On Sat, Sep 26, 2026 at 03:59:32PM +0000, [email protected] wrote: > Thank you for your contribution! Sashiko AI review found 2 potential issue(s) > to consider: > - [High] remoteproc: qcom_wcnss: DMA memory leak of metadata in wcnss_load() > - [High] remoteproc: qcom_wcnss: missing pas_ctx->use_tzmem = true prevents > required shmbridge creation > -- > > commit 77b1a378f038e55ff2a11032d9b6f243214fbb7f > Author: Mukesh Ojha <[email protected]> > Date: Fri Sep 25 15:57:30 2026 +0000 > > remoteproc: qcom_wcnss: migrate to qcom_mdt_pas_load() and > qcom_pas_prepare_and_auth_reset() > > This commit migrates the Qualcomm WCNSS remoteproc driver to use > qcom_mdt_pas_load() and qcom_pas_prepare_and_auth_reset(). It allocates a > PAS context during probe to support TrustZone authentication with shmbridge > preparation. > > > diff --git a/drivers/remoteproc/qcom_wcnss.c > > b/drivers/remoteproc/qcom_wcnss.c > > index c856a92af43ce..01f9076aa0b17 100644 > > --- a/drivers/remoteproc/qcom_wcnss.c > > +++ b/drivers/remoteproc/qcom_wcnss.c > > [ ... ] > > > @@ -157,9 +158,7 @@ static int wcnss_load(struct rproc *rproc, const struct > > firmware *fw) > > struct qcom_wcnss *wcnss = rproc->priv; > > int ret; > > > > - ret = qcom_mdt_load(wcnss->dev, fw, rproc->firmware, WCNSS_PAS_ID, > > - (__force void *)wcnss->mem_region, wcnss->mem_phys, > > - wcnss->mem_size, &wcnss->mem_reloc); > > + ret = qcom_mdt_pas_load(wcnss->pas_ctx, fw, rproc->firmware, > > &wcnss->mem_reloc); > > [Severity: High] > Does this introduce a DMA memory leak of the metadata buffer? > > By passing wcnss->pas_ctx to qcom_mdt_pas_load(), the allocated metadata > buffer is retained in wcnss->pas_ctx->ptr. In the previous qcom_mdt_load() > implementation, passing a NULL context signaled the backend to free the > buffer immediately. > > Since the remoteproc operations structure (wcnss_ops) lacks an .unprepare > hook, there doesn't seem to be a corresponding call to > qcom_pas_metadata_release(wcnss->pas_ctx) when the device is shut down. > > Will this persistently leak DMA memory each time the remoteproc is started > and stopped via sysfs?
As mentioned in the cover letter, this series has the dependency "firmware: qcom: pas: Make MDT buffer retention explicit" and as the change is already picked, this issue cannot occur and the buffer is freed immediately now as well. https://git.kernel.org/pub/scm/linux/kernel/git/next/linux-next.git/commit/?id=2aa8fafc08e777e9da356307c1a0117e7f4277cf > > > if (ret) > > return ret; > > [ ... ] > > > @@ -600,6 +599,11 @@ static int wcnss_probe(struct platform_device *pdev) > > if (ret) > > return ret; > > > > + wcnss->pas_ctx = devm_qcom_pas_context_alloc(wcnss->dev, WCNSS_PAS_ID, > > + wcnss->mem_phys, > > wcnss->mem_size); > > [Severity: High] > The commit message explicitly states the goal is to include the shmbridge > prepare step required before TrustZone authentication. However, does this > actually enable the shmbridge? > > The newly allocated wcnss->pas_ctx is zero-initialized by > devm_qcom_pas_context_alloc(), so pas_ctx->use_tzmem defaults to false. When > use_tzmem is false, qcom_scm_pas_prepare_and_auth_reset() explicitly skips > creating the shmbridge and falls back to the legacy behavior. > > Should wcnss->pas_ctx->use_tzmem be set to true here to fulfill the stated > security objective? No, it is correct. No change to use_tzmem is required; it should remain false. Perhaps the commit description needs to be updated to convey the no-functional-change behavior. > > > + if (IS_ERR(wcnss->pas_ctx)) > > + return PTR_ERR(wcnss->pas_ctx); > > + > > wcnss->pmu_cfg = mmio + data->pmu_offset; > > wcnss->spare_out = mmio + data->spare_offset; > > -- > Sashiko AI review ยท > https://sashiko.dev/#/patchset/[email protected]?part=1 -- -Mukesh Ojha
