Skip to content

Add ZTE DingHai Ethernet PF driver

[email protected] lkml netdev 10 versions – New

Get series with b4
b4 am [email protected]

Versions newest first · linked by author and title

  1. v10 Latest2 patches 0 repliesNew 1 of 2 applied
  2. v92 patches 2 repliesSuperseded
  3. v82 patches 2 repliesSuperseded
Earlier versions · 7
VersionPostedPatchesRepliesStatus
v72 patches3 repliesSuperseded
v62 patches3 repliesSuperseded
v52 patches2 repliesSuperseded
v42 patches2 repliesSuperseded
v33 patches3 repliesSuperseded
v23 patches5 repliesSuperseded
v13 patches5 repliesSuperseded

Changes in v10 from the cover letter

- Makefile: remove unnecessary ccflags-y -I$(src).
- zxdh_pf_pci_init(): dma_set_mask_and_coherent() never fails for DMA_BIT_MASK(64).Just call it once.
- zxdh_pf_probe(): reorder local variable declarations in reverse xmas tree order.Don't print dev_err on devlink_alloc() failure.
- Remove redundant initialization of fields in kzalloc'ed structs (bar_chan_valid, vepa).
- en_pf.h: drop unused includes,forward declaration of struct devlink, unused enum zxdh_coredev_type, and unused struct fields (bar_chan_valid, vepa, irq_lock, lock mutex, coredev_type).
- Reply to Sashiko's review comments:
(1) zxdh_pf_map_capability(): validate the BAR index against ZXDH_PF_MAX_BAR_VAL right after re-reading it from config space.
(2) zxdh_pf_map_capability(): bound-check length + offset (instead of minlen + offset) against the BAR size before mapping, so the reported *len never exceeds the actually mapped region and later MMIO accesses cannot run past the pci_iomap_range() boundary.
(3) Search capabilities with IORESOURCE_MEM only (common/notify/device cfg), matching the pci_request_selected_regions(IORESOURCE_MEM) call, so a capability located in an I/O BAR is never mapped without having been requested.
(4) zxdh_pf_notify_cfg_init(): rewrite
"if (notify_length + (notify_offset % PAGE_SIZE) <= PAGE_SIZE)" as a
subtraction: notify_length <= PAGE_SIZE - (notify_offset % PAGE_SIZE). This removes the addition, so there is no integer wrap a malicious length could exploit to bypass the check.

Patches in v10 1 of 2 applied

#Subject and filesSizeRepliesApplied
1/2dinghai: add ZTE network driver supportMAINTAINERS drivers/net/ethernet/Kconfig drivers/net/ethernet/Makefile +6 more+278 -00 repliesApplied
2/2dinghai: add hardware register access and PCI capability scanningdrivers/net/ethernet/zte/dinghai/dh_queue.h drivers/net/ethernet/zte/dinghai/en_pf.c drivers/net/ethernet/zte/dinghai/en_pf.h+377 -00 replies

Reviews trailers per patch

No Reviewed-by, Acked-by or Tested-by trailers for v10 yet.

Read the v10 thread

Cover letter [PATCH net-next v10 0/2] Add ZTE DingHai Ethernet PF driver

From: Junyang Han <[email protected]>

This series adds initial support for the ZTE DingHai Ethernet controller, a high-performance PCIe Ethernet device supporting SR-IOV, hardware offloading, and advanced virtualization features.

Changes from v9:
- Makefile: remove unnecessary ccflags-y -I$(src).
- zxdh_pf_pci_init(): dma_set_mask_and_coherent() never fails for DMA_BIT_MASK(64).Just call it once.
- zxdh_pf_probe(): reorder local variable declarations in reverse xmas tree order.Don't print dev_err on devlink_alloc() failure.
- Remove redundant initialization of fields in kzalloc'ed structs (bar_chan_valid, vepa).
- en_pf.h: drop unused includes,forward declaration of struct devlink, unused enum zxdh_coredev_type, and unused struct fields (bar_chan_valid, vepa, irq_lock, lock mutex, coredev_type).
- Reply to Sashiko's review comments:
(1) zxdh_pf_map_capability(): validate the BAR index against ZXDH_PF_MAX_BAR_VAL right after re-reading it from config space.
(2) zxdh_pf_map_capability(): bound-check length + offset (instead of minlen + offset) against the BAR size before mapping, so the reported *len never exceeds the actually mapped region and later MMIO accesses cannot run past the pci_iomap_range() boundary.
(3) Search capabilities with IORESOURCE_MEM only (common/notify/device cfg), matching the pci_request_selected_regions(IORESOURCE_MEM) call, so a capability located in an I/O BAR is never mapped without having been requested.
(4) zxdh_pf_notify_cfg_init(): rewrite
"if (notify_length + (notify_offset % PAGE_SIZE) <= PAGE_SIZE)" as a
subtraction: notify_length <= PAGE_SIZE - (notify_offset % PAGE_SIZE). This removes the addition, so there is no integer wrap a malicious length could exploit to bypass the check.

Changes from v8:
- Unify the three prefix variants (dh_core* / dh_pf* / zxdh_pf*) on the vendor namespace zxdh_* .
- Address the _dev / _device suffix inconsistency: zxdh_pf_device -> zxdh_pf_dev, matching zxdh_core_dev and the upstream pci_dev / net_dev / usb_dev convention.
- Keep struct zxdh_core_dev and struct zxdh_pf_dev split, with zxdh_pf_dev hung off zxdh_core_dev::priv via the container-of pattern. The same zxdh_core_dev handle will be shared by the SF (Sub-Function) driver queued for the next merge window, and by MPF drivers further out. Each device type hangs its own private struct off priv: PF -> zxdh_pf_dev, SF -> zxdh_sf_dev (planned). Merging now would force a re-split when SF lands.
- Address v8 review comment "introduce functions yet you never use them".
- Wire up zxdh_pf_modern_cfg_init() into the probe path and zxdh_pf_modern_cfg_uninit() into the remove path.
- Drop the ZXDH_CONFIG_S_* and ZXDH_MSI_NO_VECTOR macros from dh_queue.h.

Changes from v7:
- Remove the redundant depends on NET_VENDOR_ZTE since it's already implied by the enclosing if NET_VENDOR_ZTE block in the parent Kconfig.
- Check BAR 0 is an IORESOURCE_MEM resource before ioremap().
- Rework .shutdown to quiesce hardware only.
- Add NULL guards in zxdh_set_mac() and zxdh_get_mac().
- Zero the output buffer in zxdh_pf_get_vf_mac() up front.

- Two further AI-review points were investigated and require no code change:

(1) *len is written before the bounds check inside zxdh_pf_map_capability. The two fields written through *len and *bar_off are pf_dev->notify_len and pf_dev->dev_cfg_bar_off, both populated from this function. pf_dev->notify_len is only consulted when pf_dev->notify_base != NULL, which only happens when the corresponding map_capability() call returned a non-NULL pointer; pf_dev->dev_cfg_bar_off is only consumed by zxdh_pf_get_vf_mac and zxdh_pf_set_vf_mac_reg, both of which gate their access on pf_dev->pf_sriov_cap_base. So the early *len / *bar_off writes cannot produce values that end up consumed.

(2) The expression off * notify_offset_multiplier is u16 * u32; the overflow only triggers when notify_offset_multiplier exceeds UINT_MAX / 65535 (≈ 65537), which is well beyond any value a PCI config register would carry for this field. A malformed multiplier of that magnitude would also fail sane probing of the rest of the cap layout, so it cannot reach this arithmetic. The boundaries in practice are enforced by the 64-bit comparison at the top of zxdh_pf_map_vq_notify's notify_base branch, where off is promoted to u64 before the multiply. The three subsequent 32-bit multiplies in that function only feed pointer arithmetic against pf_dev->notify_base; a wrong offset would resolve to an address outside the mapped region and fail on access rather than corrupt kernel state.

Changes from v6:
- Clean up PCI device ID table (drop , 0, use { } sentinel)
- Move dh_core_alloc_priv/dh_core_free_priv from header to .c, they don't need to be static inline
- Use if (!x) and drop the "%i, x" dev_err() messages in common/notify cfg init.

Changes from v5:
- Drop dev_info() log spam.
- Propagate the real error code from dh_pf_pci_init() in dh_pf_probe() instead of hard-coding -ENOMEM.
- Register devlink only after dh_pf_pci_init() succeeds, and in dh_pf_remove()/dh_pf_shutdown() unregister devlink before tearing down PCI/mutex/priv.
- Drop the "dh_dev->priv = NULL" assignment from dh_core_free_priv().

Changes from v4:
- Fix sparse warning: add __iomem annotation to priv pointer
- Fix Clang format warning
- Use "dinghai:" as patch subject prefix
- Ensure proper patch threading

Note: Sent manually due to temporary git send-email unavailability in our environment. Will use git send-email or b4 for future submissions. Apologies for any inconvenience.

Changes from v3:
- Merged patches 1 and 2:
Combined initial framework with logging infrastructure
for better code organization and reduced patch count. This was done because the logging infrastructure now uses Linux's built-in dev_err(), dev_info(), dev_warn(), etc. macros instead of a custom logging system.
- Removed unnecessary variable initialization:
Fixed "don't initialise variables".
- Fixed variable declaration order:
Applied "Reverse Christmas tree" ordering with variables declared from longest to shortest line length.
- Code quality improvements:
Fixed all checkpatch.pl issues (alignment, formatting, etc.).

Changes from v2:
- Address maintainer feedback from v2 review:
* Remove meaningless initialization
* Change dh_pf_pci_table to static const for better encapsulation
* Simplify MODULE_DESCRIPTION for brevity
- Coding style improvements:
* Ensure all lines are within 80-column limit
* Use kernel types (u32/u8) consistently throughout
* Improve code readability with better formatting

Changes from v1 (addressing feedback from AndrewLunn):
- Update copyright years to 2022-2026
- Remove DRV_VERSION, MODULE_VERSION and related boilerplate
- Fix MODULE_AUTHOR to use person with email address
- Use module_pci_driver() instead of manual init/exit
- Remove empty suspend/resume callbacks
- Replace char priv[] flexible array with void *priv + kzalloc
- Switch logging from printk wrappers to dev_*() based macros
- Remove dh_helper.h and dh_log.c, simplify to dh_log.h only
- Fix variable declaration ordering (reverse Christmas tree)
- Remove unnecessary NULL check in remove and pf_dev=NULL in probe
- Fix indentation and remove unnecessary type casts
- Use kernel idiomatic "if (ret)" style

This is the initial submission and only includes the PF (Physical Function) driver. The VF (Virtual Function) driver will be submitted separately.

Junyang Han (2):
dinghai: add ZTE network driver support
dinghai: add hardware register access and PCI capability scanning

 MAINTAINERS                                 |   6 +
 drivers/net/ethernet/Kconfig                |   1 +
 drivers/net/ethernet/Makefile               |   1 +
 drivers/net/ethernet/zte/Kconfig            |  20 +
 drivers/net/ethernet/zte/Makefile           |   6 +
 drivers/net/ethernet/zte/dinghai/Kconfig    |  34 ++
 drivers/net/ethernet/zte/dinghai/Makefile   |   7 +
 drivers/net/ethernet/zte/dinghai/dh_queue.h |  56 +++
 drivers/net/ethernet/zte/dinghai/en_pf.c    | 449 ++++++++++++++++++++
 drivers/net/ethernet/zte/dinghai/en_pf.h    |  75 ++++
 10 files changed, 655 insertions(+)
 create mode 100644 drivers/net/ethernet/zte/Kconfig
 create mode 100644 drivers/net/ethernet/zte/Makefile
 create mode 100644 drivers/net/ethernet/zte/dinghai/Kconfig
 create mode 100644 drivers/net/ethernet/zte/dinghai/Makefile
 create mode 100644 drivers/net/ethernet/zte/dinghai/dh_queue.h
 create mode 100644 drivers/net/ethernet/zte/dinghai/en_pf.c
 create mode 100644 drivers/net/ethernet/zte/dinghai/en_pf.h
Signature · 1 line
2.27.0