kernel/linux.git/drivers/dma/idxd/cdev.c, branch v6.1.168

dmaengine: idxd: Fix freeing the allocated ida too late

2026-04-11T12:16:17+00:00

[ Upstream commit c311f5e9248471a950f0a524c2fd736414d98900 ] It can happen that when the cdev .release() is called, the driver already called ida_destroy(). Move ida_free() to the _del() path. We see with DEBUG_KOBJECT_RELEASE enabled and forcing an early PCI unbind. Fixes: 04922b7445a1 ("dmaengine: idxd: fix cdev setup and free device lifetime issues") Reviewed-by: Dave Jiang Signed-off-by: Vinicius Costa Gomes Link: https://patch.msgid.link/20260121-idxd-fix-flr-on-kernel-queues-v3-v3-9-7ed70658a9d1@intel.com Signed-off-by: Vinod Koul Signed-off-by: Sasha Levin

dmaengine: idxd: Remove usage of the deprecated ida_simple_xx() API

2026-04-11T12:16:17+00:00

[ Upstream commit 1075ee66a8c19bfa375b19c236fd6a22a867f138 ] ida_alloc() and ida_free() should be preferred to the deprecated ida_simple_get() and ida_simple_remove(). This is less verbose. Note that the upper limit of ida_simple_get() is exclusive, but the one of ida_alloc_range() is inclusive. Sothis change allows one more device. MINORMASK is ((1U << MINORBITS) - 1), so allowing MINORMASK as a maximum value makes sense. It is also consistent with other "ida_.*MINORMASK" and "ida_*MINOR()" usages. Signed-off-by: Christophe JAILLET Reviewed-by: Fenghua Yu Acked-by: Lijun Pan Link: https://lore.kernel.org/r/ac991f5f42112fa782a881d391d447529cbc4a23.1702967302.git.christophe.jaillet@wanadoo.fr Signed-off-by: Vinod Koul Stable-dep-of: c311f5e92484 ("dmaengine: idxd: Fix freeing the allocated ida too late") Signed-off-by: Sasha Levin

dmaengine: idxd: Fix passing freed memory in idxd_cdev_open()

2025-06-04T12:40:22+00:00

commit 0642287e3ecdd0d1f88e6a2e63768e16153a990c upstream. Smatch warns: drivers/dma/idxd/cdev.c:327: idxd_cdev_open() warn: 'sva' was already freed. When idxd_wq_set_pasid() fails, the current code unbinds sva and then goes to 'failed_set_pasid' where iommu_sva_unbind_device is called again causing the above warning. [ device_user_pasid_enabled(idxd) is still true when calling failed_set_pasid ] Fix this by removing additional unbind when idxd_wq_set_pasid() fails Fixes: b022f59725f0 ("dmaengine: idxd: add idxd_copy_cr() to copy user completion record during page fault handling") Signed-off-by: Harshit Mogalapalli Acked-by: Fenghua Yu Acked-by: Dave Jiang Link: https://lore.kernel.org/r/20230509060716.2830630-1-harshit.m.mogalapalli@oracle.com Signed-off-by: Vinod Koul Signed-off-by: Greg Kroah-Hartman

dmaengine: idxd: Fix ->poll() return value

2025-06-04T12:40:19+00:00

[ Upstream commit ae74cd15ade833adc289279b5c6f12e78f64d4d7 ] The fix to block access from different address space did not return a correct value for ->poll() change. kernel test bot reported that a return value of type __poll_t is expected rather than int. Fix to return POLLNVAL to indicate invalid request. Fixes: 8dfa57aabff6 ("dmaengine: idxd: Fix allowing write() from different address spaces") Reported-by: kernel test robot Closes: https://lore.kernel.org/oe-kbuild-all/202505081851.rwD7jVxg-lkp@intel.com/ Signed-off-by: Dave Jiang Link: https://lore.kernel.org/r/20250508170548.2747425-1-dave.jiang@intel.com Signed-off-by: Vinod Koul Signed-off-by: Sasha Levin

dmaengine: idxd: Fix allowing write() from different address spaces

2025-06-04T12:40:19+00:00

[ Upstream commit 8dfa57aabff625bf445548257f7711ef294cd30e ] Check if the process submitting the descriptor belongs to the same address space as the one that opened the file, reject otherwise. Fixes: 6827738dc684 ("dmaengine: idxd: add a write() method for applications to submit work") Signed-off-by: Vinicius Costa Gomes Signed-off-by: Dave Jiang Link: https://lore.kernel.org/r/20250421170337.3008875-1-dave.jiang@intel.com Signed-off-by: Vinod Koul Signed-off-by: Sasha Levin

dmaengine: idxd: add idxd_copy_cr() to copy user completion record during page fault handling

2025-06-04T12:40:19+00:00

[ Upstream commit b022f59725f0ae846191abbd6d2e611d7f60f826 ] Define idxd_copy_cr() to copy completion record to fault address in user address that is found by work queue (wq) and PASID. It will be used to write the user's completion record that the hardware device is not able to write due to user completion record page fault. An xarray is added to associate the PASID and mm with the struct idxd_user_context so mm can be found by PASID and wq. It is called when handling the completion record fault in a kernel thread context. Switch to the mm using kthread_use_vm() and copy the completion record to the mm via copy_to_user(). Once the copy is completed, switch back to the current mm using kthread_unuse_mm(). Suggested-by: Christoph Hellwig Suggested-by: Jason Gunthorpe Suggested-by: Tony Luck Tested-by: Tony Zhu Signed-off-by: Fenghua Yu Reviewed-by: Dave Jiang Link: https://lore.kernel.org/r/20230407203143.2189681-9-fenghua.yu@intel.com Signed-off-by: Vinod Koul Stable-dep-of: 8dfa57aabff6 ("dmaengine: idxd: Fix allowing write() from different address spaces") Signed-off-by: Sasha Levin

dmaengine: idxd: add per DSA wq workqueue for processing cr faults

2025-06-04T12:40:19+00:00

[ Upstream commit 2f30decd2f23a376d2ed73dfe4c601421edf501a ] Add a workqueue for user submitted completion record fault processing. The workqueue creation and destruction lifetime will be tied to the user sub-driver since it will only be used when the wq is a user type. Tested-by: Tony Zhu Signed-off-by: Dave Jiang Co-developed-by: Fenghua Yu Signed-off-by: Fenghua Yu Link: https://lore.kernel.org/r/20230407203143.2189681-7-fenghua.yu@intel.com Signed-off-by: Vinod Koul Stable-dep-of: 8dfa57aabff6 ("dmaengine: idxd: Fix allowing write() from different address spaces") Signed-off-by: Sasha Levin

dmaengine: idxd: add a write() method for applications to submit work

2024-05-17T09:56:23+00:00

commit 6827738dc684a87ad54ebba3ae7f3d7c977698eb upstream. After the patch to restrict the use of mmap() to CAP_SYS_RAWIO for the currently existing devices, most applications can no longer make use of the accelerators as in production "you don't run things as root". To keep the DSA and IAA accelerators usable, hook up a write() method so that applications can still submit work. In the write method, sufficient input validation is performed to avoid the security issue that required the mmap CAP_SYS_RAWIO check. One complication is that the DSA device allows for indirect ("batched") descriptors. There is no reasonable way to do the input validation on these indirect descriptors so the write() method will not allow these to be submitted to the hardware on affected hardware, and the sysfs enumeration of support for the opcode is also removed. Early performance data shows that the performance delta for most common cases is within the noise. Signed-off-by: Nikhil Rao Signed-off-by: Arjan van de Ven Signed-off-by: Greg Kroah-Hartman

dmaengine: idxd: add a new security check to deal with a hardware erratum

2024-05-17T09:56:23+00:00

commit e11452eb071b2a8e6ba52892b2e270bbdaa6640d upstream. On Sapphire Rapids and related platforms, the DSA and IAA devices have an erratum that causes direct access (for example, by using the ENQCMD or MOVDIR64 instructions) from untrusted applications to be a security problem. To solve this, add a flag to the PCI device enumeration and device structures to indicate the presence/absence of this security exposure. In the mmap() method of the device, this flag is then used to enforce that the user has the CAP_SYS_RAWIO capability. In a future patch, a write() based method will be added that allows untrusted applications submit work to the accelerator, where the kernel can do sanity checking on the user input to ensure secure operation of the accelerator. Signed-off-by: Arjan van de Ven Signed-off-by: Greg Kroah-Hartman

dmaengine: idxd: Do not enable user type Work Queue without Shared Virtual Addressing

2022-10-19T13:32:10+00:00

When the idxd_user_drv driver is bound to a Work Queue (WQ) device without IOMMU or with IOMMU Passthrough without Shared Virtual Addressing (SVA), the application gains direct access to physical memory via the device by programming physical address to a submitted descriptor. This allows direct userspace read and write access to arbitrary physical memory. This is inconsistent with the security goals of a good kernel API. Unlike vfio_pci driver, the IDXD char device driver does not provide any ways to pin user pages and translate the address from user VA to IOVA or PA without IOMMU SVA. Therefore the application has no way to instruct the device to perform DMA function. This makes the char device not usable for normal application usage. Since user type WQ without SVA cannot be used for normal application usage and presents the security issue, bind idxd_user_drv driver and enable user type WQ only when SVA is enabled (i.e. user PASID is enabled). Fixes: 448c3de8ac83 ("dmaengine: idxd: create user driver for wq 'device'") Cc: stable@vger.kernel.org Suggested-by: Arjan Van De Ven Signed-off-by: Fenghua Yu Reviewed-by: Dave Jiang Reviewed-by: Jerry Snitselaar Link: https://lore.kernel.org/r/20221014222541.3912195-1-fenghua.yu@intel.com Signed-off-by: Vinod Koul