSR, since MSI functionality is still needed after target is brought back, we need to reenable it.

Quoted 8 lines

Also change naming of some interfaces related.

Tested-on: QCA6390 hw2.0 PCI WLAN.HST.1.0.1-01740-QCAHSTSWPLZ_V2_TO_X86-1
Tested-on: WCN6855 hw2.0 PCI WLAN.HSP.1.1-01720.1-QCAHSPSWPL_V1_V2_SILICONZ_LITE-1

Patch applied to ath-next branch of ath.git, thanks.

96527d527b27 ath11k: Handle MSI enablement during rmmod and SSR

Signature · 3 lines
https://patchwork.kernel.org/project/linux-wireless/patch/[email protected]/

https://wireless.wiki.kernel.org/en/developers/documentation/submittingpatches
#9 lore raw
In reply to Jouni Malinen
Quoted 6 lines
drivers/net/wireless/ath/ath11k/dp_rx.c +0 -0
index 90da56316e7e..0c27eead3e02 100644

drivers/net/wireless/ath/ath11k/dp_rx.c static int ath11k_dp_rx_h_defrag_reo_reinject(struct ath11k *ar, struct dp_rx_ti · line 3440

Quoted 8 lines
drivers/net/wireless/ath/ath11k/dp_rx.c +1 -1
@@ -3434,7 +3434,7 @@ static int ath11k_dp_rx_h_defrag_reo_reinject(struct ath11k *ar, struct dp_rx_ti
 
  	paddr = dma_map_single(ab->dev, defrag_skb->data,
  			       defrag_skb->len + skb_tailroom(defrag_skb),
-			       DMA_FROM_DEVICE);
+			       DMA_TO_DEVICE);
  	if (dma_mapping_error(ab->dev, paddr))
  		return -ENOMEM;
 

Need to update this line too?

err_unmap_dma:
    dma_unmap_single(ab->dev, paddr, defrag_skb->len + skb_tailroom(defrag_skb), DMA_FROM_DEVICE);

Thanks,

Peter

Signature · 3 lines
ath11k mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/ath11k
#10 lore raw
In reply to Jouni Malinen
Quoted 8 lines · Jouni Malinen

For fragmented packets, ath11k reassembles each fragment as a normal packet and then reinjects it into HW ring. In this case, the DMA direction should be DMA_TO_DEVICE, not DMA_FROM_DEVICE, otherwise invalid payload will be reinjected to HW and then delivered to host. What is more, since arbitrary memory could be allocated to the frame, we don't know what kind of data is contained in the buffer reinjected. Thus, as a bad result, private info may be leaked.

Quoted 6 lines

Note that this issue is only found on Intel platform.

Tested-on: QCA6390 hw2.0 PCI WLAN.HST.1.0.1-01740-QCAHSTSWPLZ_V2_TO_X86-1

Dropping due to the issue Peter found.

Patch set to Changes Requested.

Signature · 9 lines
https://patchwork.kernel.org/project/linux-wireless/patch/[email protected]/

https://wireless.wiki.kernel.org/en/developers/documentation/submittingpatches


-- 
ath11k mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/ath11k
#11 lore raw
In reply to Jouni Malinen
Quoted 8 lines · Jouni Malinen

For fragmented packets, ath11k reassembles each fragment as a normal packet and then reinjects it into HW ring. In this case, the DMA direction should be DMA_TO_DEVICE, not DMA_FROM_DEVICE, otherwise invalid payload will be reinjected to HW and then delivered to host. What is more, since arbitrary memory could be allocated to the frame, we don't know what kind of data is contained in the buffer reinjected. Thus, as a bad result, private info may be leaked.

Quoted 6 lines

Note that this issue is only found on Intel platform.

Tested-on: QCA6390 hw2.0 PCI WLAN.HST.1.0.1-01740-QCAHSTSWPLZ_V2_TO_X86-1

Dropping due to the issue Peter found.

Patch set to Changes Requested.

Signature · 3 lines
https://patchwork.kernel.org/project/linux-wireless/patch/[email protected]/

https://wireless.wiki.kernel.org/en/developers/documentation/submittingpatches