| Message ID | 20260825155524.338587-1-alice.munduruca@canonical.com |
|---|---|
| Headers | show |
| Series | Resolute Stable Update v6.18.40, v7.1.5 bugs | expand |
On Tue, Aug 25, 2026 at 11:55:22AM -0400, Alice C. Munduruca via kernel-team wrote: > BugLink: https://bugs.launchpad.net/bugs/2165040 > > v2 -> Added the bug link to the cover letter > > [ Impact ] > > Due to infrastructural issues, the validation build for LP#2164666 only built tools > on each architecture, not the whole kernel. As a result, a few build failures have snuck > into the tree which need to be resolved. A list of the broken commits is presented below: > > * d14116da71c9 ("accel/amdxdna: Create shared functions for AIE2 and AIE4") > * f7688a6921ed ("bpf: Take mmap_lock in zap_pages()") > > [ Fix ] > > Resolute: Backporting and creation of the following fixes: > > * ece3e8980907 (Move to aie2_free_msg_buffer() function from the previous macro to apply a missing hunk) > * SAUCE PATCH (Add missing NULL argument to the zap_page_range_single() function) > > [ Testing ] > > Successful local build on an arm64 machine as well as validation on cloud builders for all other supported ARCHs. > > [ What could go wrong ] > > Given that the previous state of the tree was failure to build, things do not get much worse. Nonetheless, > The patches applied attempt to keep the tree as close to upstream as possible while correcting the stable update. > The only subsystems affected are: > > * accel/amdxdna > * bpf > > Alice C. Munduruca (1): > UBUNTU: SAUCE: bpf: Add missing NULL argument to zap_page_range_single > > Lizhi Hou (1): > accel/amdxdna: Allow forcing IOVA-based DMA via module parameter > > drivers/accel/amdxdna/Makefile | 1 + > drivers/accel/amdxdna/aie2_ctx.c | 6 +- > drivers/accel/amdxdna/aie2_error.c | 5 +- > drivers/accel/amdxdna/aie2_message.c | 30 +++- > drivers/accel/amdxdna/aie2_pci.h | 5 +- > drivers/accel/amdxdna/amdxdna_gem.c | 45 +++++- > drivers/accel/amdxdna/amdxdna_gem.h | 10 ++ > drivers/accel/amdxdna/amdxdna_iommu.c | 184 ++++++++++++++++++++++++ > drivers/accel/amdxdna/amdxdna_pci_drv.c | 47 +++--- > drivers/accel/amdxdna/amdxdna_pci_drv.h | 24 ++++ > kernel/bpf/arena.c | 2 +- > 11 files changed, 326 insertions(+), 33 deletions(-) > create mode 100644 drivers/accel/amdxdna/amdxdna_iommu.c > > -- > 2.53.0 > > > -- > kernel-team mailing list > kernel-team@lists.ubuntu.com > https://lists.ubuntu.com/mailman/listinfo/kernel-team Acking for the stable update reason, but there is a memory leak in amdxdna_iommu_alloc. If iommu_map fails, cpu_addr is not freed by calling free_pages. This is worth reporting upstream or making directly a submission to stable. Acked-by: Alessio Faina <alessio.faina@canonical.com>
On 8/25/26 09:55, Alice C. Munduruca via kernel-team wrote: > BugLink: https://bugs.launchpad.net/bugs/2165040 > > v2 -> Added the bug link to the cover letter > > [ Impact ] > > Due to infrastructural issues, the validation build for LP#2164666 only built tools > on each architecture, not the whole kernel. As a result, a few build failures have snuck > into the tree which need to be resolved. A list of the broken commits is presented below: > > * d14116da71c9 ("accel/amdxdna: Create shared functions for AIE2 and AIE4") > * f7688a6921ed ("bpf: Take mmap_lock in zap_pages()") > > [ Fix ] > > Resolute: Backporting and creation of the following fixes: > > * ece3e8980907 (Move to aie2_free_msg_buffer() function from the previous macro to apply a missing hunk) > * SAUCE PATCH (Add missing NULL argument to the zap_page_range_single() function) > > [ Testing ] > > Successful local build on an arm64 machine as well as validation on cloud builders for all other supported ARCHs. > > [ What could go wrong ] > > Given that the previous state of the tree was failure to build, things do not get much worse. Nonetheless, > The patches applied attempt to keep the tree as close to upstream as possible while correcting the stable update. > The only subsystems affected are: > > * accel/amdxdna > * bpf > > Alice C. Munduruca (1): > UBUNTU: SAUCE: bpf: Add missing NULL argument to zap_page_range_single > > Lizhi Hou (1): > accel/amdxdna: Allow forcing IOVA-based DMA via module parameter > > drivers/accel/amdxdna/Makefile | 1 + > drivers/accel/amdxdna/aie2_ctx.c | 6 +- > drivers/accel/amdxdna/aie2_error.c | 5 +- > drivers/accel/amdxdna/aie2_message.c | 30 +++- > drivers/accel/amdxdna/aie2_pci.h | 5 +- > drivers/accel/amdxdna/amdxdna_gem.c | 45 +++++- > drivers/accel/amdxdna/amdxdna_gem.h | 10 ++ > drivers/accel/amdxdna/amdxdna_iommu.c | 184 ++++++++++++++++++++++++ > drivers/accel/amdxdna/amdxdna_pci_drv.c | 47 +++--- > drivers/accel/amdxdna/amdxdna_pci_drv.h | 24 ++++ > kernel/bpf/arena.c | 2 +- > 11 files changed, 326 insertions(+), 33 deletions(-) > create mode 100644 drivers/accel/amdxdna/amdxdna_iommu.c > Acked-by: Rickey Castillo <rickey.castillo.valenzuela@canonical.com>
Applied to R:linux/master-next. Thanks. As reported by Alessio, this submission is missing at least these follow ups and is applied only to fix the build issues: e98bb0197afd accel/amdxdna: Return ERR_PTR on dma_alloc_noncoherent failure (a code path still returning NULL) fbe467561eda accel/amdxdna: Fix memory leak in amdxdna_iommu_alloc() On 8/25/26 18:55, Alice C. Munduruca via kernel-team wrote: > BugLink:https://bugs.launchpad.net/bugs/2165040 > > v2 -> Added the bug link to the cover letter > > [ Impact ] > > Due to infrastructural issues, the validation build for LP#2164666 only built tools > on each architecture, not the whole kernel. As a result, a few build failures have snuck > into the tree which need to be resolved. A list of the broken commits is presented below: > > * d14116da71c9 ("accel/amdxdna: Create shared functions for AIE2 and AIE4") > * f7688a6921ed ("bpf: Take mmap_lock in zap_pages()") > > [ Fix ] > > Resolute: Backporting and creation of the following fixes: > > * ece3e8980907 (Move to aie2_free_msg_buffer() function from the previous macro to apply a missing hunk) > * SAUCE PATCH (Add missing NULL argument to the zap_page_range_single() function) > > [ Testing ] > > Successful local build on an arm64 machine as well as validation on cloud builders for all other supported ARCHs. > > [ What could go wrong ] > > Given that the previous state of the tree was failure to build, things do not get much worse. Nonetheless, > The patches applied attempt to keep the tree as close to upstream as possible while correcting the stable update. > The only subsystems affected are: > > * accel/amdxdna > * bpf > > Alice C. Munduruca (1): > UBUNTU: SAUCE: bpf: Add missing NULL argument to zap_page_range_single > > Lizhi Hou (1): > accel/amdxdna: Allow forcing IOVA-based DMA via module parameter > > drivers/accel/amdxdna/Makefile | 1 + > drivers/accel/amdxdna/aie2_ctx.c | 6 +- > drivers/accel/amdxdna/aie2_error.c | 5 +- > drivers/accel/amdxdna/aie2_message.c | 30 +++- > drivers/accel/amdxdna/aie2_pci.h | 5 +- > drivers/accel/amdxdna/amdxdna_gem.c | 45 +++++- > drivers/accel/amdxdna/amdxdna_gem.h | 10 ++ > drivers/accel/amdxdna/amdxdna_iommu.c | 184 ++++++++++++++++++++++++ > drivers/accel/amdxdna/amdxdna_pci_drv.c | 47 +++--- > drivers/accel/amdxdna/amdxdna_pci_drv.h | 24 ++++ > kernel/bpf/arena.c | 2 +- > 11 files changed, 326 insertions(+), 33 deletions(-) > create mode 100644 drivers/accel/amdxdna/amdxdna_iommu.c >
Correct upstream SHAs: 140cfee588af accel/amdxdna: Return ERR_PTR on dma_alloc_noncoherent failure (a code path still returning NULL) 432fafdc9a31 accel/amdxdna: Fix memory leak in amdxdna_iommu_alloc() On 8/27/26 07:51, Edoardo Canepa wrote: > > Applied to R:linux/master-next. Thanks. > > As reported by Alessio, this submission is missing at least these > follow ups and is applied only to fix the build issues: > > e98bb0197afd accel/amdxdna: Return ERR_PTR on dma_alloc_noncoherent > failure (a code path still returning NULL) > fbe467561eda accel/amdxdna: Fix memory leak in amdxdna_iommu_alloc() > > > > On 8/25/26 18:55, Alice C. Munduruca via kernel-team wrote: >> BugLink:https://bugs.launchpad.net/bugs/2165040 >> >> v2 -> Added the bug link to the cover letter >> >> [ Impact ] >> >> Due to infrastructural issues, the validation build for LP#2164666 only built tools >> on each architecture, not the whole kernel. As a result, a few build failures have snuck >> into the tree which need to be resolved. A list of the broken commits is presented below: >> >> * d14116da71c9 ("accel/amdxdna: Create shared functions for AIE2 and AIE4") >> * f7688a6921ed ("bpf: Take mmap_lock in zap_pages()") >> >> [ Fix ] >> >> Resolute: Backporting and creation of the following fixes: >> >> * ece3e8980907 (Move to aie2_free_msg_buffer() function from the previous macro to apply a missing hunk) >> * SAUCE PATCH (Add missing NULL argument to the zap_page_range_single() function) >> >> [ Testing ] >> >> Successful local build on an arm64 machine as well as validation on cloud builders for all other supported ARCHs. >> >> [ What could go wrong ] >> >> Given that the previous state of the tree was failure to build, things do not get much worse. Nonetheless, >> The patches applied attempt to keep the tree as close to upstream as possible while correcting the stable update. >> The only subsystems affected are: >> >> * accel/amdxdna >> * bpf >> >> Alice C. Munduruca (1): >> UBUNTU: SAUCE: bpf: Add missing NULL argument to zap_page_range_single >> >> Lizhi Hou (1): >> accel/amdxdna: Allow forcing IOVA-based DMA via module parameter >> >> drivers/accel/amdxdna/Makefile | 1 + >> drivers/accel/amdxdna/aie2_ctx.c | 6 +- >> drivers/accel/amdxdna/aie2_error.c | 5 +- >> drivers/accel/amdxdna/aie2_message.c | 30 +++- >> drivers/accel/amdxdna/aie2_pci.h | 5 +- >> drivers/accel/amdxdna/amdxdna_gem.c | 45 +++++- >> drivers/accel/amdxdna/amdxdna_gem.h | 10 ++ >> drivers/accel/amdxdna/amdxdna_iommu.c | 184 ++++++++++++++++++++++++ >> drivers/accel/amdxdna/amdxdna_pci_drv.c | 47 +++--- >> drivers/accel/amdxdna/amdxdna_pci_drv.h | 24 ++++ >> kernel/bpf/arena.c | 2 +- >> 11 files changed, 326 insertions(+), 33 deletions(-) >> create mode 100644 drivers/accel/amdxdna/amdxdna_iommu.c >>
BugLink: https://bugs.launchpad.net/bugs/2165040 v2 -> Added the bug link to the cover letter [ Impact ] Due to infrastructural issues, the validation build for LP#2164666 only built tools on each architecture, not the whole kernel. As a result, a few build failures have snuck into the tree which need to be resolved. A list of the broken commits is presented below: * d14116da71c9 ("accel/amdxdna: Create shared functions for AIE2 and AIE4") * f7688a6921ed ("bpf: Take mmap_lock in zap_pages()") [ Fix ] Resolute: Backporting and creation of the following fixes: * ece3e8980907 (Move to aie2_free_msg_buffer() function from the previous macro to apply a missing hunk) * SAUCE PATCH (Add missing NULL argument to the zap_page_range_single() function) [ Testing ] Successful local build on an arm64 machine as well as validation on cloud builders for all other supported ARCHs. [ What could go wrong ] Given that the previous state of the tree was failure to build, things do not get much worse. Nonetheless, The patches applied attempt to keep the tree as close to upstream as possible while correcting the stable update. The only subsystems affected are: * accel/amdxdna * bpf Alice C. Munduruca (1): UBUNTU: SAUCE: bpf: Add missing NULL argument to zap_page_range_single Lizhi Hou (1): accel/amdxdna: Allow forcing IOVA-based DMA via module parameter drivers/accel/amdxdna/Makefile | 1 + drivers/accel/amdxdna/aie2_ctx.c | 6 +- drivers/accel/amdxdna/aie2_error.c | 5 +- drivers/accel/amdxdna/aie2_message.c | 30 +++- drivers/accel/amdxdna/aie2_pci.h | 5 +- drivers/accel/amdxdna/amdxdna_gem.c | 45 +++++- drivers/accel/amdxdna/amdxdna_gem.h | 10 ++ drivers/accel/amdxdna/amdxdna_iommu.c | 184 ++++++++++++++++++++++++ drivers/accel/amdxdna/amdxdna_pci_drv.c | 47 +++--- drivers/accel/amdxdna/amdxdna_pci_drv.h | 24 ++++ kernel/bpf/arena.c | 2 +- 11 files changed, 326 insertions(+), 33 deletions(-) create mode 100644 drivers/accel/amdxdna/amdxdna_iommu.c