跳到主要内容
主页代理入口联系客服

Hardware guide

DMA Firmware: The Part That Actually Determines Behaviour

Two cards built on the same FPGA can behave completely differently because firmware decides what the host sees: the vendor and device ID, the class code, the BAR layout, and how the device responds during enumeration. Understanding firmware explains both why cards stop working after driver updates and why "which card" is a less useful question than "whose firmware, maintained how".

Last reviewed: 2026-11-14Published by Heino HMA Hardware Lab

Firmware is the identity layer

The silicon provides capability. The bitstream provides identity. If firmware reports a device ID that the host operating system no longer has a driver for, or that a checker treats as anomalous, the card stops being useful regardless of what it can physically do.

Stored inConfiguration flash on the board
ControlsVendor/device ID, class code, BARs, behaviour
Typical sizeTens to hundreds of KB of bitstream
Failure modeStops enumerating after host update

What firmware actually controls during enumeration

When the host boots, the PCIe configuration space is read for every endpoint. Firmware determines what is returned: the vendor ID, device ID, subsystem IDs, class code, revision, and the BAR size and alignment requests. All of that happens before any operating system driver is involved.

After the BAR is assigned, firmware decides how the device responds to memory transactions from the host. A minimal implementation acknowledges and returns data from the link; a more elaborate one emulates a specific device class so that a stock driver binds to it.

FieldSet by firmwareWhy it matters
Vendor / device IDYesDetermines which driver binds, and what a checker sees
Class codeYesGroups the device into a category
BAR size / typeYesMemory or I/O mapping the OS must allocate
Subsystem IDYesOEM identification, often overlooked
RevisionYesDistinguishes firmware versions

Why a working card stops working after a Windows update

The most common failure is not hardware: it is that the host OS changed how it classifies or initialises a device. A firmware reporting a class code that Windows previously accepted without a driver may now trigger a different initialisation path, or fail enumeration entirely.

The second common failure is IOMMU enforcement becoming active. Windows 11 ships with Virtualization-Based Security more aggressively enabled than Windows 10 did; once Memory Integrity and Kernel DMA Protection are on, a bus master that assumed permissive physical addressing stops being able to transfer.

Neither failure is fixed by reinstalling drivers, because the card has no driver to reinstall. Both are resolved by changing firmware or changing the host configuration — which is why firmware maintenance, not card specification, predicts longevity.

How to verify what firmware is actually running

On Windows, Device Manager shows the vendor and device ID the card reports under Properties → Details → Hardware Ids. That string is the firmware identity, and it is the first thing to capture when something stops working.

Cross-check the reported IDs against what the vendor claims to ship. A mismatch means either the card was re-flashed or the vendor changed firmware without updating documentation — both are worth knowing before you spend time debugging elsewhere.

  • Device Manager → Properties → Details → Hardware Ids reveals the reported identity.
  • Compare against the vendor's stated firmware version for that board.
  • Record the string before applying any host update, so you can prove what changed.

The economics of firmware updates

Firmware maintenance is ongoing work: a vendor has to watch how operating systems and detection systems classify devices, then adjust. Vendors that ship once and stop are providing hardware with a fixed expiry date.

For buyers the practical question is not "does it work today" but "who fixes it in eight months". Ask for the release history. A vendor with no releases in the last year is either perfect or absent, and in this category it is almost never the former.

Re-flashing risk and recovery

Flashing firmware incorrectly can leave a board that no longer enumerates at all. Recovery usually requires an external programmer connected to the flash chip, which is not something most buyers have on hand.

Before flashing anything, confirm that the vendor provides a documented recovery procedure and a known-good image. If they do not, treat the board as unrecoverable if the flash fails and price that risk accordingly.

Frequently asked questions

01What does DMA firmware control?

Device identity during PCIe enumeration — vendor and device ID, class code, BAR layout, revision — plus how the device responds to memory transactions. It is what determines how the host OS classifies the card.

02Why did my DMA card stop working after a Windows update?

Usually one of two things: the OS changed how it initialises that device class, or Virtualization-Based Security and Kernel DMA Protection became active and the IOMMU started enforcing mappings.

03Can I update firmware myself?

Sometimes, but a failed flash can leave the board unable to enumerate at all, and recovery typically needs an external flash programmer. Only proceed if the vendor documents the procedure and supplies a known-good image.

04How do I see which firmware identity my card reports?

Device Manager → Properties → Details → Hardware Ids. The vendor and device ID shown there is what the firmware reports.

Official references

  1. PCI-SIG: PCI Express specifications
  2. Microsoft: Kernel DMA Protection
  3. Microsoft: Memory integrity
  4. pcileech: firmware repository
HMA 安装支持

售后支持已迁移

IM 售后频道已关闭,售后服务已转移至 Discord。

前往 Discord