Skip to contentSkip to conversation

Fix several bugs in error handling during device probe

PATCH Superseded Jiang Liu 4 participants 18 messages Jan 5, 2025 – Jan 8, 2025

Subsystems (3)

v2 of 6 Superseded Latest is v6 →

lore
Share & tools
lore
author
#1 lore raw

PATCH v2 1/6 amdgpu: fix possible resource leakage in kfd_cleanup_nodes()

In reply to Jiang Liu

Fix possible resource leakage on error recovery path in function kgd2kfd_device_init().

---
 drivers/gpu/drm/amd/amdkfd/kfd_device.c | 9 +++++----
 1 file changed, 5 insertions(+), 4 deletions(-)
drivers/gpu/drm/amd/amdkfd/kfd_device.c +5 -4
index a29374c86405..fa5054940486 100644
@@ -898,15 +898,15 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
 		if (kfd->adev->xcp_mgr)
 			kfd_setup_interrupt_bitmap(node, i);
 
+		spin_lock_init(&node->watch_points_lock);
+
+		kfd->nodes[i] = node;
+
 		/* Initialize the KFD node */
 		if (kfd_init_node(node)) {
 			dev_err(kfd_device, "Error initializing KFD node\n");
 			goto node_init_error;
 		}
-
-		spin_lock_init(&node->watch_points_lock);
-
-		kfd->nodes[i] = node;
 	}
 
 	svm_range_set_max_pages(kfd->adev);
@@ -921,6 +921,7 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
 	goto out;
 
 node_init_error:
+	i++;
 node_alloc_error:
 	kfd_cleanup_nodes(kfd, i);
 	kfd_doorbell_fini(kfd);
Signature · 1 line
2.43.5
author
#7 lore raw

PATCH v2 2/6 amdgpu: clear adev->in_suspend flag when fails to suspend

In reply to Jiang Liu

Clear adev->in_suspend flag when fails to suspend, otherwise it will cause too much warnings like:

[ 1802.212027] ------------[ cut here ]------------
[ 1802.212028] WARNING: CPU: 97 PID: 11282 at drivers/gpu/drm/amd/amdgpu/amdgpu_object.c:452 amdgpu_bo_free_kernel+0xf9/0x120 [amdgpu]
[ 1802.212198] Modules linked in: amdgpu(E-) tcp_diag(E) inet_diag(E) rfkill(E) intel_rapl_msr(E) intel_rapl_common(E) intel_uncore_frequency(E) intel_uncore_frequency_common(E) i10nm_edac(E) nfit(E) x86_pkg_temp_thermal(E) intel_powerclamp(E) coretemp(E) kvm_intel(E) kvm(E) crct10dif_pclmul(E) crc32_pclmul(E) ghash_clmulni_intel(E) rapl(E) iTCO_wdt(E) pmt_telemetry(E) iTCO_vendor_support(E) pmt_class(E) intel_cstate(E) snd_hda_intel(E) ipmi_ssif(E) snd_intel_dspcfg(E) snd_hda_codec(E) snd_hda_core(E) snd_hwdep(E) snd_pcm(E) snd_timer(E) intel_uncore(E) snd(E) cdc_ether(E) pcspkr(E) i2c_i801(E) idxd(E) mei_me(E) usbnet(E) ses(E) isst_if_mmio(E) isst_if_mbox_pci(E) mii(E) joydev(E) soundcore(E) isst_if_common(E) idxd_bus(E) mei(E) enclosure(E) intel_vsec(E) i2c_smbus(E) i2c_ismt(E) sunrpc(E) acpi_power_meter(E) ipmi_si(E) acpi_ipmi(E) ipmi_devintf(E) acpi_pad(E) ipmi_msghandler(E) vfat(E) fat(E) sg(E) video(E) amdxcp(E) drm_ttm_helper(E) ttm(E) drm_exec(E) gpu_sched(E) drm_suballoc_helper(E) crc32c_intel(E) drm_buddy(E)
[ 1802.212223]  ast(E) drm_shmem_helper(E) drm_display_helper(E) i2c_algo_bit(E) drm_kms_helper(E) virtio_net(E) mpt3sas(E) drm(E) net_failover(E) ahci(E) raid_class(E) failover(E) libahci(E) scsi_transport_sas(E) dimlib(E) libata(E) i2c_core(E) wmi(E) pinctrl_emmitsburg(E) [last unloaded: amdgpu(E)]
[ 1802.212231] CPU: 97 PID: 11282 Comm: rmmod Kdump: loaded Tainted: G        W   E      6.10.0+ #2
[ 1802.212232] Hardware name: Alibaba Alibaba Cloud ECS/Alibaba Cloud ECS, BIOS 3.0.ES.AL.P.087.05 04/07/2024
[ 1802.212232] RIP: 0010:amdgpu_bo_free_kernel+0xf9/0x120 [amdgpu]
[ 1802.212401] Code: 00 00 00 4d 85 e4 74 08 49 c7 04 24 00 00 00 00 48 85 ed 74 08 48 c7 45 00 00 00 00 00 5b 5d 41 5c 41 5d 41 5e c3 cc cc cc cc <0f> 0b e9 3b ff ff ff 3d 00 fe ff ff 74 b3 49 8b be b0 10 ff ff 4c
[ 1802.212402] RSP: 0018:ffffbe0e92087cb0 EFLAGS: 00010202
[ 1802.212403] RAX: 0000000000000206 RBX: ffff9cbc06c37f10 RCX: 000000000020000d
[ 1802.212404] RDX: ffff9cbc06c37f18 RSI: ffff9cbc06c37f58 RDI: ffff9cbc06c37f10
[ 1802.212407] RBP: ffff9cbc06c37f18 R08: ffff9cba42388800 R09: 000000000020000d
[ 1802.212409] R10: 0000000000040000 R11: 0000000000000006 R12: ffff9cbc06c37f58
[ 1802.212410] R13: ffff9cba4238dc00 R14: ffff9cbc06c0ef50 R15: ffff9cbc06c00000
[ 1802.212411] FS:  00007f927dc4e740(0000) GS:ffff9db47e880000(0000) knlGS:0000000000000000
[ 1802.212412] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[ 1802.212413] CR2: 000056342b2be1a8 CR3: 00000003f1546005 CR4: 0000000000f70ef0
[ 1802.212413] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[ 1802.212414] DR3: 0000000000000000 DR6: 00000000fffe07f0 DR7: 0000000000000400
[ 1802.212415] PKRU: 55555554
[ 1802.212415] Call Trace:
[ 1802.212416]  <TASK>
[ 1802.212417]  ? __warn+0x83/0x130
[ 1802.212419]  ? amdgpu_bo_free_kernel+0xf9/0x120 [amdgpu]
[ 1802.212586]  ? __report_bug+0xea/0x100
[ 1802.212588]  ? report_bug+0x24/0x70
[ 1802.212589]  ? handle_bug+0x3c/0x70
[ 1802.212590]  ? exc_invalid_op+0x18/0x70
[ 1802.212591]  ? asm_exc_invalid_op+0x1a/0x20
[ 1802.212594]  ? amdgpu_bo_free_kernel+0xf9/0x120 [amdgpu]
[ 1802.212746]  amdgpu_ring_fini+0x91/0x120 [amdgpu]
[ 1802.212901]  amdgpu_jpeg_sw_fini+0xb2/0xe0 [amdgpu]
[ 1802.213106]  amdgpu_device_ip_fini.isra.0+0xb1/0x1c0 [amdgpu]
[ 1802.213247]  amdgpu_device_fini_sw+0x49/0x290 [amdgpu]
[ 1802.213413]  amdgpu_driver_release_kms+0x16/0x30 [amdgpu]
[ 1802.213576]  devm_drm_dev_init_release+0x4e/0x70 [drm]
[ 1802.213602]  release_nodes+0x35/0xb0
[ 1802.213605]  devres_release_all+0x8f/0xd0
[ 1802.213606]  device_unbind_cleanup+0xe/0x70
[ 1802.213609]  device_release_driver_internal+0x1bc/0x200
[ 1802.213611]  driver_detach+0x48/0x90
[ 1802.213613]  bus_remove_driver+0x6d/0xf0
[ 1802.213615]  pci_unregister_driver+0x2e/0xb0
[ 1802.213618]  amdgpu_exit+0x15/0x1c4 [amdgpu]
[ 1802.213861]  __do_sys_delete_module.constprop.0+0x176/0x310
[ 1802.213863]  do_syscall_64+0x5d/0x170
[ 1802.213866]  entry_SYSCALL_64_after_hwframe+0x76/0x7e
[ 1802.213869] RIP: 0033:0x7f927d6620cb
[ 1802.213870] Code: 73 01 c3 48 8b 0d a5 6d 19 00 f7 d8 64 89 01 48 83 c8 ff c3 66 2e 0f 1f 84 00 00 00 00 00 90 f3 0f 1e fa b8 b0 00 00 00 0f 05 <48> 3d 01 f0 ff ff 73 01 c3 48 8b 0d 75 6d 19 00 f7 d8 64 89 01 48
[ 1802.213871] RSP: 002b:00007fff031d9d78 EFLAGS: 00000206 ORIG_RAX: 00000000000000b0
[ 1802.213873] RAX: ffffffffffffffda RBX: 000055d8d6df69e0 RCX: 00007f927d6620cb
[ 1802.213873] RDX: 000000000000000a RSI: 0000000000000800 RDI: 000055d8d6df6a48
[ 1802.213874] RBP: 0000000000000000 R08: 0000000000000000 R09: 0000000000000000
[ 1802.213875] R10: 00007f927d7aaac0 R11: 0000000000000206 R12: 00007fff031d9fa0
[ 1802.213876] R13: 00007fff031da5f1 R14: 000055d8d6df62a0 R15: 000055d8d6df69e0
[ 1802.213878]  </TASK>
[ 1802.213879] ---[ end trace 0000000000000000 ]---
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 13 +++++++++----
 1 file changed, 9 insertions(+), 4 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +9 -4
index 0a121aab5c74..5ff53a3b9851 100644
@@ -4910,7 +4910,7 @@ int amdgpu_device_suspend(struct drm_device *dev, bool notify_clients)
 		amdgpu_virt_fini_data_exchange(adev);
 		r = amdgpu_virt_request_full_gpu(adev, false);
 		if (r)
-			return r;
+			goto error_out;
 	}
 
 	if (amdgpu_acpi_smart_shift_update(dev, AMDGPU_SS_DEV_D3))
@@ -4930,7 +4930,7 @@ int amdgpu_device_suspend(struct drm_device *dev, bool notify_clients)
 
 	r = amdgpu_device_evict_resources(adev);
 	if (r)
-		return r;
+		goto error_out;
 
 	amdgpu_ttm_set_buffer_funcs_status(adev, false);
 
@@ -4943,9 +4943,12 @@ int amdgpu_device_suspend(struct drm_device *dev, bool notify_clients)
 
 	r = amdgpu_dpm_notify_rlc_state(adev, false);
 	if (r)
-		return r;
+		goto error_out;
 
 	return 0;
+error_out:
+	adev->in_suspend = false;
+	return r;
 }
 
 /**
@@ -5007,8 +5010,10 @@ int amdgpu_device_resume(struct drm_device *dev, bool notify_clients)
 		amdgpu_virt_release_full_gpu(adev, true);
 	}
 
-	if (r)
+	if (r) {
+		adev->in_suspend = false;
 		return r;
+	}
 
 	/* Make sure IB tests flushed */
 	flush_delayed_work(&adev->delayed_init_work);
Signature · 1 line
2.43.5

Jiang Liu

author
#0 lore raw

v2 cover Fix several bugs in error handling during device probe

This patchset tries to fix several memory leakages/invalid memory accesses on error handling path during GPU driver loading/unloading. They applies to:
https://gitlab.freedesktop.org/agd5f/linux.git amd-staging-drm-next

v2:
1) rebased to https://gitlab.freedesktop.org/agd5f/linux.git branch amd-staging-drm-next.
2) removed the first patch, which is unnecessary.
3) add amdgpu_xcp_drm_dev_free() in patch 0003 to enhance amdxcp driver to better support device remove and error handling.
4) reworked patch 0005 to fix it in amdgpu instead of drm core.

Jiang Liu (6):
amdgpu: fix invalid memory access in kfd_cleanup_nodes()
amdgpu: clear adev->in_suspend flag when fails to suspend drm/amdxcp: introduce new API amdgpu_xcp_drm_dev_free()
amdgpu: fix use after free bug related to amdgpu_driver_release_kms()
amdgpu: fix invalid memory access in amdgpu_fence_driver_sw_fini()
amdgpu: get rid of false warnings caused by amdgpu_irq_put()

 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c  | 14 +++--
 drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c   | 13 +++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c     |  4 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h    |  1 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_xcp.c     | 11 +++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_xcp.h     |  1 +
 drivers/gpu/drm/amd/amdkfd/kfd_device.c     |  9 +--
 drivers/gpu/drm/amd/amdxcp/amdgpu_xcp_drv.c | 70 +++++++++++++++++----
 drivers/gpu/drm/amd/amdxcp/amdgpu_xcp_drv.h |  1 +
 9 files changed, 99 insertions(+), 25 deletions(-)
Signature · 1 line
2.43.5
author
#8 lore raw

PATCH v2 3/6 drm/amdxcp: introduce new API amdgpu_xcp_drm_dev_free()

Introduce new interface amdgpu_xcp_drm_dev_free() to free a specific drm_device crreated by amdgpu_xcp_drm_dev_alloc(), which will be used to do error recovery.

---
 drivers/gpu/drm/amd/amdgpu/amdgpu_xcp.c     | 11 +++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_xcp.h     |  1 +
 drivers/gpu/drm/amd/amdxcp/amdgpu_xcp_drv.c | 70 +++++++++++++++++----
 drivers/gpu/drm/amd/amdxcp/amdgpu_xcp_drv.h |  1 +
 4 files changed, 70 insertions(+), 13 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_xcp.c +10 -1
index e209b5e101df..401fbaa0b6b8 100644
@@ -359,6 +359,8 @@ int amdgpu_xcp_dev_register(struct amdgpu_device *adev,
 		ret = drm_dev_register(adev->xcp_mgr->xcp[i].ddev, ent->driver_data);
 		if (ret)
 			return ret;
+
+		adev->xcp_mgr->xcp[i].registered = true;
 	}
 
 	return 0;
@@ -376,12 +378,19 @@ void amdgpu_xcp_dev_unplug(struct amdgpu_device *adev)
 		if (!adev->xcp_mgr->xcp[i].ddev)
 			break;
 
+		// Restore and free the original drm_device.
 		p_ddev = adev->xcp_mgr->xcp[i].ddev;
-		drm_dev_unplug(p_ddev);
+		if (adev->xcp_mgr->xcp[i].registered) {
+			drm_dev_unplug(p_ddev);
+			adev->xcp_mgr->xcp[i].registered = false;
+		}
 		p_ddev->render->dev = adev->xcp_mgr->xcp[i].rdev;
 		p_ddev->primary->dev = adev->xcp_mgr->xcp[i].pdev;
 		p_ddev->driver =  adev->xcp_mgr->xcp[i].driver;
 		p_ddev->vma_offset_manager = adev->xcp_mgr->xcp[i].vma_offset_manager;
+		amdgpu_xcp_drm_dev_free(p_ddev);
+
+		adev->xcp_mgr->xcp[i].ddev = NULL;
 	}
 }
 
drivers/gpu/drm/amd/amdgpu/amdgpu_xcp.h +1 -0
index b63f53242c57..cd06a4a232be 100644
@@ -101,6 +101,7 @@ struct amdgpu_xcp {
 	uint8_t id;
 	uint8_t mem_id;
 	bool valid;
+	bool registered;
 	atomic_t	ref_cnt;
 	struct drm_device *ddev;
 	struct drm_device *rdev;
drivers/gpu/drm/amd/amdxcp/amdgpu_xcp_drv.c +58 -12

107 lines not on this page · raw message

drivers/gpu/drm/amd/amdxcp/amdgpu_xcp_drv.h +1 -0
index c1c4b679bf95..580a1602c8e3 100644
@@ -25,5 +25,6 @@
 #define _AMDGPU_XCP_DRV_H_
 
 int amdgpu_xcp_drm_dev_alloc(struct drm_device **ddev);
+void amdgpu_xcp_drm_dev_free(struct drm_device *ddev);
 void amdgpu_xcp_drv_release(void);
 #endif /* _AMDGPU_XCP_DRV_H_ */
Signature · 1 line
2.43.5
#12 lore raw

Re: [PATCH v2 4/6] amdgpu: fix use after free bug related to amdgpu_driver_release_kms()

In reply to Jiang Liu
Quoted 23 lines · Jiang Liu

If some GPU device failed to probe, rmmod amdgpu will trigger a use after free bug related to amdgpu_driver_release_kms() as:

2024-12-26 16:17:45 [16002.085540] BUG: kernel NULL pointer dereference, address: 0000000000000000
2024-12-26 16:17:45 [16002.093792] #PF: supervisor read access in kernel mode
2024-12-26 16:17:45 [16002.099993] #PF: error_code(0x0000) - not-present page
2024-12-26 16:17:45 [16002.106188] PGD 0 P4D 0
2024-12-26 16:17:45 [16002.109464] Oops: Oops: 0000 [#1] PREEMPT SMP NOPTI
2024-12-26 16:17:45 [16002.115372] CPU: 2 PID: 14375 Comm: rmmod Kdump: loaded Tainted: G        W   E      6.10.0+ #2
2024-12-26 16:17:45 [16002.125577] Hardware name: Alibaba Alibaba Cloud ECS/Alibaba Cloud ECS, BIOS 3.0.ES.AL.P.087.05 04/07/2024
2024-12-26 16:17:45 [16002.136858] RIP: 0010:drm_sched_fini+0x3f/0xe0 [gpu_sched]
2024-12-26 16:17:45 [16002.143463] Code: 60 c6 87 be 00 00 00 01 e8 ce e0 90 d8 48 8d bb 80 00 00 00 e8 c2 e0 90 d8 8b 43 20 85 c0 74 51 45 31 e4 48 8b
43 28 4d 63 ec <4a> 8b 2c e8 48 89 ef e8 f5 0e 59 d9 48 8b 45 10 48 8d 55 10 48 39
2024-12-26 16:17:45 [16002.164992] RSP: 0018:ffffb570dbbb7da8 EFLAGS: 00010246
2024-12-26 16:17:45 [16002.171316] RAX: 0000000000000000 RBX: ffff96b0fdadc878 RCX: 0000000000000000
2024-12-26 16:17:46 [16002.179784] RDX: 000fffffffe00000 RSI: 0000000000000000 RDI: ffff96b0fdadc8f8
2024-12-26 16:17:46 [16002.188252] RBP: ffff96b0fdadc800 R08: ffff97abbd035040 R09: ffffffff9ac52540
2024-12-26 16:17:46 [16002.196722] R10: 0000000000000000 R11: 0000000000000000 R12: 0000000000000000
2024-12-26 16:17:46 [16002.205179] R13: 0000000000000000 R14: ffff96b0fdadfc00 R15: 0000000000000000
2024-12-26 16:17:46 [16002.213648] FS:  00007f2737000740(0000) GS:ffff97abbd100000(0000) knlGS:0000000000000000
2024-12-26 16:17:46 [16002.223189] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
2024-12-26 16:17:46 [16002.230103] CR2: 0000000000000000 CR3: 000000011be3a005 CR4: 0000000000f70ef0
2024-12-26 16:17:46 [16002.238581] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
2024-12-26 16:17:46 [16002.247053] DR3: 0000000000000000 DR6: 00000000fffe07f0 DR7: 0000000000000400
Quoted 1 line

e024se+0x3c/0x90 [amdxcp]

This calltrace seems corrupted.
You can use this one from my test.

BUG: unable to handle page fault for address: 000000010000008f
#PF: supervisor read access in kernel mode
#PF: error_code(0x0000) - not-present page PGD 0 P4D 0
Oops: Oops: 0000 [#1] PREEMPT SMP NOPTI
CPU: 139 PID: 10172 Comm: rmmod Kdump: loaded Tainted: G            E      6.10.0 #5
RIP: 0010:amdgpu_fence_driver_sw_fini+0x3b/0xd0 [amdgpu]
Code: 00 41 55 49 89 fd 41 54 49 81 c5 78 07 01 00 41 bc ff ff ff ff 55 53 eb 09 49 83 c6 08 4d 39 ee 74 7d 49 8b 2e 48 85 ed 74 ef <80> 7d 28 00 74 e9 48 83 7d 78 00 74 09 48 8d 7d 78 e8 ff 65 e1 ff
RSP: 0018:ffffb8e77badfde0 EFLAGS: 00010202
RAX: ffffffffc09e13b0 RBX: ffffa1461a4287f0 RCX: 00000000000000c0
RDX: 00000000000000c0 RSI: 00000000000000c0 RDI: ffffa1461a4287f0
RBP: 0000000100000067 R08: 00000000000000c0 R09: ffffa146112a6c48
R10: 0000000000040000 R11: 0000000000000007 R12: 00000000ffffffff
R13: ffffa1461a438f68 R14: ffffa1461a438b88 R15: 0000000000000000
FS:  00007f98ed908740(0000) GS:ffffa141ffd80000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 000000010000008f CR3: 0000000127d40003 CR4: 0000000000f70ef0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000fffe07f0 DR7: 0000000000000400
PKRU: 55555554
Call Trace:
 <TASK>
 ? __die+0x24/0x70
 ? page_fault_oops+0x65/0x140
 ? exc_page_fault+0x68/0x140
 ? asm_exc_page_fault+0x26/0x30
 ? __pfx_amdgpu_driver_release_kms+0x10/0x10 [amdgpu]
 ? amdgpu_fence_driver_sw_fini+0x3b/0xd0 [amdgpu]
 amdgpu_device_fini_sw+0x16/0x220 [amdgpu]
 amdgpu_driver_release_kms+0x16/0x30 [amdgpu]
 devm_drm_dev_init_release+0x4e/0x70 [drm]
 release_nodes+0x35/0xb0
 devres_release_group+0xa0/0xf0
 amdgpu_xcp_drv_release+0x3c/0x90 [amdxcp]

shuo

Quoted 33 lines
2024-12-26 16:17:46 [16002.337645]  __do_sys_delete_module.constprop.0+0x176/0x310
2024-12-26 16:17:46 [16002.344324]  do_syscall_64+0x5d/0x170
2024-12-26 16:17:46 [16002.348858]  entry_SYSCALL_64_after_hwframe+0x76/0x7e
2024-12-26 16:17:46 [16002.354956] RIP: 0033:0x7f2736a620cb-12-26

Fix it by unplugging xcp drm devices when failed to probe GPU devices.

---
drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c | 4 +++-

1 file changed, 3 insertions(+), 1 deletion(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c +0 -0
index d2a046736edd..9ebc0d47d1cb 100644
@@ -165,8 +165,10 @@ int amdgpu_driver_load_kms(struct amdgpu_device *adev, unsigned long flags)
 		DRM_WARN("smart shift update failed\n");
 

out:

drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c +2 -1
-	if (r)
+	if (r) {
+		amdgpu_xcp_dev_unplug(adev);
		amdgpu_driver_unload_kms(dev);
drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c +1 -0
+	}
	return r;
}
Signature · 1 line
2.43.5
author
#17 lore raw

PATCH v2 6/6 amdgpu: get rid of false warnings caused by amdgpu_irq_put()

In reply to Jiang Liu

If error happens before amdgpu_fence_driver_hw_init() gets called during device probe, it will trigger a false warning in amdgpu_irq_put() as below:

[ 1209.300996] ------------[ cut here ]------------
[ 1209.301061] WARNING: CPU: 48 PID: 293 at /tmp/amd.Rc9jFrl7/amd/amdgpu/amdgpu_irq.c:633 amdgpu_irq_put+0x45/0x70 [amdgpu]
[ 1209.301062] Modules linked in: ...
[ 1209.301093] CPU: 48 PID: 293 Comm: kworker/48:1 Kdump: loaded Tainted: G        W  OE     5.10.134-17.2.al8.x86_64 #1
[ 1209.301094] Hardware name: Alibaba Alibaba Cloud ECS/Alibaba Cloud ECS, BIOS 3.0.ES.AL.P.087.05 04/07/2024
[ 1209.301095] Workqueue: events work_for_cpu_fn
[ 1209.301159] RIP: 0010:amdgpu_irq_put+0x45/0x70 [amdgpu]
[ 1209.301160] Code: 48 8b 4e 10 48 83 39 00 74 2c 89 d1 48 8d 04 88 8b 08 85 c9 74 14 f0 ff 08 b8 00 00 00 00 74 05 c3 cc cc cc cc e9 8b fd ff ff <0f> 0b b8 ea ff ff ff c3 cc cc cc cc b8 ea ff ff ff c3 cc cc cc cc
[ 1209.301162] RSP: 0018:ffffb08a99c8fd88 EFLAGS: 00010246
[ 1209.301162] RAX: ffff9efe1bcbf500 RBX: ffff9efe1cc3e400 RCX: 0000000000000000
[ 1209.301163] RDX: 0000000000000000 RSI: ffff9efe1cc3b108 RDI: ffff9efe1cc00000
[ 1209.301163] RBP: ffff9efe1cc10818 R08: 0000000000000001 R09: 000000000000000d
[ 1209.301164] R10: ffffb08a99c8fb48 R11: ffffffffa2068018 R12: ffff9efe1cc109d0
[ 1209.301164] R13: ffff9efe1cc00010 R14: ffff9efe1cc00000 R15: ffff9efe1cc3b108
[ 1209.301165] FS:  0000000000000000(0000) GS:ffff9ff9fce00000(0000) knlGS:0000000000000000
[ 1209.301165] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[ 1209.301165] CR2: 00007fd0f6e860d0 CR3: 0000010092baa003 CR4: 0000000002770ee0
[ 1209.301166] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[ 1209.301166] DR3: 0000000000000000 DR6: 00000000fffe07f0 DR7: 0000000000000400
[ 1209.301167] PKRU: 55555554
[ 1209.301167] Call Trace:
[ 1209.301225]  amdgpu_fence_driver_hw_fini+0xda/0x110 [amdgpu]
[ 1209.301284]  amdgpu_device_fini_hw+0xaf/0x200 [amdgpu]
[ 1209.301342]  amdgpu_driver_load_kms+0x7f/0xc0 [amdgpu]
[ 1209.301400]  amdgpu_pci_probe+0x1cd/0x4a0 [amdgpu]
[ 1209.301401]  local_pci_probe+0x40/0xa0
[ 1209.301402]  work_for_cpu_fn+0x13/0x20
[ 1209.301403]  process_one_work+0x1ad/0x380
[ 1209.301404]  worker_thread+0x1c8/0x310
[ 1209.301405]  ? process_one_work+0x380/0x380
[ 1209.301406]  kthread+0x118/0x140
[ 1209.301407]  ? __kthread_bind_mask+0x60/0x60
[ 1209.301408]  ret_from_fork+0x1f/0x30
[ 1209.301410] ---[ end trace 733f120fe2ab13e5 ]---
[ 1209.301418] ------------[ cut here ]------------
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c | 9 +++++++--
 drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h  | 1 +
 2 files changed, 8 insertions(+), 2 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c +7 -2
index b5e87b515139..0e41a535e05f 100644
@@ -614,9 +614,11 @@ void amdgpu_fence_driver_hw_fini(struct amdgpu_device *adev)
 
 		if (!drm_dev_is_unplugged(adev_to_drm(adev)) &&
 		    ring->fence_drv.irq_src &&
-		    amdgpu_fence_need_ring_interrupt_restore(ring))
+		    ring->fence_drv.irq_enabled) {
 			amdgpu_irq_put(adev, ring->fence_drv.irq_src,
 				       ring->fence_drv.irq_type);
+		        ring->fence_drv.irq_enabled = false;
+		}
 
 		del_timer_sync(&ring->fence_drv.fallback_timer);
 	}
@@ -693,9 +695,12 @@ void amdgpu_fence_driver_hw_init(struct amdgpu_device *adev)
 
 		/* enable the interrupt */
 		if (ring->fence_drv.irq_src &&
-		    amdgpu_fence_need_ring_interrupt_restore(ring))
+		    !ring->fence_drv.irq_enabled &&
+		    amdgpu_fence_need_ring_interrupt_restore(ring)) {
 			amdgpu_irq_get(adev, ring->fence_drv.irq_src,
 				       ring->fence_drv.irq_type);
+		        ring->fence_drv.irq_enabled = true;
+		}
 	}
 }
 
drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h +1 -0
index dee5a1b4e572..959d474a0516 100644
@@ -118,6 +118,7 @@ struct amdgpu_fence_driver {
 	uint32_t			sync_seq;
 	atomic_t			last_seq;
 	bool				initialized;
+	bool				irq_enabled;
 	struct amdgpu_irq_src		*irq_src;
 	unsigned			irq_type;
 	struct timer_list		fallback_timer;
Signature · 1 line
2.43.5
#2 lore raw

Re: [PATCH v2 1/6] amdgpu: fix possible resource leakage in kfd_cleanup_nodes()

In reply to Jiang Liu

Hi Gerry,

Quoted 36 lines · Jiang Liu

Fix possible resource leakage on error recovery path in function kgd2kfd_device_init().

---
drivers/gpu/drm/amd/amdkfd/kfd_device.c | 9 +++++----

1 file changed, 5 insertions(+), 4 deletions(-)
drivers/gpu/drm/amd/amdkfd/kfd_device.c +4 -4
index a29374c86405..fa5054940486 100644
@@ -898,15 +898,15 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
 		if (kfd->adev->xcp_mgr)
 			kfd_setup_interrupt_bitmap(node, i);
 
+		spin_lock_init(&node->watch_points_lock);
+
+		kfd->nodes[i] = node;
+
 		/* Initialize the KFD node */
 		if (kfd_init_node(node)) {
 			dev_err(kfd_device, "Error initializing KFD node\n");
 			goto node_init_error;
 		}
-
-		spin_lock_init(&node->watch_points_lock);
-
-		kfd->nodes[i] = node;
 	}
 
 	svm_range_set_max_pages(kfd->adev);
@@ -921,6 +921,7 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
 	goto out;
 

node_init_error:

drivers/gpu/drm/amd/amdkfd/kfd_device.c bool kgd2kfd_device_init(struct kfd_dev *kfd, · line 924

Quoted 1 line
drivers/gpu/drm/amd/amdkfd/kfd_device.c +1 -0
+	i++;

The err cleanup path can release node itself. So the following kfd_cleanup_nodes() may do a double free?

Quoted 3 lines

node_alloc_error:

	kfd_cleanup_nodes(kfd, i);
	kfd_doorbell_fini(kfd);

shuo

author
#15 lore raw

PATCH v2 5/6 amdgpu: fix invalid memory access in amdgpu_fence_driver_sw_fini()

In reply to Jiang Liu

Function detects initialization status by checking sched->ops, so set sched->ops to non-NULL just before return in function drm_sched_init() to avoid possible invalid memory access on error recover path.

---
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 1 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c  | 4 +++-
 2 files changed, 4 insertions(+), 1 deletion(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +1 -0
index 5ff53a3b9851..475ab635c699 100644
@@ -2857,6 +2857,7 @@ static int amdgpu_device_init_schedulers(struct amdgpu_device *adev)
 		if (r) {
 			DRM_ERROR("Failed to create scheduler on ring %s.\n",
 				  ring->name);
+			ring->sched.ops = NULL;
 			return r;
 		}
 		r = amdgpu_uvd_entity_init(adev, ring);
drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c +3 -1
index 2f24a6aa13bf..b5e87b515139 100644
@@ -656,8 +656,10 @@ void amdgpu_fence_driver_sw_fini(struct amdgpu_device *adev)
 		 * The natural check would be sched.ready, which is
 		 * set as drm_sched_init() finishes...
 		 */
-		if (ring->sched.ops)
+		if (ring->sched.ops) {
 			drm_sched_fini(&ring->sched);
+			ring->sched.ops = NULL;
+		}
 
 		for (j = 0; j <= ring->fence_drv.num_fences_mask; ++j)
 			dma_fence_put(ring->fence_drv.fences[j]);
Signature · 1 line
2.43.5
author
#3 lore raw

Re: [PATCH v2 1/6] amdgpu: fix possible resource leakage in kfd_cleanup_nodes()

In reply to Shuo Liu
Quoted 38 lines

2025年1月5日 13:22,Shuo Liu <[email protected]> 写道:

Hi Gerry,

Quoted 31 lines · Jiang Liu

Fix possible resource leakage on error recovery path in function kgd2kfd_device_init().

---
drivers/gpu/drm/amd/amdkfd/kfd_device.c | 9 +++++----

1 file changed, 5 insertions(+), 4 deletions(-)
drivers/gpu/drm/amd/amdkfd/kfd_device.c +4 -4
index a29374c86405..fa5054940486 100644
@@ -898,15 +898,15 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
 		if (kfd->adev->xcp_mgr)
 			kfd_setup_interrupt_bitmap(node, i);
 
+		spin_lock_init(&node->watch_points_lock);
+
+		kfd->nodes[i] = node;
+
 		/* Initialize the KFD node */
 		if (kfd_init_node(node)) {
 			dev_err(kfd_device, "Error initializing KFD node\n");
 			goto node_init_error;
 		}
-
-		spin_lock_init(&node->watch_points_lock);
-
-		kfd->nodes[i] = node;
 	}
 
Quoted 8 lines

drivers/gpu/drm/amd/amdkfd/kfd_device.c bool kgd2kfd_device_init(struct kfd_dev *kfd, · line 924

Quoted 6 lines
drivers/gpu/drm/amd/amdkfd/kfd_device.c +0 -0
 	svm_range_set_max_pages(kfd->adev);
@@ -921,6 +921,7 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
 	goto out;
 

node_init_error:

drivers/gpu/drm/amd/amdkfd/kfd_device.c +1 -0
+	i++;

The err cleanup path can release node itself. So the following kfd_cleanup_nodes() may do a double free?

The code in function kgd2kfd_device_exit() checks flag kfd->init_complete.
And if error happens within function kgd2kfd_device_init(), the flag kfd->init_complete will be set to false, thus avoid double free.

Quoted 4 lines
Quoted 3 lines

node_alloc_error:

	kfd_cleanup_nodes(kfd, i);
	kfd_doorbell_fini(kfd);

shuo

author
#11 lore raw

PATCH v2 4/6 amdgpu: fix use after free bug related to amdgpu_driver_release_kms()

In reply to Jiang Liu

If some GPU device failed to probe, rmmod amdgpu will trigger a use after free bug related to amdgpu_driver_release_kms() as:

2024-12-26 16:17:45 [16002.085540] BUG: kernel NULL pointer dereference, address: 0000000000000000
2024-12-26 16:17:45 [16002.093792] #PF: supervisor read access in kernel mode
2024-12-26 16:17:45 [16002.099993] #PF: error_code(0x0000) - not-present page
2024-12-26 16:17:45 [16002.106188] PGD 0 P4D 0
2024-12-26 16:17:45 [16002.109464] Oops: Oops: 0000 [#1] PREEMPT SMP NOPTI
2024-12-26 16:17:45 [16002.115372] CPU: 2 PID: 14375 Comm: rmmod Kdump: loaded Tainted: G        W   E      6.10.0+ #2
2024-12-26 16:17:45 [16002.125577] Hardware name: Alibaba Alibaba Cloud ECS/Alibaba Cloud ECS, BIOS 3.0.ES.AL.P.087.05 04/07/2024
2024-12-26 16:17:45 [16002.136858] RIP: 0010:drm_sched_fini+0x3f/0xe0 [gpu_sched]
2024-12-26 16:17:45 [16002.143463] Code: 60 c6 87 be 00 00 00 01 e8 ce e0 90 d8 48 8d bb 80 00 00 00 e8 c2 e0 90 d8 8b 43 20 85 c0 74 51 45 31 e4 48 8b
43 28 4d 63 ec <4a> 8b 2c e8 48 89 ef e8 f5 0e 59 d9 48 8b 45 10 48 8d 55 10 48 39
2024-12-26 16:17:45 [16002.164992] RSP: 0018:ffffb570dbbb7da8 EFLAGS: 00010246
2024-12-26 16:17:45 [16002.171316] RAX: 0000000000000000 RBX: ffff96b0fdadc878 RCX: 0000000000000000
2024-12-26 16:17:46 [16002.179784] RDX: 000fffffffe00000 RSI: 0000000000000000 RDI: ffff96b0fdadc8f8
2024-12-26 16:17:46 [16002.188252] RBP: ffff96b0fdadc800 R08: ffff97abbd035040 R09: ffffffff9ac52540
2024-12-26 16:17:46 [16002.196722] R10: 0000000000000000 R11: 0000000000000000 R12: 0000000000000000
2024-12-26 16:17:46 [16002.205179] R13: 0000000000000000 R14: ffff96b0fdadfc00 R15: 0000000000000000
2024-12-26 16:17:46 [16002.213648] FS:  00007f2737000740(0000) GS:ffff97abbd100000(0000) knlGS:0000000000000000
2024-12-26 16:17:46 [16002.223189] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
2024-12-26 16:17:46 [16002.230103] CR2: 0000000000000000 CR3: 000000011be3a005 CR4: 0000000000f70ef0
2024-12-26 16:17:46 [16002.238581] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
2024-12-26 16:17:46 [16002.247053] DR3: 0000000000000000 DR6: 00000000fffe07f0 DR7: 0000000000000400
e024se+0x3c/0x90 [amdxcp]
2024-12-26 16:17:46 [16002.337645]  __do_sys_delete_module.constprop.0+0x176/0x310
2024-12-26 16:17:46 [16002.344324]  do_syscall_64+0x5d/0x170
2024-12-26 16:17:46 [16002.348858]  entry_SYSCALL_64_after_hwframe+0x76/0x7e
2024-12-26 16:17:46 [16002.354956] RIP: 0033:0x7f2736a620cb-12-26

Fix it by unplugging xcp drm devices when failed to probe GPU devices.

---
 drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c +3 -1
index d2a046736edd..9ebc0d47d1cb 100644
@@ -165,8 +165,10 @@ int amdgpu_driver_load_kms(struct amdgpu_device *adev, unsigned long flags)
 		DRM_WARN("smart shift update failed\n");
 
 out:
-	if (r)
+	if (r) {
+		amdgpu_xcp_dev_unplug(adev);
 		amdgpu_driver_unload_kms(dev);
+	}
 
 	return r;
 }
Signature · 1 line
2.43.5
#9 lore raw

Re: [PATCH v2 3/6] drm/amdxcp: introduce new API amdgpu_xcp_drm_dev_free()

In reply to Jiang Liu
Quoted 63 lines · Jiang Liu

Introduce new interface amdgpu_xcp_drm_dev_free() to free a specific drm_device crreated by amdgpu_xcp_drm_dev_alloc(), which will be used to do error recovery.

---
 drivers/gpu/drm/amd/amdgpu/amdgpu_xcp.c     | 11 +++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_xcp.h     |  1 +
 drivers/gpu/drm/amd/amdxcp/amdgpu_xcp_drv.c | 70 +++++++++++++++++----
 drivers/gpu/drm/amd/amdxcp/amdgpu_xcp_drv.h |  1 +
 4 files changed, 70 insertions(+), 13 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_xcp.c +10 -1
index e209b5e101df..401fbaa0b6b8 100644
@@ -359,6 +359,8 @@ int amdgpu_xcp_dev_register(struct amdgpu_device *adev,
 		ret = drm_dev_register(adev->xcp_mgr->xcp[i].ddev, ent->driver_data);
 		if (ret)
 			return ret;
+
+		adev->xcp_mgr->xcp[i].registered = true;
 	}
 
 	return 0;
@@ -376,12 +378,19 @@ void amdgpu_xcp_dev_unplug(struct amdgpu_device *adev)
 		if (!adev->xcp_mgr->xcp[i].ddev)
 			break;
 
+		// Restore and free the original drm_device.
 		p_ddev = adev->xcp_mgr->xcp[i].ddev;
-		drm_dev_unplug(p_ddev);
+		if (adev->xcp_mgr->xcp[i].registered) {
+			drm_dev_unplug(p_ddev);
+			adev->xcp_mgr->xcp[i].registered = false;
+		}
 		p_ddev->render->dev = adev->xcp_mgr->xcp[i].rdev;
 		p_ddev->primary->dev = adev->xcp_mgr->xcp[i].pdev;
 		p_ddev->driver =  adev->xcp_mgr->xcp[i].driver;
 		p_ddev->vma_offset_manager = adev->xcp_mgr->xcp[i].vma_offset_manager;
+		amdgpu_xcp_drm_dev_free(p_ddev);
+
+		adev->xcp_mgr->xcp[i].ddev = NULL;
 	}
 }
 
drivers/gpu/drm/amd/amdgpu/amdgpu_xcp.h +1 -0
index b63f53242c57..cd06a4a232be 100644
@@ -101,6 +101,7 @@ struct amdgpu_xcp {
 	uint8_t id;
 	uint8_t mem_id;
 	bool valid;
+	bool registered;
 	atomic_t	ref_cnt;
 	struct drm_device *ddev;
 	struct drm_device *rdev;
drivers/gpu/drm/amd/amdxcp/amdgpu_xcp_drv.c +0 -0
index faed84172dd4..9058d71b4756 100644

drivers/gpu/drm/amd/amdxcp/amdgpu_xcp_drv.c static const struct drm_driver amdgpu_xcp_driver = { · line 48

Quoted 4 lines
drivers/gpu/drm/amd/amdxcp/amdgpu_xcp_drv.c +1 -0
@@ -45,18 +45,26 @@ static const struct drm_driver amdgpu_xcp_driver = {
 
 static int8_t pdev_num;
 static struct xcp_device *xcp_dev[MAX_XCP_PLATFORM_DEVICE];
+static struct mutex xcp_mutex;

I think this needs to be static DEFINE_MUTEX(xcp_mutex).

drivers/gpu/drm/amd/amdxcp/amdgpu_xcp_drv.c static const struct drm_driver amdgpu_xcp_driver = { · line 58

Quoted 11 lines
drivers/gpu/drm/amd/amdxcp/amdgpu_xcp_drv.c +3 -1
 int amdgpu_xcp_drm_dev_alloc(struct drm_device **ddev)
 {
 	struct platform_device *pdev;
 	struct xcp_device *pxcp_dev;
 	char dev_name[20];
-	int ret;
+	int ret, index;
 
+	mutex_lock(&xcp_mutex);
+	ret = -ENODEV;

Preference would be do this inside the below if() to associate the error with the condition.

Quoted 4 lines
drivers/gpu/drm/amd/amdxcp/amdgpu_xcp_drv.c +1 -1
 	if (pdev_num >= MAX_XCP_PLATFORM_DEVICE)
-		return -ENODEV;
+		goto out_unlock;
 

drivers/gpu/drm/amd/amdxcp/amdgpu_xcp_drv.c static const struct drm_driver amdgpu_xcp_driver = { · line 70

Quoted 10 lines
drivers/gpu/drm/amd/amdxcp/amdgpu_xcp_drv.c +6 -1
-	snprintf(dev_name, sizeof(dev_name), "amdgpu_xcp_%d", pdev_num);
+	for (index = 0; index < MAX_XCP_PLATFORM_DEVICE; index++) {
+		if (!xcp_dev[index])
+			break;
+	}
+
+	snprintf(dev_name, sizeof(dev_name), "amdgpu_xcp_%d", index);
 	pdev = platform_device_register_simple(dev_name, -1, NULL, 0);
 	if (IS_ERR(pdev))
 		return PTR_ERR(pdev);

Seems mutex is left locked here.

Quoted 61 lines
drivers/gpu/drm/amd/amdxcp/amdgpu_xcp_drv.c +33 -8
@@ -72,10 +80,11 @@ int amdgpu_xcp_drm_dev_alloc(struct drm_device **ddev)
 		goto out_devres;
 	}
 
-	xcp_dev[pdev_num] = pxcp_dev;
-	xcp_dev[pdev_num]->pdev = pdev;
+	xcp_dev[index] = pxcp_dev;
+	xcp_dev[index]->pdev = pdev;
 	*ddev = &pxcp_dev->drm;
 	pdev_num++;
+	mutex_unlock(&xcp_mutex);
 
 	return 0;
 
@@ -83,21 +92,58 @@ int amdgpu_xcp_drm_dev_alloc(struct drm_device **ddev)
 	devres_release_group(&pdev->dev, NULL);
 out_unregister:
 	platform_device_unregister(pdev);
+out_unlock:
+	mutex_unlock(&xcp_mutex);
 
 	return ret;
 }
 EXPORT_SYMBOL(amdgpu_xcp_drm_dev_alloc);
 
-void amdgpu_xcp_drv_release(void)
+static void amdgpu_xcp_drm_dev_destroy(int index)
+{
+	struct platform_device *pdev;
+
+	pdev = xcp_dev[index]->pdev;
+	devres_release_group(&pdev->dev, NULL);
+	platform_device_unregister(pdev);
+	xcp_dev[index] = NULL;
+	pdev_num--;
+}
+
+void amdgpu_xcp_drm_dev_free(struct drm_device *ddev)
 {
-	for (--pdev_num; pdev_num >= 0; --pdev_num) {
-		struct platform_device *pdev = xcp_dev[pdev_num]->pdev;
+	int index;
+	struct xcp_device *pxcp_dev;
+
+	if (ddev == NULL)
+		return;
 
-		devres_release_group(&pdev->dev, NULL);
-		platform_device_unregister(pdev);
-		xcp_dev[pdev_num] = NULL;
+	pxcp_dev = container_of(ddev, struct xcp_device, drm);
+
+	mutex_lock(&xcp_mutex);
+	for (index = 0; index < MAX_XCP_PLATFORM_DEVICE; index++) {
+		if (xcp_dev[index] == pxcp_dev) {
+			amdgpu_xcp_drm_dev_destroy(index);
+			break;
+		}
+	}
+	mutex_unlock(&xcp_mutex);
+}

drivers/gpu/drm/amd/amdxcp/amdgpu_xcp_drv.c int amdgpu_xcp_drm_dev_alloc(struct drm_device **ddev) · line 141

Quoted 10 lines
drivers/gpu/drm/amd/amdxcp/amdgpu_xcp_drv.c +10 -0
+EXPORT_SYMBOL(amdgpu_xcp_drm_dev_free);
+
+void amdgpu_xcp_drv_release(void)
+{
+	int index;
+
+	mutex_lock(&xcp_mutex);
+	for (index = 0; index < MAX_XCP_PLATFORM_DEVICE; index++) {
+		if (xcp_dev[index]) {
+			WARN_ON(xcp_dev[index]);

Why is this WARN check needed? There is already a if() check for valid index.

Also, would suggest to separate out amdgpu_xcp.c from xcp_drv.c. xcp_drv introducing a new interface may be kept in a separate patch.

Thanks,
Lijo

Quoted 20 lines
drivers/gpu/drm/amd/amdxcp/amdgpu_xcp_drv.c +4 -1
+			amdgpu_xcp_drm_dev_destroy(index);
+		}
 	}
-	pdev_num = 0;
+	WARN_ON(pdev_num != 0);
+	mutex_unlock(&xcp_mutex);
 }
 EXPORT_SYMBOL(amdgpu_xcp_drv_release);
 
drivers/gpu/drm/amd/amdxcp/amdgpu_xcp_drv.h +1 -0
index c1c4b679bf95..580a1602c8e3 100644
@@ -25,5 +25,6 @@
 #define _AMDGPU_XCP_DRV_H_
 
 int amdgpu_xcp_drm_dev_alloc(struct drm_device **ddev);
+void amdgpu_xcp_drm_dev_free(struct drm_device *ddev);
 void amdgpu_xcp_drv_release(void);
 #endif /* _AMDGPU_XCP_DRV_H_ */
author
#4 lore raw

Re: [PATCH v2 1/6] amdgpu: fix possible resource leakage in kfd_cleanup_nodes()

In reply to Shuo Liu
Quoted 38 lines

2025年1月5日 13:22,Shuo Liu <[email protected]> 写道:

Hi Gerry,

Quoted 31 lines · Jiang Liu

Fix possible resource leakage on error recovery path in function kgd2kfd_device_init().

---
drivers/gpu/drm/amd/amdkfd/kfd_device.c | 9 +++++----

1 file changed, 5 insertions(+), 4 deletions(-)
drivers/gpu/drm/amd/amdkfd/kfd_device.c +4 -4
index a29374c86405..fa5054940486 100644
@@ -898,15 +898,15 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
 		if (kfd->adev->xcp_mgr)
 			kfd_setup_interrupt_bitmap(node, i);
 
+		spin_lock_init(&node->watch_points_lock);
+
+		kfd->nodes[i] = node;
+
 		/* Initialize the KFD node */
 		if (kfd_init_node(node)) {
 			dev_err(kfd_device, "Error initializing KFD node\n");
 			goto node_init_error;
 		}
-
-		spin_lock_init(&node->watch_points_lock);
-
-		kfd->nodes[i] = node;
 	}
 
Quoted 8 lines

drivers/gpu/drm/amd/amdkfd/kfd_device.c bool kgd2kfd_device_init(struct kfd_dev *kfd, · line 924

Quoted 6 lines
drivers/gpu/drm/amd/amdkfd/kfd_device.c +0 -0
 	svm_range_set_max_pages(kfd->adev);
@@ -921,6 +921,7 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
 	goto out;
 

node_init_error:

drivers/gpu/drm/amd/amdkfd/kfd_device.c +1 -0
+	i++;

The err cleanup path can release node itself. So the following kfd_cleanup_nodes() may do a double free?

Hi Shuo,
I missed the kfree() in function kfd_init_node(), so this patch is wrong and should be dropped.

Thanks,
Gerry

Quoted 4 lines
Quoted 3 lines

node_alloc_error:

	kfd_cleanup_nodes(kfd, i);
	kfd_doorbell_fini(kfd);

shuo

author
#10 lore raw

Re: [PATCH v2 3/6] drm/amdxcp: introduce new API amdgpu_xcp_drm_dev_free()

In reply to Lazar, Lijo
Quoted 96 lines

2025年1月6日 14:51,Lazar, Lijo <[email protected]> 写道:

Quoted 63 lines · Jiang Liu

Introduce new interface amdgpu_xcp_drm_dev_free() to free a specific drm_device crreated by amdgpu_xcp_drm_dev_alloc(), which will be used to do error recovery.

---

drivers/gpu/drm/amd/amdgpu/amdgpu_xcp.c     | 11 +++-
drivers/gpu/drm/amd/amdgpu/amdgpu_xcp.h     |  1 +
drivers/gpu/drm/amd/amdxcp/amdgpu_xcp_drv.c | 70 +++++++++++++++++----
drivers/gpu/drm/amd/amdxcp/amdgpu_xcp_drv.h |  1 +
4 files changed, 70 insertions(+), 13 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_xcp.c +10 -1
index e209b5e101df..401fbaa0b6b8 100644
@@ -359,6 +359,8 @@ int amdgpu_xcp_dev_register(struct amdgpu_device *adev,
 		ret = drm_dev_register(adev->xcp_mgr->xcp[i].ddev, ent->driver_data);
 		if (ret)
 			return ret;
+
+		adev->xcp_mgr->xcp[i].registered = true;
 	}
 
 	return 0;
@@ -376,12 +378,19 @@ void amdgpu_xcp_dev_unplug(struct amdgpu_device *adev)
 		if (!adev->xcp_mgr->xcp[i].ddev)
 			break;
 
+		// Restore and free the original drm_device.
 		p_ddev = adev->xcp_mgr->xcp[i].ddev;
-		drm_dev_unplug(p_ddev);
+		if (adev->xcp_mgr->xcp[i].registered) {
+			drm_dev_unplug(p_ddev);
+			adev->xcp_mgr->xcp[i].registered = false;
+		}
 		p_ddev->render->dev = adev->xcp_mgr->xcp[i].rdev;
 		p_ddev->primary->dev = adev->xcp_mgr->xcp[i].pdev;
 		p_ddev->driver =  adev->xcp_mgr->xcp[i].driver;
 		p_ddev->vma_offset_manager = adev->xcp_mgr->xcp[i].vma_offset_manager;
+		amdgpu_xcp_drm_dev_free(p_ddev);
+
+		adev->xcp_mgr->xcp[i].ddev = NULL;
 	}
 }
 
drivers/gpu/drm/amd/amdgpu/amdgpu_xcp.h +1 -0
index b63f53242c57..cd06a4a232be 100644
@@ -101,6 +101,7 @@ struct amdgpu_xcp {
 	uint8_t id;
 	uint8_t mem_id;
 	bool valid;
+	bool registered;
 	atomic_t	ref_cnt;
 	struct drm_device *ddev;
 	struct drm_device *rdev;
drivers/gpu/drm/amd/amdxcp/amdgpu_xcp_drv.c +0 -0
index faed84172dd4..9058d71b4756 100644

drivers/gpu/drm/amd/amdxcp/amdgpu_xcp_drv.c static const struct drm_driver amdgpu_xcp_driver = { · line 48

Quoted 4 lines
drivers/gpu/drm/amd/amdxcp/amdgpu_xcp_drv.c +1 -0
@@ -45,18 +45,26 @@ static const struct drm_driver amdgpu_xcp_driver = {
 
 static int8_t pdev_num;
 static struct xcp_device *xcp_dev[MAX_XCP_PLATFORM_DEVICE];
+static struct mutex xcp_mutex;

I think this needs to be static DEFINE_MUTEX(xcp_mutex).

drivers/gpu/drm/amd/amdxcp/amdgpu_xcp_drv.c static const struct drm_driver amdgpu_xcp_driver = { · line 58

Quoted 11 lines
drivers/gpu/drm/amd/amdxcp/amdgpu_xcp_drv.c +3 -1
 int amdgpu_xcp_drm_dev_alloc(struct drm_device **ddev)
 {
 	struct platform_device *pdev;
 	struct xcp_device *pxcp_dev;
 	char dev_name[20];
-	int ret;
+	int ret, index;
 
+	mutex_lock(&xcp_mutex);
+	ret = -ENODEV;

Preference would be do this inside the below if() to associate the error with the condition.

Quoted 4 lines
drivers/gpu/drm/amd/amdxcp/amdgpu_xcp_drv.c +1 -1
 	if (pdev_num >= MAX_XCP_PLATFORM_DEVICE)
-		return -ENODEV;
+		goto out_unlock;
 
Quoted 12 lines

drivers/gpu/drm/amd/amdxcp/amdgpu_xcp_drv.c static const struct drm_driver amdgpu_xcp_driver = { · line 70

Quoted 10 lines
drivers/gpu/drm/amd/amdxcp/amdgpu_xcp_drv.c +6 -1
-	snprintf(dev_name, sizeof(dev_name), "amdgpu_xcp_%d", pdev_num);
+	for (index = 0; index < MAX_XCP_PLATFORM_DEVICE; index++) {
+		if (!xcp_dev[index])
+			break;
+	}
+
+	snprintf(dev_name, sizeof(dev_name), "amdgpu_xcp_%d", index);
 	pdev = platform_device_register_simple(dev_name, -1, NULL, 0);
 	if (IS_ERR(pdev))
 		return PTR_ERR(pdev);

Seems mutex is left locked here.

Will fixed in next version:)

Quoted 76 lines
Quoted 62 lines
drivers/gpu/drm/amd/amdxcp/amdgpu_xcp_drv.c +3 -2
@@ -72,10 +80,11 @@ int amdgpu_xcp_drm_dev_alloc(struct drm_device **ddev)
 		goto out_devres;
 	}
 
-	xcp_dev[pdev_num] = pxcp_dev;
-	xcp_dev[pdev_num]->pdev = pdev;
+	xcp_dev[index] = pxcp_dev;
+	xcp_dev[index]->pdev = pdev;
 	*ddev = &pxcp_dev->drm;
 	pdev_num++;
+	mutex_unlock(&xcp_mutex);
 
 	return 0;
 
@@ -83,21 +92,58 @@ int amdgpu_xcp_drm_dev_alloc(struct drm_device **ddev)
 	devres_release_group(&pdev->dev, NULL);

out_unregister:

	platform_device_unregister(pdev);
drivers/gpu/drm/amd/amdxcp/amdgpu_xcp_drv.c +2 -0
+out_unlock:
+	mutex_unlock(&xcp_mutex);
	return ret;
}

EXPORT_SYMBOL(amdgpu_xcp_drm_dev_alloc);

drivers/gpu/drm/amd/amdxcp/amdgpu_xcp_drv.c +12 -1
-void amdgpu_xcp_drv_release(void)
+static void amdgpu_xcp_drm_dev_destroy(int index)
+{
+	struct platform_device *pdev;
+
+	pdev = xcp_dev[index]->pdev;
+	devres_release_group(&pdev->dev, NULL);
+	platform_device_unregister(pdev);
+	xcp_dev[index] = NULL;
+	pdev_num--;
+}
+
+void amdgpu_xcp_drm_dev_free(struct drm_device *ddev)

{

drivers/gpu/drm/amd/amdxcp/amdgpu_xcp_drv.c +16 -5
-	for (--pdev_num; pdev_num >= 0; --pdev_num) {
-		struct platform_device *pdev = xcp_dev[pdev_num]->pdev;
+	int index;
+	struct xcp_device *pxcp_dev;
+
+	if (ddev == NULL)
+		return;
 
-		devres_release_group(&pdev->dev, NULL);
-		platform_device_unregister(pdev);
-		xcp_dev[pdev_num] = NULL;
+	pxcp_dev = container_of(ddev, struct xcp_device, drm);
+
+	mutex_lock(&xcp_mutex);
+	for (index = 0; index < MAX_XCP_PLATFORM_DEVICE; index++) {
+		if (xcp_dev[index] == pxcp_dev) {
+			amdgpu_xcp_drm_dev_destroy(index);
+			break;
+		}
+	}
+	mutex_unlock(&xcp_mutex);
+}

drivers/gpu/drm/amd/amdxcp/amdgpu_xcp_drv.c int amdgpu_xcp_drm_dev_alloc(struct drm_device **ddev) · line 141

Quoted 10 lines
drivers/gpu/drm/amd/amdxcp/amdgpu_xcp_drv.c +10 -0
+EXPORT_SYMBOL(amdgpu_xcp_drm_dev_free);
+
+void amdgpu_xcp_drv_release(void)
+{
+	int index;
+
+	mutex_lock(&xcp_mutex);
+	for (index = 0; index < MAX_XCP_PLATFORM_DEVICE; index++) {
+		if (xcp_dev[index]) {
+			WARN_ON(xcp_dev[index]);

Why is this WARN check needed? There is already a if() check for valid index.

Quoted 2 lines

Also, would suggest to separate out amdgpu_xcp.c from xcp_drv.c. xcp_drv introducing a new interface may be kept in a separate patch.

With new implementation, all xcp devices should have already be removed when amdgpu_xcp_drv_release() gets called, So hope to verify whether it works as expected.

Thanks!

Quoted 24 lines

Thanks,
Lijo

Quoted 20 lines
drivers/gpu/drm/amd/amdxcp/amdgpu_xcp_drv.c +2 -0
+			amdgpu_xcp_drm_dev_destroy(index);
+		}
	}
drivers/gpu/drm/amd/amdxcp/amdgpu_xcp_drv.c +2 -1
-	pdev_num = 0;
+	WARN_ON(pdev_num != 0);
+	mutex_unlock(&xcp_mutex);

}
EXPORT_SYMBOL(amdgpu_xcp_drv_release);

drivers/gpu/drm/amd/amdxcp/amdgpu_xcp_drv.h +1 -0
index c1c4b679bf95..580a1602c8e3 100644
@@ -25,5 +25,6 @@
 #define _AMDGPU_XCP_DRV_H_
 
 int amdgpu_xcp_drm_dev_alloc(struct drm_device **ddev);
+void amdgpu_xcp_drm_dev_free(struct drm_device *ddev);
 void amdgpu_xcp_drv_release(void);
 #endif /* _AMDGPU_XCP_DRV_H_ */
#5 lore raw

Re: [PATCH v2 1/6] amdgpu: fix possible resource leakage in kfd_cleanup_nodes()

In reply to Jiang Liu
Quoted 33 lines · Jiang Liu

Fix possible resource leakage on error recovery path in function kgd2kfd_device_init().

---
  drivers/gpu/drm/amd/amdkfd/kfd_device.c | 9 +++++----
  1 file changed, 5 insertions(+), 4 deletions(-)
drivers/gpu/drm/amd/amdkfd/kfd_device.c +4 -4
index a29374c86405..fa5054940486 100644
@@ -898,15 +898,15 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
  		if (kfd->adev->xcp_mgr)
  			kfd_setup_interrupt_bitmap(node, i);
 
+		spin_lock_init(&node->watch_points_lock);
+
+		kfd->nodes[i] = node;
+
  		/* Initialize the KFD node */
  		if (kfd_init_node(node)) {
  			dev_err(kfd_device, "Error initializing KFD node\n");
  			goto node_init_error;
  		}
-
-		spin_lock_init(&node->watch_points_lock);
-
-		kfd->nodes[i] = node;
  	}
 
  	svm_range_set_max_pages(kfd->adev);

drivers/gpu/drm/amd/amdkfd/kfd_device.c bool kgd2kfd_device_init(struct kfd_dev *kfd, · line 927

Quoted 7 lines
drivers/gpu/drm/amd/amdkfd/kfd_device.c +1 -0
@@ -921,6 +921,7 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
  	goto out;
 
  node_init_error:
+	i++;
  node_alloc_error:
  	kfd_cleanup_nodes(kfd, i);
  	kfd_doorbell_fini(kfd);

I think this change is not right: if kfd_init_node fail it does clean up for the kfd_node that it is initializing internally. kfd_cleanup_nodes does not need to clean up the kfd node i which failed to init, just clean up the kfd_nodes that were initialized previously.

Regard

Xiaogang

#13 lore raw

Re: [PATCH v2 4/6] amdgpu: fix use after free bug related to amdgpu_driver_release_kms()

In reply to Jiang Liu
Quoted 43 lines · Jiang Liu

If some GPU device failed to probe, rmmod amdgpu will trigger a use after free bug related to amdgpu_driver_release_kms() as:

2024-12-26 16:17:45 [16002.085540] BUG: kernel NULL pointer dereference, address: 0000000000000000
2024-12-26 16:17:45 [16002.093792] #PF: supervisor read access in kernel mode
2024-12-26 16:17:45 [16002.099993] #PF: error_code(0x0000) - not-present page
2024-12-26 16:17:45 [16002.106188] PGD 0 P4D 0
2024-12-26 16:17:45 [16002.109464] Oops: Oops: 0000 [#1] PREEMPT SMP NOPTI
2024-12-26 16:17:45 [16002.115372] CPU: 2 PID: 14375 Comm: rmmod Kdump: loaded Tainted: G        W   E      6.10.0+ #2
2024-12-26 16:17:45 [16002.125577] Hardware name: Alibaba Alibaba Cloud ECS/Alibaba Cloud ECS, BIOS 3.0.ES.AL.P.087.05 04/07/2024
2024-12-26 16:17:45 [16002.136858] RIP: 0010:drm_sched_fini+0x3f/0xe0 [gpu_sched]
2024-12-26 16:17:45 [16002.143463] Code: 60 c6 87 be 00 00 00 01 e8 ce e0 90 d8 48 8d bb 80 00 00 00 e8 c2 e0 90 d8 8b 43 20 85 c0 74 51 45 31 e4 48 8b
43 28 4d 63 ec <4a> 8b 2c e8 48 89 ef e8 f5 0e 59 d9 48 8b 45 10 48 8d 55 10 48 39
2024-12-26 16:17:45 [16002.164992] RSP: 0018:ffffb570dbbb7da8 EFLAGS: 00010246
2024-12-26 16:17:45 [16002.171316] RAX: 0000000000000000 RBX: ffff96b0fdadc878 RCX: 0000000000000000
2024-12-26 16:17:46 [16002.179784] RDX: 000fffffffe00000 RSI: 0000000000000000 RDI: ffff96b0fdadc8f8
2024-12-26 16:17:46 [16002.188252] RBP: ffff96b0fdadc800 R08: ffff97abbd035040 R09: ffffffff9ac52540
2024-12-26 16:17:46 [16002.196722] R10: 0000000000000000 R11: 0000000000000000 R12: 0000000000000000
2024-12-26 16:17:46 [16002.205179] R13: 0000000000000000 R14: ffff96b0fdadfc00 R15: 0000000000000000
2024-12-26 16:17:46 [16002.213648] FS:  00007f2737000740(0000) GS:ffff97abbd100000(0000) knlGS:0000000000000000
2024-12-26 16:17:46 [16002.223189] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
2024-12-26 16:17:46 [16002.230103] CR2: 0000000000000000 CR3: 000000011be3a005 CR4: 0000000000f70ef0
2024-12-26 16:17:46 [16002.238581] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
2024-12-26 16:17:46 [16002.247053] DR3: 0000000000000000 DR6: 00000000fffe07f0 DR7: 0000000000000400
e024se+0x3c/0x90 [amdxcp]
2024-12-26 16:17:46 [16002.337645]  __do_sys_delete_module.constprop.0+0x176/0x310
2024-12-26 16:17:46 [16002.344324]  do_syscall_64+0x5d/0x170
2024-12-26 16:17:46 [16002.348858]  entry_SYSCALL_64_after_hwframe+0x76/0x7e
2024-12-26 16:17:46 [16002.354956] RIP: 0033:0x7f2736a620cb-12-26

Fix it by unplugging xcp drm devices when failed to probe GPU devices.

---
  drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c | 4 +++-
  1 file changed, 3 insertions(+), 1 deletion(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c +0 -0
index d2a046736edd..9ebc0d47d1cb 100644

drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c int amdgpu_driver_load_kms(struct amdgpu_device *adev, unsigned long flags) · line 169

Quoted 6 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c +2 -1
@@ -165,8 +165,10 @@ int amdgpu_driver_load_kms(struct amdgpu_device *adev, unsigned long flags)
  		DRM_WARN("smart shift update failed\n");
 
  out:
-	if (r)
+	if (r) {
+		amdgpu_xcp_dev_unplug(adev);

You have made amdgpu_xcp_drm_dev_free, why still use amdgpu_xcp_dev_unplug here? I think you want undo amdgpu_xcp_drm_dev_alloc in error path. Why involve adev device unplug? It is a different scenario.

Regards

Xiaogang

Quoted 5 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c +1 -0
  		amdgpu_driver_unload_kms(dev);
+	}
 
  	return r;
  }
#16 lore raw

Re: [PATCH v2 5/6] amdgpu: fix invalid memory access in amdgpu_fence_driver_sw_fini()

In reply to Jiang Liu
Quoted 2 lines · Jiang Liu

Function detects initialization status by checking sched->ops, so set sched->ops to non-NULL just before return in function drm_sched_init()

This commit message is not what the change did: you set sched->ops to NULL just after return from drm_sched_init.

Other than that this change looks good to me.

Regards

Xiaogang

Quoted 36 lines

to avoid possible invalid memory access on error recover path.

---
  drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 1 +
  drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c  | 4 +++-
  2 files changed, 4 insertions(+), 1 deletion(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +1 -0
index 5ff53a3b9851..475ab635c699 100644
@@ -2857,6 +2857,7 @@ static int amdgpu_device_init_schedulers(struct amdgpu_device *adev)
  		if (r) {
  			DRM_ERROR("Failed to create scheduler on ring %s.\n",
  				  ring->name);
+			ring->sched.ops = NULL;
  			return r;
  		}
  		r = amdgpu_uvd_entity_init(adev, ring);
drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c +3 -1
index 2f24a6aa13bf..b5e87b515139 100644
@@ -656,8 +656,10 @@ void amdgpu_fence_driver_sw_fini(struct amdgpu_device *adev)
  		 * The natural check would be sched.ready, which is
  		 * set as drm_sched_init() finishes...
  		 */
-		if (ring->sched.ops)
+		if (ring->sched.ops) {
  			drm_sched_fini(&ring->sched);
+			ring->sched.ops = NULL;
+		}
 
  		for (j = 0; j <= ring->fence_drv.num_fences_mask; ++j)
  			dma_fence_put(ring->fence_drv.fences[j]);
author
#6 lore raw

Re: [PATCH v2 1/6] amdgpu: fix possible resource leakage in kfd_cleanup_nodes()

In reply to Chen, Xiaogang
Quoted 42 lines

2025年1月8日 06:53,Chen, Xiaogang <[email protected]> 写道:

Quoted 35 lines · Jiang Liu

Fix possible resource leakage on error recovery path in function kgd2kfd_device_init().

---
 drivers/gpu/drm/amd/amdkfd/kfd_device.c | 9 +++++----
 1 file changed, 5 insertions(+), 4 deletions(-)
drivers/gpu/drm/amd/amdkfd/kfd_device.c +4 -4
index a29374c86405..fa5054940486 100644
@@ -898,15 +898,15 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
 		if (kfd->adev->xcp_mgr)
 			kfd_setup_interrupt_bitmap(node, i);
 
+		spin_lock_init(&node->watch_points_lock);
+
+		kfd->nodes[i] = node;
+
 		/* Initialize the KFD node */
 		if (kfd_init_node(node)) {
 			dev_err(kfd_device, "Error initializing KFD node\n");
 			goto node_init_error;
 		}
-
-		spin_lock_init(&node->watch_points_lock);
-
-		kfd->nodes[i] = node;
 	}
 
 	svm_range_set_max_pages(kfd->adev);
@@ -921,6 +921,7 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
 	goto out;
 
Quoted 7 lines

drivers/gpu/drm/amd/amdkfd/kfd_device.c bool kgd2kfd_device_init(struct kfd_dev *kfd, · line 927

Quoted 5 lines
drivers/gpu/drm/amd/amdkfd/kfd_device.c +1 -0
 node_init_error:
+	i++;
 node_alloc_error:
 	kfd_cleanup_nodes(kfd, i);
 	kfd_doorbell_fini(kfd);

I think this change is not right: if kfd_init_node fail it does clean up for the kfd_node that it is initializing internally. kfd_cleanup_nodes does not need to clean up the kfd node i which failed to init, just clean up the kfd_nodes that were initialized previously.

Yes, I made a mistake here and will drop this patch in next version.

Thanks,
Gerry

Quoted 4 lines

Regard

Xiaogang

author
#14 lore raw

Re: [PATCH v2 4/6] amdgpu: fix use after free bug related to amdgpu_driver_release_kms()

In reply to Chen, Xiaogang
Quoted 52 lines

2025年1月8日 06:55,Chen, Xiaogang <[email protected]> 写道:

Quoted 45 lines · Jiang Liu

If some GPU device failed to probe, rmmod amdgpu will trigger a use after free bug related to amdgpu_driver_release_kms() as:

2024-12-26 16:17:45 [16002.085540] BUG: kernel NULL pointer dereference, address: 0000000000000000
2024-12-26 16:17:45 [16002.093792] #PF: supervisor read access in kernel mode
2024-12-26 16:17:45 [16002.099993] #PF: error_code(0x0000) - not-present page
2024-12-26 16:17:45 [16002.106188] PGD 0 P4D 0
2024-12-26 16:17:45 [16002.109464] Oops: Oops: 0000 [#1] PREEMPT SMP NOPTI
2024-12-26 16:17:45 [16002.115372] CPU: 2 PID: 14375 Comm: rmmod Kdump: loaded Tainted: G        W   E      6.10.0+ #2
2024-12-26 16:17:45 [16002.125577] Hardware name: Alibaba Alibaba Cloud ECS/Alibaba Cloud ECS, BIOS 3.0.ES.AL.P.087.05 04/07/2024
2024-12-26 16:17:45 [16002.136858] RIP: 0010:drm_sched_fini+0x3f/0xe0 [gpu_sched]
2024-12-26 16:17:45 [16002.143463] Code: 60 c6 87 be 00 00 00 01 e8 ce e0 90 d8 48 8d bb 80 00 00 00 e8 c2 e0 90 d8 8b 43 20 85 c0 74 51 45 31 e4 48 8b
43 28 4d 63 ec <4a> 8b 2c e8 48 89 ef e8 f5 0e 59 d9 48 8b 45 10 48 8d 55 10 48 39
2024-12-26 16:17:45 [16002.164992] RSP: 0018:ffffb570dbbb7da8 EFLAGS: 00010246
2024-12-26 16:17:45 [16002.171316] RAX: 0000000000000000 RBX: ffff96b0fdadc878 RCX: 0000000000000000
2024-12-26 16:17:46 [16002.179784] RDX: 000fffffffe00000 RSI: 0000000000000000 RDI: ffff96b0fdadc8f8
2024-12-26 16:17:46 [16002.188252] RBP: ffff96b0fdadc800 R08: ffff97abbd035040 R09: ffffffff9ac52540
2024-12-26 16:17:46 [16002.196722] R10: 0000000000000000 R11: 0000000000000000 R12: 0000000000000000
2024-12-26 16:17:46 [16002.205179] R13: 0000000000000000 R14: ffff96b0fdadfc00 R15: 0000000000000000
2024-12-26 16:17:46 [16002.213648] FS:  00007f2737000740(0000) GS:ffff97abbd100000(0000) knlGS:0000000000000000
2024-12-26 16:17:46 [16002.223189] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
2024-12-26 16:17:46 [16002.230103] CR2: 0000000000000000 CR3: 000000011be3a005 CR4: 0000000000f70ef0
2024-12-26 16:17:46 [16002.238581] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
2024-12-26 16:17:46 [16002.247053] DR3: 0000000000000000 DR6: 00000000fffe07f0 DR7: 0000000000000400
e024se+0x3c/0x90 [amdxcp]
2024-12-26 16:17:46 [16002.337645]  __do_sys_delete_module.constprop.0+0x176/0x310
2024-12-26 16:17:46 [16002.344324]  do_syscall_64+0x5d/0x170
2024-12-26 16:17:46 [16002.348858]  entry_SYSCALL_64_after_hwframe+0x76/0x7e
2024-12-26 16:17:46 [16002.354956] RIP: 0033:0x7f2736a620cb-12-26

Fix it by unplugging xcp drm devices when failed to probe GPU devices.

---
 drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c +0 -0
index d2a046736edd..9ebc0d47d1cb 100644
@@ -165,8 +165,10 @@ int amdgpu_driver_load_kms(struct amdgpu_device *adev, unsigned long flags)
 		DRM_WARN("smart shift update failed\n");
 
Quoted 6 lines

drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c int amdgpu_driver_load_kms(struct amdgpu_device *adev, unsigned long flags) · line 169

Quoted 4 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c +2 -1
 out:
-	if (r)
+	if (r) {
+		amdgpu_xcp_dev_unplug(adev);

You have made amdgpu_xcp_drm_dev_free, why still use amdgpu_xcp_dev_unplug here? I think you want undo amdgpu_xcp_drm_dev_alloc in error path. Why involve adev device unplug? It is a different scenario.

Hi xiaogang,
I didn’t get your point yet. Current work flow is:
amdgpu_pci_probe()

	-> amdgpu_driver_load_kms()
		-> amdgpu_device_init()
			->amdgpu_discovery_set_ip_blocks()
				-> amdgpu_xcp_mgr_init()
					-> amdgpu_xcp_alloc_dev()
	-> amdgpu_xcp_dev_register()

And amdgpu_xcp_dev_unplug() undos work done by amdgpu_xcp_dev_register() and part of amdgpu_xcp_mgr_ini().
How about splitting amdgpu_xcp_dev_unplug() into amdgpuxcp_dev_deregister() and amdgpu_xcp_mgr_fini(), then things should get much clear. And it seems the xcp_mgr object is leaked current, and there are resource leakage on error handling path of amdgpu_pci_probe. Above proposal may also help to fix those resource leakages.
Thanks,
Gerry

Quoted 9 lines

Regards

Xiaogang

Quoted 5 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c +1 -0
 		amdgpu_driver_unload_kms(dev);
+	}
 
 	return r;
 }