Skip to contentSkip to conversation

drm/amd/display: Avoid MPC infinite loop

PATCH New Sasha Levin 1 participant 11 messages Aug 25, 2022

Subsystems (6)
lore
Share & tools
lore
author maintainer
#1 lore raw

PATCH 2/11 drm/amd/display: For stereo keep "FLIP_ANY_FRAME"

In reply to Sasha Levin

From: Alvin Lee <[email protected]>

[ Upstream commit 84ef99c728079dfd21d6bc70b4c3e4af20602b3c ]

[Description]
Observed in stereomode that programming FLIP_LEFT_EYE can cause hangs. Keep FLIP_ANY_FRAME in stereo mode so the surface flip can take place before left or right eye

---
 drivers/gpu/drm/amd/display/dc/dcn30/dcn30_hubp.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/gpu/drm/amd/display/dc/dcn30/dcn30_hubp.c +1 -1
index af462fe4260d..b0fd8859bd2f 100644
@@ -86,7 +86,7 @@ bool hubp3_program_surface_flip_and_addr(
 			VMID, address->vmid);
 
 	if (address->type == PLN_ADDR_TYPE_GRPH_STEREO) {
-		REG_UPDATE(DCSURF_FLIP_CONTROL, SURFACE_FLIP_MODE_FOR_STEREOSYNC, 0x1);
+		REG_UPDATE(DCSURF_FLIP_CONTROL, SURFACE_FLIP_MODE_FOR_STEREOSYNC, 0);
 		REG_UPDATE(DCSURF_FLIP_CONTROL, SURFACE_FLIP_IN_STEREOSYNC, 0x1);
 
 	} else {
Signature · 1 line
2.35.1

Sasha Levin

author maintainer
#0 lore raw

PATCH 1/11 drm/amd/display: Avoid MPC infinite loop

From: Josip Pavic <[email protected]>

[ Upstream commit 8de297dc046c180651c0500f8611663ae1c3828a ]

[why]
In some cases MPC tree bottom pipe ends up point to itself. This causes iterating from top to bottom to hang the system in an infinite loop.

[how]
When looping to next MPC bottom pipe, check that the pointer is not same as current to avoid infinite loop.

---
 drivers/gpu/drm/amd/display/dc/dcn10/dcn10_mpc.c | 6 ++++++
 drivers/gpu/drm/amd/display/dc/dcn20/dcn20_mpc.c | 6 ++++++
 2 files changed, 12 insertions(+)
drivers/gpu/drm/amd/display/dc/dcn10/dcn10_mpc.c +6 -0
index 3fcd408e9103..855682590c1b 100644
@@ -125,6 +125,12 @@ struct mpcc *mpc1_get_mpcc_for_dpp(struct mpc_tree *tree, int dpp_id)
 	while (tmp_mpcc != NULL) {
 		if (tmp_mpcc->dpp_id == dpp_id)
 			return tmp_mpcc;
+
+		/* avoid circular linked list */
+		ASSERT(tmp_mpcc != tmp_mpcc->mpcc_bot);
+		if (tmp_mpcc == tmp_mpcc->mpcc_bot)
+			break;
+
 		tmp_mpcc = tmp_mpcc->mpcc_bot;
 	}
 	return NULL;
drivers/gpu/drm/amd/display/dc/dcn20/dcn20_mpc.c +6 -0
index 99cc095dc33c..a701ea56c0aa 100644
@@ -533,6 +533,12 @@ struct mpcc *mpc2_get_mpcc_for_dpp(struct mpc_tree *tree, int dpp_id)
 	while (tmp_mpcc != NULL) {
 		if (tmp_mpcc->dpp_id == 0xf || tmp_mpcc->dpp_id == dpp_id)
 			return tmp_mpcc;
+
+		/* avoid circular linked list */
+		ASSERT(tmp_mpcc != tmp_mpcc->mpcc_bot);
+		if (tmp_mpcc == tmp_mpcc->mpcc_bot)
+			break;
+
 		tmp_mpcc = tmp_mpcc->mpcc_bot;
 	}
 	return NULL;
Signature · 1 line
2.35.1
author maintainer
#2 lore raw

PATCH 3/11 drm/amd/display: clear optc underflow before turn off odm clock

From: Fudong Wang <[email protected]>

[ Upstream commit b2a93490201300a749ad261b5c5d05cb50179c44 ]

[Why]
After ODM clock off, optc underflow bit will be kept there always and clear not work. We need to clear that before clock off.

[How]
Clear that if have when clock off.

---
 drivers/gpu/drm/amd/display/dc/dcn10/dcn10_optc.c | 5 +++++
 1 file changed, 5 insertions(+)
drivers/gpu/drm/amd/display/dc/dcn10/dcn10_optc.c +5 -0
index 800be2693fac..963d72f96dca 100644
@@ -464,6 +464,11 @@ void optc1_enable_optc_clock(struct timing_generator *optc, bool enable)
 				OTG_CLOCK_ON, 1,
 				1, 1000);
 	} else  {
+
+		//last chance to clear underflow, otherwise, it will always there due to clock is off.
+		if (optc->funcs->is_optc_underflow_occurred(optc) == true)
+			optc->funcs->clear_optc_underflow(optc);
+
 		REG_UPDATE_2(OTG_CLOCK_CONTROL,
 				OTG_CLOCK_GATE_DIS, 0,
 				OTG_CLOCK_EN, 0);
Signature · 1 line
2.35.1
author maintainer
#5 lore raw

PATCH 6/11 drm/amd/pm: add missing ->fini_microcode interface for Sienna Cichlid

In reply to Sasha Levin

From: Evan Quan <[email protected]>

[ Upstream commit 0a2d922a5618377cdf8fa476351362733ef55342 ]

To avoid any potential memory leak.

---
 drivers/gpu/drm/amd/pm/swsmu/smu11/sienna_cichlid_ppt.c | 1 +
 1 file changed, 1 insertion(+)
drivers/gpu/drm/amd/pm/swsmu/smu11/sienna_cichlid_ppt.c +1 -0
index 8556c229ff59..49d7fa1d0842 100644
@@ -2759,6 +2759,7 @@ static const struct pptable_funcs sienna_cichlid_ppt_funcs = {
 	.dump_pptable = sienna_cichlid_dump_pptable,
 	.init_microcode = smu_v11_0_init_microcode,
 	.load_microcode = smu_v11_0_load_microcode,
+	.fini_microcode = smu_v11_0_fini_microcode,
 	.init_smc_tables = sienna_cichlid_init_smc_tables,
 	.fini_smc_tables = smu_v11_0_fini_smc_tables,
 	.init_power = smu_v11_0_init_power,
Signature · 1 line
2.35.1
author maintainer
#6 lore raw

PATCH 7/11 drm/amd/display: Fix pixel clock programming

In reply to Sasha Levin

From: Ilya Bakoulin <[email protected]>

[ Upstream commit 04fb918bf421b299feaee1006e82921d7d381f18 ]

[Why]
Some pixel clock values could cause HDMI TMDS SSCPs to be misaligned between different HDMI lanes when using YCbCr420 10-bit pixel format.

BIOS functions for transmitter/encoder control take pixel clock in kHz increments, whereas the function for setting the pixel clock is in 100Hz increments. Setting pixel clock to a value that is not on a kHz boundary will cause the issue.

[How]
Round pixel clock down to nearest kHz in 10/12-bpc cases.

---
 drivers/gpu/drm/amd/display/dc/dce/dce_clock_source.c | 2 ++
 1 file changed, 2 insertions(+)
drivers/gpu/drm/amd/display/dc/dce/dce_clock_source.c +2 -0
index bae3a146b2cc..89cc852cb27c 100644
@@ -546,9 +546,11 @@ static void dce112_get_pix_clk_dividers_helper (
 		switch (pix_clk_params->color_depth) {
 		case COLOR_DEPTH_101010:
 			actual_pixel_clock_100hz = (actual_pixel_clock_100hz * 5) >> 2;
+			actual_pixel_clock_100hz -= actual_pixel_clock_100hz % 10;
 			break;
 		case COLOR_DEPTH_121212:
 			actual_pixel_clock_100hz = (actual_pixel_clock_100hz * 6) >> 2;
+			actual_pixel_clock_100hz -= actual_pixel_clock_100hz % 10;
 			break;
 		case COLOR_DEPTH_161616:
 			actual_pixel_clock_100hz = actual_pixel_clock_100hz * 2;
Signature · 1 line
2.35.1
author maintainer
#7 lore raw

PATCH 8/11 drm/amdgpu: Increase tlb flush timeout for sriov

In reply to Sasha Levin

From: Dusica Milinkovic <[email protected]>

[ Upstream commit 373008bfc9cdb0f050258947fa5a095f0657e1bc ]

[Why]
During multi-vf executing benchmark (Luxmark) observed kiq error timeout. It happenes because all of VFs do the tlb invalidation at the same time. Although each VF has the invalidate register set, from hardware side the invalidate requests are queue to execute.

[How]
In case of 12 VF increase timeout on 12*100ms

---
 drivers/gpu/drm/amd/amdgpu/amdgpu.h    | 2 +-
 drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c | 3 ++-
 drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c  | 3 ++-
 3 files changed, 5 insertions(+), 3 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu.h +1 -1
index d949d6c52f24..ff5555353eb4 100644
@@ -283,7 +283,7 @@ enum amdgpu_kiq_irq {
 	AMDGPU_CP_KIQ_IRQ_DRIVER0 = 0,
 	AMDGPU_CP_KIQ_IRQ_LAST
 };
-
+#define SRIOV_USEC_TIMEOUT  1200000 /* wait 12 * 100ms for SRIOV */
 #define MAX_KIQ_REG_WAIT       5000 /* in usecs, 5ms */
 #define MAX_KIQ_REG_BAILOUT_INTERVAL   5 /* in msecs, 5ms */
 #define MAX_KIQ_REG_TRY 80 /* 20 -> 80 */
drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c +2 -1
index 150fa5258fb6..2aa9242c58ab 100644
@@ -371,6 +371,7 @@ static int gmc_v10_0_flush_gpu_tlb_pasid(struct amdgpu_device *adev,
 	uint32_t seq;
 	uint16_t queried_pasid;
 	bool ret;
+	u32 usec_timeout = amdgpu_sriov_vf(adev) ? SRIOV_USEC_TIMEOUT : adev->usec_timeout;
 	struct amdgpu_ring *ring = &adev->gfx.kiq.ring;
 	struct amdgpu_kiq *kiq = &adev->gfx.kiq;
 
@@ -389,7 +390,7 @@ static int gmc_v10_0_flush_gpu_tlb_pasid(struct amdgpu_device *adev,
 
 		amdgpu_ring_commit(ring);
 		spin_unlock(&adev->gfx.kiq.ring_lock);
-		r = amdgpu_fence_wait_polling(ring, seq, adev->usec_timeout);
+		r = amdgpu_fence_wait_polling(ring, seq, usec_timeout);
 		if (r < 1) {
 			dev_err(adev->dev, "wait for kiq fence error: %ld.\n", r);
 			return -ETIME;
drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c +2 -1
index 3a864041968f..1673bf3bae55 100644
@@ -839,6 +839,7 @@ static int gmc_v9_0_flush_gpu_tlb_pasid(struct amdgpu_device *adev,
 	uint32_t seq;
 	uint16_t queried_pasid;
 	bool ret;
+	u32 usec_timeout = amdgpu_sriov_vf(adev) ? SRIOV_USEC_TIMEOUT : adev->usec_timeout;
 	struct amdgpu_ring *ring = &adev->gfx.kiq.ring;
 	struct amdgpu_kiq *kiq = &adev->gfx.kiq;
 
@@ -878,7 +879,7 @@ static int gmc_v9_0_flush_gpu_tlb_pasid(struct amdgpu_device *adev,
 
 		amdgpu_ring_commit(ring);
 		spin_unlock(&adev->gfx.kiq.ring_lock);
-		r = amdgpu_fence_wait_polling(ring, seq, adev->usec_timeout);
+		r = amdgpu_fence_wait_polling(ring, seq, usec_timeout);
 		if (r < 1) {
 			dev_err(adev->dev, "wait for kiq fence error: %ld.\n", r);
 			up_read(&adev->reset_sem);
Signature · 1 line
2.35.1
author maintainer
#3 lore raw

PATCH 4/11 neigh: fix possible DoS due to net iface start/stop loop

In reply to Sasha Levin

From: "Denis V. Lunev" <[email protected]>

[ Upstream commit 66ba215cb51323e4e55e38fd5f250e0fae0cbc94 ]

Normal processing of ARP request (usually this is Ethernet broadcast packet) coming to the host is looking like the following:
* the packet comes to arp_process() call and is passed through routing procedure
* the request is put into the queue using pneigh_enqueue() if corresponding ARP record is not local (common case for container records on the host)
* the request is processed by timer (within 80 jiffies by default) and ARP reply is sent from the same arp_process() using NEIGH_CB(skb)->flags & LOCALLY_ENQUEUED condition (flag is set inside pneigh_enqueue())

And here the problem comes. Linux kernel calls pneigh_queue_purge() which destroys the whole queue of ARP requests on ANY network interface start/stop event through __neigh_ifdown().

This is actually not a problem within the original world as network interface start/stop was accessible to the host 'root' only, which could do more destructive things. But the world is changed and there are Linux containers available. Here container 'root' has an access to this API and could be considered as untrusted user in the hosting (container's) world.

Thus there is an attack vector to other containers on node when container's root will endlessly start/stop interfaces. We have observed similar situation on a real production node when docker container was doing such activity and thus other containers on the node become not accessible.

The patch proposed doing very simple thing. It drops only packets from the same namespace in the pneigh_queue_purge() where network interface state change is detected. This is enough to prevent the problem for the whole node preserving original semantics of the code.

v2:
- do del_timer_sync() if queue is empty after pneigh_queue_purge()
v3:
- rebase to net tree

Cc "David S. Miller" <[email protected]>
Cc Eric Dumazet <[email protected]>
Cc Jakub Kicinski <[email protected]>
Cc Paolo Abeni <[email protected]>
Cc Daniel Borkmann <[email protected]>
Cc David Ahern <[email protected]>
Cc Yajun Deng <[email protected]>
Cc Roopa Prabhu <[email protected]>
Cc Christian Brauner <[email protected]>
Cc Alexey Kuznetsov <[email protected]>
Cc Alexander Mikhalitsyn <[email protected]>
Cc Konstantin Khorenko <[email protected]>
---
 net/core/neighbour.c | 25 +++++++++++++++++--------
 1 file changed, 17 insertions(+), 8 deletions(-)
net/core/neighbour.c +17 -8
index 52a1c8725337..e7dcdad5876b 100644
@@ -280,14 +280,23 @@ static int neigh_del_timer(struct neighbour *n)
 	return 0;
 }
 
-static void pneigh_queue_purge(struct sk_buff_head *list)
+static void pneigh_queue_purge(struct sk_buff_head *list, struct net *net)
 {
+	unsigned long flags;
 	struct sk_buff *skb;
 
-	while ((skb = skb_dequeue(list)) != NULL) {
-		dev_put(skb->dev);
-		kfree_skb(skb);
+	spin_lock_irqsave(&list->lock, flags);
+	skb = skb_peek(list);
+	while (skb != NULL) {
+		struct sk_buff *skb_next = skb_peek_next(skb, list);
+		if (net == NULL || net_eq(dev_net(skb->dev), net)) {
+			__skb_unlink(skb, list);
+			dev_put(skb->dev);
+			kfree_skb(skb);
+		}
+		skb = skb_next;
 	}
+	spin_unlock_irqrestore(&list->lock, flags);
 }
 
 static void neigh_flush_dev(struct neigh_table *tbl, struct net_device *dev,
@@ -358,9 +367,9 @@ static int __neigh_ifdown(struct neigh_table *tbl, struct net_device *dev,
 	write_lock_bh(&tbl->lock);
 	neigh_flush_dev(tbl, dev, skip_perm);
 	pneigh_ifdown_and_unlock(tbl, dev);
-
-	del_timer_sync(&tbl->proxy_timer);
-	pneigh_queue_purge(&tbl->proxy_queue);
+	pneigh_queue_purge(&tbl->proxy_queue, dev_net(dev));
+	if (skb_queue_empty_lockless(&tbl->proxy_queue))
+		del_timer_sync(&tbl->proxy_timer);
 	return 0;
 }
 
@@ -1743,7 +1752,7 @@ int neigh_table_clear(int index, struct neigh_table *tbl)
 	/* It is not clean... Fix it to unload IPv6 module safely */
 	cancel_delayed_work_sync(&tbl->gc_work);
 	del_timer_sync(&tbl->proxy_timer);
-	pneigh_queue_purge(&tbl->proxy_queue);
+	pneigh_queue_purge(&tbl->proxy_queue, NULL);
 	neigh_ifdown(tbl, NULL);
 	if (atomic_read(&tbl->entries))
 		pr_crit("neighbour leakage\n");
Signature · 1 line
2.35.1
author maintainer
#4 lore raw

PATCH 5/11 s390/hypfs: avoid error message under KVM

In reply to Sasha Levin

From: Juergen Gross <[email protected]>

[ Upstream commit 7b6670b03641ac308aaa6fa2e6f964ac993b5ea3 ]

When booting under KVM the following error messages are issued:

hypfs.7f5705: The hardware system does not support hypfs hypfs.7a79f0: Initialization of hypfs failed with rc=-61

Demote the severity of first message from "error" to "info" and issue the second message only in other error cases.

---
 arch/s390/hypfs/hypfs_diag.c | 2 +-
 arch/s390/hypfs/inode.c      | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)
arch/s390/hypfs/hypfs_diag.c +1 -1
index f0bc4dc3e9bf..6511d15ace45 100644
@@ -437,7 +437,7 @@ __init int hypfs_diag_init(void)
 	int rc;
 
 	if (diag204_probe()) {
-		pr_err("The hardware system does not support hypfs\n");
+		pr_info("The hardware system does not support hypfs\n");
 		return -ENODATA;
 	}
 
arch/s390/hypfs/inode.c +1 -1
index 5c97f48cea91..ee919bfc8186 100644
@@ -496,9 +496,9 @@ static int __init hypfs_init(void)
 	hypfs_vm_exit();
 fail_hypfs_diag_exit:
 	hypfs_diag_exit();
+	pr_err("Initialization of hypfs failed with rc=%i\n", rc);
 fail_dbfs_exit:
 	hypfs_dbfs_exit();
-	pr_err("Initialization of hypfs failed with rc=%i\n", rc);
 	return rc;
 }
 device_initcall(hypfs_init)
Signature · 1 line
2.35.1
author maintainer
#8 lore raw

PATCH 9/11 netfilter: conntrack: NF_CONNTRACK_PROCFS should no longer default to y

In reply to Sasha Levin

From: Geert Uytterhoeven <[email protected]>

[ Upstream commit aa5762c34213aba7a72dc58e70601370805fa794 ]

NF_CONNTRACK_PROCFS was marked obsolete in commit 54b07dca68557b09 ("netfilter: provide config option to disable ancient procfs parts") in v3.3.

---
 net/netfilter/Kconfig | 1 -
 1 file changed, 1 deletion(-)
net/netfilter/Kconfig +0 -1
index 6bafd3876aff..8bf70ce03f95 100644
@@ -118,7 +118,6 @@ config NF_CONNTRACK_ZONES
 
 config NF_CONNTRACK_PROCFS
 	bool "Supply CT list in procfs (OBSOLETE)"
-	default y
 	depends on PROC_FS
 	help
 	This option enables for the list of known conntrack entries
Signature · 1 line
2.35.1
author maintainer
#9 lore raw

PATCH 10/11 btrfs: tree-checker: check for overlapping extent items

In reply to Sasha Levin

From: Josef Bacik <[email protected]>

[ Upstream commit 899b7f69f244e539ea5df1b4d756046337de44a5 ]

We're seeing a weird problem in production where we have overlapping extent items in the extent tree. It's unclear where these are coming from, and in debugging we realized there's no check in the tree checker for this sort of problem. Add a check to the tree-checker to make sure that the extents do not overlap each other.

---
 fs/btrfs/tree-checker.c | 25 +++++++++++++++++++++++--
 1 file changed, 23 insertions(+), 2 deletions(-)
fs/btrfs/tree-checker.c +23 -2
index 32f1b15b25dc..1aadf9a43ef3 100644
@@ -1189,7 +1189,8 @@ static void extent_err(const struct extent_buffer *eb, int slot,
 }
 
 static int check_extent_item(struct extent_buffer *leaf,
-			     struct btrfs_key *key, int slot)
+			     struct btrfs_key *key, int slot,
+			     struct btrfs_key *prev_key)
 {
 	struct btrfs_fs_info *fs_info = leaf->fs_info;
 	struct btrfs_extent_item *ei;
@@ -1400,6 +1401,26 @@ static int check_extent_item(struct extent_buffer *leaf,
 			   total_refs, inline_refs);
 		return -EUCLEAN;
 	}
+
+	if ((prev_key->type == BTRFS_EXTENT_ITEM_KEY) ||
+	    (prev_key->type == BTRFS_METADATA_ITEM_KEY)) {
+		u64 prev_end = prev_key->objectid;
+
+		if (prev_key->type == BTRFS_METADATA_ITEM_KEY)
+			prev_end += fs_info->nodesize;
+		else
+			prev_end += prev_key->offset;
+
+		if (unlikely(prev_end > key->objectid)) {
+			extent_err(leaf, slot,
+	"previous extent [%llu %u %llu] overlaps current extent [%llu %u %llu]",
+				   prev_key->objectid, prev_key->type,
+				   prev_key->offset, key->objectid, key->type,
+				   key->offset);
+			return -EUCLEAN;
+		}
+	}
+
 	return 0;
 }
 
@@ -1568,7 +1589,7 @@ static int check_leaf_item(struct extent_buffer *leaf,
 		break;
 	case BTRFS_EXTENT_ITEM_KEY:
 	case BTRFS_METADATA_ITEM_KEY:
-		ret = check_extent_item(leaf, key, slot);
+		ret = check_extent_item(leaf, key, slot, prev_key);
 		break;
 	case BTRFS_TREE_BLOCK_REF_KEY:
 	case BTRFS_SHARED_DATA_REF_KEY:
Signature · 1 line
2.35.1
author maintainer
#10 lore raw

PATCH 11/11 ftrace: Fix NULL pointer dereference in is_ftrace_trampoline when ftrace is dead

In reply to Sasha Levin

From: Yang Jihong <[email protected]>

[ Upstream commit c3b0f72e805f0801f05fa2aa52011c4bfc694c44 ]

ftrace_startup does not remove ops from ftrace_ops_list when ftrace_startup_enable fails:

register_ftrace_function
ftrace_startup

    __register_ftrace_function
      ...
      add_ftrace_ops(&ftrace_ops_list, ops)
      ...
    ...
    ftrace_startup_enable // if ftrace failed to modify, ftrace_disabled is set to 1
    ...

return 0 // ops is in the ftrace_ops_list.

When ftrace_disabled = 1, unregister_ftrace_function simply returns without doing anything:
unregister_ftrace_function
ftrace_shutdown

    if (unlikely(ftrace_disabled))
            return -ENODEV;  // return here, __unregister_ftrace_function is not executed,
                             // as a result, ops is still in the ftrace_ops_list
    __unregister_ftrace_function
    ...

If ops is dynamically allocated, it will be free later, in this case, is_ftrace_trampoline accesses NULL pointer:

is_ftrace_trampoline
ftrace_ops_trampoline
do_for_each_ftrace_op(op, ftrace_ops_list) // OOPS! op may be NULL!

Syzkaller reports as follows:

[ 1203.506103] BUG: kernel NULL pointer dereference, address: 000000000000010b
[ 1203.508039] #PF: supervisor read access in kernel mode
[ 1203.508798] #PF: error_code(0x0000) - not-present page
[ 1203.509558] PGD 800000011660b067 P4D 800000011660b067 PUD 130fb8067 PMD 0
[ 1203.510560] Oops: 0000 [#1] SMP KASAN PTI
[ 1203.511189] CPU: 6 PID: 29532 Comm: syz-executor.2 Tainted: G    B   W         5.10.0 #8
[ 1203.512324] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.14.0-0-g155821a1990b-prebuilt.qemu.org 04/01/2014
[ 1203.513895] RIP: 0010:is_ftrace_trampoline+0x26/0xb0
[ 1203.514644] Code: ff eb d3 90 41 55 41 54 49 89 fc 55 53 e8 f2 00 fd ff 48 8b 1d 3b 35 5d 03 e8 e6 00 fd ff 48 8d bb 90 00 00 00 e8 2a 81 26 00 <48> 8b ab 90 00 00 00 48 85 ed 74 1d e8 c9 00 fd ff 48 8d bb 98 00
[ 1203.518838] RSP: 0018:ffffc900012cf960 EFLAGS: 00010246
[ 1203.520092] RAX: 0000000000000000 RBX: 000000000000007b RCX: ffffffff8a331866
[ 1203.521469] RDX: 0000000000000000 RSI: 0000000000000008 RDI: 000000000000010b
[ 1203.522583] RBP: 0000000000000000 R08: 0000000000000000 R09: ffffffff8df18b07
[ 1203.523550] R10: fffffbfff1be3160 R11: 0000000000000001 R12: 0000000000478399
[ 1203.524596] R13: 0000000000000000 R14: ffff888145088000 R15: 0000000000000008
[ 1203.525634] FS:  00007f429f5f4700(0000) GS:ffff8881daf00000(0000) knlGS:0000000000000000
[ 1203.526801] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[ 1203.527626] CR2: 000000000000010b CR3: 0000000170e1e001 CR4: 00000000003706e0
[ 1203.528611] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[ 1203.529605] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400

Therefore, when ftrace_startup_enable fails, we need to rollback registration process and remove ops from ftrace_ops_list.

---
 kernel/trace/ftrace.c | 10 ++++++++++
 1 file changed, 10 insertions(+)
kernel/trace/ftrace.c +10 -0
index a63713dcd05d..d868df6f13c8 100644
@@ -2899,6 +2899,16 @@ int ftrace_startup(struct ftrace_ops *ops, int command)
 
 	ftrace_startup_enable(command);
 
+	/*
+	 * If ftrace is in an undefined state, we just remove ops from list
+	 * to prevent the NULL pointer, instead of totally rolling it back and
+	 * free trampoline, because those actions could cause further damage.
+	 */
+	if (unlikely(ftrace_disabled)) {
+		__unregister_ftrace_function(ops);
+		return -ENODEV;
+	}
+
 	ops->flags &= ~FTRACE_OPS_FL_ADDING;
 
 	return 0;
Signature · 1 line
2.35.1