From: "Lai, Yi" <yi1.lai@intel.com>
To: Dan Williams <dan.j.williams@intel.com>
Cc: linux-coco@lists.linux.dev, linux-pci@vger.kernel.org,
gregkh@linuxfoundation.org, aik@amd.com, aneesh.kumar@kernel.org,
yilun.xu@linux.intel.com, bhelgaas@google.com,
alistair23@gmail.com, lukas@wunner.de, jgg@nvidia.com,
yi1.lai@intel.com
Subject: Re: [PATCH v2 16/19] samples/devsec: Introduce a "Device Security TSM" sample driver
Date: Fri, 27 Mar 2026 16:44:39 +0800 [thread overview]
Message-ID: <acZDd2AtDoSD9/UV@ly-workstation> (raw)
In-Reply-To: <20260303000207.1836586-17-dan.j.williams@intel.com>
On Mon, Mar 02, 2026 at 04:02:04PM -0800, Dan Williams wrote:
> There are 2 sides to a TEE Security Manager (TSM), the 'link' TSM, and the
> 'devsec' TSM. The 'link' TSM, outside the TEE, establishes physical link
> confidentiality and integerity, and a secure session for transporting
> commands the manage the security state of devices. The 'devsec' TSM, within
> the TEE, issues requests for confidential devices to lock their
> configuration and transition to secure operation.
>
> Implement a sample implementation of a 'devsec' TSM. This leverages the PCI
> core's ability to register multiple TSMs at a time to load a sample
> devsec_tsm module alongside the existing devsec_link_tsm module. When both
> are loaded the TSM personality is selected by choosing to 'connect' vs
> 'lock' the device.
>
> Drivers like tdx_guest, sev_guest, or arm-cca-guest are examples of "Device
> Security TSM" drivers.
>
> A devsec_pci driver is included to test the device_cc_probe() helper for
> drivers that need to coordinate some configuration before 'lock' and
> 'accept'.
>
> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
> ---
> samples/devsec/Makefile | 6 ++
> samples/devsec/pci.c | 39 +++++++++++++
> samples/devsec/tsm.c | 124 ++++++++++++++++++++++++++++++++++++++++
> 3 files changed, 169 insertions(+)
> create mode 100644 samples/devsec/pci.c
> create mode 100644 samples/devsec/tsm.c
>
> diff --git a/samples/devsec/Makefile b/samples/devsec/Makefile
> index da122eb8d23d..0c52448a629f 100644
> --- a/samples/devsec/Makefile
> +++ b/samples/devsec/Makefile
> @@ -8,3 +8,9 @@ devsec_bus-y := bus.o
>
> obj-$(CONFIG_SAMPLE_DEVSEC) += devsec_link_tsm.o
> devsec_link_tsm-y := link_tsm.o
> +
> +obj-$(CONFIG_SAMPLE_DEVSEC) += devsec_tsm.o
> +devsec_tsm-y := tsm.o
> +
> +obj-$(CONFIG_SAMPLE_DEVSEC) += devsec_pci.o
> +devsec_pci-y := pci.o
> diff --git a/samples/devsec/pci.c b/samples/devsec/pci.c
> new file mode 100644
> index 000000000000..50519be412ed
> --- /dev/null
> +++ b/samples/devsec/pci.c
> @@ -0,0 +1,39 @@
> +// SPDX-License-Identifier: GPL-2.0-only
> +/* Copyright (C) 2024 - 2026 Intel Corporation */
> +#include <linux/device.h>
> +#include <linux/module.h>
> +#include <linux/pci.h>
> +
> +static int devsec_pci_probe(struct pci_dev *pdev,
> + const struct pci_device_id *id)
> +{
> + void __iomem *base;
> + int rc;
> +
> + rc = pcim_enable_device(pdev);
> + if (rc)
> + return dev_err_probe(&pdev->dev, rc, "enable failed\n");
> +
> + base = pcim_iomap_region(pdev, 0, KBUILD_MODNAME);
> + if (IS_ERR(base))
> + return dev_err_probe(&pdev->dev, PTR_ERR(base),
> + "iomap failed\n");
> +
> + dev_dbg(&pdev->dev, "attach\n");
> + return 0;
> +}
> +
> +static const struct pci_device_id devsec_pci_ids[] = {
> + { PCI_DEVICE(0x8086, 0xffff), .override_only = 1, },
> + { }
> +};
> +
> +static struct pci_driver devsec_pci_driver = {
> + .name = "devsec_pci",
> + .probe = devsec_pci_probe,
> + .id_table = devsec_pci_ids,
> +};
> +
> +module_pci_driver(devsec_pci_driver);
> +MODULE_LICENSE("GPL");
> +MODULE_DESCRIPTION("Device Security Sample Infrastructure: Secure PCI Driver");
> diff --git a/samples/devsec/tsm.c b/samples/devsec/tsm.c
> new file mode 100644
> index 000000000000..46dbe668945a
> --- /dev/null
> +++ b/samples/devsec/tsm.c
> @@ -0,0 +1,124 @@
> +// SPDX-License-Identifier: GPL-2.0-only
> +/* Copyright (C) 2024 - 2026 Intel Corporation */
> +
> +#define dev_fmt(fmt) "devsec: " fmt
> +#include <linux/device/faux.h>
> +#include <linux/pci-tsm.h>
> +#include <linux/module.h>
> +#include <linux/pci.h>
> +#include <linux/tsm.h>
> +#include "devsec.h"
> +
> +struct devsec_dev_data {
> + struct pci_tsm_devsec pci;
> +};
> +
> +static struct devsec_dev_data *to_devsec_data(struct pci_tsm *tsm)
> +{
> + return container_of(tsm, struct devsec_dev_data, pci.base_tsm);
> +}
> +
> +static struct pci_tsm *devsec_tsm_lock(struct tsm_dev *tsm_dev, struct pci_dev *pdev)
> +{
> + int rc;
> +
> + struct devsec_dev_data *devsec_data __free(kfree) =
> + kzalloc(sizeof(*devsec_data), GFP_KERNEL);
> + if (!devsec_data)
> + return ERR_PTR(-ENOMEM);
> +
> + rc = pci_tsm_devsec_constructor(pdev, &devsec_data->pci, tsm_dev);
> + if (rc)
> + return ERR_PTR(rc);
> +
> + return &no_free_ptr(devsec_data)->pci.base_tsm;
> +}
> +
> +static void devsec_tsm_unlock(struct pci_tsm *tsm)
> +{
> + struct devsec_dev_data *devsec_data = to_devsec_data(tsm);
> + struct pci_tsm_devsec *devsec_tsm = to_pci_tsm_devsec(tsm);
> +
> + pci_tsm_mmio_teardown(devsec_tsm->mmio);
> + kfree(devsec_tsm->mmio);
> + kfree(devsec_data);
> +}
> +
Hi Dan,
While validating devsec mode transitions, I hit a reproducible crash in the
sample devsec driver.
Reproducer:
1. lock with devsec tsm
2. unlock
Observed: NULL pointer dereference in the MMIO teardown path
Expected: unlock from LOCKED should return to UNLOCKED safely.
My understanding is that this is a sample driver implementation bug - missing
NULL guard before MMIO teardown.
A follow-up question: do you prefer current design and each device
security TSM driver is responsible for MMIO check, or should tsm/core
adds a NULL guard to avoid potential crash?
Regards,
Yi Lai
> +static int devsec_tsm_accept(struct pci_dev *pdev)
> +{
> + struct pci_tsm_devsec *devsec_tsm = to_pci_tsm_devsec(pdev->tsm);
> + int rc;
> +
> + struct pci_tsm_mmio *mmio __free(kfree) =
> + kzalloc(struct_size(mmio, mmio, PCI_NUM_RESOURCES), GFP_KERNEL);
> + if (!mmio)
> + return -ENOMEM;
> +
> + /*
> + * Typically this range request would come from the TDISP Interface
> + * Report. For this sample, just request all BARs be marked encrypted
> + */
> + for (int i = 0; i < PCI_NUM_RESOURCES; i++) {
> + struct resource *res = pci_tsm_mmio_resource(mmio, mmio->nr);
> +
> + if (pci_resource_len(pdev, i) == 0 ||
> + !(pci_resource_flags(pdev, i) & IORESOURCE_MEM))
> + continue;
> + res->start = pci_resource_start(pdev, i);
> + res->end = pci_resource_end(pdev, i);
> + mmio->nr++;
> + }
> +
> + rc = pci_tsm_mmio_setup(pdev, mmio);
> + if (rc)
> + return rc;
> + devsec_tsm->mmio = no_free_ptr(mmio);
> + return 0;
> +}
> +
> +static struct pci_tsm_ops devsec_pci_ops = {
> + .lock = devsec_tsm_lock,
> + .unlock = devsec_tsm_unlock,
> + .accept = devsec_tsm_accept,
> +};
> +
> +static void devsec_tsm_remove(void *tsm_dev)
> +{
> + tsm_unregister(tsm_dev);
> +}
> +
> +static int devsec_tsm_probe(struct faux_device *fdev)
> +{
> + struct tsm_dev *tsm_dev;
> +
> + tsm_dev = tsm_register(&fdev->dev, &devsec_pci_ops);
> + if (IS_ERR(tsm_dev))
> + return PTR_ERR(tsm_dev);
> +
> + return devm_add_action_or_reset(&fdev->dev, devsec_tsm_remove,
> + tsm_dev);
> +}
> +
> +static struct faux_device *devsec_tsm;
> +
> +static const struct faux_device_ops devsec_device_ops = {
> + .probe = devsec_tsm_probe,
> +};
> +
> +static int __init devsec_tsm_init(void)
> +{
> + devsec_tsm = faux_device_create("devsec_tsm", NULL, &devsec_device_ops);
> + if (!devsec_tsm)
> + return -ENOMEM;
> + return 0;
> +}
> +module_init(devsec_tsm_init);
> +
> +static void __exit devsec_tsm_exit(void)
> +{
> + faux_device_destroy(devsec_tsm);
> +}
> +module_exit(devsec_tsm_exit);
> +
> +MODULE_LICENSE("GPL");
> +MODULE_DESCRIPTION("Device Security Sample Infrastructure: Device Security TSM Driver");
> --
> 2.52.0
>
next prev parent reply other threads:[~2026-03-27 8:44 UTC|newest]
Thread overview: 83+ messages / expand[flat|nested] mbox.gz Atom feed top
2026-03-03 0:01 [PATCH v2 00/19] PCI/TSM: TEE I/O infrastructure Dan Williams
2026-03-03 0:01 ` [PATCH v2 01/19] PCI/TSM: Report active IDE streams per host bridge Dan Williams
2026-03-09 16:36 ` Jonathan Cameron
2026-03-03 0:01 ` [PATCH v2 02/19] device core: Fix kernel-doc warnings in base.h Dan Williams
2026-03-09 16:39 ` Jonathan Cameron
2026-03-12 14:45 ` Greg KH
2026-03-03 0:01 ` [PATCH v2 03/19] device core: Introduce confidential device acceptance Dan Williams
2026-03-09 16:42 ` Jonathan Cameron
2026-03-12 14:44 ` Greg KH
2026-03-13 4:11 ` Dan Williams
2026-03-13 12:18 ` Greg KH
2026-03-13 18:53 ` Dan Williams
2026-03-13 19:07 ` Jason Gunthorpe
2026-03-13 13:32 ` Jason Gunthorpe
2026-03-13 19:56 ` Dan Williams
2026-03-13 20:24 ` Jason Gunthorpe
2026-03-14 1:32 ` Dan Williams
2026-03-23 18:14 ` Jason Gunthorpe
2026-03-24 2:18 ` Dan Williams
2026-03-24 12:36 ` Jason Gunthorpe
2026-03-25 4:13 ` Dan Williams
2026-03-25 11:56 ` Jason Gunthorpe
2026-03-26 1:27 ` Dan Williams
2026-03-26 12:00 ` Jason Gunthorpe
2026-03-26 15:00 ` Greg KH
2026-03-26 18:31 ` Dan Williams
2026-03-26 19:28 ` Jason Gunthorpe
2026-03-03 0:01 ` [PATCH v2 04/19] modules: Document the global async_probe parameter Dan Williams
2026-03-03 0:01 ` [PATCH v2 05/19] device core: Autoprobe considered harmful? Dan Williams
2026-03-09 16:58 ` Jonathan Cameron
2026-03-03 0:01 ` [PATCH v2 06/19] PCI/TSM: Add Device Security (TVM Guest) LOCK operation support Dan Williams
2026-03-03 0:01 ` [PATCH v2 07/19] PCI/TSM: Add Device Security (TVM Guest) ACCEPT " Dan Williams
2026-03-03 7:15 ` Baolu Lu
2026-03-03 0:01 ` [PATCH v2 08/19] PCI/TSM: Add "evidence" support Dan Williams
2026-03-03 3:14 ` kernel test robot
2026-03-03 10:16 ` Aneesh Kumar K.V
2026-03-03 16:38 ` Aneesh Kumar K.V
2026-03-13 10:07 ` Xu Yilun
2026-03-13 18:06 ` Dan Williams
2026-03-14 18:12 ` Jakub Kicinski
2026-03-17 1:45 ` Dan Williams
2026-03-19 0:00 ` Jakub Kicinski
2026-03-20 2:50 ` Dan Williams
2026-03-17 18:14 ` Lukas Wunner
2026-03-18 7:56 ` Dan Williams
2026-03-23 18:18 ` Jason Gunthorpe
2026-03-14 18:37 ` Lukas Wunner
2026-03-16 20:13 ` Dan Williams
2026-03-16 23:02 ` Dan Williams
2026-03-17 14:13 ` Lukas Wunner
2026-03-18 7:22 ` Dan Williams
2026-03-17 18:24 ` Lukas Wunner
2026-03-18 7:41 ` Dan Williams
2026-03-03 0:01 ` [PATCH v2 09/19] PCI/TSM: Support creating encrypted MMIO descriptors via TDISP Report Dan Williams
2026-03-04 17:14 ` dan.j.williams
2026-03-13 9:57 ` Xu Yilun
2026-03-05 4:46 ` Aneesh Kumar K.V
2026-03-13 10:23 ` Xu Yilun
2026-03-13 13:36 ` Jason Gunthorpe
2026-03-17 5:13 ` Xu Yilun
2026-03-24 3:26 ` Dan Williams
2026-03-24 12:38 ` Jason Gunthorpe
2026-03-16 5:19 ` Alexey Kardashevskiy
2026-03-23 18:20 ` Jason Gunthorpe
2026-03-26 23:38 ` Alexey Kardashevskiy
2026-03-27 11:49 ` Jason Gunthorpe
2026-03-03 0:01 ` [PATCH v2 10/19] x86, swiotlb: Teach swiotlb to skip "accepted" devices Dan Williams
2026-03-03 9:07 ` Aneesh Kumar K.V
2026-03-13 10:26 ` Xu Yilun
2026-03-03 0:01 ` [PATCH v2 11/19] x86, dma: Allow accepted devices to map private memory Dan Williams
2026-03-03 7:36 ` Alexey Kardashevskiy
2026-03-03 0:02 ` [PATCH v2 12/19] x86, ioremap, resource: Support IORES_DESC_ENCRYPTED for encrypted PCI MMIO Dan Williams
2026-03-19 15:34 ` Borislav Petkov
2026-03-03 0:02 ` [PATCH v2 13/19] samples/devsec: Introduce a PCI device-security bus + endpoint sample Dan Williams
2026-03-03 0:02 ` [PATCH v2 14/19] samples/devsec: Add sample IDE establishment Dan Williams
2026-03-03 0:02 ` [PATCH v2 15/19] samples/devsec: Add sample TSM bind and guest_request flows Dan Williams
2026-03-03 0:02 ` [PATCH v2 16/19] samples/devsec: Introduce a "Device Security TSM" sample driver Dan Williams
2026-03-27 8:44 ` Lai, Yi [this message]
2026-03-03 0:02 ` [PATCH v2 17/19] tools/testing/devsec: Add a script to exercise samples/devsec/ Dan Williams
2026-03-03 0:02 ` [PATCH v2 18/19] samples/devsec: Add evidence support Dan Williams
2026-03-03 0:02 ` [PATCH v2 19/19] tools/testing/devsec: Add basic evidence retrieval validation Dan Williams
2026-03-03 9:23 ` [PATCH v2 00/19] PCI/TSM: TEE I/O infrastructure Aneesh Kumar K.V
2026-03-03 22:01 ` dan.j.williams
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=acZDd2AtDoSD9/UV@ly-workstation \
--to=yi1.lai@intel.com \
--cc=aik@amd.com \
--cc=alistair23@gmail.com \
--cc=aneesh.kumar@kernel.org \
--cc=bhelgaas@google.com \
--cc=dan.j.williams@intel.com \
--cc=gregkh@linuxfoundation.org \
--cc=jgg@nvidia.com \
--cc=linux-coco@lists.linux.dev \
--cc=linux-pci@vger.kernel.org \
--cc=lukas@wunner.de \
--cc=yilun.xu@linux.intel.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox