class ArbitrationControl2

Maximum number of write requests accepted from WD (display capture).

This field has a non-trivial encoding. It should be used via the helpers

`display_capture_write_request_limit()` and

`set_display_capture_write_request_limit()`.

Maximum number of write requests accepted from WD (display capture).

This field has a non-trivial encoding. It should be used via the helpers

`display_capture_write_request_limit()` and

`set_display_capture_write_request_limit()`.

Maximum number of write requests accepted from DSB (Display State Buffer).

This field has a non-trivial encoding. It should be used via the helpers

`display_state_buffer_write_request_limit_tiger_lake()` and

`set_display_state_buffer_write_request_limit_tiger_lake()`.

Maximum number of write requests accepted from DSB (Display State Buffer).

This field has a non-trivial encoding. It should be used via the helpers

`display_state_buffer_write_request_limit_tiger_lake()` and

`set_display_state_buffer_write_request_limit_tiger_lake()`.

Maximum number of requests accepted from "par5".

"par5" does not appear to be documented.

This field does not exist on Kaby Lake or Skylake.

This field has a non-trivial encoding.

Maximum number of requests accepted from "par5".

"par5" does not appear to be documented.

This field does not exist on Kaby Lake or Skylake.

This field has a non-trivial encoding.

Maximum number of requests accepted from FBC (framebuffer compression).

This field has a non-trivial encoding. It should be used via the helpers

`framebuffer_compression_request_limit_tiger_lake()` and

`set_framebuffer_compression_request_limit_tiger_lake()`.

Maximum number of requests accepted from FBC (framebuffer compression).

This field has a non-trivial encoding. It should be used via the helpers

`framebuffer_compression_request_limit_tiger_lake()` and

`set_framebuffer_compression_request_limit_tiger_lake()`.

If true, a HP request causes the arbiter to accept trickle feed requests.

Trickle feed requests are sent by DBUF (Display Buffer) clients that

react immediately when a DBUF block that they own frees up. By contrast,

DBUF clients may send batched requests after a few blocks free up. The

concept was most recently documented in IHD-OS-VLV-Vol10-04.14 (the

Valleyview Display Engine PRM).

If this bit is enabled, the arbiter allows trickle feed requests from all

clients when it receives a HP (high-priority) request from one client.

If true, a HP request causes the arbiter to accept trickle feed requests.

Trickle feed requests are sent by DBUF (Display Buffer) clients that

react immediately when a DBUF block that they own frees up. By contrast,

DBUF clients may send batched requests after a few blocks free up. The

concept was most recently documented in IHD-OS-VLV-Vol10-04.14 (the

Valleyview Display Engine PRM).

If this bit is enabled, the arbiter allows trickle feed requests from all

clients when it receives a HP (high-priority) request from one client.

The maximum number of in-flight LP (low-priority) read transactions.

This field has a non-trivial encoding. It should be used via the helpers

`max_inflight_low_priority_read_requests()` and

`set_max_inflight_low_priority_read_requests()`.

The maximum number of in-flight LP (low-priority) read transactions.

This field has a non-trivial encoding. It should be used via the helpers

`max_inflight_low_priority_read_requests()` and

`set_max_inflight_low_priority_read_requests()`.

The maximum number of in-flight HP (high-priority) read transactions.

This field has a non-trivial encoding. It should be used via the helpers

`max_inflight_high_priority_read_requests()` and

`set_max_inflight_high_priority_read_requests()`.

The maximum number of in-flight HP (high-priority) read transactions.

This field has a non-trivial encoding. It should be used via the helpers

`max_inflight_high_priority_read_requests()` and

`set_max_inflight_high_priority_read_requests()`.

If false, the Display Engine always sends requests at isochronous priority.

If true, the Display Engine may issue request at lower priority, which may

improve performance for other memory-bound workloads. The Display Engine's

requests will be demoted once the relevant transition watermark is reached.

If the transition watermark is not enabled, the Display Engine's requests

are demoted once the DBUF (Display Buffer) is full.

The hardware only reads this bit from the global arbiter's configuration.

Other arbiters, such as Arbiter 1 and 2 on Tiger Lake, use the global

arbiter's setting.

If false, the Display Engine always sends requests at isochronous priority.

If true, the Display Engine may issue request at lower priority, which may

improve performance for other memory-bound workloads. The Display Engine's

requests will be demoted once the relevant transition watermark is reached.

If the transition watermark is not enabled, the Display Engine's requests

are demoted once the DBUF (Display Buffer) is full.

The hardware only reads this bit from the global arbiter's configuration.

Other arbiters, such as Arbiter 1 and 2 on Tiger Lake, use the global

arbiter's setting.

Minimum RTID queue size required to start HP (high-priority) transactions.

This field has a non-trivial encoding. It should be used via the helpers

`request_transaction_id_queue_watermark()` and

`set_request_transaction_id_queue_watermark()`.

Minimum RTID queue size required to start HP (high-priority) transactions.

This field has a non-trivial encoding. It should be used via the helpers

`request_transaction_id_queue_watermark()` and

`set_request_transaction_id_queue_watermark()`.

Maximum number of write requests accepted from WD (display capture).

WD (Wireless Display / display capture) transcoders issue LP (low-priority)

write requests. Re-arbitration occurs after this limit is hit.

See `display_capture_write_request_limit()` for details.

Maximum number of write requests accepted from DSB (Display State Buffer).

DSB (Display State Buffer) engines issues LP (low-priority) write requests.

Re-arbitration occurs after this limit is hit.

This field does not exist on Kaby Lake or Skylake.

See `display_state_buffer_write_request_limit_tiger_lake()` for details.

Maximum number of requests accepted from "par5".

The "par5" client / feature does not appear to be documented. Nevertheless,

re-arbitration occurs after this limit is hit.

This field does not exist on Kaby Lake or Skylake.

See `par5_request_limit_tiger_lake()` for details.

Maximum number of requests accepted from FBC (framebuffer compression).

The FBC (framebuffer compression) feature issues LP (low-priority) write

requests to update the compressed framebuffer. Re-arbitration occurs after

this limit is hit.

This field does not exist on Kaby Lake or Skylake.

See `framebuffer_compression_request_limit_tiger_lake()` for details.

The maximum number of in-flight LP (low-priority) read transactions.

See `max_inflight_low_priority_read_requests()` for details.

The maximum number of in-flight HP (high-priority) read transactions.

See `max_inflight_high_priority_read_requests()` for details.

Minimum RTID queue size required to start HP (high-priority) transactions.

RTIDs (Request Transaction IDs) are credits issued by the per-processor

cache coherence coordinator, which is the CHA (combined Caching and Home

Agent) on recent CPU models. These credits track outstanding transactions

across the cache coherence bus, which is UPI (Ultra-Path Interconnect) on

recent CPU models, and QPI (Quick-Path Interconnect) on older models. The

"RTID" acronym is defined in some processor datasheets, such as document

735086 (Ice Lake Xeon processor datasheet, volume 2). The CHA is described

at a very high level in uncore performance monitoring manuals, such as

document

This watermark is the minimum level of the RTID FIFO (queue) required for

the arbiter to start HP (high-priority) transactions.

The pseudocode in IHD-OS-TGL-Vol 12-1.22-Rev2.0 "Bandwidth Restrictions" >

"Available Memory Bandwidth Calculation" on pages 163-166 suggests that

each memory transaction (on each DRAM channel) requires an RTID.

See `request_transaction_id_queue_watermark()` for details.

`arbiter_index` zero (0) represents the global arbiter. The following

indexes represent the numbered arbiters.

Kaby Lake and Skylake only have a global arbiter.