12. General purpose direct memory access controller (GPDMA)

12.1 GPDMA introduction

The general purpose direct memory access (GPDMA) controller is a bus master and system peripheral.

The GPDMA is used to perform programmable data transfers between memory-mapped peripherals and/or memories via linked-lists, upon the control of an off-loaded CPU.

12.2 GPDMA main features

• • Dual bidirectional AHB master
• • Memory-mapped data transfers from a source to a destination:
  • – Peripheral-to-memory
  • – Memory-to-peripheral
  • – Memory-to-memory
  • – Peripheral-to-peripheral
  Transfer arbitration based on a 4-grade programmed priority at channel level:
  • – One high-priority traffic class, for time-sensitive channels (queue 3)
  • – Three low-priority traffic classes, with a weighted round-robin allocation for non time-sensitive channels (queues 0, 1, 2)
• • Per channel event generation, on any of the following events: transfer complete, half-transfer complete, data transfer error, user setting error, link transfer error, completed suspension, and trigger overrun
• • Per channel interrupt generation, with separately programmed interrupt enable per event
• • 16 concurrent GPDMA channels:
  • – Per channel FIFO for queuing source and destination transfers (see Section 12.3.1 )
  • – Intra-channel GPDMA transfers chaining via programmable linked-list into memory, supporting two execution modes: run-to-completion and link step mode
  • – Intra-channel and inter-channel GPDMA transfers chaining via programmable GPDMA input triggers connection to GPDMA task completion events
• • Per linked-list item within a channel:
  • – Separately programmed source and destination transfers
  • – Programmable data handling between source and destination: byte-based reordering, packing or unpacking, padding or truncation, sign extension and left/right realignment
  • – Programmable number of data bytes to be transferred from the source, defining the block level
  • – Linear source and destination addressing: either fixed or contiguously incremented addressing, programmed at a block level, between successive burst transfers

• – 2D source and destination addressing: programmable signed address offsets between successive burst transfers (non-contiguous addressing within a block, combined with programmable signed address offsets between successive blocks, at a second 2D/repeated block level, for a reduced set of channels (see Section 12.3.1 ))
• – Support for scatter-gather (multi-buffer transfers), data interleaving and deinterleaving via 2D addressing
• – Programmable GPDMA request and trigger selection
• – Programmable GPDMA half-transfer and transfer complete events generation
• – Pointer to the next linked-list item and its data structure in memory, with automatic update of the GPDMA linked-list control registers
• – Channel abort and restart
• • Debug:
  • – Channel suspend and resume support
  • – Channel status reporting, including FIFO level, and event flags
• • Privileged/unprivileged support:
  • – Support for privileged and unprivileged GPDMA transfers, independently at a channel level
  • – Privileged-aware AHB slave port

12.3 GPDMA implementation

12.3.1 GPDMA channels

A given GPDMA channel x is implemented with the following features and intended usage. To make the best use of the GPDMA performances, the table below lists some general recommendations, allowing the user to select and allocate a channel, given its implemented FIFO size and the requested GPDMA transfer.

Table 88. GPDMA1 channel implementation

12.3.2 GPDMA in low-power modes

The GPDMA wake-up feature is implemented in the device low-power modes as per the table below.

Table 89. GPDMA1 wake-up in low-power modes

12.3.3 GPDMA requests

A GPDMA request from a peripheral can be assigned to a GPDMA channel x, via REQSEL[6:0] in GPDMA_CxTR2, provided that SWREQ = 0.

The GPDMA requests mapping is specified in the table below.

Table 90. Programmed GPDMA1 request

Table 90. Programmed GPDMA1 request (continued)

Table 90. Programmed GPDMA1 request (continued)

Table 90. Programmed GPDMA1 request (continued)

12.3.4 GPDMA block requests

Some GPDMA requests must be programmed as a block request, and not as a burst request. Then BREQ in GPDMA_CxTR2 must be set for a correct GPDMA execution of the requested peripheral transfer at the hardware level.

Table 91. Programmed GPDMA1 request as a block request

12.3.5 GPDMA channels with peripheral early termination

A GPDMA channel, if implemented with this feature, can support the early termination of the data transfer from the peripheral which does also support this feature.

Table 92. GPDMA1 channel with peripheral early termination

This GPDMA support is activated when the channel x is programmed with GPDMA_CxTR2.PFREQ = 1. Then, the peripheral itself can initiate and request a data transfer completion, before the GPDMA has transferred the whole block (see Section 12.4.14 for more details).

Table 93. Programmed GPDMA1 request with peripheral early termination

12.3.6 GPDMA triggers

A GPDMA trigger can be assigned to a GPDMA channel x, via TRIGSEL[5:0] in GPDMA_CxTR2, provided that TRIGPOL[1:0] defines a rising or a falling edge of the selected trigger (TRIGPOL[1:0] = 01 or TRIGPOL[1:0] = 10).

Table 94. Programmed GPDMA1 trigger

Table 94. Programmed GPDMA1 trigger (continued)

12.4 GPDMA functional description

12.4.1 GPDMA block diagram

Figure 87. GPDMA block diagram

12.4.2 GPDMA channel state and direct programming without any linked-list

After a GPDMA reset, a GPDMA channel x is in idle state. When the software writes 1 in GPDMA_CxCR.EN, the channel takes into account the value of the different channel configuration registers (GPDMA_CxXXX), switches to the active/non-idle state and starts to execute the corresponding requested data transfers.

After enabling/starting a GPDMA channel transfer by writing 1 in GPDMA_CxCR.EN, a GPDMA channel interrupt on a complete transfer notifies the software that the GPDMA channel is back in idle state (EN is then deasserted by hardware) and that the channel is ready to be reconfigured then enabled again.

Figure 88 illustrates this GPDMA direct programming without any linked-list (GPDMA_CxLLR = 0).

Figure 88. GPDMA channel direct programming without linked-list (GPDMA_CxLLR = 0)

graph TD
    subgraph Idle [Channel state = Idle]
        Init[Initialize DMA channel
(keeping DMA_CxLLR[31:0] = 0)]
        Reconfig[Reconfigure DMA channel
(keeping DMA_CxLLR[31:0] = 0)]
    end
    Init --> Enable[Enable DMA channel]
    Reconfig --> Enable
    Enable --> Active [Channel state = Active]
    subgraph Active
        Valid{Valid user setting?}
        Valid -- N --> USEF[Setting USEF = 1
Disabling DMA channel]
        Valid -- Y --> Transfer[Executing the data transfer
from the register file]
        Transfer --> NoError{No transfer error?}
        NoError -- N --> DTEF[Setting DTEF = 1
Disabling DMA channel]
        NoError -- Y --> TCF[Setting TCF = 1
Disabling DMA channel]
    end
    USEF --> End[/End/]
    DTEF --> End
    TCF --> End
  

12.4.3 GPDMA channel suspend and resume

The software can suspend on its own a channel still active, with the following sequence:

1. 1. The software writes 1 into the GPDMA_CxCR.SUSP bit.
2. 2. The software polls the suspended flag GPDMA_CxSR.SUSPF until SUSPF = 1, or waits for an interrupt previously enabled by writing 1 to GPDMA_CxCR.SUSPIE. Wait for the channel to be effectively in suspended state means wait for the completion of any ongoing GPDMA transfer over its master ports. Then the software can observe, in a steady state, any read register or register field that is hardware modifiable.

Note: An ongoing GPDMA transfer can be a data transfer (a source/destination burst transfer) or a link transfer for the internal update of the linked-list register file from the next linked-list item.

1. 3. The software safely resumes the suspended channel by writing 0 to GPDMA_CxCR.SUSP.

graph TD; A[Channel state = Active] --> B[Suspend the DMA channel (write 1 to CxCR.SUSP)]; B --> C{SUSPF=1?}; C -- N --> B; C -- Y --> D[Channel state = Suspended and Idle]; D --> E[Receiving suspended interrupt]; E --> F[Resume the DMA channel (write 0 to CxCR.SUSP)]; F --> G[Channel state = Active];

Note: A suspend and resume sequence does not impact the GPDMA_CxCR.EN bit. Suspending a channel (transfer) does not suspend a started trigger detection.

12.4.4 GPDMA channel abort and restart

Alternatively, like for aborting a continuous GPDMA transfer with a circular buffering or a double buffering, the software can abort, on its own, a still active channel with the following sequence:

1. 1. The software writes 1 into the GPDMA_CxCR.SUSP bit.
2. 2. The software polls suspended flag GPDMA_CxSR.SUSPF until SUSPF = 1, or waits for an interrupt previously enabled by writing 1 to GPDMA_CxCR.SUSPIE. Wait for the channel to be effectively in suspended state means wait for the completion of any ongoing GPDMA transfer over its master port.
3. 3. The software resets the channel by writing 1 to GPDMA_CxCR.RESET. This causes the reset of the FIFO, the reset of the channel internal state, the reset of the GPDMA_CxCR.EN bit, and the reset of the GPDMA_CxCR.SUSP bit.
4. 4. The software safely reconfigures the channel. The software must reprogram the hardware-modified GPDMA_CxBR1, GPDMA_CxSAR, and GPDMA_CxDAR registers.
5. 5. In order to restart the aborted then reprogrammed channel, the software enables it again by writing 1 to the GPDMA_CxCR.EN bit.

Figure 90. GPDMA channel abort and restart sequence

graph TD
    Start(( )) -.-> Active1[Channel state = Active]
    subgraph Active_State
        SUSP[Suspend the DMA channel<br/>(write 1 to CxCR.SUSP)]
        SUSPF{SUSPF=1?}
        SUSP -- or --> SUSPF
        SUSPF -- N --> SUSPF
    end
    SUSPF -- Y --> Suspended[Channel state = Suspended<br/>(and Idle)]
    subgraph Suspended_State
        Interrupt[Receiving suspended interrupt]
        RESET[Reset the DMA channel<br/>(write 1 to CxCR.RESET)]
        Interrupt --> RESET
    end
    RESET --> Idle[Channel state = Idle]
    subgraph Idle_State
        Reconfig[Reconfigure the DMA channel]
        Enable[Enable the DMA channel]
        Reconfig --> Enable
    end
    Enable --> Active2[Channel state = Active]
    Active2 -.-> End(( ))
  

12.4.5 GPDMA linked-list data structure

Alternatively to the direct programming mode, a channel can be programmed by a list of transfers, known as a list of linked-list items (LLI). Each LLI is defined by its data structure.

The base address in memory of the data structure of a next LLI _n+1 of a channel x is the sum of the following:

• • the link base address of the channel x (in GPDMA_CxLBAR)
• • the link address offset (LA[15:2] field in GPDMA_CxLLR). The linked-list register GPDMA_CxLLR is the updated result from the data structure of the previous LLI _n of the channel x.

The data structure for each LLI may be specific.

A linked-list data structure is addressed following the value of the following UT1, UT2, UB1, USA, UDA, UT3 and UB2 if present, and ULL bits in GPDMA_CxLLR.

In linked-list mode, each GPDMA linked-list register (GPDMA_CxTR1, GPDMA_CxTR2, GPDMA_CxBR1, GPDMA_CxSAR, GPDMA_CxDAR, GPDMA_CxTR3 and GPDMA_CxBR2 if present, and GPDMA_CxLLR) is conditionally and automatically updated from the next linked-list data structure in the memory, following the current value of the

GPDMA_CxLLR register that was conditionally updated from the linked-list data structure of the previous LLI.

Caution: The user must program the pointer to the next linked-list data structure (GPDMA_CxLLR[15:0]) not to exceed the 64-Kbyte addressable space defined by the link base address register (GPDMA_CxLBAR). The complete linked-list data structure must be included in the 64-Kbyte addressable space.

Static linked-list data structure

For example, when the update bits (UT1, UT2, UB1, USA, UDA, plus UB2 and UT3 if present, and ULL) in GPDMA_CxLLR are all asserted, the linked-list data structure in the memory is maximal with:

• • channel x (x = 0 to 11) contiguous 32-bit locations, including GPDMA_CxTR1, GPDMA_CxTR2, GPDMA_CxBR1, GPDMA_CxSAR, GPDMA_CxDAR and GPDMA_CxLLR (see Figure 91) and including the first linked-list register file (LLI ₀ ) and the next LLIs (such as LLI ₁ , LLI ₂ ) in the memory
• • channel x (x = 12 to 15), contiguous 32-bit locations, including GPDMA_CxTR1, GPDMA_CxTR2, GPDMA_CxBR1, GPDMA_CxSAR, GPDMA_CxDAR, GPDMA_CxTR3, GPDMA_CxBR2 and GPDMA_CxLLR, (see Figure 92), and including the first linked-list register file (LLI ₀ ) and the next LLIs (such as LLI ₁ , LLI ₂ ) in the memory

Figure 91. Static linked-list data structure (all Uxx = 1) of a linear addressing channel x

Figure 92. Static linked-list data structure (all Uxx = 1)
of a 2D addressing channel x

Dynamic linked-list data structure

Alternatively, the memory organization for the full list of LLIs can be compacted with specific data structure for each LLI.

If UT1 = 0 and UT2 = 1, the link address offset of the register GPDMA_CxLLR is pointing to the updated value of the GPDMA_CxTR2 instead of the GPDMA_CxTR1 which is not to be modified (see Figure 93 ).

Example: if UT1 = UB1 = USA = 0 and if UT3 = UB2 = 0, when channel x is with 2D addressing, and if UT2 = UDA = ULL = 1, the next LLI does not contain an (updated) value for GPDMA_CxTR1, nor GPDMA_CxBR1, nor GPDMA_CxSAR, nor GPDMA_CxTR3, nor GPDMA_CxBR2 when channel x is with 2D addressing. The next LLI contains an updated value for GPDMA_CxTR2, GPDMA_CxDAR, and GPDMA_CxLLR, as shown in Figure 94 .

Figure 93. GPDMA dynamic linked-list data structure of a linear addressing channel x

Figure 94. GPDMA dynamic linked-list data structure of a 2D addressing channel x

12.4.6 Linked-list item transfer execution

A LLI _n transfer is the sequence of:

1. 1. a data transfer: GPDMA executes the data transfer as described by the GPDMA internal register file (this data transfer can be void/null for LLI ₀ )
2. 2. a conditional link transfer: GPDMA automatically and conditionally updates its internal register file by the data structure of the next LLI _n+1 , as defined by the GPDMA_CxLLR value of the LLI _n .

Note: The initial data transfer as defined by the internal register file (LLI ₀ ) can be null (GPDMA_CxBR1.BNDT[15:0] = 0 and GPDMA_CxTR2.PFREQ = 0) provided that the conditional update bit UB1 in GPDMA_CxLLR is set (meaning there is a non-null data transfer described by the next LLI ₁ in the memory to be executed).

Depending on the intended GPDMA usage, a GPDMA channel x can be executed as described by the full linked-list (run-to-completion mode, GPDMA_CxCR.LSM = 0) or a GPDMA channel x can be programmed for a single execution of a LLI (link step mode, GPDMA_CxCR.LSM = 1), as described in the next sections.

12.4.7 GPDMA channel state and linked-list programming in run-to-completion mode

When GPDMA_CxCR.LSM = 0 (in full list execution mode, execution of the full sequence of LLIs, named run-to-completion mode), a GPDMA channel x is initially programmed, started

by writing 1 to GPDMA_CxCR.EN, and after completed at channel level. The channel transfer is:

• • configured with at least the following:
  • – the first LLI ₀ , internal linked-list register file: GPDMA_CxTR1, GPDMA_CxTR2, GPDMA_CxBR1, GPDMA_CxSAR, GPDMA_CxDAR, and GPDMA_CxLLR, plus GPDMA_CxTR3 and GPDMA_CxBR2
  • – the last LLI _N , described by the linked-list data structure in memory, as defined by the GPDMA_CxLLR reflecting the before last LLI _N-1
• • completed when GPDMA_CxLLR[31:0] = 0, GPDMA_CxBR1.BRC[10:0] = 0, and GPDMA_CxBR1.BNDT[15:0] = 0, at the end of the last LLI _N-1 transfer

GPDMA_CxLLR[31:0] = 0 is the condition of a linked-list based channel completion and means the following:

• • The 16 low significant bits GPDMA_CxLLR.LA[15:0] of the next link address are null.
• • All the update bits GPDMA_CxLLR.Uxx are null (UT1, UT2, UB1, USA, UDA and ULL, plus UB2 and UT3).

The channel may never be completed when GPDMA_CxLLR.LSM = 0:

• • If the last LLI _N is recursive, pointing to itself as a next LLI:
  • – either GPDMA_CxLLR.ULL = 1 and GPDMA_CxLLR.LA[15:2] is updated with the same value
  • – or GPDMA_CxLLR.ULL = 0
• • If LLI _N is pointing to a previous LLI

In the regular data transfer completion at a block level, GPDMA_CxBR1.BNDT[15:0] = 0 and GPDMA_CxBR1.BRC[10:0] = 0 (if present). Alternatively, a block transfer may be early completed by a peripheral (such as an I3C in Rx mode), and then BNDT[15:0] is not null (see Section 12.4.14 for more details).

In the typical run-to-completion mode, the allocation of a GPDMA channel, including its fine programming, is done once during the GPDMA initialization. In order to have a reserved data communication link and GPDMA service during run-time, for continuously repeated transfers (from/to a peripheral respectively to/from memory or for memory-to-memory transfers). This reserved data communication link can consist of a channel, or the channel can be shared and a repeated transfer consists of a sequence of LLIs.

Figure 95 depicts the GPDMA channel execution and its registers programming in run-to-completion mode.

Note:

Figure 95 is not intended to illustrate how often a TCEF can be raised, depending on the programmed value of TCEM[1:0] in GPDMA_CxTR2. It can be raised at (each) block completion, at (each) 2D block completion, at (each) LLI completion, or only at channel completion. In run-to-completion mode, whatever is the value of TCEM[1:0], at the channel completion, the hardware always set TCEF = 1 and disables the channel.

In Figure 95 , BNDT ≠ 0 is the typical condition for starting the first data transfer in this figure. This condition becomes (BNDT ≠ 0 and PFREQ = 1) if the peripheral requests a data transfer with early termination (see Section 12.3.5 ).

Figure 95. GPDMA channel execution and linked-list programming in run-to-completion mode (GPDMA_CxCR.LSM = 0)

graph TD
    subgraph Idle [Channel state = Idle]
        Init[Initialize DMA channel] --> Enable[Enable DMA channel]
    end
    Enable --> Active
    subgraph Active [Channel state = Active]
        Valid1{Valid user setting?} -- N --> Reconfig[Reconfigure DMA channel]
        Valid1 -- Y --> BNDT{BNDT ≠ 0?}
        BNDT -- N --> USEF1[Setting USEF = 1 Disabling DMA channel]
        BNDT -- Y --> Exec[Executing once the data transfer from the register file]
        Exec --> NoError1{No transfer error?}
        NoError1 -- N --> DTEF[Setting DTEF = 1 Disabling DMA channel]
        NoError1 -- Y --> LLR{LLR ≠ 0?}
        LLR -- N --> TCF[Setting TCF = 1 Disabling DMA channel]
        LLR -- Y --> LoadLLI[Loading next LLI into the register file]
        LoadLLI --> NoError2{No transfer error?}
        NoError2 -- N --> ULEF[Setting ULEF = 1 Disabling DMA channel]
        NoError2 -- Y --> Valid2{Valid user setting?}
        Valid2 -- N --> USEF2[Setting USEF = 1 Disabling DMA channel]
        Valid2 -- Y --> TCF
    end
    Reconfig --> Init
    USEF1 --> End[/End/]
    DTEF --> End
    ULEF --> End
    USEF2 --> End
    TCF --> End
    

MSV62631V1

Run-time inserting a \( LLI_n \) via an auxiliary channel, in run-to-completion mode

The start of the link transfer of the \( LLI_{n-1} \) (start of the \( LLI_n \) loading) can be conditioned by the occurrence of a trigger, when programming the following fields of the GPDMA_CxTR2 in the data structure of the \( LLI_{n-1} \) :

• • TRIGM[1:0] = 10 (link transfer triggering mode)
• • TRIGPOL[1:0] = 01 or 10 (rising or falling edge)
• • TRIGSEL[5:0] (see Section 12.3.6 for the trigger selection details)

Another auxiliary channel y can be used to store the channel x \( LLI_n \) in the memory and to generate a transfer complete event gpdma_chy_tc. By selecting this event as the input trigger of the link transfer of the \( LLI_{n-1} \) of the channel x, the software can pause the primary channel x after its \( LLI_{n-1} \) data transfer, until it is indeed written the \( LLI_n \) .

Figure 96 depicts such a dynamic elaboration of a linked-list of a primary channel x, via another auxiliary channel y.

Caution: This use case is restricted to an application with a \( LLI_{n-1} \) data transfer that does not need a trigger. The triggering mode of this \( LLI_{n-1} \) is used to load the next \( LLI_n \) .

12.4.8 GPDMA channel state and linked-list programming in link step mode

When \( GPDMA\_CxCR.LSM = 1 \) (in link step execution mode, single execution of one \( LLI \) ), a channel transfer is executed and completed after each single execution of a \( LLI \) , including its (conditional) data transfer and its (conditional) link transfer.

A GPDMA channel transfer can be programmed at \( LLI \) level, started by writing 1 into \( GPDMA\_CxCR.EN \) , and after completed at \( LLI \) level:

• • The current \( LLI_n \) transfer is described with:
  • – \( GPDMA\_CxTR1 \) defines the source/destination elementary single/burst transfers.
  • – \( GPDMA\_CxBR1 \) defines the number of bytes at a block level ( \( BNDT[15:0] \) ) and, for channel \( x \) ( \( x = 12 \) to \( 15 \) ), the number of blocks at a 2D/repeated block level ( \( BRC[10:0]+1 \) ) and the incrementing/decrementing mode for address offsets.

• – GPDMA_CxTR2 defines the input control (request, trigger) and the output control (transfer complete event) of the transfer.
• – GPDMA_CxSAR/GPDMA_CxDAR define the source/destination transfer start address.
• – GPDMA_CxTR3 for channel x (x = 12 to 15) defines the source/destination additional address offset between burst transfers.
• – GPDMA_CxBR2 for channel x (x = 12 to 15) defines the source/destination additional address offset between blocks at a 2D/repeated block level.
• – GPDMA_CxLLR defines the data structure and the address offset of the next \( LLI_{n+1} \) in the memory.
• • The current \( LLI_n \) transfer is completed after the single execution of the current \( LLI_n \) :
  • – after the (conditional) data transfer completion (when \( GPDMA\_CxBR1.BRC[10:0] = 0 \) , and \( GPDMA\_CxBR1.BNDT[15:0] = 0 \) )
  • – after the (conditional) update of the GPDMA link register file from the data structure of the next \( LLI_{n+1} \) in memory

Note: If a LLI is recursive (pointing to itself as a next LLI, either \( GPDMA\_CxLLR.ULL = 1 \) and \( GPDMA\_CxLLR.LA[15:2] \) is updated with the same value, or \( GPDMA\_CxLLR.ULL = 0 \) ), a channel in link step mode is completed after each repeated single execution of this LLI.

In the regular data transfer completion at a block level, \( GPDMA\_CxBR1.BNDT[15:0] = 0 \) and \( GPDMA\_CxBR1.BRC[10:0] = 0 \) . Alternatively, a block transfer may be early completed by a peripheral (such as an I3C in Rx mode), and then \( BNDT[15:0] \) is not null (see Section 12.4.14 for more details).

The link step mode can be used to elaborate dynamically LLIs in memory during run-time. The software can be facilitated by using a static data structure for any \( LLI_n \) (all update bits of GPDMA_CxLLR have a static value, \( LLI_n.LLR.LA = LLI_{n-1}.LLR.LA + \text{constant} \) ).

Figure 97 depicts the GPDMA channel execution mode, and its programming in link step mode.

Note: Figure 97 is not intended to illustrate how often a TCEF can be raised, depending on the programmed value of \( TCEM[1:0] \) in GPDMA_CxTR2. It can be raised at (each) block completion, at (each) 2D block completion, at (each) LLI completion, or only at the last LLI data transfer completion. In link step mode, the channel is disabled after each single execution of a LLI, and depending on the value of \( TCEM[1:0] \) a TCEF is raised or not. In Figure 97 , \( BNDT \neq 0 \) is the typical condition for starting the first data transfer. This condition becomes ( \( BNDT \neq 0 \) and \( PFREQ = 1 \) ) if the peripheral requests a data transfer with early termination (see Section 12.3.5 ).

Figure 97. GPDMA channel execution and linked-list programming in link step mode (GPDMA_CxCR.LSM = 1)

graph TD
    subgraph Idle
        Start([Start]) --> Init[Initialize DMA channel]
        Init --> Enable[Enable DMA channel]
    end
    Enable --> Active
    subgraph Active
        Valid1{Valid user setting?} -- N --> Disable1[Setting USEF = 1
Disabling DMA channel]
        Valid1 -- Y --> BNDT{BNDT ≠ 0?}
        BNDT -- N --> Disable2[Setting DTEF = 1
Disabling DMA channel]
        BNDT -- Y --> Transfer[Executing once the data transfer from the register file]
        Transfer --> NoError1{No transfer error?}
        NoError1 -- N --> Disable2
        NoError1 -- Y --> LLR{LLR ≠ 0?}
        LLR -- N --> Disable3[Setting TCF = 1
Disabling DMA channel]
        LLR -- Y --> LoadLLI[Loading next LLI into the register file]
        LoadLLI --> NoError2{No transfer error?}
        NoError2 -- N --> Disable4[Setting ULEF = 1
Disabling DMA channel]
        NoError2 -- Y --> Valid2{Valid user setting?}
        Valid2 -- N --> Disable5[Setting USEF = 1
Disabling DMA channel]
        Valid2 -- Y --> Disable3
    end
    Disable1 --> End[/End/]
    Disable2 --> End
    Disable3 --> End
    Disable4 --> End
    Disable5 --> End
    End --> Reconfigure[Reconfigure DMA channel]
    Reconfigure --> Init
    

MSv62633V1

Run-time adding a \( LLI_{n+1} \) in link step mode

During run-time, the software can defer the elaboration of the \( LLI_{n+1} \) (and next LLIs), until/after GPDMA executed the transfer from the \( LLI_{n-1} \) and loaded the \( LLI_n \) from the memory, as shown in Figure 98.

Figure 98. Building \( LLI_{n+1} \) : GPDMA dynamic linked-lists in link step mode

Run-time replacing a \( LLI_n \) with a new \( LLI_{n'} \) in link step mode (in linked-list register file)

In this link step mode, during run-time, the software can build and insert a new \( LLI_{n'} \) , after GPDMA executed the transfer from the \( LLI_{n-1} \) and loaded a formerly elaborated \( LLI_n \) from the memory by overwriting directly the linked-list register file with the new \( LLI_{n'} \) , as shown in Figure 99.

sequenceDiagram
    participant DMA channel
    participant CPU

    Note over DMA channel: LLI_{n-1} transfer
    DMA channel->>DMA channel: Executing LLI_{n-1} data transfer
    DMA channel->>DMA channel: Loading LLI_n
    DMA channel-->>CPU: Transfer complete interrupt
    
    CPU->>CPU: Build LLI_{n'} and overwrite linked-list register file
    CPU->>CPU: Enable DMA channel

    Note over DMA channel: LLI_{n'} transfer
    DMA channel->>DMA channel: Executing LLI_{n'} data transfer
    DMA channel->>DMA channel: Loading LLI_{n+1'}
    DMA channel-->>CPU: Transfer complete interrupt

    CPU->>CPU: Build LLI_{n+1'} and overwrite linked-list register file
    CPU->>CPU: Enable DMA channel

    Note over DMA channel: LLI_{n+1'} transfer
    DMA channel->>DMA channel: LLI_{n+1'} transfer
    DMA channel-->>CPU: Transfer complete interrupt
  

The software can build and insert a new \( LLI_{n'} \) and \( LLI_{n+1'} \) in the memory, after GPDMA executed the transfer from the \( LLI_{n-1} \) and loaded a formerly elaborated \( LLI_n \) from the memory, by overwriting partly the linked-list register file (GPDMA_CxBR1.BNDT[15:0] to be null and GPDMA_CxLLR to point to new \( LLI_{n'} \) ) as shown in Figure 100 .

Figure 100. Replace with a new \( LLI_n \) and \( LLI_{n+1} \) in memory in link step mode (option 1)

sequenceDiagram
    participant DMA Channel
    participant CPU

    Note over DMA Channel: LLI_{n-1} transfer
    DMA Channel->>DMA Channel: Executing LLI_{n-1} data transfer
    DMA Channel->>DMA Channel: Loading LLI_n
    DMA Channel-->>CPU: Transfer complete interrupt
    
    CPU->>CPU: Build LLI_{n'} and LLI_{n+1'} in memory
    CPU->>CPU: Write DMA_CxBR1.BNDT = 0
    CPU->>CPU: Write DMA_CxLLR to point to new LLI_{n'}
    CPU->>DMA Channel: Enable DMA channel

    Note over DMA Channel: LLI_{n'} transfer
    DMA Channel->>DMA Channel: Loading LLI_{n'}
    DMA Channel-->>CPU: Transfer complete interrupt
    CPU->>DMA Channel: Enable DMA channel

    Note over DMA Channel: LLI_{n+1'} transfer
    DMA Channel->>DMA Channel: Executing LLI_{n+1'} data transfer
    DMA Channel->>DMA Channel: Loading LLI_{n+1'}
    DMA Channel-->>CPU: Transfer complete interrupt

Run-time replacing a \( LLI_n \) with a new \( LLI_{n'} \) in link step mode

Other software implementations exist. Meanwhile GPDMA executes the transfer from the \( LLI_{n-1} \) and loads a formerly elaborated \( LLI_n \) from the memory (or even earlier), the software can do the following:

1. 1. Disable the NVIC for not being interrupted by the interrupt handling.
2. 2. Build a new \( LLI_{n'} \) and a new \( LLI_{n+1'} \) .
3. 3. Enable again the NVIC for the channel interrupt (transfer complete) notification.

The software in the interrupt handler for \( LLI_{n-1} \) is then restricted to overwrite GPDMA_CxBR1.BNDT[15:0] to be null and GPDMA_CxLLR to point to new \( LLI_{n'} \) , as shown in Figure 101.

Figure 101. Replace with a new \( LLI_{n'} \) and \( LLI_{n+1'} \) in memory in link step mode (option 2)

sequenceDiagram
    participant DMA as DMA channel
    participant CPU as CPU

    Note over DMA, CPU: LSM = 1 with 1-stage linked-list programming...

    rect rgb(240, 240, 240)
    Note left of DMA: LLI_{n-1} transfer
    DMA->>DMA: Executing LLI_{n-1} data transfer
    DMA->>DMA: Loading LLI_n
    end

    CPU->>CPU: Disable NVIC DMA irq channel
    CPU->>CPU: Build LLI_{n'} & LLI_{n+1'} in memory
    CPU->>CPU: Enable NVIC DMA irq channel

    DMA-->>CPU: Transfer complete interrupt

    CPU->>CPU: Write DMA_CxBR1.BNDT = 0
Write DMA_CxLLR to point to new LLI_{n'}
    CPU->>CPU: Enable DMA channel

    rect rgb(240, 240, 240)
    Note left of DMA: LLI_{n'} transfer
    DMA->>DMA: Loading LLI_{n'}
    end

    DMA-->>CPU: Transfer complete interrupt

    CPU->>CPU: Enable DMA channel

    rect rgb(240, 240, 240)
    Note left of DMA: LLI_{n+1'} transfer
    DMA->>DMA: Executing LLI_{n+1'} data transfer
    DMA->>DMA: Loading LLI_{n+1'}
    end

    DMA-->>CPU: Transfer complete interrupt

12.4.9 GPDMA channel state and linked-list programming

The software can reconfigure a channel when the channel is disabled (GPDMA_CxCR.EN = 0) and update the execution mode (GPDMA_CxCR.LSM) to change from/to run-to-completion mode to/from link step mode.

In any execution mode, the software can:

• • reprogram \( LLI_{n+1} \) in the memory to finally complete the channel by this \( LLI_{n+1} \) (clear the GPDMA_CxLLR of this \( LLI_{n+1} \) ), before this \( LLI_{n+1} \) is loaded/used by the GPDMA channel
• • abort and reconfigure the channel with a LSM update (see Section 12.4.4 .)

In link step mode, the software can clear LSM after each a single execution of any LLI, during \( LLI_{n-1} \) .

Figure 102 shows the overall and unified GPDMA linked-list programming, whatever is the execution mode.

Note: Figure 102 is not intended to illustrate how often a TCEF can be raised, depending on the programmed value of TCEM[1:0] in GPDMA_CxTR2. It can be raised at (each) block completion, at (each) 2D block completion, at (each) LLI completion, or only at the last LLI data transfer completion. In run-to-completion mode, whatever is the value of TCEM[1:0], at the channel completion the hardware always set TCEF = 1 and disables the channel. In link step mode, the channel is disabled after each single execution of a LLI, and depending on the value of TCEM[1:0] a TCEF is raised or not.

In Figure 102 , BNDT \( \neq \) 0 is the typical condition for starting the first data transfer. This condition becomes (BNDT \( \neq \) 0 and PFREQ = 1) if the peripheral requests a data transfer with early termination (see Section 12.3.5 ).

Figure 102. GPDMA channel execution and linked-list programming

graph TD
    subgraph Idle [Channel state = Idle]
        Init[Initialize DMA channel] --> Enable[Enable DMA channel]
    end
    Enable --> Active
    subgraph Active [Channel state = Active]
        V1{Valid user setting?} -- N --> USEF1[Setting USEF = 1
Disabling DMA channel]
        V1 -- Y --> BNDT{BNDT ≠ 0?}
        BNDT -- N --> USEF1
        BNDT -- Y --> Exec[Executing once the data transfer from the register file]
        Exec --> NE1{No transfer error?}
        NE1 -- N --> DTEF[Setting DTEF = 1
Disabling DMA channel]
        NE1 -- Y --> LLR{LLR ≠ 0?}
        LLR -- N --> TCF[Setting TCF = 1
Disabling DMA channel]
        LLR -- Y --> LLI[Loading next LLI into the register file]
        LLI --> NE2{No transfer error?}
        NE2 -- N --> ULEF[Setting ULEF = 1
Disabling DMA channel]
        NE2 -- Y --> V2{Valid user setting?}
        V2 -- N --> USEF2[Setting USEF = 1
Disabling DMA channel]
        V2 -- Y --> LSM{LSM = 1?}
        LSM -- N --> Exec
        LSM -- Y --> TCF
    end
    USEF1 --> End[/End/]
    DTEF --> End
    ULEF --> End
    USEF2 --> End
    TCF --> End
    TCF --> Reconfig[Reconfigure DMA channel]
    Reconfig --> Init
    

MSv62638V1

12.4.10 GPDMA FIFO-based transfers

There is a single transfer operation mode: the FIFO mode. There are FIFO-based transfers. Any channel x is implemented with a dedicated FIFO whose size is defined by dma_fifo_size[x] (see Section 12.3.1 for more details).

GPDMA burst

A programmed transfer at the lowest level is a GPDMA burst.

A GPDMA burst is a burst of data received from the source, or a burst of data sent to the destination. A source (and destination) burst is programmed with a burst length by the field SBL_1[5:0] (respectively DBL_1[5:0] ), and with a data width defined by the field SDW_LOG2[1:0] (respectively DDW_LOG2[1:0] ) in the GPDMA_CxTR1 register.

The addressing mode after each data (named beat) of a GPDMA burst is defined by SINC and DINC in GPDMA_CxTR1 , for source and destination respectively: either a fixed addressing or an incremented addressing with contiguous data.

The start and next addresses of a GPDMA source/destination burst (defined by GPDMA_CxSAR and GPDMA_CxDAR ) must be aligned with the respective data width.

The table below lists the main characteristics of a GPDMA burst.

Table 95. Programmed GPDMA source/destination burst

1. When S/DBL_1[5:0] = 0, burst is of length 1. Then burst can be also named as single.

The next burst address in the above table is the next source/destination default address pointed by GPDMA_CxSAR or GPDMA_CxDAR , once the programmed source/destination burst is completed. This default value refers to the fixed/contiguously incremented address.

When the channel has additional 2D addressing feature, this default value refers to the value without taking into account the two programmed incremented or decremented offsets. These two additional offsets (with a null default value) are applied:

• • after each completed source/destination burst, as defined respectively by GPDMA_CxTR2.SAO[12:0]/DAO[12:0] and GPDMA_CxBR1.SDEC/DDEC
• • after each completed block, as defined respectively by GPDMA_CxBR2.BRSAO[15:0]/BRDAO[15:0] and GPDMA_CxBR1.BRSDEC/BRDDEC)

Then, a 2D/repeated block can be addressed with a first programmed address jump after each completed burst, and with a second programmed address jump after each block, as depicted by Figure 103 with a 2D destination buffer.

Figure 103. Programmed 2D addressing

GPDMA FIFO-based burst

In FIFO-mode, a transfer generally consists of two pipelined and separated burst transfers:

• • one burst from the source to the FIFO over the allocated source master port, as defined by GPDMA_CxTR1.SAP
• • one burst from the FIFO to the destination over the allocated destination master port, as defined by GPDMA_CxTR1.DAP

GPDMA source burst

The requested source burst transfer to the FIFO can be scheduled as early as possible over the allocated port, depending on the current FIFO level versus the programmed burst size (when the FIFO is ready to get one new burst from the source):

where:

• • FIFO level is the current filling level of the FIFO, in bytes.
• • \( 2^{\text{dma\_fifo\_size}[x]} \) is the half the FIFO size of channel x, in bytes (see Section 12.3.1 for the implementation details and dma_fifo_size[x] value).
• • \( (\text{SBL\_1}[5:0]+1) * 2^{\text{SDW\_LOG2}[1:0]} \) is the size of the programmed source burst transfer, in bytes.

Based on the channel priority (GPDMA_CxCR.PRIO[1:0]), this ready FIFO-based source transfer is internally arbitrated versus the other requested and active channels.

GPDMA destination burst

The requested destination burst transfer from the FIFO can be scheduled as early as possible over the allocated port, depending on the current FIFO level versus the programmed burst size (when the FIFO is ready to push one new burst to the destination):

where:

• • FIFO level is the current filling level of the FIFO, in bytes.
• • \( (\text{DBL\_1}[5:0]+1) * 2^{\text{DDW\_LOG2}[1:0]} \) is the size of the programmed destination burst transfer, in bytes.

Based on the channel priority, this ready FIFO-based destination transfer is internally arbitrated versus the other requested and active channels.

GPDMA burst vs source block size, 1-Kbyte address boundary and FIFO size

The programmed source/destination GPDMA burst is implemented with an AHB burst as is, unless one of the following conditions is met:

• • When half of the FIFO size of the channel x is lower than the programmed source/destination burst size, the programmed source/destination GPDMA burst is implemented with a series of singles or bursts of a lower size. Each transfer is of a size that is lower than or equal to half of the FIFO size, without any user constraint.
• • If the source block size (GPDMA_CxBR1.BNDT[15:0]) is not a multiple of the source burst size but is a multiple of the data width of the source burst (GPDMA_CxTR1.SDW_LOG2[1:0]), the GPDMA modifies and shortens bursts into singles or bursts of lower length, in order to transfer exactly the source block size, without any user constraint.

• • if the source/destination burst transfer have crossed the 1-Kbyte address boundary on a AHB transfer, the GPDMA modifies and shortens the programmed burst into singles or bursts of lower length, to be compliant with the AHB protocol, without any user constraint.
• • If the source/destination burst length exceeds 16 on a AHB transfer, the GPDMA modifies and shortens the programmed burst into singles or bursts of lower length, to be compliant with the AHB protocol, without any user constraint.

In any case, the GPDMA keeps ensuring source/destination data (and address) integrity without any user constraint. The current FIFO level (software readable in GPDMA_CxSR) is compared to and updated with the effective transfer size, and the GPDMA re-arbitrates between each AHB single or burst transfer, possibly modified.

Based on the channel priority, each single or burst of a lower burst size versus the programmed burst, is internally arbitrated versus the other requested and active channels.

Note: In linked-list mode, the GPDMA read transfers related to the update of the linked-list parameters from the memory to the internal GPDMA registers, are scheduled over the link allocated port, as programmed by GPDMA_CxCR.LAP.

GPDMA data handling: byte-based reordering, packing/unpacking, padding/truncation, sign extension and left/right alignment

The data handling is controlled by GPDMA_CxTR1. The source/destination data width of the programmed burst is byte, half-word or word, as per the SDW_LOG2[1:0] and DDW_LOG2[1:0] fields (see Table 96 ).

The user can configure the data handling between transferred data from the source and transfer to the destination. More specifically, programmed data handling is orderly performed with:

1. 1. Byte-based source reordering
   • – If SBX = 1 and if source data width is a word, the two bytes of the unaligned half-word at the middle of each source data word are exchanged.
2. 2. Data width conversion by packing, unpacking, padding or truncation, if destination data width is different than the source data width, depending on PAM[1:0]:
   • – If destination data width > source data width, the post SBX source data is either right-aligned and padded with 0 s, or sign extended up to the destination data width, or is FIFO queued and packed up to the destination data width.
   • – If destination data width < source data width, the post SBX data is either right-aligned and left-truncated down to the destination data width, or is FIFO queued and unpacked and streamed down to the destination data width.
3. 3. Byte-based destination re-ordering:
   • – If DBX = 1 and if the destination data width is not a byte, the two bytes are exchanged within the aligned post PAM[1:0] half-words.
   • – If DHX = 1 and if the destination data width is neither a byte nor a half-word, the two aligned half-words are exchanged within the aligned post PAM[1:0] words.

Note: Left-alignment with 0s-padding can be achieved by programming both a right-alignment with a 0s-padding and a destination byte-based re-ordering.

The table below lists the possible data handling from the source to the destination.

Table 96. Programmed data handling

Table 96. Programmed data handling (continued)

Table 96. Programmed data handling (continued)

1. Data stream is timely ordered starting from the byte with the lowest index (B ₀ ).

2. RA= right aligned, LA = left aligned, RT = right truncated, LT = left truncated, 0P = zero bit padding up to the destination data width, SE = sign bit extended up to the destination data width.

12.4.11 GPDMA transfer request and arbitration

GPDMA transfer request

As defined by GPDMA_CxTR2, a programmed GPDMA data transfer is requested with one of the following:

• • a software request if the control bit SWREQ = 1: This is used typically by the CPU for a data transfer from a memory-mapped address to another memory mapped address (memory-to-memory, GPIO to/from memory)
• • an input hardware request coming from a peripheral if SWREQ = 0: The selection of the GPDMA hardware peripheral request is driven by the REQSEL[6:0] field (see Section 12.3.3 ). The selected hardware request can be one of the following:
  • – an hardware request from a peripheral configured in GPDMA mode (for a transfer from/to the peripheral data register respectively to/from the memory)
  • – an hardware request from a peripheral for its control registers update from the memory
  • – an hardware request from a peripheral for a read of its status registers transferred to the memory

Caution: The user must not assign a same input hardware peripheral GPDMA request via GPDMA_CxTR.REQSEL[6:0] to two different channels, if at a given time this request is asserted by the peripheral and each channel is ready to execute this requested data transfer. There is no user setting error reporting.

GPDMA transfer request for arbitration

A ready FIFO-based GPDMA source single/burst transfer (from the source address to the FIFO) to be scheduled over the allocated master port (GPDMA_CxTR1.SAP) is arbitrated based on the channel priority (GPDMA_CxCR.PRIO[1:0]) versus the other simultaneous requested GPDMA transfers to the same master port.

A ready FIFO-based GPDMA destination single/burst transfer (from the FIFO to the destination address) to be scheduled over the allocated master port (GPDMA_CxTR1.DAP) is arbitrated based on the channel priority (GPDMA_CxCR.PRIO[1:0]) versus the other simultaneous requested GPDMA transfers to the same master port.

An arbitrated GPDMA requested link transfer consists of one 32-bit read from the linked-list data structure in memory to one of the linked-list registers (GPDMA_CxTR1, GPDMA_CxTR2, GPDMA_CxBR1, GPDMA_CxSAR, GPDMA_CxDAR or GPDMA_CxLLR, plus GPDMA_CxTR3, GPDMA_CxBR2). Each 32-bit read from memory is arbitrated with the same channel priority as for data transfers, in order to be scheduled over the allocated master port (GPDMA_CxCR.LAP).

Whatever the requested data transfer is programmed with a software request for a memory-to-memory transfer (GPDMA_CxTR2.SWREQ = 1), or with a hardware request (GPDMA_CxTR2.SWREQ = 0) for a memory-to-peripheral transfer or a peripheral-to-memory transfer and whatever is the hardware request type, re-arbitration occurs after each granted single/burst transfer.

When an hardware request is programmed from a destination peripheral (GPDMA_CxTR2.SWREQ = 0 and GPDMA_CxTR2.DREQ = 1), the first memory read of a (possibly 2D/repeated) block (the first ready FIFO-based source burst request), is gated by the occurrence of the corresponding and selected hardware request. This first read request to memory is not taken into account earlier by the arbiter (not as soon as the block transfer is enabled and executable).

GPDMA arbitration

The GPDMA arbitration is directed from the 4-grade assigned channel priority (GPDMA_CxCR.PRIO[1:0]). The arbitration policy, as illustrated in Figure 104 , is defined by:

• one high-priority traffic class (queue 3), dedicated to the assigned channels with priority 3, for time-sensitive channels

This traffic class is granted via a fixed-priority arbitration against any other low-priority traffic class. Within this class, requested single/burst transfers are round-robin arbitrated.

• three low-priority traffic classes (queues 0, 1, or 2) for non time-sensitive channels with priority 0, 1, or 2

Each requested single/burst transfer within this class is round-robin arbitrated, with a weight that is monotonically driven from the programmed priority:

• – Requests with priority 0 are allocated to the queue 0.
• – Requests with priority 1 are allocated and replicated to the queue 0 and queue 1.

• – Requests with priority 2 are allocated and replicated to the queue 0, queue 1, and queue 2.
• – Any queue 0, 1, or 2 equally grants any of its active input requests in a round-robin manner, provided there are simultaneous requests.
• – Additionally, there is a second stage for the low-traffic with a round-robin arbiter that fairly alternates between simultaneous selected requests from queue 0, queue 1, and queue 2.

Figure 104. GPDMA arbitration policy

GPDMA arbitration and bandwidth

With this arbitration policy, the following is guaranteed:

• • Equal maximum bandwidth between requests with same priority
• • Reserved bandwidth (noted as \( B_{Q3} \) ) to the time-sensitive requests (with priority 3)
• • Residual weighted bandwidth between different low-priority requests (priority 0 versus priority 1 versus priority 2).

The two following examples highlight that the weighted round-robin arbitration is driven by the programmed priorities:

• • Example 1: basic application with two non time-sensitive GPDMA requests: req0 and req1. There are the following programming possibilities:
  • – If they are assigned with same priority, the allocated bandwidth by the arbiter to req0 ( \( B_{req0} \) ) is equal to the allocated bandwidth to req1 ( \( B_{req1} \) ).
    \[ B_{req0} = B_{req1} = 1/2 * (1 - B_{Q3}) \]
  • – If req0 is assigned to priority 0 and req1 to priority 1, the allocated bandwidth to req0 ( \( B_{P0} \) ) is 3 times less than the allocated bandwidth to req1 ( \( B_{P1} \) ).
    \[ B_{req0} = B_{P0} = 1/2 * 1/2 * (1 - B_{Q3}) = 1/4 * (1 - B_{Q3}) \]
    \[ B_{req1} = B_{P1} = (1/2 + 1) * 1/2 * (1 - B_{Q3}) = 3/4 * (1 - B_{Q3}) \]
  • – If req0 is assigned to priority 0 and req1 to priority 2, the allocated bandwidth to req0 ( \( B_{P0} \) ) is 5 times less than the allocated bandwidth to req1 ( \( B_{P2} \) ).
    \[ B_{req0} = B_{P0} = 1/2 * 1/3 * (1 - B_{Q3}) = 1/6 * (1 - B_{Q3}) \]
    \[ B_{req1} = B_{P2} = (1/2 + 1 + 1) * 1/3 * (1 - B_{Q3}) = 5/6 * (1 - B_{Q3}) \]

The above computed bandwidth calculation is based on a theoretical input request, always active for any GPDMA clock cycle. This computed bandwidth from the arbiter must be weighted by the frequency of the request given by the application, that cannot be always active and may be quite much variable from one GPDMA client (example I2C at 400 kHz) to another one (PWM at 1 kHz) than the above x3 and x5 ratios.

• Example 2: application where the user distributes a same non-null N number of GPDMA requests to every non time-sensitive priority 0, 1 and 2. The bandwidth calculation is then the following:
  • – The allocated bandwidth to the set of requests of priority 0 ( \( B_{P0} \) ) is
    \[ B_{P0} = 1/3 * 1/3 * (1 - B_{Q3}) = 1/9 * (1 - B_{Q3}) \]
  • – The allocated bandwidth to the set of requests of priority 1 ( \( B_{P1} \) ) is
    \[ B_{P1} = (1/3 + 1/2) * 1/3 * (1 - B_{Q3}) = 5/18 * (1 - B_{Q3}) \]
  • – The allocated bandwidth to the set of requests of priority 2 ( \( B_{P2} \) ) is
    \[ B_{P2} = (1/3 + 1/2 + 1) * 1/3 * (1 - B_{Q3}) = 11/18 * (1 - B_{Q3}) \]
  • – The allocated bandwidth to any request n ( \( B_n \) ) among the N requests of that priority \( P_i \) ( \( i = 0 \) to \( 2 \) ) is \( B_n = 1/N * B_{P_i} \)
  • – The allocated bandwidth to any request n of priority \( 0_i \) ( \( B_{n, P_i} \) ) is
    \[ B_{n, P0} = 1/N * 1/9 * (1 - B_{Q3}) \]
    \[ B_{n, P1} = 1/N * 5/18 * (1 - B_{Q3}) \]
    \[ B_{n, P2} = 1/N * 11/18 * (1 - B_{Q3}) \]

In this example, when the master port bus bandwidth is not totally consumed by the time-sensitive queue 3, the residual bandwidth is such that 2.5 times less bandwidth is allocated to any request of priority 0 versus priority 1, and 5.5 times less bandwidth is allocated to any request of priority 0 versus priority 2.

More generally, assume that the following requests are present:

• • I requests ( \( I \geq 0 \) ) assigned to priority 0
  If \( I > 0 \) , these requests are noted from \( i = 0 \) to \( I-1 \) .
• • J requests ( \( J \geq 0 \) ) assigned to priority 1
  If \( J > 0 \) , these requests are noted from \( j = 0 \) to \( J-1 \) .
• • K requests ( \( K > 0 \) ) assigned to priority 2
  These requests are noted from \( k = 0 \) to \( K-1 \)
• • L requests ( \( L \geq 0 \) ) assigned to priority 3
  If \( L > 0 \) , these requests are noted from \( l = 0 \) to \( L-1 \) .

As \( B_{Q3} \) is the reserved bandwidth to time-sensitive requests, the bandwidth for each request L with priority 3 is:

• • \( B_l = B_{Q3} / L \) for \( L > 0 \) (else: \( B_l = 0 \) )

The bandwidth for each non-time sensitive queue is:

• • \( B_{Q0} = 1/3 * (1 - B_{Q3}) \)
• • \( B_{Q1} = 1/3 * (1 - B_{Q3}) \)
• • \( B_{Q2} = 1/3 * (1 - B_{Q3}) \)

The bandwidth for the set of requests with priority 0 is:

• • \( B_{P0} = I / (I + J + K) * B_{Q0} \)

The bandwidth for each request i with priority 0 is:

• • \( B_i = B_{P0} / I \) for \( L > 0 \) (else \( B_{P0} = 0 \) )

The bandwidth for the set of requests with priority 1 and routed to queue 0 is:

• • \( B_{P1,Q0} = J / (I + J + K) * B_{Q0} \)

The bandwidth for the set of requests with priority 1 and routed to queue 1 is:

• \( B_{P1,Q1} = J / (J + K) * B_{Q1} \)

The total bandwidth for the set of requests with priority 1 is:

• \( B_{P1} = B_{P1,Q0} + B_{P1,Q1} \)

The bandwidth for each request j with priority 1 is:

• \( B_j = B_{P1} / J \) for \( J > 0 \) (else \( B_j = 0 \) )

The bandwidth for the set of requests with priority 2 and routed to queue 0 is:

• \( B_{P2,Q0} = K / (I + J + K) * B_{Q0} \)

The bandwidth for the set of requests with priority 2 and routed to queue 1 is:

• \( B_{P2,Q1} = K / (J + K) * B_{Q1} \)

The bandwidth for the set of requests with priority 2 and routed to queue 2 is:

• \( B_{P2,Q2} = B_{Q2} \)

The total bandwidth for the set of requests with priority 2 is:

• \( B_{P2} = B_{P2,Q0} + B_{P2,Q1} + B_{P2,Q2} \)

The bandwidth for each request k with priority 2 is:

• \( B_k = B_{P2} / K \) ( \( K > 0 \) in the general case)

Thus finally the maximum allocated residual bandwidths for any i, j, k non-time sensitive request are:

• in the general case (when there is at least one request k with a priority 2 ( \( K > 0 \) )):
  • \( B_i = 1/I * 1/3 * I/(I + J + K) * (1 - B_{Q3}) \)
  • \( B_j = 1/J * 1/3 * [J/(I + J + K) + J/(J + K)] * (1 - B_{Q3}) \)
  • \( B_k = 1/K * 1/3 * [K/(I + J + K) + K/(J + K) + 1] * (1 - B_{Q3}) \)
• in the specific case (when there is no request k with a priority 2 ( \( K = 0 \) )):
  • \( B_i = 1/I * 1/2 * I/(I + J) * (1 - B_{Q3}) \)
  • \( B_j = 1/J * 1/2 * [J/(I + J) + 1] * (1 - B_{Q3}) \)

Consequently, the GPDMA arbiter can be used as a programmable weighted bandwidth limiter, for each queue and more generally for each request/channel. The different weights are monotonically resulting from the programmed channel priorities.

12.4.12 GPDMA triggered transfer

A programmed GPDMA transfer can be triggered by a rising/falling edge of a selected input trigger event, as defined by GPDMA_CxTR2.TRIGPOL[1:0] and GPDMA_CxTR2.TRIGSEL[5:0] (see Section 12.3.6 for the trigger selection).

The triggered transfer, as defined by the trigger mode in GPDMA_CxTR2.TRIGM[1:0], can be at LLI data transfer level, to condition the first burst read of a block, the first burst read of a 2D/repeated block for channel x ( \( x = 12 \) to \( 15 \) ), or each programmed single read. The trigger mode can also be programmed to condition the LLI link transfer (see TRIGM[1:0] in GPDMA_CxTR2 for more details).

Trigger hit memorization and trigger overrun flag generation

The GPDMA monitoring of a trigger for a channel x is started when the channel is enabled/loaded with a new active trigger configuration: rising or falling edge on a selected trigger (respectively TRIGPOL[1:0] = 01 or TRIGPOL[1:0] = 10).

The monitoring of this trigger is kept active during the triggered and uncompleted (data or link) transfer. If a new trigger is detected, this hit is internally memorized to grant the next transfer, as long as the defined rising/falling edge and TRIGSEL[5:0] are not modified, and the channel is enabled.

Transferring a next \( LLI_{n+1} \) , that updates the GPDMA_CxTR2 with a new value for any of TRIGSEL[5:0] or TRIGPOL[1:0], resets the monitoring, trashing the possible memorized hit of the formerly defined \( LLI_n \) trigger.

Caution: After a first new trigger hit _n+1 is memorized, if another trigger hit _n+2 is detected and if the hit _n triggered transfer is still not completed, hit _n+2 is lost and not memorized. A trigger overrun flag is reported ( GPDMA_CxSR.TOF = 1) and an interrupt is generated if enabled (if GPDMA_CxCR.TOIE = 1). The channel is not automatically disabled by hardware due to a trigger overrun.

Figure 105 illustrates the trigger hit, memorization and overrun in the configuration example with a block-level trigger mode and a rising edge trigger polarity.

Figure 105. Trigger hit, memorization, and overrun waveform

Note: The user can assign the same input trigger event to different channels. This can be used to trigger different channels on a broadcast trigger event.

12.4.13 GPDMA circular buffering with linked-list programming

GPDMA circular buffering for memory-to-peripheral and peripheral-to-memory transfers, with a linear addressing channel

For a circular buffering, with a continuous memory-to-peripheral (or peripheral-to-memory) transfer, the software must set up a channel with half-transfer and complete transfer

events/interrupts generation (GPDMA_CxCR.HTIE = 1 and GPDMA_CxCR.TCIE = 1), in order to enable a concurrent buffer software processing.

LLI ₀ is configured for the first block transfer with the linear addressing channel. A continuously-executed LLI ₁ is needed to restore the memory source (or destination) start address, for the memory-to-peripheral transfer (respectively the peripheral-to-memory transfer). GPDMA automatically reloads the initially programmed GPDMA_CxBR1.BNDT[15:0] when a block transfer is completed, and there is no need to restore GPDMA_CxBR1.

Figure 106 illustrates this programming with a linear addressing GPDMA channel and a source circular buffer.

Figure 106. GPDMA circular buffer programming: update of the memory start address with a linear addressing channel

Note:

With a 2D addressing channel, the user may use a single LLI with GPDMA_CxBR1.BRC[10:0] = 1, and program a negative memory block address offset with GPDMA_CxBR2 and GPDMA_CxBR1, in order to jump back to the memory source or the destination start address.

If circular buffering must be executed after some other transfers over the shared GPDMA channel x, the before-last LLI _N-1 in memory is needed to configure the first block transfer. And the last LLI _N restores the memory source (or destination) start address in memory-to-peripheral transfer (respectively in peripheral-to-memory transfer).

Figure 107 illustrates this programming with a linear addressing shared GPDMA channel, and a source circular buffer.

Figure 107. Shared GPDMA channel with circular buffering: update of the memory start address with a linear addressing channel

12.4.14 GPDMA transfer in peripheral flow-control mode

A peripheral with the peripheral flow-control mode feature can decide to early terminate a GPDMA block transfer, provided that the allocated channel is implemented with this feature (see Section 12.3.5 ).

If the related GPDMA channel x is also programmed in peripheral flow-control mode (GPDMA_CxTR2.PFREQ = 1):

• • The GPDMA block transfer starts as follows:
  • – If GPDMA_CxBR1.BNDT[15:0] \( \neq \) 0, the programmed value is internally taken into account by the GPDMA hardware.
  • – If GPDMA_CxBR1.BNDT[15:0] = 0, the GPDMA hardware internally considers a 64-Kbyte value for the maximum source block size to be transferred.
• • The GPDMA block transfer is completed as soon as the first occurrence of one of the following condition occurs:
  • – when GPDMA_CxBR1.BNDT[15:0] = 0
  • – when the peripheral early terminates the block. The complete transfer event is generated if programmed, depending on GPDMA_CxTR2 (see GPDMA channel x transfer register 2 (GPDMA_CxTR2) ). Then the software can read the current number of transferred bytes from the source (GPDMA_CxBR1.BNDT[15:0]), and/or read the current source or destination address of the buffer in memory (GPDMA_CxSAR[31:0] or GPDMA_CxDAR[31:0]).

In peripheral flow-control mode:

• • a destination peripheral with a hardware requested transfer is not supported: memory-to-peripheral transfer is not supported.
• • Data packing from a source peripheral is not supported.
• • 2D/repeated block is not supported.
• • GPDMA_CxBR1.BNDT[15:0] must be programmed as a multiple of the source (peripheral) burst size.

12.4.15 GPDMA privileged/unprivileged channel

Any channel x is a privileged or unprivileged hardware resource, as configured by a privileged agent via GPDMA_PRIVCFGR.PRIVx.

When a channel x is configured in a privileged state by a privileged agent, the following access control rules are applied:

• • An unprivileged read access to a register field of this channel is forced to return 0, except for GPDMA_PRIVCFGR.GPDMA_RCFGLOCKR that is readable by an unprivileged agent.
• • An unprivileged write access to a register field of this channel has no impact.

When a channel is configured in a privileged (or unprivileged) state, the source and destination data transfers are privileged (respectively unprivileged) transfers over the AHB master port.

When a channel is configured in a privileged (or unprivileged) state and in linked-list mode, the loading of the next linked-list data structure from the GPDMA memory into its register file, is automatically performed with privileged (respectively unprivileged) transfers, via the GPDMA_CxCR.LAP allocated master port.

The GPDMA generates a privileged bus that reflects GPDMA_PRIVCFGR, to keep the other peripherals informed of the privileged/unprivileged state of each GPDMA channel x.

When the privileged software must switch a channel from a privileged state to an unprivileged state, the privileged software must abort the channel or wait until that the privileged channel is completed before switching. This is needed to dynamically re-allocate a channel to a next unprivileged transfer as an unprivileged software is not allowed to do so, and must have GPDMA_CxCR.EN = 0 before the unprivileged software can reprogram the GPDMA_CxCR for a next transfer. The privileged software may reset not only the channel x (GPDMA_CxCR.RESET = 1) but also the full channel x register file to its reset value.

12.4.16 GPDMA error management

The GPDMA is able to manage and report to the user a transfer error, as follows, depending on the root cause.

Data transfer error

On a bus access (as a AHB single or a burst) to the source or the destination:

• • The source or destination target reports an AHB error.
• • The programmed channel transfer is stopped (GPDMA_CxCR.EN cleared by the GPDMA hardware). The channel status register reports an idle state (GPDMA_CxSR.IDLEF = 1) and the data error (GPDMA_CxSR.DTEF = 1).

• • After a GPDMA data transfer error, the user must perform a debug session, taking care of the product-defined memory mapping of the source and destination, including the protection attributes.
• • After a GPDMA data transfer error, the user must issue a channel reset (set GPDMA_CxCR.RESET) to reset the hardware GPDMA channel data path and the content of the FIFO, before the user enables again the same channel for a next transfer.

Link transfer error

On a tentative update of a GPDMA channel register from the programmed LLI in the memory:

• • The linked-list memory reports an AHB error.
• • The programmed channel transfer is stopped (GPDMA_CxCR.EN cleared by the GPDMA hardware), the channel status register reports an idle state (GPDMA_CxSR.IDLEF = 1) and the link error (GPDMA_CxSR.ULEF = 1).
• • After a GPDMA link error, the user must perform a debug session, taking care of the product-defined memory mapping of the linked-list data structure (GPDMA_CxLBAR and GPDMA_CxLLR), including the protection attributes.
• • After a GPDMA link error, the user must explicitly write the linked-list register file (GPDMA_CxTR1, GPDMA_CxTR2, GPDMA_CxBR1, GPDMA_CxSAR, GPDMA_CxDAR and GPDMA_CxLLR, plus GPDMA_CxTR3 and GPDMA_CxBR2), before the user enables again the same channel for a next transfer.

User setting error

On a tentative execution of a GPDMA transfer with an unauthorized user setting:

• • The programmed channel transfer is disabled (GPDMA_CxCR.EN forced and cleared by the GPDMA hardware) preventing the next unauthorized programmed data transfer from being executed. The channel status register reports an idle state (GPDMA_CxSR.IDLEF = 1) and a user setting error (GPDMA_CxSR.USEF = 1).
• • After a GPDMA user setting error, the user must perform a debug session, taking care of the GPDMA channel programming. A user setting error can be caused by one of the following:
  • – a programmed null source block size without a programmed update of this value from the next LLI _i (GPDMA_CxBR1.BNDT[15:0] = 0 and GPDMA_CxLLR.UB1 = 0)
  • – a programmed non-null source block size being not a multiple of the programmed data width of a source burst transfer (GPDMA_CxBR1.BNDT[2:0] versus GPDMA_CxTR1.SDW_LOG2[1:0])
  • – when in packing/unpacking mode (if PAM[1] = 1), a programmed non-null source block size being not a multiple of the programmed data width of a destination burst transfer (GPDMA_CxBR1.BNDT[2:0] versus GPDMA_CxTR1.DDW_LOG2[1:0])
  • – a programmed unaligned source start address, being not a multiple of the programmed data width of a source burst transfer (GPDMA_CxSAR[2:0] versus GPDMA_CxTR1.SDW_LOG2[1:0])
  • – for channel x (x = 12 to 15): a programmed unaligned source address offset being not a multiple of the programmed data width of a source burst transfer (GPDMA_CxTR3.SAO[2:0] versus GPDMA_CxTR1.SDW_LOG2[1:0])

• – for channel x (x = 12 to 15): a programmed unaligned block repeated source address offset being not a multiple of the programmed data width of a source burst transfer (GPDMA_CxBR2.BRSAO[2:0] versus GPDMA_CxTR1.SDW_LOG2[1:0])
• – a programmed unaligned destination start address, being not a multiple of the programmed data width of a destination burst transfer (GPDMA_CxDAR[2:0] versus GPDMA_CxTR1.DDW_LOG2[1:0])
• – for channel x (x = 12 to 15): a programmed unaligned destination address offset being not a multiple of the programmed data width of a destination burst transfer (GPDMA_CxTR3.DAO[2:0] versus GPDMA_CxTR1.DDW_LOG2[1:0])
• – for channel x (x = 12 to 15): a programmed unaligned block repeated destination address offset being not a multiple of the programmed data width of a destination burst transfer (GPDMA_CxBR2.BRDAO[2:0] versus GPDMA_CxTR1.DDW_LOG2[1:0])
• – a programmed double-word source data width (GPDMA_CxTR1.SDW_LOG2[1:0] = 11)
• – a programmed double-word destination data width (GPDMA_CxTR1.DDW_LOG2[1:0] = 11)
• – a programmed linked-list item \( LLI_{n+1} \) with a null data transfer (GPDMA_CxLLR.UB1 = 1 and GPDMA_CxBR1.BNDT = 0)

12.5 GPDMA in debug mode

When the microcontroller enters debug mode (core halted), any channel x can be individually either continued (default) or suspended, depending on the programmable control bit in the DBGMCU module.

Note: In debug mode, GPDMA_CxSR.SUSPF is not altered by a suspension from the programmable control bit in the DBGMCU module. In this case, GPDMA_CxSR.IDLEF can be checked to know the completion status of the channel suspension.

12.6 GPDMA in low-power modes

Table 97. Effect of low-power modes on GPDMA

1. Refer to Section 12.3.2 to know if any Stop mode is supported.

12.7 GPDMA interrupts

There is one GPDMA interrupt line for each channel, and separately for each CPU (if several ones in the devices).

Table 98. GPDMA interrupt requests

A GPDMA channel x event may be:

• • a transfer complete
• • a half-transfer complete
• • a transfer error, due to either:
  • – a data transfer error
  • – an update link error
  • – a user setting error completed suspension
• • a trigger overrun

Note: When a channel x transfer complete event occurs, the output signal gpdma_chx_tc is generated as a high pulse of one clock cycle.

An interrupt is generated following any xx event, provided that both:

• • the corresponding interrupt event xx is enabled (GPDMA_CxCR.xxIE = 1)
• • the corresponding event flag is cleared (GPDMA_CxSR.xxF = 0). This means that, after a previous same xx event occurrence, a software agent must have written 1 into the corresponding xx flag clear control bit (write 1 into GPDMA_CxFCR.xxF).

TCF (transfer complete) and HTF (half transfer) events generation is controlled by GPDMA_CxTR2.TCEM[1:0] as follows:

• • A transfer complete event is a block transfer complete, a 2D/repeated block transfer complete, or a LLI transfer complete including the upload of the next LLI if any, or the full linked-list completion, depending on the transfer complete event mode GPDMA_CxTR2.TCEM[1:0].

• • A half-transfer event is a half-block transfer or a half 2D/repeated block transfer, depending on the transfer complete event mode GPDMA_CxTR2.TCEM[1:0].

A half-block transfer occurs when half of the source block size bytes (rounded-up integer of GPDMA_CxBR1.BNDT[15:0] / 2) is transferred to the destination.

A half 2D/repeated block transfer occurs when half of the repeated blocks (rounded-up integer of (GPDMA_CxBR1.BRC[10:0] + 1) / 2) is transferred to the destination.

See GPDMA channel x transfer register 2 (GPDMA_CxTR2) for more details.

A transfer error rises in one of the following situations:

1. • • during a single/burst data transfer from the source or to the destination (DTEF)
   • • during an update of a GPDMA channel register from the programmed LLI in memory (ULEF)
   • • during a tentative execution of a GPDMA channel with an unauthorized setting (USEF)
2. The user must perform a debug session to correct the GPDMA channel programming versus the USEF root causes list (see Section 12.4.16 ).

A trigger overrun is described in Trigger hit memorization and trigger overrun flag generation .

12.8 GPDMA registers

The GPDMA registers must be accessed with an aligned 32-bit word data access.

12.8.1 GPDMA privileged configuration register (GPDMA_PRIVCFGR)

Address offset: 0x04

Reset value: 0x0000 0000

A write access to this register must be privileged. A read access can be privileged or unprivileged.

This register must be written when GPDMA_CxCR.EN = 0.

This register is read-only when GPDMA_CxCR.EN = 1.

This register must be programmed at a bit level, at the initialization/closure of a GPDMA channel (GPDMA_CxCR.EN = 0), to individually allocate any channel x to the privileged or unprivileged world.

Bits 31:16 Reserved, must be kept at reset value.

Bits 15:0 PRIVx : privileged state of channel x (x = 15 to 0)

0: unprivileged

1: privileged

12.8.2 GPDMA configuration lock register (GPDMA_RCFGLOCKR)

Address offset: 0x08

Reset value: 0x0000 0000

This register can be written by a software agent with privileged attributes in order to individually lock, for example at boot time, the privileged attributes of any GPDMA channel/resource (to lock the setting of GPDMA_CxPRIVCFGR for any channel x, for example at boot time).

A read access may be privileged or unprivileged.

Bits 31:16 Reserved, must be kept at reset value.

Bits 15:0 LOCKx : lock the configuration of GPDMA_PRIVCFGR.PRIVx until a global GPDMA reset (x = 15 to 0)
This bit is cleared after reset and, once set, it cannot be reset until a global GPDMA reset.
0: privilege configuration of the channel x is writable.
1: privilege configuration of the channel x is not writable.

12.8.3 GPDMA masked interrupt status register (GPDMA_MISR)

Address offset: 0x0C

Reset value: 0x0000 0000

This is a read register.

This register contains the masked interrupt status bit MISx for each channel x. It is a logical OR of all the GPDMA_CxSR flags, each source flag being enabled by the corresponding GPDMA_CxCR interrupt enable bit.

Every bit is deasserted by hardware when writing 1 to the corresponding GPDMA_CxFCR flag clear bit.

This register can mix privileged and unprivileged information, depending on the privileged state of each channel GPDMA_PRIVCFGR.PRIVx. A privileged software can read the full interrupt status. An unprivileged software is restricted to read the status of unprivileged channels, other privileged bit fields returning zero.

Bits 31:16 Reserved, must be kept at reset value.

Bits 15:0 MISx : masked interrupt status of channel x (x = 15 to 0)

0: no interrupt occurred on channel x

1: an interrupt occurred on channel x

12.8.4 GPDMA channel x linked-list base address register (GPDMA_CxLBAR)

Address offset: \( 0x50 + 0x80 \times x \) (x = 0 to 15)

Reset value: 0x0000 0000

This register must be written by a privileged software. It is either privileged readable or not, depending on the privileged state of the channel x GPDMA_PRIVCFG.PRIVx.

This register must be written when GPDMA_CxCR.EN = 0.

This register is read-only when GPDMA_CxCR.EN = 1.

This channel-based register is the linked-list base address of the memory region, for a given channel x, from which the LLIs describing the programmed sequence of the GPDMA transfers, are conditionally and automatically updated.

This 64-Kbyte aligned channel x linked-list base address is offset by the 16-bit GPDMA_CxLLR register that defines the word-aligned address offset for each LLI.

Bits 31:16 LBA[31:16] : linked-list base address of GPDMA channel x

Bits 15:0 Reserved, must be kept at reset value.

12.8.5 GPDMA channel x flag clear register (GPDMA_CxFCR)

Address offset: 0x5C+ 0x80 * x (x = 0 to 15)

Reset value: 0x0000 0000

This is a write register privileged or unprivileged, depending on the privileged state of the channel x (GPDMA_PRIVCFGR.PRIVx).

Bits 31:15 Reserved, must be kept at reset value.

Bit 14 TOF : trigger overrun flag clear

0: no effect

1: corresponding TOF flag cleared

Bit 13 SUSPF : completed suspension flag clear

0: no effect

1: corresponding SUSPF flag cleared

Bit 12 USEF : user setting error flag clear

0: no effect

1: corresponding USEF flag cleared

Bit 11 ULEF : update link transfer error flag clear

0: no effect

1: corresponding ULEF flag cleared

Bit 10 DTEF : data transfer error flag clear

0: no effect

1: corresponding DTEF flag cleared

Bit 9 HTF : half-transfer flag clear

0: no effect

1: corresponding HTF flag cleared

Bit 8 TCF : transfer complete flag clear

0: no effect

1: corresponding TCF flag cleared

Bits 7:0 Reserved, must be kept at reset value.

12.8.6 GPDMA channel x status register (GPDMA_CxSR)

Address offset: \( 0x60 + 0x80 * x \) ( \( x = 0 \) to \( 15 \) )

Reset value: \( 0x0000\ 0001 \)

This is a read register, reporting the channel status.

This register is privileged or unprivileged, depending on the privileged state of the channel (GPDMA_PRIVCFG.R.PRIVx).

Bits 31:24 Reserved, must be kept at reset value.

Bits 23:16 FIFOL[7:0] : monitored FIFO level

Number of available write beats in the FIFO, in units of the programmed destination data width (see GPDMA_CxTR1.DDW_LOG2[1:0], in units of bytes, half-words, or words).

Note: After having suspended an active transfer, the user may need to read FIFOL[7:0], additionally to GPDMA_CxBR1.BDNT[15:0] and GPDMA_CxBR1.BRC[10:0], to know how many data have been transferred to the destination. Before reading, the user may wait for the transfer to be suspended (GPDMA_CxSR.SUSPF = 1).

Bit 15 Reserved, must be kept at reset value.

Bit 14 TOF : trigger overrun flag

0: no trigger overrun event

1: a trigger overrun event occurred

Bit 13 SUSPF : completed suspension flag

0: no completed suspension event

1: a completed suspension event occurred

Bit 12 USEF : user setting error flag

0: no user setting error event

1: a user setting error event occurred

Bit 11 ULEF : update link transfer error flag

0: no update link transfer error event

1: a master bus error event occurred while updating a linked-list register from memory

Bit 10 DTEF : data transfer error flag

0: no data transfer error event

1: a master bus error event occurred on a data transfer

Bit 9 HTF : half-transfer flag

0: no half-transfer event

1: a half-transfer event occurred

A half-transfer event is either a half-block transfer or a half 2D/repeated block transfer, depending on the transfer complete event mode (GPDMA_CxTR2.TCEM[1:0]).

A half-block transfer occurs when half of the bytes of the source block size (rounded up integer of GPDMA_CxBR1.BNDT[15:0]/2) has been transferred to the destination.

A half 2D/repeated block transfer occurs when half of the repeated blocks (rounded up integer of (GPDMA_CxBR1.BRC[10:0] + 1) / 2)) has been transferred to the destination.

Bit 8 TCF : transfer complete flag

0: no transfer complete event

1: a transfer complete event occurred

A transfer complete event is either a block transfer complete, a 2D/repeated block transfer complete, or a LLI transfer complete including the upload of the next LLI if any, or the full linked-list completion, depending on the transfer complete event mode (GPDMA_CxTR2.TCEM[1:0]).

Bits 7:1 Reserved, must be kept at reset value.

Bit 0 IDLEF : idle flag

0: channel not in idle state

1: channel in idle state

This idle flag is deasserted by hardware when the channel is enabled (GPDMA_CxCR.EN = 1) with a valid channel configuration (no USEF to be immediately reported).

This idle flag is asserted after hard reset or by hardware when the channel is back in idle state (in suspended or disabled state).

12.8.7 GPDMA channel x control register (GPDMA_CxCR)

Address offset: 0x64 + 0x80 * x (x = 0 to 15)

Reset value: 0x0000 0000

This register is privileged or unprivileged, depending on the privileged state of the channel x (GPDMA_PRIVCFGR.PRIVx).

This register is used to control a channel (activate, suspend, abort or disable it).

Bits 31:24 Reserved, must be kept at reset value.

Bits 23:22 PRIO[1:0] : priority level of the channel x GPDMA transfer versus others

00: low priority, low weight

01: low priority, mid weight

10: low priority, high weight

11: high priority

Note: This bit must be written when EN = 0. This bit is read-only when EN = 1.

Bits 21:18 Reserved, must be kept at reset value.

Bit 17 LAP : linked-list allocated port

This bit is used to allocate the master port for the update of the GPDMA linked-list registers from the memory.

0: port 0 (AHB) allocated

1: port 1 (AHB) allocated

Note: This bit must be written when EN = 0. This bit is read-only when EN = 1.

Bit 16 LSM : Link step mode

0: channel executed for the full linked-list and completed at the end of the last LLI (GPDMA_CxLLR = 0). The 16 low-significant bits of the link address are null (LA[15:0] = 0) and all the update bits are null (UT1 = UB1 = UT2 = USA = UDA = ULL = 0 and UT3 = UB2 = 0). Then GPDMA_CxBR1.BNDT[15:0] = 0 and GPDMA_CxBR1.BRC[10:0] = 0.

1: channel executed once for the current LLI

First the (possible 1D/repeated) block transfer is executed as defined by the current internal register file until GPDMA_CxBR1.BNDT[15:0] = 0 and GPDMA_CxBR1.BRC[10:0] = 0.

Secondly the next linked-list data structure is conditionally uploaded from memory as defined by GPDMA_CxLLR. Then channel execution is completed.

Note: This bit must be written when EN = 0. This bit is read-only when EN = 1.

Bit 15 Reserved, must be kept at reset value.

Bit 14 TOIE : trigger overrun interrupt enable

0: interrupt disabled

1: interrupt enabled

Bit 13 SUSPIE : completed suspension interrupt enable

0: interrupt disabled

1: interrupt enabled

Bit 12 USEIE : user setting error interrupt enable

0: interrupt disabled

1: interrupt enabled

Bit 11 ULEIE : update link transfer error interrupt enable

0: interrupt disabled

1: interrupt enabled

Bit 10 DTEIE : data transfer error interrupt enable

0: interrupt disabled

1: interrupt enabled

Bit 9 HTIE : half-transfer complete interrupt enable

0: interrupt disabled

1: interrupt enabled

Bit 8 TCIE : transfer complete interrupt enable

0: interrupt disabled

1: interrupt enabled

Bits 7:3 Reserved, must be kept at reset value.

Bit 2 SUSP : suspend

Writing 1 into the field RESET (bit 1) causes the hardware to de-assert this bit, whatever is written into this bit 2. Else:
Software must write 1 in order to suspend an active channel (channel with an ongoing GPDMA transfer over its master ports).

The software must write 0 in order to resume a suspended channel, following the programming sequence detailed in Figure 89 .

0: write: resume channel, read: channel not suspended

1: write: suspend channel, read: channel suspended.

Bit 1 RESET : reset

This bit is write only. Writing 0 has no impact. Writing 1 implies the reset of the following: the FIFO, the channel internal state, SUSP and EN bits (whatever is written receptively in bit 2 and bit 0).
The reset is effective when the channel is in steady state, meaning one of the following:

- active channel in suspended state (GPDMA_CxSR.SUSPF = 1 and

GPDMA_CxSR.IDLEF = GPDMA_CxCR.EN = 1)

- channel in disabled state (GPDMA_CxSR.IDLEF = 1 and GPDMA_CxCR.EN = 0).

After writing a RESET, to continue using this channel, the user must explicitly reconfigure the channel including the hardware-modified configuration registers (GPDMA_CxBR1, GPDMA_CxSAR, and GPDMA_CxDAR) before enabling again the channel (see the programming sequence in Figure 90 ).

0: no channel reset

1: channel reset

Bit 0 EN : enable

Writing 1 into the field RESET (bit 1) causes the hardware to de-assert this bit, whatever is written into this bit 0. Else:

this bit is deasserted by hardware when there is a transfer error (master bus error or user setting error) or when there is a channel transfer complete (channel ready to be configured, for example if LSM = 1 at the end of a single execution of the LLI).

Else, this bit can be asserted by software.

Writing 0 into this EN bit is ignored.

0: write: ignored, read: channel disabled

1: write: enable channel, read: channel enabled

12.8.8 GPDMA channel x transfer register 1 (GPDMA_CxTR1)

Address offset: \( 0x90 + 0x80 \times x \) ( \( x = 0 \) to 15)

Reset value: 0x0000 0000

This register is privileged or unprivileged, depending on the privileged state of the channel x in GPDMA_PRIVCFGR.PRIVx.

This register controls the transfer of a channel x.

This register must be written when GPDMA_CxCR.EN = 0.

This register is read-only when GPDMA_CxCR.EN = 1.

This register must be written when the channel is completed. Then the hardware has deasserted GPDMA_CxCR.EN). A channel transfer can be completed and programmed at different levels: block, 2D/repeated block, LLI or full linked-list.

In linked-list mode, during the link transfer, this register is automatically updated by GPDMA from the memory if GPDMA_CxLLR.UT1 = 1.

Bit 31 Reserved, must be kept at reset value.

Bit 30 DAP : destination allocated port

This bit is used to allocate the master port for the destination transfer

0: port 0 (AHB) allocated

1: port 1 (AHB) allocated

Note: This bit must be written when EN = 0. This bit is read-only when EN = 1.

Bits 29:28 Reserved, must be kept at reset value.

Bit 27 DHX : destination half-word exchange

If the destination data size is shorter than a word, this bit is ignored.

If the destination data size is a word:

0: no halfword-based exchanged within word

1: the two consecutive (post PAM) half-words are exchanged in each destination word.

Bit 26 DBX : destination byte exchange

If the destination data size is a byte, this bit is ignored.

If the destination data size is not a byte:

0: no byte-based exchange within half-word

1: the two consecutive (post PAM) bytes are exchanged in each destination half-word.

Bits 25:20 DBL_1[5:0] : destination burst length minus 1, between 0 and 63

The burst length unit is one data named beat within a burst. If DBL_1[5:0] = 0, the burst can be named as single. Each data/beat has a width defined by the destination data width DDW_LOG2[1:0].

Note: If a burst transfer crossed a 1-Kbyte address boundary on a AHB transfer, the GPDMA modifies and shortens the programmed burst into singles or bursts of lower length, to be compliant with the AHB protocol.

If a burst transfer is of length greater than the FIFO size of the channel x, the GPDMA modifies and shortens the programmed burst into singles or bursts of lower length, to be compliant with the FIFO size. Transfer performance is lower, with GPDMA re-arbitration between effective and lower singles/bursts, but the data integrity is guaranteed.

Bit 19 DINC : destination incrementing burst

0: fixed burst

1: contiguously incremented burst

The destination address, pointed by GPDMA_CxDAR, is kept constant after a burst beat/single transfer, or is incremented by the offset value corresponding to a contiguous data after a burst beat/single transfer.

Bit 18 Reserved, must be kept at reset value.

Bits 17:16 DDW_LOG2[1:0] : binary logarithm of the destination data width of a burst, in bytes

00: byte

01: half-word (2 bytes)

10: word (4 bytes)

11: user setting error reported and no transfer issued

Note: A destination burst transfer must have an aligned address with its data width (start address GPDMA_CxDAR[2:0] and if present address offset GPDMA_CxTR3.DAO[2:0], versus DDW_LOG2[1:0]). Otherwise a user setting error is reported and no transfer is issued.

When configured in packing mode (PAM[1] = 1 and destination data width different from source data width), a source block size must be a multiple of the destination data width (see GPDMA_CxBR1.BNDT[2:0] versus DDW_LOG2[1:0]). Else a user setting error is reported and none transfer is issued.

Setting a 8-byte data width causes a user setting error to be reported and none transfer is issued.

Bit 15 Reserved, must be kept at reset value.

Bit 14 SAP : source allocated port

This bit is used to allocate the master port for the source transfer

0: port 0 (AHB) allocated

1: port 1 (AHB) allocated

Note: This bit must be written when EN = 0. This bit is read-only when EN = 1.

Bit 13 SBX : source byte exchange within the unaligned half-word of each source word

If the source data width is shorter than a word, this bit is ignored.

If the source data width is a word:

0: no byte-based exchange within the unaligned half-word of each source word

1: the two consecutive bytes within the unaligned half-word of each source word are exchanged.

Bits 12:11 PAM[1:0] : padding/alignment mode

If DDW_LOG2[1:0] = SDW_LOG2[1:0]: if the data width of a burst destination transfer is equal to the data width of a burst source transfer, these bits are ignored.

Else, in the following enumerated values, the condition PAM_1 is when destination data width is higher than source data width, and the condition PAM_2 is when source data width is higher than destination data width.

Condition: PAM_1

00: source data is transferred as right aligned, padded with 0s up to the destination data width

01: source data is transferred as right aligned, sign extended up to the destination data width

10-11: successive source data are FIFO queued and packed at the destination data width, in a left (LSB) to right (MSB) order (named little endian), before a destination transfer

Condition: PAM_2

00: source data is transferred as right aligned, left-truncated down to the destination data width

01: source data is transferred as left-aligned, right-truncated down to the destination data width

10-11: source data is FIFO queued and unpacked at the destination data width, to be transferred in a left (LSB) to right (MSB) order (named little endian) to the destination

Note: If the transfer from the source peripheral is configured with peripheral flow-control mode (SWREQ = 0 and PFREQ = 1 and DREQ = 0), and if the destination data width > the source data width, packing is not supported.

Bit 10 Reserved, must be kept at reset value.

Bits 9:4 SBL_1[5:0] : source burst length minus 1, between 0 and 63

The burst length unit is one data named beat within a burst. If SBL_1[5:0] = 0, the burst can be named as single. Each data/beat has a width defined by the destination data width SDW_LOG2[1:0].

Note: If a burst transfer crossed a 1-Kbyte address boundary on a AHB transfer, the GPDMA modifies and shortens the programmed burst into singles or bursts of lower length, to be compliant with the AHB protocol.

If a burst transfer is of length greater than the FIFO size of the channel x, the GPDMA modifies and shortens the programmed burst into singles or bursts of lower length, to be compliant with the FIFO size. Transfer performance is lower, with GPDMA re-arbitration between effective and lower singles/bursts, but the data integrity is guaranteed.

Bit 3 SINC : source incrementing burst

0: fixed burst

1: contiguously incremented burst

The source address, pointed by GPDMA_CxSAR, is kept constant after a burst beat/single transfer or is incremented by the offset value corresponding to a contiguous data after a burst beat/single transfer.

Bit 2 Reserved, must be kept at reset value.

Bits 1:0 SDW_LOG2[1:0] : binary logarithm of the source data width of a burst in bytes

00: byte

01: half-word (2 bytes)

10: word (4 bytes)

11: user setting error reported and no transfer issued

Note: A source block size must be a multiple of the source data width (GPDMA_CxBR1.BNDT[2:0] versus SDW_LOG2[1:0]). Otherwise, a user setting error is reported and no transfer is issued.

A source burst transfer must have an aligned address with its data width (start address GPDMA_CxSAR[2:0] versus SDW_LOG2[1:0]). Otherwise, a user setting error is reported and none transfer is issued.

Setting a 8-byte data width causes a user setting error to be reported and no transfer is issued.

12.8.9 GPDMA channel x transfer register 2 (GPDMA_CxTR2)

Address offset: 0x94 + 0x80 * x (x = 0 to 15)

Reset value: 0x0000 0000

This register is privileged or unprivileged, depending on the privileged state of channel x (GPDMA_PRIVCFGR.PRIVx).

This register controls the transfer of a channel x. This register must be written when GPDMA_CxCR.EN = 0.

This register is read-only when GPDMA_CxCR.EN = 1.

This register must be written when the channel is completed (the hardware deasserted GPDMA_CxCR.EN). A channel transfer can be completed and programmed at different levels: block or LLI or full linked-list.

In linked-list mode, during the link transfer, this register is automatically updated by GPDMA from the memory, if GPDMA_CxLLR.UT2 = 1.

Bits 31:30 TCEM[1:0]: transfer complete event mode

These bits define the transfer granularity for the transfer complete and half-transfer complete events generation.

00: at block level (when GPDMA_CxBR1.BNDT[15:0] = 0): the complete (and the half) transfer event is generated at the (respectively half of the) end of a block.

Note: If the initial LLI ₀ data transfer is null/void (directly programmed by the internal register file with GPDMA_CxBR1.BNDT[15:0] = 0), then neither the complete transfer event nor the half-transfer event is generated.

01: channel x (x = 0 to 11), same as 00, channel x (x = 12 to 15), at 2D/repeated block level (when GPDMA_CxBR1.BRC[10:0] = 0 and GPDMA_CxBR1.BNDT[15:0] = 0). The complete (and the half) transfer event is generated at the end (respectively half of the end) of the 2D/repeated block.

Note: If the initial LLI ₀ data transfer is null/void (directly programmed by the internal register file with GPDMA_CxBR1.BNDT[15:0] = 0), then neither the complete transfer event nor the half-transfer event is generated.

10: at LLI level: the complete transfer event is generated at the end of the LLI transfer, including the update of the LLI if any. The half-transfer event is generated at the half of the LLI data transfer. The LLI data transfer is a block transfer or a 2D/repeated block transfer for channel x (x = 12 to 15), if any data transfer.

Note: If the initial LLI ₀ data transfer is null/void (directly programmed by the internal register file with GPDMA_CxBR1.BNDT[15:0] = 0), then the half-transfer event is not generated, and the transfer complete event is generated when is completed the loading of the LLI ₁ .

11: at channel level: the complete transfer event is generated at the end of the last LLI transfer. The half-transfer event is generated at the half of the data transfer of the last LLI. The last LLI updates the link address GPDMA_CxLLR.LA[15:2] to zero and clears all the GPDMA_CxLLR update bits (UT1, UT2, UB1, USA, UDA and ULL, plus UT3 and UB2). If the channel transfer is continuous/infinite, no event is generated.

Bits 29:26 Reserved, must be kept at reset value.

Bits 25:24 TRIGPOL[1:0]: trigger event polarity

These bits define the polarity of the selected trigger event input defined by TRIGSEL[5:0].

00: no trigger (masked trigger event)

01: trigger on the rising edge

10: trigger on the falling edge

11: same as 00

Bits 23:22 Reserved, must be kept at reset value.

Bits 21:16 TRIGSEL[5:0]: trigger event input selection

These bits select the trigger event input of the GPDMA transfer (as per Section 12.3.6 ), with an active trigger event if TRIGPOL[1:0] ≠ 00.

Bits 15:14 TRIGM[1:0] : trigger mode

These bits define the transfer granularity for its conditioning by the trigger.

If the channel \( x \) is enabled (GPDMA_CxCR.EN asserted) with TRIGPOL[1:0] = 00 or 11, these TRIGM[1:0] bits are ignored.

Else, a GPDMA transfer is conditioned by one trigger hit:

00: at block level: the first burst read of each block transfer is conditioned by one hit trigger (channel \( x \) ( \( x = 12 \) to \( 15 \) ): same, each block also if a 2D/repeated block is configured with GPDMA_CxBR1.BRC[10:0] \( \neq 0 \) ).

01: channel \( x \) ( \( x = 0 \) to \( 11 \) ): same as 00; channel \( x \) ( \( x = 12 \) to \( 15 \) ): at 2D/repeated block level: the first burst read of the 2D/repeated block transfer is conditioned by one hit trigger.

10: at link level: a LLI link transfer is conditioned by one hit trigger. LLI data transfer is not conditioned.

11: at programmed burst level: If SWREQ = 1, each programmed burst read is conditioned by one hit trigger. If SWREQ = 0, each programmed burst that is requested by the selected peripheral, is conditioned by one hit trigger:

• – If the peripheral is the source (DREQ = 0) of the LLI data transfer, each programmed burst read is conditioned.

• – If the peripheral is the destination (DREQ = 1) of the LLI data transfer, each programmed burst write is conditioned. The first memory burst read of a (possibly 2D/repeated) block, also named as the first ready FIFO-based source burst, is gated by the occurrence of both the hardware request and one trigger hit.

As shown in Figure 105: Trigger hit, memorization, and overrun waveform , the GPDMA monitoring of a trigger for channel \( x \) is started when the channel is enabled/loading with a new active trigger configuration: rising or falling edge on a selected trigger (TRIGPOL[1:0] = 01 or respectively TRIGPOL[1:0] = 10). The monitoring of this trigger is kept active during the triggered and uncompleted (data or link) transfer; and if a new trigger is detected then, this hit is internally memorized to grant the next transfer, as long as the defined rising or falling edge is not modified, and the TRIGSEL[5:0] is not modified, and the channel is enabled. After a first new trigger hit _n+1 is memorized, if another second trigger hit _n+2 is detected and if the hit _n triggered transfer is still not completed, hit _n+2 is lost and not memorized. A trigger overrun flag is reported (GPDMA_CxSR.TOF = 1), and an interrupt is generated if enabled (GPDMA_CxCR.TOIE = 1). The channel is not automatically disabled by hardware due to a trigger overrun.

Transferring a next LLI _n+1 that updates the GPDMA_CxTR2 with a new value for any of TRIGSEL[5:0] or TRIGPOL[1:0], resets the monitoring, trashing the memorized hit of the formerly defined LLI _n trigger.

Note: When the source block size is not a multiple of the source burst size and is a multiple of the source data width, then the last programmed source burst is not completed and is internally shorten to match the block size. In this case, if TRIGM[1:0] = 11 and (SWREQ = 1 or (SWREQ = 0 and DREQ = 0)), the shortened burst transfer (by singles or/and by bursts of lower length) is conditioned once by the trigger.

When the programmed destination burst is internally shortened by singles or/and by bursts of lower length (versus FIFO size, versus block size, 1-Kbyte boundary address crossing): if the trigger is conditioning the programmed destination burst (if TRIGM[1:0] = 11 and SWREQ = 0 and DREQ = 1), this shortened destination burst transfer is conditioned once by the trigger.

Bit 13 Reserved, must be kept at reset value.

Bit 12 PFREQ : Hardware request in peripheral flow control mode

Important: If a given channel x is not implemented with this feature, this bit is reserved and PFREQ is not present (see Section 12.3.1 for the list of the implemented channels with this feature).

If the channel x is activated (GPDMA_CxCR.EN asserted) with SWREQ = 1 (software request for a memory-to-memory transfer), this bit is ignored. Else:

0: the selected hardware request is driven by a peripheral with a hardware request/acknowledge protocol in GPDMA control mode. The GPDMA is programmed with GPDMA_CxTR1.BNDT[15:0] and this is internally used by the hardware for the block transfer completion.

1: the selected hardware request is driven by a peripheral with a hardware request/acknowledge protocol in peripheral control mode. The GPDMA block transfer can be early completed by the peripheral itself (see Section 12.3.5 for more details).

Note: In peripheral flow control mode, there are the following restrictions:

• - no 2D/repeated block support (GPDMA_CxBR1.BRC[10:0] must be set to 0)
• - the peripheral must be set as the source of the transfer (DREQ = 0).
• - data packing to a wider destination width is not supported (if destination width > source data width, GPDMA_CxTR1.PAM[1] must be set to 0).
• - GPDMA_CxBR1.BNDT[15:0] must be programmed as a multiple of the source (peripheral) burst size.

Bit 11 BREQ : Block hardware request

If the channel x is activated (GPDMA_CxCR.EN asserted) with SWREQ = 1 (software request for a memory-to-memory transfer), this bit is ignored. Else:

0: the selected hardware request is driven by a peripheral with a hardware request/acknowledge protocol at a burst level.

1: the selected hardware request is driven by a peripheral with a hardware request/acknowledge protocol at a block level (see Section 12.3.3 ).

Bit 10 DREQ : destination hardware request

This bit is ignored if channel x is activated (GPDMA_CxCR.EN asserted) with SWREQ = 1 (software request for a memory-to-memory transfer). Else:

0: selected hardware request driven by a source peripheral (request signal taken into account by the GPDMA transfer scheduler over the source/read port)

1: selected hardware request driven by a destination peripheral (request signal taken into account by the GPDMA transfer scheduler over the destination/write port)

Note: If the channel x is activated (GPDMA_CxCR.EN is asserted) with SWREQ = 0 and PFREQ = 1 (peripheral hardware request with peripheral flow-control mode), any software assertion to this DREQ bit is ignored: in peripheral flow-control mode, only a peripheral-to-memory transfer is supported.

Bit 9 SWREQ : software request

This bit is internally taken into account when GPDMA_CxCR.EN is asserted.

0: no software request. The selected hardware request REQSEL[6:0] is taken into account.

1: software request for a memory-to-memory transfer. The default selected hardware request as per REQSEL[6:0] is ignored.

Bits 8:7 Reserved, must be kept at reset value.

Bits 6:0 REQSEL[6:0] : GPDMA hardware request selection

These bits are ignored if channel x is activated (GPDMA_CxCR.EN asserted) with SWREQ = 1 (software request for a memory-to-memory transfer). Else, the selected hardware request is internally taken into account as per Section 12.3.3 .

Caution: The user must not assign a same input hardware request (same REQSEL[6:0] value) to different active GPDMA channels (GPDMA_CxCR.EN = 1 and GPDMA_CxTR2.SWREQ = 0 for these channels). GPDMA is not intended to hardware support the case of simultaneous enabled channels incorrectly configured with a same hardware peripheral request signal, and there is no user setting error reporting.

12.8.10 GPDMA channel x block register 1 (GPDMA_CxBR1)

Address offset: 0x98 + 0x80 * x (x = 0 to 11)

Reset value: 0x0000 0000

This register is privileged or unprivileged, depending on the privileged state of channel x (GPDMA_PRIVCFGR.PRIVx).

This register controls the transfer of a channel x at a block level.

This register must be written when GPDMA_CxCR.EN = 0.

This register is read-only when GPDMA_CxCR.EN = 1.

This register must be written when channel x is completed (then the hardware has deasserted GPDMA_CxCR.EN). A channel transfer can be completed and programmed at different levels: block, or LLI or full linked-list.

In linked-list mode, during the link transfer:

• • if GPDMA_CxLLR.UB1 = 1, this register is automatically updated by the GPDMA from the next LLI in memory.
• • If GPDMA_CxLLR.UB1 = 0 and if there is at least one linked-list register to be updated from the next LLI in memory, this register is automatically and internally restored with the programmed value for the field BNDT[15:0].
• • If all the update bits GPDMA_CxLLR.Uxx are null and if GPDMA_CxLLR.LA[15:0] ≠ 0, the current LLI is the last one and is continuously executed: this register is automatically and internally restored with the programmed value for BNDT[15:0] after each execution of this final LLI
• • If GPDMA_CxLLR = 0, this register and BNDT[15:0] are kept as null, channel x is completed.

Bits 31:16 Reserved, must be kept at reset value.

Bits 15:0 BNDT[15:0] : block number of data bytes to transfer from the source

Block size transferred from the source. When the channel is enabled, this field becomes read-only and is decremented, indicating the remaining number of data items in the current source block to be transferred. BNDT[15:0] is programmed in number of bytes, maximum source block size is 64 Kbytes -1.

Once the last data transfer is completed (BNDT[15:0] = 0):

- if GPDMA_CxLLR.UB1 = 1, this field is updated with the LLI in the memory.

- if GPDMA_CxLLR.UB1 = 0 and if there is at least one non null Uxx update bit, this field is internally restored to the programmed value.

- if all GPDMA_CxLLR.Uxx = 0 and if GPDMA_CxLLR.LA[15:0] = 0, this field is internally restored to the programmed value (infinite/continuous last LLI).

- if GPDMA_CxLLR = 0, this field is kept as zero following the last LLI data transfer.

Note: A non-null source block size must be a multiple of the source data width (BNDT[2:0] versus GPDMA_CxTR1.SDW_LOG2[1:0]). Else a user setting error is reported and no transfer is issued.

When configured in packing mode (GPDMA_CxTR1.PAM[1] = 1 and destination data width different from source data width), a non-null source block size must be a multiple of the destination data width (BNDT[2:0] versus GPDMA_CxTR1.DDW_LOG2[1:0]). Else a user setting error is reported and no transfer is issued.

12.8.11 GPDMA channel x alternate block register 1 (GPDMA_CxBR1)

Address offset: 0x98 + 0x80 * x (x = 12 to 15)

Reset value: 0x0000 0000

This register is privileged or unprivileged, depending on the privileged state of channel x (GPDMA_PRIVCFGR.PRIVx).

This register controls the transfer of a channel x at a block level.

This register must be written when GPDMA_CxCR.EN = 0.

This register is read-only when GPDMA_CxCR.EN = 1.

This register must be written when channel x is completed (then the hardware has deasserted GPDMA_CxCR.EN). A channel transfer can be completed and programmed at different levels: block, or LLI or full linked-list.

In linked-list mode, during the link transfer:

• • if GPDMA_CxLLR.UB1 = 1, this register is automatically updated by the GPDMA from the next LLI in memory.
• • If GPDMA_CxLLR.UB1 = 0 and if there is at least one linked-list register to be updated from the next LLI in memory, this register is automatically and internally restored with the programmed value for the fields BNDT[15:0] and BRC[10:0].
• • If all the update bits GPDMA_CxLLR.Uxx are null and if GPDMA_CxLLR.LA[15:0] ≠ 0, the current LLI is the last one and is continuously executed: this register is

automatically and internally restored with the programmed value for the fields BNDT[15:0] and BRC[10:0] after each execution of this final LLI

• • If GPDMA_CxLLR = 0, BNDT[15:0] and BRC[10:0] are kept as null, channel x is completed.

Bit 31 BRDDEC: Block repeat destination address decrement

0: at the end of a block transfer, the GPDMA_CxDAR register is updated by adding the programmed offset GPDMA_CxBR2.BRDAO to the current GPDMA_CxDAR value (current destination address)

1: at the end of a block transfer, the GPDMA_CxDAR register is updated by subtracting the programmed offset GPDMA_CxBR2.BRDAO from the current GPDMA_CxDAR value (current destination address)

Note: On top of this increment/decrement (depending on BRDDEC), GPDMA_CxDAR is in the same time also updated by the increment/decrement (depending on DDEC) of the GPDMA_CxTR3.DAO value, as it is usually done at the end of each programmed burst transfer.

Bit 30 BRSDDEC: Block repeat source address decrement

0: at the end of a block transfer, the GPDMA_CxSAR register is updated by adding the programmed offset GPDMA_CxBR2.BRSAO to the current GPDMA_CxSAR value (current source address)

1: at the end of a block transfer, the GPDMA_CxSAR register is updated by subtracting the programmed offset GPDMA_CxBR2.BRSAO from the current GPDMA_CxSAR value (current source address)

Note: On top of this increment/decrement (depending on BRSDDEC), GPDMA_CxSAR is in the same time also updated by the increment/decrement (depending on SDEC) of the GPDMA_CxTR3.SAO value, as it is done after any programmed burst transfer.

Bit 29 DDEC: destination address decrement

0: At the end of a programmed burst transfer to the destination, the GPDMA_CxDAR register is updated by adding the programmed offset GPDMA_CxTR3.DAO to the current GPDMA_CxDAR value (current destination address)

1: At the end of a programmed burst transfer to the destination, the GPDMA_CxDAR register is updated by subtracting the programmed offset GPDMA_CxTR3.DAO to the current GPDMA_CxDAR value (current destination address)

Bit 28 SDEC: source address decrement

0: At the end of a programmed burst transfer from the source, the GPDMA_CxSAR register is updated by adding the programmed offset GPDMA_CxTR3.SAO to the current GPDMA_CxSAR value (current source address)

1: At the end of a programmed burst transfer from the source, the GPDMA_CxSAR register is updated by subtracting the programmed offset GPDMA_CxTR3.SAO to the current GPDMA_CxSAR value (current source address)

Bit 27 Reserved, must be kept at reset value.

Bits 26:16 BRC[10:0] : Block repeat counter

This field contains the number of repetitions of the current block (0 to 2047).

When the channel is enabled, this field becomes read-only. After decrements, this field indicates the remaining number of blocks, excluding the current one. This counter is hardware decremented for each completed block transfer.

Once the last block transfer is completed ( \( BRC[10:0] = BNDT[15:0] = 0 \) ):

• – If \( GPDMA\_CxLLR.UB1 = 1 \) , all \( GPDMA\_CxBR1 \) fields are updated by the next LLI in the memory.
• – If \( GPDMA\_CxLLR.UB1 = 0 \) and if there is at least one not null Uxx update bit, this field is internally restored to the programmed value.
• – if all \( GPDMA\_CxLLR.Uxx = 0 \) and if \( GPDMA\_CxLLR.LA[15:0] \neq 0 \) , this field is internally restored to the programmed value (infinite/continuous last LLI).
• – if \( GPDMA\_CxLLR = 0 \) , this field is kept as zero following the last LLI and data transfer.

Bits 15:0 BNDT[15:0] : block number of data bytes to transfer from the source

Block size transferred from the source. When the channel is enabled, this field becomes read-only and is decremented, indicating the remaining number of data items in the current source block to be transferred. \( BNDT[15:0] \) is programmed in number of bytes, maximum source block size is 64 Kbytes -1.

Once the last data transfer is completed ( \( BNDT[15:0] = 0 \) ):

• - if \( GPDMA\_CxLLR.UB1 = 1 \) , this field is updated with the LLI in the memory.
• - if \( GPDMA\_CxLLR.UB1 = 0 \) and if there is at least one not null Uxx update bit, this field is internally restored to the programmed value.
• - if all \( GPDMA\_CxLLR.Uxx = 0 \) and if \( GPDMA\_CxLLR.LA[15:0] \neq 0 \) , this field is internally restored to the programmed value (infinite/continuous last LLI).
• - if \( GPDMA\_CxLLR = 0 \) , this field is kept as zero following the last LLI data transfer.

Note: A non-null source block size must be a multiple of the source data width ( \( BNDT[2:0] \) versus \( GPDMA\_CxTR1.SDW\_LOG2[1:0] \) ). Else a user setting error is reported and no transfer is issued.

When configured in packing mode ( \( GPDMA\_CxTR1.PAM[1] = 1 \) and destination data width different from source data width), a non-null source block size must be a multiple of the destination data width ( \( BNDT[2:0] \) versus \( GPDMA\_CxTR1.DDW\_LOG2[1:0] \) ). Else a user setting error is reported and no transfer is issued.

12.8.12 GPDMA channel x source address register (GPDMA_CxSAR)

Address offset: \( 0x9C + 0x80 * x \) ( \( x = 0 \) to \( 15 \) )

Reset value: \( 0x0000\ 0000 \)

This register is privileged or unprivileged, depending on the privileged state of channel x (GPDMA_PRIVCFGR.PRIVx).

This register configures the source start address of a transfer.

This register must be written when GPDMA_CxCR.EN = 0.

This register is read-only when GPDMA_CxCR.EN = 1, and continuously updated by hardware, in order to reflect the address of the next burst transfer from the source.

This register must be written when the channel is completed (then the hardware has deasserted GPDMA_CxCR.EN). A channel transfer can be completed and programmed at different levels: block, 2D/repeated block, LLI or full linked-list.

In linked-list mode, during the link transfer, this register is automatically updated by the GPDMA from the memory if GPDMA_CxLLR.USA = 1.

Bits 31:0 SA[31:0] : source address

This field is the pointer to the address from which the next data is read.

During the channel activity, depending on the source addressing mode (GPDMA_CxTR1.SINC), this field is kept fixed or incremented by the data width (GPDMA_CxTR1.SDW_LOG2[1:0]) after each burst source data, reflecting the next address from which data is read.

During the channel activity, this address is updated after each completed source burst, consequently to:

• – the programmed source burst; either in fixed addressing mode or in contiguous-data incremented mode. If contiguously incremented (GPDMA_CxTR1.SINC = 1), then the additional address offset value is the programmed burst size, as defined by GPDMA_CxTR1.SBL_1[5:0] and GPDMA_CxTR1.SDW_LOG2[1:0]
• – the additional source incremented/decremented offset value as programmed by GPDMA_CxBR1.SDEC and GPDMA_CxTR3.SAO[12:0].
• – once/if completed source block transfer, for a channel x with 2D addressing capability ( \( x = 12 \) to \( 15 \) ), additional block repeat source incremented/decremented offset value as programmed by GPDMA_CxBR1.BRSDEC and GPDMA_CxBR2.BRSAO[15:0]

In linked-list mode, after a LLI data transfer is completed, this register is automatically updated by GPDMA from the memory, provided the LLI is set with GPDMA_CxLLR.USA = 1.

Note: A source address must be aligned with the programmed data width of a source burst (SA[2:0] versus GPDMA_CxTR1.SDW_LOG2[1:0]). Else, a user setting error is reported and no transfer is issued.

When the source block size is not a multiple of the source burst size and is a multiple of the source data width, the last programmed source burst is not completed and is internally shorten to match the block size. In this case, the additional GPDMA_CxTR3.SAO[12:0] is not applied.

12.8.13 GPDMA channel x destination address register (GPDMA_CxDAR)

Address offset: \( 0xA0 + 0x80 \cdot x \) ( \( x = 0 \) to 15)

Reset value: 0x0000 0000

This register is privileged or unprivileged, depending on the privileged state of channel x (GPDMA_PRIVCFGR.PRIVx).

This register configures the destination start address of a transfer.

This register must be written when GPDMA_CxCR.EN = 0.

This register is read-only when GPDMA_CxCR.EN = 1, and continuously updated by hardware, in order to reflect the address of the next burst transfer to the destination.

This register must be written when the channel is completed (then the hardware has deasserted GPDMA_CxCR.EN). A channel transfer can be completed and programmed at different levels: block, 2D/repeated block, LLI or full linked-list.

In linked-list mode, during the link transfer, this register is automatically updated by GPDMA from the memory if GPDMA_CxLLR.UDA = 1.

Bits 31:0 DA[31:0] : destination address

This field is the pointer to the address from which the next data is written.

During the channel activity, depending on the destination addressing mode (GPDMA_CxTR1.DINC), this field is kept fixed or incremented by the data width (GPDMA_CxTR1.DDW_LOG2[1:0]) after each burst destination data, reflecting the next address from which data is written.

During the channel activity, this address is updated after each completed destination burst, consequently to:

• – the programmed destination burst; either in fixed addressing mode or in contiguous-data incremented mode. If contiguously incremented (GPDMA_CxTR1.DINC = 1), then the additional address offset value is the programmed burst size, as defined by GPDMA_CxTR1.DB1_1[5:0] and GPDMA_CxTR1.DDW_LOG2[1:0]
• – the additional destination incremented/decremented offset value as programmed by GPDMA_CxBR1.DDEC and GPDMA_CxTR3.DAO[12:0].
• – once/if completed destination block transfer, for a channel x with 2D addressing capability ( \( x = 12 \) to 15), the additional block repeat destination incremented/decremented offset value as programmed by GPDMA_CxBR1.BRDDEC and GPDMA_CxBR2.BRDAO[15:0]

In linked-list mode, after a LLI data transfer is completed, this register is automatically updated by the GPDMA from the memory, provided the LLI is set with GPDMA_CxLLR.UDA = 1.

Note: A destination address must be aligned with the programmed data width of a destination burst (DA[2:0] versus GPDMA_CxTR1.DDW_LOG2[1:0]). Else, a user setting error is reported and no transfer is issued.

12.8.14 GPDMA channel x transfer register 3 (GPDMA_CxTR3)

Address offset: 0xA4 + 0x80 * x (x = 12 to 15)

Reset value: 0x0000 0000

This register is privileged or unprivileged, depending on the privileged state of channel x (GPDMA_PRIVCFGR.PRIVx).

This register controls the transfer of a channel x.

This register must be written when GPDMA_CxCR.EN = 0.

This register is read-only when GPDMA_CxCR.EN = 1.

This register must be written when the channel is completed (then the hardware has deasserted GPDMA_CxCR.EN). A channel transfer can be completed and programmed at different levels: block or LLI or full linked-list.

In linked-list mode, during the link transfer, this register is automatically updated by the GPDMA from the memory if GPDMA_CxLLR.UT3 = 1.

Bits 31:29 Reserved, must be kept at reset value.

Bits 28:16 DAO[12:0] : destination address offset increment

The destination address, pointed by GPDMA_CxDAR, is incremented or decremented (depending on GPDMA_CxBR1.DDEC) by this offset DAO[12:0] for each programmed destination burst. This offset is not including and is added to the programmed burst size when the completed burst is addressed in incremented mode (GPDMA_CxTR1.DINC = 1).

Note: A destination address offset must be aligned with the programmed data width of a destination burst (DAO[2:0] versus GPDMA_CxTR1.DDW_LOG2[1:0]). Else, a user setting error is reported and no transfer is issued.

Bits 15:13 Reserved, must be kept at reset value.

Bits 12:0 SAO[12:0] : source address offset increment

The source address, pointed by GPDMA_CxSAR, is incremented or decremented (depending on GPDMA_CxBR1.SDEC) by this offset SAO[12:0] for each programmed source burst. This offset is not including and is added to the programmed burst size when the completed burst is addressed in incremented mode (GPDMA_CxTR1.SINC = 1).

Note: A source address offset must be aligned with the programmed data width of a source burst (SAO[2:0] versus GPDMA_CxTR1.SDW_LOG2[1:0]). Else a user setting error is reported and none transfer is issued.

When the source block size is not a multiple of the destination burst size, and is a multiple of the source data width, then the last programmed source burst is not completed and is internally shorten to match the block size. In this case, the additional GPDMA_CxTR3.SAO[12:0] is not applied.

12.8.15 GPDMA channel x block register 2 (GPDMA_CxBR2)

Address offset: \( 0xA8 + 0x80 * x \) ( \( x = 12 \) to \( 15 \) )

Reset value: \( 0x0000\ 0000 \)

This register is privileged or unprivileged, depending on the privileged state of channel x (GPDMA_PRIVCFGR.PRIVx).

This register controls the transfer of a channel x at a 2D/repeated block level.

This register must be written when GPDMA_CxCR.EN = 0.

This register is read-only when GPDMA_CxCR.EN = 1.

This register must be written when the channel is completed (then the hardware has deasserted GPDMA_CxCR.EN). A channel transfer can be completed and programmed at different levels: block, 2D/repeated block, LLI or full linked-list.

In linked-list mode, during the link transfer, this register is automatically updated by the GPDMA from the memory if GPDMA_CxLLR.UB2 = 1.

Bits 31:16 BRDAO[15:0] : Block repeated destination address offset

For a channel with 2D addressing capability, this field is used to update (by addition or subtraction depending on GPDMA_CxBR1.BRDDEC) the current destination address (GPDMA_CxDAR) at the end of a block transfer.

Note: A block repeated destination address offset must be aligned with the programmed data width of a destination burst (BRDAO[2:0] versus GPDMA_CxTR1.DDW_LOG2[1:0]). Else a user setting error is reported and no transfer is issued. BRDAO[15:0] must be set to 0 in peripheral flow-control mode (if GPDMA_CxTR2.PFREQ = 1).

Bits 15:0 BRSAO[15:0] : Block repeated source address offset

For a channel with 2D addressing capability, this field is used to update (by addition or subtraction depending on GPDMA_CxBR1.BRSDEC) the current source address (GPDMA_CxSAR) at the end of a block transfer.

Note: A block repeated source address offset must be aligned with the programmed data width of a source burst (BRSAO[2:0] versus GPDMA_CxTR1.SDW_LOG2[1:0]). Else a user setting error is reported and no transfer is issued. BRSAO[15:0] must be set to 0 in peripheral flow-control mode (if GPDMA_CxTR2.PFREQ = 1).

12.8.16 GPDMA channel x linked-list address register (GPDMA_CxLLR)

Address offset: \( 0xCC + 0x80 * x \) ( \( x = 0 \) to \( 11 \) )

Reset value: \( 0x0000\ 0000 \)

This register is privileged or unprivileged, depending on the privileged state of channel x (GPDMA_PRIVCFG.PRIVx).

This register configures the data structure of the next LLI in the memory and its address pointer. A channel transfer is completed when this register is null.

This register must be written when GPDMA_CxCR.EN = 0.

This register is read-only when GPDMA_CxCR.EN = 1.

This register must be written when the channel is completed (then the hardware has deasserted GPDMA_CxCR.EN). A channel transfer can be completed and programmed at different levels: block or LLI or full linked-list.

In linked-list mode, during the link transfer, this register is automatically updated by the GPDMA from the memory if GPDMA_CxLLR.ULL = 1.

Bit 31 UT1 : Update GPDMA_CxTR1 from memory

This bit controls the update of GPDMA_CxTR1 from the memory during the link transfer.

0: no GPDMA_CxTR1 update

1: GPDMA_CxTR1 update

Bit 30 UT2 : Update GPDMA_CxTR2 from memory

This bit controls the update of GPDMA_CxTR2 from the memory during the link transfer.

0: no GPDMA_CxTR2 update

1: GPDMA_CxTR2 update

Bit 29 UB1 : Update GPDMA_CxBR1 from memory

This bit controls the update of GPDMA_CxBR1 from the memory during the link transfer. If UB1 = 0 and if GPDMA_CxLLR ≠ 0, the linked-list is not completed.

GPDMA_CxBR1.BNDT[15:0] is then restored to the programmed value after data transfer is completed and before the link transfer.

0: no GPDMA_CxBR1 update from memory (GPDMA_CxBR1.BNDT[15:0] restored if any link transfer)

1: GPDMA_CxBR1 update

Bit 28 USA : update GPDMA_CxSAR from memory

This bit controls the update of GPDMA_CxSAR from the memory during the link transfer.

0: no GPDMA_CxSAR update

1: GPDMA_CxSAR update

Bit 27 UDA : Update GPDMA_CxDAR register from memory

This bit is used to control the update of GPDMA_CxDAR from the memory during the link transfer.

0: no GPDMA_CxDAR update

1: GPDMA_CxDAR update

Bits 26:17 Reserved, must be kept at reset value.

Bit 16 ULL : Update GPDMA_CxLLR register from memory

This bit is used to control the update of GPDMA_CxLLR from the memory during the link transfer.

0: no GPDMA_CxLLR update

1: GPDMA_CxLLR update

Bits 15:2 LA[15:2] : pointer (16-bit low-significant address) to the next linked-list data structure

If UT1 = UT2 = UB1 = USA = UDA = ULL = 0 and if LA[15:2] = 0, the current LLI is the last one. The channel transfer is completed without any update of the linked-list GPDMA register file.

Else, this field is the pointer to the memory address offset from which the next linked-list data structure is automatically fetched from, once the data transfer is completed, in order to conditionally update the linked-list GPDMA internal register file (GPDMA_CxTR1, GPDMA_CxTR2, GPDMA_CxBR1, GPDMA_CxSAR, GPDMA_CxDAR, and GPDMA_CxLLR).

Caution: The user must program this pointer not to exceed the 64-Kbyte addressable space defined by the link base address register GPDMA_CxLBAR. The complete linked-list data structure must be included in the 64-Kbyte addressable space.

Note: The user must program the pointer to be 32-bit aligned. The two low-significant bits are write ignored.

Bits 1:0 Reserved, must be kept at reset value.

12.8.17 GPDMA channel x alternate linked-list address register (GPDMA_CxLLR)

Address offset: 0xCC + 0x80 * x (x = 12 to 15)

Reset value: 0x0000 0000

This register is privileged or unprivileged, depending on the privileged state of channel x (GPDMA_PRIVCFG.R.PRIVx).

This register configures the data structure of the next LLI in the memory and its address pointer. A channel transfer is completed when this register is null.

This register must be written when GPDMA_CxCR.EN = 0.

This register is read-only when GPDMA_CxCR.EN = 1.

This register must be written when the channel is completed (then the hardware has deasserted GPDMA_CxCR.EN). A channel transfer can be completed and programmed at different levels: block or LLI or full linked-list.

In linked-list mode, during the link transfer, this register is automatically updated by the GPDMA from the memory if GPDMA_CxLLR.ULL = 1.

Bit 31 UT1 : Update GPDMA_CxTR1 from memory

This bit controls the update of GPDMA_CxTR1 from the memory during the link transfer.

0: no GPDMA_CxTR1 update

1: GPDMA_CxTR1 update

Bit 30 UT2 : Update GPDMA_CxTR2 from memory

This bit controls the update of GPDMA_CxTR2 from the memory during the link transfer.

0: no GPDMA_CxTR2 update

1: GPDMA_CxTR2 update

Bit 29 UB1 : Update GPDMA_CxBR1 from memory

This bit controls the update of GPDMA_CxBR1 from the memory during the link transfer.

If UB1 = 0 and if GPDMA_CxLLR ≠ 0, the linked-list is not completed.

GPDMA_CxBR1.BNDT[15:0] is then restored to the programmed value after data transfer is completed and before the link transfer.

0: no GPDMA_CxBR1 update from memory (GPDMA_CxBR1.BNDT[15:0] restored if any link transfer)

1: GPDMA_CxBR1 update

Bit 28 USA : update GPDMA_CxSAR from memory

This bit controls the update of GPDMA_CxSAR from the memory during the link transfer.

0: no GPDMA_CxSAR update

1: GPDMA_CxSAR update

Bit 27 UDA : Update GPDMA_CxDAR register from memory

This bit is used to control the update of GPDMA_CxDAR from the memory during the link transfer.
0: no GPDMA_CxDAR update
1: GPDMA_CxDAR update

Bit 26 UT3 : Update GPDMA_CxTR3 from memory

This bit controls the update of GPDMA_CxTR3 from the memory during the link transfer.
0: no GPDMA_CxTR3 update
1: GPDMA_CxTR3 update

Bit 25 UB2 : Update GPDMA_CxBR2 from memory

This bit controls the update of GPDMA_CxBR2 from the memory during the link transfer.
0: no GPDMA_CxBR2 update
1: GPDMA_CxBR2 update

Bits 24:17 Reserved, must be kept at reset value.

Bit 16 ULL : Update GPDMA_CxLLR register from memory

This bit is used to control the update of GPDMA_CxLLR from the memory during the link transfer.
0: no GPDMA_CxLLR update
1: GPDMA_CxLLR update

Bits 15:2 LA[15:2] : pointer (16-bit low-significant address) to the next linked-list data structure

If UT1 = UT2 = UB1 = USA = UDA = ULL = 0 and if LA[15:2] = 0, the current LLI is the last one. The channel transfer is completed without any update of the linked-list GPDMA register file.

Else, this field is the pointer to the memory address offset from which the next linked-list data structure is automatically fetched from, once the data transfer is completed, in order to conditionally update the linked-list GPDMA internal register file (GPDMA_CxTR1, GPDMA_CxTR2, GPDMA_CxBR1, GPDMA_CxSAR, GPDMA_CxDAR, and GPDMA_CxLLR).

Caution: The user must program this pointer not to exceed the 64-Kbyte addressable space defined by the link base address register GPDMA_CxLBAR. The complete linked-list data structure must be included in the 64-Kbyte addressable space.

Note: The user must program the pointer to be 32-bit aligned. The two low-significant bits are write ignored.

Bits 1:0 Reserved, must be kept at reset value.

12.8.18 GPDMA register map

Table 99. GPDMA register map and reset values

Table 99. GPDMA register map and reset values (continued)