- Linux-stable-mirror - lists.linaro.org

Re: [PATCH] unshare: Fix nsproxy leak on set_cred_ucounts() error path

by Alexey Gladkov

On Tue, Nov 18, 2025 at 02:45:50PM +0800, Pavel Tikhomirov wrote: > If unshare_nsproxy_namespaces() successfully creates the new_nsproxy, > but then set_cred_ucounts() fails, on its error path there is no cleanup > for new_nsproxy, so it is leaked. Let's fix that by freeing new_nsproxy > if it's not NULL on this error path. > > Fixes: 905ae01c4ae2a ("Add a reference to ucounts for each cred") > Signed-off-by: Pavel Tikhomirov <ptikhomirov(a)virtuozzo.com> Cc: stable(a)vger.kernel.org Acked-by: Alexey Gladkov <legion(a)kernel.org> > --- > kernel/fork.c | 5 ++++- > 1 file changed, 4 insertions(+), 1 deletion(-) > > diff --git a/kernel/fork.c b/kernel/fork.c > index 3da0f08615a95..6f7332e3e0c8c 100644 > --- a/kernel/fork.c > +++ b/kernel/fork.c > @@ -3133,8 +3133,11 @@ int ksys_unshare(unsigned long unshare_flags) > > if (new_cred) { > err = set_cred_ucounts(new_cred); > - if (err) > + if (err) { > + if (new_nsproxy) > + free_nsproxy(new_nsproxy); > goto bad_unshare_cleanup_cred; > + } > } > > if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) { > -- > 2.51.1 > -- Rgrds, legion

18 hours, 3 minutes

1
0
0 0

[PATCH] counter: interrupt-cnt: Drop IRQF_NO_THREAD flag

by A. Sverdlin

From: Alexander Sverdlin <alexander.sverdlin(a)siemens.com> An IRQ handler can either be IRQF_NO_THREAD or acquire spinlock_t, as CONFIG_PROVE_RAW_LOCK_NESTING warns: ============================= [ BUG: Invalid wait context ] 6.18.0-rc1+git... #1 ----------------------------- some-user-space-process/1251 is trying to lock: (&counter->events_list_lock){....}-{3:3}, at: counter_push_event [counter] other info that might help us debug this: context-{2:2} no locks held by some-user-space-process/.... stack backtrace: CPU: 0 UID: 0 PID: 1251 Comm: some-user-space-process 6.18.0-rc1+git... #1 PREEMPT Call trace: show_stack (C) dump_stack_lvl dump_stack __lock_acquire lock_acquire _raw_spin_lock_irqsave counter_push_event [counter] interrupt_cnt_isr [interrupt_cnt] __handle_irq_event_percpu handle_irq_event handle_simple_irq handle_irq_desc generic_handle_domain_irq gpio_irq_handler handle_irq_desc generic_handle_domain_irq gic_handle_irq call_on_irq_stack do_interrupt_handler el0_interrupt __el0_irq_handler_common el0t_64_irq_handler el0t_64_irq ... and Sebastian correctly points out. Remove IRQF_NO_THREAD as an alternative to switching to raw_spinlock_t, because the latter would limit all potential nested locks to raw_spinlock_t only. Cc: Sebastian Andrzej Siewior <bigeasy(a)linutronix.de> Cc: stable(a)vger.kernel.org Link: https://lore.kernel.org/all/20251117151314.xwLAZrWY@linutronix.de/ Fixes: a55ebd47f21f ("counter: add IRQ or GPIO based counter") Signed-off-by: Alexander Sverdlin <alexander.sverdlin(a)siemens.com> --- drivers/counter/interrupt-cnt.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-) diff --git a/drivers/counter/interrupt-cnt.c b/drivers/counter/interrupt-cnt.c index 6c0c1d2d7027d..e6100b5fb082e 100644 --- a/drivers/counter/interrupt-cnt.c +++ b/drivers/counter/interrupt-cnt.c @@ -229,8 +229,7 @@ static int interrupt_cnt_probe(struct platform_device *pdev) irq_set_status_flags(priv->irq, IRQ_NOAUTOEN); ret = devm_request_irq(dev, priv->irq, interrupt_cnt_isr, - IRQF_TRIGGER_RISING | IRQF_NO_THREAD, - dev_name(dev), counter); + IRQF_TRIGGER_RISING, dev_name(dev), counter); if (ret) return ret; -- 2.51.1

18 hours, 3 minutes

4
5
0 0

[PATCH] Bluetooth: HCI: Fix hci0 not release in usb disconnect process

by zhangchen200426＠163.com

From: zhangchen <zhangchen01(a)kylinos.cn> If hci_resume_dev before hci_unregister_dev, the hci command will timeout and the reference count of hdev will not reset to zero. Then the node "hci0" will not release. The output in question is as follows: [ 3391.553518][ 7] [T247244] Bluetooth: hci0: command 0x0c01 tx timeout [ 3391.553588][ 7] [T264732] Bluetooth: hci0: Opcode 0x0c01 failed: -110 [ 3393.569514][ 3] [T247244] Bluetooth: hci0: command 0x0c01 tx timeout [ 3393.569515][ 3] [T264732] Bluetooth: hci0: Opcode 0x0c1a failed: -110 [ 3393.709645][ 6] [T104579] usb 10-1: new full-speed USB device number 95 using xhci-hcd [ 3393.862194][ 6] [T104579] usb 10-1: New USB device found, idVendor=13d3, idProduct=3570, bcdDevice= 0.00 [ 3393.862205][ 6] [T104579] usb 10-1: New USB device strings: Mfr=1, Product=2, SerialNumber=3 [ 3393.862208][ 6] [T104579] usb 10-1: Product: Bluetooth Radio [ 3393.862210][ 6] [T104579] usb 10-1: Manufacturer: Realtek [ 3393.862212][ 6] [T104579] usb 10-1: SerialNumber: 00e04c000001 [ 3393.867589][ 6] [T247244] Bluetooth: hci1: RTL: examining hci_ver=0b hci_rev=000b lmp_ver=0b lmp_subver=8852 [ 3393.868573][ 6] [T247244] Bluetooth: hci1: RTL: rom_version status=0 version=1 [ 3393.868583][ 6] [T247244] Bluetooth: hci1: RTL: loading rtl_bt/rtl8852bu_fw.bin [ 3393.868672][ 6] [T247244] Bluetooth: hci1: RTL: loading rtl_bt/rtl8852bu_config.bin [ 3393.869699][ 6] [T247244] Bluetooth: hci1: RTL: cfg_sz 6, total sz 65603 The call sequence in question is as follows: usb disconnect: btusb_disconnect hci_unregister_dev hci_dev_set_flag hci_cmd_sync_clear hci_unregister_suspend_notifier hci_dev_do_close device_del device resume: hci_suspend_notifier hci_resume_dev hci_resume_sync hci_set_event_mask_sync __hci_cmd_sync_status __hci_cmd_sync_status_sk __hci_cmd_sync_sk wait_event_interruptible_timeout The output after adding debug information in question is as follows: [ 6378.366215][ 6] [T434033] hci_resume_dev hci name hci0 [ 6378.366218][ 6] [T434033] hci_resume_sync set event mask sync [ 6378.366219][ 6] [T434033] hci_set_event_mask_sync [ 6378.366220][ 6] [T434033] __hci_cmd_sync_sk hci name hci0 Opcode 0x0c01 [ 6378.366227][ 6] [T434033] __hci_cmd_sync_sk wait event interruptible timeout [ 6378.367632][ 6] [T420012] btusb_disconnect intf 0000000024117fc1 [ 6378.367637][ 6] [T420012] btusb_disconnect hci_unregister_dev [ 6378.367638][ 6] [T420012] hci_unregister_dev 0000000064bfd783 name hci0 bus 1 [ 6378.367641][ 6] [T420012] hci_unregister_dev set flag [ 6378.367804][ 6] [T420012] hci_unregister_dev cmd sync clear [ 6378.367807][ 6] [T420012] hci_unregister_dev unregister suspend notifier [ 6380.367544][ 6] [T434033] __hci_cmd_sync_sk cmd timeout [ 6380.367542][ 6] [T197498] Bluetooth: hci0: command 0x0c01 tx timeout [ 6380.367550][ 6] [T434033] __hci_cmd_sync_sk hci0 end: err -110 [ 6380.367552][ 6] [T434033] Bluetooth: hci0: Opcode 0x0c01 failed: -110 [ 6380.367555][ 6] [T434033] hci_resume_sync clear event filter [ 6380.367556][ 6] [T434033] hci_resume_sync resume scan sync [ 6380.367558][ 6] [T434033] __hci_cmd_sync_sk hci name hci0 Opcode 0x0c1a [ 6380.367561][ 6] [T434033] __hci_cmd_sync_sk wait event interruptible timeout [ 6382.383538][ 6] [T197498] Bluetooth: hci0: command 0x0c01 tx timeout [ 6382.383593][ 6] [T434033] __hci_cmd_sync_sk hci0 end: err -110 [ 6382.383597][ 6] [T434033] Bluetooth: hci0: Opcode 0x0c1a failed: -110 The output after adding debug information in normal is as follows: [50.039156][ 6] [ T8360] btusb_disconnect intf 00000000fca35842 [50.039160][ 6] [ T8360] btusb_disconnect hci_unregister_dev [50.039162][ 6] [ T8360] hci_unregister_dev 000000002422b946 name hci0 bus 1 [50.039164][ 6] [ T8360] hci_unregister_dev set flag [50.039224][ 6] [ T8360] hci_unregister_dev cmd sync clear [50.039227][ 5] [ T8360] hci_unregister_dev unregister suspend notifier [50.043542][ 5] [ T8284] hci_resume_dev hci name hci0 This patch add hci_cancel_cmd_sync in hci_unregister_dev to wake up hdev->req_wait_q, and stop __hci_cmd_sync_sk by judging the HCI_UNREGISTER flag. Then stopping hci_resume_dev process based on the returned error code. Fixes: 182ee45da083 ("Bluetooth: hci_sync: Rework hci_suspend_notifier") Cc: stable(a)vger.kernel.org Signed-off-by: zhangchen <zhangchen01(a)kylinos.cn> --- net/bluetooth/hci_core.c | 6 ++++++ net/bluetooth/hci_sync.c | 22 ++++++++++++++++------ 2 files changed, 22 insertions(+), 6 deletions(-) diff --git a/net/bluetooth/hci_core.c b/net/bluetooth/hci_core.c index 3418d7b964a1..c977bcba3e76 100644 --- a/net/bluetooth/hci_core.c +++ b/net/bluetooth/hci_core.c @@ -50,6 +50,7 @@ static void hci_rx_work(struct work_struct *work); static void hci_cmd_work(struct work_struct *work); static void hci_tx_work(struct work_struct *work); +static void hci_cancel_cmd_sync(struct hci_dev *hdev, int err); /* HCI device list */ LIST_HEAD(hci_dev_list); @@ -2695,6 +2696,8 @@ void hci_unregister_dev(struct hci_dev *hdev) hci_dev_set_flag(hdev, HCI_UNREGISTER); mutex_unlock(&hdev->unregister_lock); + hci_cancel_cmd_sync(hdev, EINTR); + write_lock(&hci_dev_list_lock); list_del(&hdev->list); write_unlock(&hci_dev_list_lock); @@ -2877,6 +2880,9 @@ int hci_resume_dev(struct hci_dev *hdev) ret = hci_resume_sync(hdev); hci_req_sync_unlock(hdev); + if (ret && hci_dev_test_flag(hdev, HCI_UNREGISTER)) + return 0; + mgmt_resuming(hdev, hdev->wake_reason, &hdev->wake_addr, hdev->wake_addr_type); diff --git a/net/bluetooth/hci_sync.c b/net/bluetooth/hci_sync.c index 6e76798ec786..f48d34fbfff2 100644 --- a/net/bluetooth/hci_sync.c +++ b/net/bluetooth/hci_sync.c @@ -174,10 +174,11 @@ struct sk_buff *__hci_cmd_sync_sk(struct hci_dev *hdev, u16 opcode, u32 plen, return ERR_PTR(err); err = wait_event_interruptible_timeout(hdev->req_wait_q, - hdev->req_status != HCI_REQ_PEND, + hdev->req_status != HCI_REQ_PEND || + hci_dev_test_flag(hdev, HCI_UNREGISTER), timeout); - if (err == -ERESTARTSYS) + if (err == -ERESTARTSYS || hci_dev_test_flag(hdev, HCI_UNREGISTER)) return ERR_PTR(-EINTR); switch (hdev->req_status) { @@ -6296,6 +6297,7 @@ static int hci_resume_scan_sync(struct hci_dev *hdev) */ int hci_resume_sync(struct hci_dev *hdev) { + int err; /* If not marked as suspended there nothing to do */ if (!hdev->suspended) return 0; @@ -6303,10 +6305,14 @@ int hci_resume_sync(struct hci_dev *hdev) hdev->suspended = false; /* Restore event mask */ - hci_set_event_mask_sync(hdev); + err = hci_set_event_mask_sync(hdev); + if (err && hci_dev_test_flag(hdev, HCI_UNREGISTER)) + return err; /* Clear any event filters and restore scan state */ - hci_clear_event_filter_sync(hdev); + err = hci_clear_event_filter_sync(hdev); + if (err && hci_dev_test_flag(hdev, HCI_UNREGISTER)) + return err; /* Resume scanning */ hci_resume_scan_sync(hdev); @@ -6315,10 +6321,14 @@ int hci_resume_sync(struct hci_dev *hdev) hci_resume_monitor_sync(hdev); /* Resume other advertisements */ - hci_resume_advertising_sync(hdev); + err = hci_resume_advertising_sync(hdev); + if (err && hci_dev_test_flag(hdev, HCI_UNREGISTER)) + return err; /* Resume discovery */ - hci_resume_discovery_sync(hdev); + err = hci_resume_discovery_sync(hdev); + if (err && hci_dev_test_flag(hdev, HCI_UNREGISTER)) + return err; return 0; } -- 2.25.1

18 hours, 23 minutes

1
0
0 0

[PATCH v4 00/22] drm/msm/adreno: Introduce Adreno 8xx family support

by Akhil P Oommen

This series adds the A8xx HWL along with Adreno 840 GPU support to the drm-msm driver. A8x is the next generation in the Adreno family, featuring a significant hardware design change. A major update to the design is the introduction of 'Slice' architecture. Slices are sort of mini-GPUs within the GPU which are more independent in processing Graphics and compute workloads. Also, in addition to the BV and BR pipe we saw in A7x, CP has more concurrency with additional pipes. From KMD-HW SWI perspective, there is significant register shuffling in some of the blocks. For slice or aperture related registers which are virtualized now, KMD/crashdumper has to configure an aperture register to access them. On the GMU front, there are some shuffling in register offsets, but it is manageable as of now. There is a new HFI message to transfer data tables and new power related features to support higher peak currents and thermal mitigations. Adreno 840 GPU is the second generation architecture in the A8x family present in Kaanapali (a.k.a Snapdragon 8 Elite Gen 5) chipset [1]. It has a maximum of 3 slices with 2 SPs per slice. Along with the 3-slice configuration, there is also another 2-slice SKU (Partial Slice SKU). A840 GPU has a bigger 18MB of GMEM which can be utilized for graphics and compute workload. It also features improved Concurrent binning support, UBWC v6 etc. Adreno X2-85 GPU present in Glymur chipset is very similar to A840 architecturally. So adding initial support for it requires just an additional entry in the catalog with the necessary register lists. This series adds only the driver side support along with a few dt bindings updates. Devicetree patches will be sent separately, but those who are interested can take look at it from the Qualcomm's public tree [2]. Features like coredump, gmu power features, ifpc, preemption etc will be added in a future series. Initial few patches are for improving code sharing between a6xx/a7xx and a8x routines. Then there is a patch to rebase GMU register offsets from GPU's base. Rest of the patches add A8x HWL and Adreno 840/X2-85 GPU support. Mesa support for A8x/A840 GPU is WIP and will be posted in the near future. The last patch in the series ("drm/msm/a8xx: Add UBWC v6 support") has a compile time dependency on the below patch from the qcom-soc tree ("soc: qcom: ubwc: Add config for Kaanapali"): https://lore.kernel.org/lkml/20250930-kaana-gpu-support-v1-1-73530b0700ed@o… [1] https://www.qualcomm.com/products/mobile/snapdragon/smartphones/snapdragon-… [2] https://git.codelinaro.org/clo/linux-kernel/kernel-qcom/-/commit/5fb72c2790… Signed-off-by: Akhil P Oommen <akhilpo(a)oss.qualcomm.com> --- Changes in v4: - Rebase on top of msm-next - Clean up AQE bo during a6xx_destroy (Konrad) - Split out UBWC v6 support into a separate patch to ease merge (Rob) - Rebase gmu register list's offsets in a6xx_gpu_state - Add a new patch#1 to fix Out of boud register access - Link to v3: https://lore.kernel.org/r/20251114-kaana-gpu-support-v3-0-92300c7ec8ff@oss.… Changes in v3: - Squash gpu smmu bindings patches for Kaana and Glymur (Krzysztof) - Reuse a6xx_flush() and drop the patch that added submit_flush callback - Fix GBIF configs for a640 and a650 family (Konrad) - Add partial SKU detection support - Correct Chipids in the catalog - Add a new patch to drop SCRATCH reg dumps (Rob) - Read slice info right after CX gdsc is up - Don't drop raytracing support if preemption is unsupported - Drop the unused A840 pwrup list (Konrad) - Updates to A840 nonctxt list (Rob) - Capture trailers - Link to v2: https://lore.kernel.org/r/20251110-kaana-gpu-support-v2-0-bef18acd5e94@oss.… Changes in v2: - Rebase on top of next-20251110 tag - Include support for Glymur chipset - Drop the ubwc_config driver patch as it is picked up - Sync the latest a6xx register definitions from Rob's tree - New patch to do LRZ flush to fix pagefaults - Reuse a7xx_cx_mem_init(). Dropped related patch (Connor) - Few changes around cp protect configuration to align it with downstream - Fix the incorrect register usage at few places - Updates to non-ctxt register list - Serialize aperture updates (Rob) - More helpful cp error irq logging - Split A8x GMU support patch (Dmitry) - Use devm_platform_get_and_ioremap_resource in GMU init (Konrad) - Link to v1: https://lore.kernel.org/r/20250930-kaana-gpu-support-v1-0-73530b0700ed@oss.… --- Akhil P Oommen (22): drm/msm/a6xx: Fix out of bound IO access in a6xx_get_gmu_registers drm/msm/a6xx: Flush LRZ cache before PT switch drm/msm/a6xx: Fix the gemnoc workaround drm/msm/a6xx: Skip dumping SCRATCH registers drm/msm/adreno: Common-ize PIPE definitions drm/msm/adreno: Move adreno_gpu_func to catalogue drm/msm/adreno: Move gbif_halt() to adreno_gpu_func drm/msm/adreno: Add MMU fault handler to adreno_gpu_func drm/msm/a6xx: Sync latest register definitions drm/msm/a6xx: Rebase GMU register offsets drm/msm/a8xx: Add support for A8x GMU drm/msm/a6xx: Improve MX rail fallback in RPMH vote init drm/msm/a6xx: Share dependency vote table with GMU drm/msm/adreno: Introduce A8x GPU Support drm/msm/adreno: Support AQE engine drm/msm/a8xx: Add support for Adreno 840 GPU drm/msm/adreno: Do CX GBIF config before GMU start drm/msm/a8xx: Add support for Adreno X2-85 GPU dt-bindings: arm-smmu: Add Kaanapali and Glymur GPU SMMU dt-bindings: display/msm/gmu: Add Adreno 840 GMU dt-bindings: display/msm/gmu: Add Adreno X2-85 GMU drm/msm/a8xx: Add UBWC v6 support .../devicetree/bindings/display/msm/gmu.yaml | 60 +- .../devicetree/bindings/iommu/arm,smmu.yaml | 2 + drivers/gpu/drm/msm/Makefile | 2 + drivers/gpu/drm/msm/adreno/a2xx_catalog.c | 7 +- drivers/gpu/drm/msm/adreno/a2xx_gpu.c | 50 +- drivers/gpu/drm/msm/adreno/a2xx_gpu.h | 2 + drivers/gpu/drm/msm/adreno/a3xx_catalog.c | 13 +- drivers/gpu/drm/msm/adreno/a3xx_gpu.c | 52 +- drivers/gpu/drm/msm/adreno/a3xx_gpu.h | 2 + drivers/gpu/drm/msm/adreno/a4xx_catalog.c | 7 +- drivers/gpu/drm/msm/adreno/a4xx_gpu.c | 54 +- drivers/gpu/drm/msm/adreno/a4xx_gpu.h | 2 + drivers/gpu/drm/msm/adreno/a5xx_catalog.c | 17 +- drivers/gpu/drm/msm/adreno/a5xx_gpu.c | 61 +- drivers/gpu/drm/msm/adreno/a5xx_gpu.h | 1 + drivers/gpu/drm/msm/adreno/a6xx_catalog.c | 371 +++- drivers/gpu/drm/msm/adreno/a6xx_gmu.c | 287 ++- drivers/gpu/drm/msm/adreno/a6xx_gmu.h | 25 +- drivers/gpu/drm/msm/adreno/a6xx_gpu.c | 399 ++-- drivers/gpu/drm/msm/adreno/a6xx_gpu.h | 31 +- drivers/gpu/drm/msm/adreno/a6xx_gpu_state.c | 2 +- drivers/gpu/drm/msm/adreno/a6xx_gpu_state.h | 74 +- drivers/gpu/drm/msm/adreno/a6xx_hfi.c | 53 + drivers/gpu/drm/msm/adreno/a6xx_hfi.h | 17 + drivers/gpu/drm/msm/adreno/a8xx_gpu.c | 1205 ++++++++++++ drivers/gpu/drm/msm/adreno/adreno_device.c | 4 +- .../gpu/drm/msm/adreno/adreno_gen7_0_0_snapshot.h | 420 ++--- .../gpu/drm/msm/adreno/adreno_gen7_2_0_snapshot.h | 332 ++-- .../gpu/drm/msm/adreno/adreno_gen7_9_0_snapshot.h | 470 ++--- drivers/gpu/drm/msm/adreno/adreno_gpu.h | 38 +- drivers/gpu/drm/msm/registers/adreno/a6xx.xml | 1954 +++++++++++++++----- .../gpu/drm/msm/registers/adreno/a6xx_enums.xml | 2 +- drivers/gpu/drm/msm/registers/adreno/a6xx_gmu.xml | 283 +-- .../gpu/drm/msm/registers/adreno/a7xx_enums.xml | 7 - .../drm/msm/registers/adreno/a8xx_descriptors.xml | 120 ++ .../gpu/drm/msm/registers/adreno/a8xx_enums.xml | 289 +++ .../gpu/drm/msm/registers/adreno/adreno_common.xml | 12 + 37 files changed, 5043 insertions(+), 1684 deletions(-) --- base-commit: 50a0b122cfc8a7dc35009ef9bf33cf6034c7bd69 change-id: 20250929-kaana-gpu-support-11d21c8fa1dc prerequisite-message-id: <20250930-kaana-gpu-support-v1-1-73530b0700ed(a)oss.qualcomm.com> prerequisite-patch-id: f15bd99b078d228da892fb1224e10cac31f4a5c2 prerequisite-patch-id: 5b3d152595fbcce7c118d42c00f89160bbf03d41 prerequisite-patch-id: 4387aff0073a3217132ae5da358e5d4b2cb23cb3 prerequisite-patch-id: e047a6ea27db881db0089923af688c38729a7dad prerequisite-patch-id: e686f7f592194f7d5e943858ce4dab49da6f4d18 prerequisite-patch-id: 638bc6f946cb2c1a2c68c3713a1ce7e6839c3465 prerequisite-patch-id: a85a264e87f79e9ac34dc22124153b050f97dded prerequisite-patch-id: 8bba83cdb88cb7a8851978590cb24033d95c21de prerequisite-patch-id: 9f08bcf9e33501478a2312e7a317f730f167652d prerequisite-patch-id: 65a2884909f6f0e3f111412388fde0c18a4a3334 prerequisite-patch-id: 3e9a011409f3461e3de7b1a8a4e99de6fbf02abf prerequisite-patch-id: 0ae4c8dc17fd54c84d903badccdf7a2018ec5606 prerequisite-patch-id: 6e0829024fb62bfc4510ef4c5472392dc76efcbf prerequisite-patch-id: 5e5e177cb37fd1c0151568744565483809f357ba prerequisite-patch-id: c2236f76a9fda88c41ea535708be1b51fd4d444c prerequisite-patch-id: 6e26922186365d994987026b674baa66f9ac0139 prerequisite-patch-id: 784df303a9e75f062c1e069d2bdb88578a76ba0e Best regards, -- Akhil P Oommen <akhilpo(a)oss.qualcomm.com>

18 hours, 24 minutes

2
2
0 0

Bug: Performance regression in 1013af4f585f: mm/hugetlb: fix huge_pmd_unshare() vs GUP-fast race

by Uschakow, Stanislav

Hello. We have observed a huge latency increase using `fork()` after ingesting the CVE-2025-38085 fix which leads to the commit `1013af4f585f: mm/hugetlb: fix huge_pmd_unshare() vs GUP-fast race`. On large machines with 1.5TB of memory with 196 cores, we identified mmapping of 1.2TB of shared memory and forking itself dozens or hundreds of times we see a increase of execution times of a factor of 4. The reproducer is at the end of the email. Comparing the a kernel without this patch with a kernel with this patch applied when spawning 1000 children we see those execution times: Patched kernel: $ time make stress ... real 0m11.275s user 0m0.177s sys 0m23.905s Original kernel : $ time make stress ...real 0m2.475s user 0m1.398s sys 0m2.501s The patch in question: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/commit/?id… My observation/assumption is: each child touches 100 random pages and despawns on each despawn `huge_pmd_unshare()` is called each call to `huge_pmd_unshare()` syncrhonizes all threads using `tlb_remove_table_sync_one()` leading to the regression I'm happy to provide more information. Thank you Stanislav Uschakow === Reproducer === Setup: #!/bin/bash echo "Setting up hugepages for reproduction..." # hugepages (1.2TB / 2MB = 614400 pages) REQUIRED_PAGES=614400 # Check current hugepage allocation CURRENT_PAGES=$(cat /proc/sys/vm/nr_hugepages) echo "Current hugepages: $CURRENT_PAGES" if [ "$CURRENT_PAGES" -lt "$REQUIRED_PAGES" ]; then echo "Allocating $REQUIRED_PAGES hugepages..." echo $REQUIRED_PAGES | sudo tee /proc/sys/vm/nr_hugepages ALLOCATED=$(cat /proc/sys/vm/nr_hugepages) echo "Allocated hugepages: $ALLOCATED" if [ "$ALLOCATED" -lt "$REQUIRED_PAGES" ]; then echo "Warning: Could not allocate all required hugepages" echo "Available: $ALLOCATED, Required: $REQUIRED_PAGES" fi fi echo never | sudo tee /sys/kernel/mm/transparent_hugepage/enabled echo -e "\nHugepage information:" cat /proc/meminfo | grep -i huge echo -e "\nSetup complete. You can now run the reproduction test." Makefile: CXX = gcc CXXFLAGS = -O2 -Wall TARGET = hugepage_repro SOURCE = hugepage_repro.c $(TARGET): $(SOURCE) $(CXX) $(CXXFLAGS) -o $(TARGET) $(SOURCE) clean: rm -f $(TARGET) setup: chmod +x setup_hugepages.sh ./setup_hugepages.sh test: $(TARGET) ./$(TARGET) 20 3 stress: $(TARGET) ./$(TARGET) 1000 1 .PHONY: clean setup test stress hugepage_repro.c: #include <sys/mman.h> #include <sys/wait.h> #include <unistd.h> #include <stdlib.h> #include <string.h> #include <time.h> #include <stdio.h> #define HUGEPAGE_SIZE (2 * 1024 * 1024) // 2MB #define TOTAL_SIZE (1200ULL * 1024 * 1024 * 1024) // 1.2TB #define NUM_HUGEPAGES (TOTAL_SIZE / HUGEPAGE_SIZE) void* create_hugepage_mapping() { void* addr = mmap(NULL, TOTAL_SIZE, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANONYMOUS | MAP_HUGETLB, -1, 0); if (addr == MAP_FAILED) { perror("mmap hugepages failed"); exit(1); } return addr; } void touch_random_pages(void* addr, int num_touches) { char* base = (char*)addr; for (int i = 0; i < num_touches; ++i) { size_t offset = (rand() % NUM_HUGEPAGES) * HUGEPAGE_SIZE; volatile char val = base[offset]; (void)val; } } void child_process(void* shared_mem, int child_id) { struct timespec start, end; clock_gettime(CLOCK_MONOTONIC, &start); touch_random_pages(shared_mem, 100); clock_gettime(CLOCK_MONOTONIC, &end); long duration = (end.tv_sec - start.tv_sec) * 1000000 + (end.tv_nsec - start.tv_nsec) / 1000; printf("Child %d completed in %ld μs\n", child_id, duration); } int main(int argc, char* argv[]) { int num_processes = argc > 1 ? atoi(argv[1]) : 50; int iterations = argc > 2 ? atoi(argv[2]) : 5; printf("Creating %lluGB hugepage mapping...\n", TOTAL_SIZE / (1024*1024*1024)); void* shared_mem = create_hugepage_mapping(); for (int iter = 0; iter < iterations; ++iter) { printf("\nIteration %d: Forking %d processes\n", iter + 1, num_processes); pid_t children[num_processes]; struct timespec iter_start, iter_end; clock_gettime(CLOCK_MONOTONIC, &iter_start); for (int i = 0; i < num_processes; ++i) { pid_t pid = fork(); if (pid == 0) { child_process(shared_mem, i); exit(0); } else if (pid > 0) { children[i] = pid; } } for (int i = 0; i < num_processes; ++i) { waitpid(children[i], NULL, 0); } clock_gettime(CLOCK_MONOTONIC, &iter_end); long iter_duration = (iter_end.tv_sec - iter_start.tv_sec) * 1000 + (iter_end.tv_nsec - iter_start.tv_nsec) / 1000000; printf("Iteration completed in %ld ms\n", iter_duration); } munmap(shared_mem, TOTAL_SIZE); return 0; } Amazon Web Services Development Center Germany GmbH Tamara-Danz-Str. 13 10243 Berlin Geschaeftsfuehrung: Christian Schlaeger, Jonathan Weiss Eingetragen am Amtsgericht Charlottenburg unter HRB 257764 B Sitz: Berlin Ust-ID: DE 365 538 597

18 hours, 30 minutes

6
28
0 0

[PATCH 09/15] dmaengine: stm32: dmamux: fix device leak on route allocation

by Johan Hovold

Make sure to drop the reference taken when looking up the DMA mux platform device during route allocation. Note that holding a reference to a device does not prevent its driver data from going away so there is no point in keeping the reference. Fixes: df7e762db5f6 ("dmaengine: Add STM32 DMAMUX driver") Cc: stable(a)vger.kernel.org # 4.15 Cc: Pierre-Yves MORDRET <pierre-yves.mordret(a)foss.st.com> Signed-off-by: Johan Hovold <johan(a)kernel.org> --- drivers/dma/stm32/stm32-dmamux.c | 18 ++++++++++++------ 1 file changed, 12 insertions(+), 6 deletions(-) diff --git a/drivers/dma/stm32/stm32-dmamux.c b/drivers/dma/stm32/stm32-dmamux.c index 8d77e2a7939a..791179760782 100644 --- a/drivers/dma/stm32/stm32-dmamux.c +++ b/drivers/dma/stm32/stm32-dmamux.c @@ -90,23 +90,25 @@ static void *stm32_dmamux_route_allocate(struct of_phandle_args *dma_spec, struct stm32_dmamux_data *dmamux = platform_get_drvdata(pdev); struct stm32_dmamux *mux; u32 i, min, max; - int ret; + int ret = -EINVAL; unsigned long flags; if (dma_spec->args_count != 3) { dev_err(&pdev->dev, "invalid number of dma mux args\n"); - return ERR_PTR(-EINVAL); + goto err_put_pdev; } if (dma_spec->args[0] > dmamux->dmamux_requests) { dev_err(&pdev->dev, "invalid mux request number: %d\n", dma_spec->args[0]); - return ERR_PTR(-EINVAL); + goto err_put_pdev; } mux = kzalloc(sizeof(*mux), GFP_KERNEL); - if (!mux) - return ERR_PTR(-ENOMEM); + if (!mux) { + ret = -ENOMEM; + goto err_put_pdev; + } spin_lock_irqsave(&dmamux->lock, flags); mux->chan_id = find_first_zero_bit(dmamux->dma_inuse, @@ -133,7 +135,6 @@ static void *stm32_dmamux_route_allocate(struct of_phandle_args *dma_spec, dma_spec->np = of_parse_phandle(ofdma->of_node, "dma-masters", i - 1); if (!dma_spec->np) { dev_err(&pdev->dev, "can't get dma master\n"); - ret = -EINVAL; goto error; } @@ -160,6 +161,8 @@ static void *stm32_dmamux_route_allocate(struct of_phandle_args *dma_spec, dev_dbg(&pdev->dev, "Mapping DMAMUX(%u) to DMA%u(%u)\n", mux->request, mux->master, mux->chan_id); + put_device(&pdev->dev); + return mux; error: @@ -167,6 +170,9 @@ static void *stm32_dmamux_route_allocate(struct of_phandle_args *dma_spec, error_chan_id: kfree(mux); +err_put_pdev: + put_device(&pdev->dev); + return ERR_PTR(ret); } -- 2.51.0

19 hours, 2 minutes

2
2
0 0

[PATCH 06/15] dmaengine: lpc18xx-dmamux: fix device leak on route allocation

by Johan Hovold

Make sure to drop the reference taken when looking up the DMA mux platform device during route allocation. Note that holding a reference to a device does not prevent its driver data from going away so there is no point in keeping the reference. Fixes: e5f4ae84be74 ("dmaengine: add driver for lpc18xx dmamux") Cc: stable(a)vger.kernel.org # 4.3 Signed-off-by: Johan Hovold <johan(a)kernel.org> --- drivers/dma/lpc18xx-dmamux.c | 19 ++++++++++++++----- 1 file changed, 14 insertions(+), 5 deletions(-) diff --git a/drivers/dma/lpc18xx-dmamux.c b/drivers/dma/lpc18xx-dmamux.c index 2b6436f4b193..d3ff521951b8 100644 --- a/drivers/dma/lpc18xx-dmamux.c +++ b/drivers/dma/lpc18xx-dmamux.c @@ -57,30 +57,31 @@ static void *lpc18xx_dmamux_reserve(struct of_phandle_args *dma_spec, struct lpc18xx_dmamux_data *dmamux = platform_get_drvdata(pdev); unsigned long flags; unsigned mux; + int ret = -EINVAL; if (dma_spec->args_count != 3) { dev_err(&pdev->dev, "invalid number of dma mux args\n"); - return ERR_PTR(-EINVAL); + goto err_put_pdev; } mux = dma_spec->args[0]; if (mux >= dmamux->dma_master_requests) { dev_err(&pdev->dev, "invalid mux number: %d\n", dma_spec->args[0]); - return ERR_PTR(-EINVAL); + goto err_put_pdev; } if (dma_spec->args[1] > LPC18XX_DMAMUX_MAX_VAL) { dev_err(&pdev->dev, "invalid dma mux value: %d\n", dma_spec->args[1]); - return ERR_PTR(-EINVAL); + goto err_put_pdev; } /* The of_node_put() will be done in the core for the node */ dma_spec->np = of_parse_phandle(ofdma->of_node, "dma-masters", 0); if (!dma_spec->np) { dev_err(&pdev->dev, "can't get dma master\n"); - return ERR_PTR(-EINVAL); + goto err_put_pdev; } spin_lock_irqsave(&dmamux->lock, flags); @@ -89,7 +90,8 @@ static void *lpc18xx_dmamux_reserve(struct of_phandle_args *dma_spec, dev_err(&pdev->dev, "dma request %u busy with %u.%u\n", mux, mux, dmamux->muxes[mux].value); of_node_put(dma_spec->np); - return ERR_PTR(-EBUSY); + ret = -EBUSY; + goto err_put_pdev; } dmamux->muxes[mux].busy = true; @@ -106,7 +108,14 @@ static void *lpc18xx_dmamux_reserve(struct of_phandle_args *dma_spec, dev_dbg(&pdev->dev, "mapping dmamux %u.%u to dma request %u\n", mux, dmamux->muxes[mux].value, mux); + put_device(&pdev->dev); + return &dmamux->muxes[mux]; + +err_put_pdev: + put_device(&pdev->dev); + + return ERR_PTR(ret); } static int lpc18xx_dmamux_probe(struct platform_device *pdev) -- 2.51.0

19 hours, 3 minutes

2
2
0 0

[PATCH 01/15] dmaengine: at_hdmac: fix device leak on of_dma_xlate()

by Johan Hovold

Make sure to drop the reference taken when looking up the DMA platform device during of_dma_xlate() when releasing channel resources. Note that commit 3832b78b3ec2 ("dmaengine: at_hdmac: add missing put_device() call in at_dma_xlate()") fixed the leak in a couple of error paths but the reference is still leaking on successful allocation. Fixes: bbe89c8e3d59 ("at_hdmac: move to generic DMA binding") Fixes: 3832b78b3ec2 ("dmaengine: at_hdmac: add missing put_device() call in at_dma_xlate()") Cc: stable(a)vger.kernel.org # 3.10: 3832b78b3ec2 Cc: Yu Kuai <yukuai3(a)huawei.com> Signed-off-by: Johan Hovold <johan(a)kernel.org> --- drivers/dma/at_hdmac.c | 9 +++++++-- 1 file changed, 7 insertions(+), 2 deletions(-) diff --git a/drivers/dma/at_hdmac.c b/drivers/dma/at_hdmac.c index 2d147712cbc6..dffe5becd6c3 100644 --- a/drivers/dma/at_hdmac.c +++ b/drivers/dma/at_hdmac.c @@ -1765,6 +1765,7 @@ static int atc_alloc_chan_resources(struct dma_chan *chan) static void atc_free_chan_resources(struct dma_chan *chan) { struct at_dma_chan *atchan = to_at_dma_chan(chan); + struct at_dma_slave *atslave; BUG_ON(atc_chan_is_enabled(atchan)); @@ -1774,8 +1775,12 @@ static void atc_free_chan_resources(struct dma_chan *chan) /* * Free atslave allocated in at_dma_xlate() */ - kfree(chan->private); - chan->private = NULL; + atslave = chan->private; + if (atslave) { + put_device(atslave->dma_dev); + kfree(atslave); + chan->private = NULL; + } dev_vdbg(chan2dev(chan), "free_chan_resources: done\n"); } -- 2.51.0

19 hours, 4 minutes

2
2
0 0

[PATCH 04/15] dmaengine: dw: dmamux: fix OF node leak on route allocation failure

by Johan Hovold

Make sure to drop the reference taken to the DMA master OF node also on late route allocation failures. Fixes: 134d9c52fca2 ("dmaengine: dw: dmamux: Introduce RZN1 DMA router support") Cc: stable(a)vger.kernel.org # 5.19 Cc: Miquel Raynal <miquel.raynal(a)bootlin.com> Signed-off-by: Johan Hovold <johan(a)kernel.org> --- drivers/dma/dw/rzn1-dmamux.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) diff --git a/drivers/dma/dw/rzn1-dmamux.c b/drivers/dma/dw/rzn1-dmamux.c index deadf135681b..cbec277af4dd 100644 --- a/drivers/dma/dw/rzn1-dmamux.c +++ b/drivers/dma/dw/rzn1-dmamux.c @@ -90,7 +90,7 @@ static void *rzn1_dmamux_route_allocate(struct of_phandle_args *dma_spec, if (test_and_set_bit(map->req_idx, dmamux->used_chans)) { ret = -EBUSY; - goto free_map; + goto put_dma_spec_np; } mask = BIT(map->req_idx); @@ -103,6 +103,8 @@ static void *rzn1_dmamux_route_allocate(struct of_phandle_args *dma_spec, clear_bitmap: clear_bit(map->req_idx, dmamux->used_chans); +put_dma_spec_np: + of_node_put(dma_spec->np); free_map: kfree(map); put_device: -- 2.51.0

19 hours, 12 minutes

3
3
0 0

[PATCH v2] phy: HiSilicon: Fix error handling in hi3670_pcie_get_resources_from_pcie

by Ma Ke

In hi3670_pcie_get_resources_from_pcie(), the reference obtained via bus_find_device_by_of_node() is never released with put_device(). Each call to this function increments the reference count of the PCIe device without decrementing it, preventing proper device cleanup. And the device node obtained via of_get_child_by_name() is never released with of_node_put(). This could cause a leak of the node reference. Add proper resource cleanup using goto labels to ensure all acquired references are released before function return, regardless of the exit path. Found by code review. Cc: stable(a)vger.kernel.org Fixes: 73075011ffff ("phy: HiSilicon: Add driver for Kirin 970 PCIe PHY") Signed-off-by: Ma Ke <make24(a)iscas.ac.cn> --- Changes in v2: - modified the patch for the warning that variable 'ret' is set but not used. --- drivers/phy/hisilicon/phy-hi3670-pcie.c | 15 +++++++++++---- 1 file changed, 11 insertions(+), 4 deletions(-) diff --git a/drivers/phy/hisilicon/phy-hi3670-pcie.c b/drivers/phy/hisilicon/phy-hi3670-pcie.c index dbc7dcce682b..9960c9da9b4a 100644 --- a/drivers/phy/hisilicon/phy-hi3670-pcie.c +++ b/drivers/phy/hisilicon/phy-hi3670-pcie.c @@ -560,7 +560,8 @@ static int hi3670_pcie_get_resources_from_pcie(struct hi3670_pcie_phy *phy) { struct device_node *pcie_port; struct device *dev = phy->dev; - struct device *pcie_dev; + struct device *pcie_dev = NULL; + int ret = 0; pcie_port = of_get_child_by_name(dev->parent->of_node, "pcie"); if (!pcie_port) { @@ -572,7 +573,8 @@ static int hi3670_pcie_get_resources_from_pcie(struct hi3670_pcie_phy *phy) pcie_dev = bus_find_device_by_of_node(&platform_bus_type, pcie_port); if (!pcie_dev) { dev_err(dev, "Didn't find pcie device\n"); - return -ENODEV; + ret = -ENODEV; + goto out_put_node; } /* @@ -586,10 +588,15 @@ static int hi3670_pcie_get_resources_from_pcie(struct hi3670_pcie_phy *phy) phy->apb = dev_get_regmap(pcie_dev, "kirin_pcie_apb"); if (!phy->apb) { dev_err(dev, "Failed to get APB regmap\n"); - return -ENODEV; + ret = -ENODEV; + goto out_put_device; } - return 0; +out_put_device: + put_device(pcie_dev); +out_put_node: + of_node_put(pcie_port); + return ret; } static int kirin_pcie_clk_ctrl(struct hi3670_pcie_phy *phy, bool enable) -- 2.17.1

19 hours, 24 minutes

2
1
0 0

2025

2024

2023

2022

2021

2020

2019

2018

2017

Linux-stable-mirror