* [PATCH v5 0/2] Add AMD MDB Endpoint and non-LL mode Support
@ 2025-10-28 11:28 Devendra K Verma
2025-10-28 11:28 ` [PATCH v5 1/2] dmaengine: dw-edma: Add AMD MDB Endpoint Support Devendra K Verma
2025-10-28 11:28 ` [PATCH v5 2/2] dmaengine: dw-edma: Add non-LL mode Devendra K Verma
0 siblings, 2 replies; 7+ messages in thread
From: Devendra K Verma @ 2025-10-28 11:28 UTC (permalink / raw)
To: bhelgaas, mani, vkoul
Cc: dmaengine, linux-pci, linux-kernel, michal.simek, Devendra.Verma
This series of patch support the following:
- AMD MDB Endpoint Support, as part of this patch following are
added:
o AMD supported device ID and vendor ID (Xilinx)
o AMD MDB specific driver data
o AMD specific VSEC capabilities to retrieve the base of
phys address of MDB side DDR
o Logic to assign the offsets to LL and data blocks if
more number of channels are enabled than configured
in the given pci_data struct.
- Addition of non-LL mode
o The IP supported non-LL mode functions
o Flexibility to choose non-LL mode via dma_slave_config
param peripheral_config, by the client for all the vendors
using HDMA IP.
o Allow IP utilization if LL mode is not available
Devendra K Verma (2):
dmaengine: dw-edma: Add AMD MDB Endpoint Support
dmaengine: dw-edma: Add non-LL mode
drivers/dma/dw-edma/dw-edma-core.c | 41 ++++++++-
drivers/dma/dw-edma/dw-edma-core.h | 1 +
drivers/dma/dw-edma/dw-edma-pcie.c | 168 ++++++++++++++++++++++++++++++++--
drivers/dma/dw-edma/dw-hdma-v0-core.c | 62 ++++++++++++-
include/linux/dma/edma.h | 1 +
5 files changed, 259 insertions(+), 14 deletions(-)
--
1.8.3.1
^ permalink raw reply [flat|nested] 7+ messages in thread
* [PATCH v5 1/2] dmaengine: dw-edma: Add AMD MDB Endpoint Support
2025-10-28 11:28 [PATCH v5 0/2] Add AMD MDB Endpoint and non-LL mode Support Devendra K Verma
@ 2025-10-28 11:28 ` Devendra K Verma
2025-10-28 13:28 ` Ilpo Järvinen
2025-10-28 11:28 ` [PATCH v5 2/2] dmaengine: dw-edma: Add non-LL mode Devendra K Verma
1 sibling, 1 reply; 7+ messages in thread
From: Devendra K Verma @ 2025-10-28 11:28 UTC (permalink / raw)
To: bhelgaas, mani, vkoul
Cc: dmaengine, linux-pci, linux-kernel, michal.simek, Devendra.Verma
AMD MDB PCIe endpoint support. For AMD specific support
added the following
- AMD supported PCIe Device IDs and Vendor ID (Xilinx).
- AMD MDB specific driver data
- AMD MDB specific VSEC capability to retrieve the device DDR
base address.
Signed-off-by: Devendra K Verma <devendra.verma@amd.com>
---
Changes in v5:
Added the definitions for Xilinx specific VSEC header id,
revision, and register offsets.
Corrected the error type when no physical offset found for
device side memory.
Corrected the order of variables.
Changes in v4:
Configured 8 read and 8 write channels for Xilinx vendor
Added checks to validate vendor ID for vendor
specific vsec id.
Added Xilinx specific vendor id for vsec specific to Xilinx
Added the LL and data region offsets, size as input params to
function dw_edma_set_chan_region_offset().
Moved the LL and data region offsets assignment to function
for Xilinx specific case.
Corrected comments.
Changes in v3:
Corrected a typo when assigning AMD (Xilinx) vsec id macro
and condition check.
Changes in v2:
Reverted the devmem_phys_off type to u64.
Renamed the function appropriately to suit the
functionality for setting the LL & data region offsets.
Changes in v1:
Removed the pci device id from pci_ids.h file.
Added the vendor id macro as per the suggested method.
Changed the type of the newly added devmem_phys_off variable.
Added to logic to assign offsets for LL and data region blocks
in case more number of channels are enabled than given in
amd_mdb_data struct.
---
drivers/dma/dw-edma/dw-edma-pcie.c | 138 ++++++++++++++++++++++++++++++++++++-
1 file changed, 136 insertions(+), 2 deletions(-)
diff --git a/drivers/dma/dw-edma/dw-edma-pcie.c b/drivers/dma/dw-edma/dw-edma-pcie.c
index 3371e0a7..7b991a0 100644
--- a/drivers/dma/dw-edma/dw-edma-pcie.c
+++ b/drivers/dma/dw-edma/dw-edma-pcie.c
@@ -17,12 +17,27 @@
#include "dw-edma-core.h"
+/* Synopsys */
#define DW_PCIE_VSEC_DMA_ID 0x6
#define DW_PCIE_VSEC_DMA_BAR GENMASK(10, 8)
#define DW_PCIE_VSEC_DMA_MAP GENMASK(2, 0)
#define DW_PCIE_VSEC_DMA_WR_CH GENMASK(9, 0)
#define DW_PCIE_VSEC_DMA_RD_CH GENMASK(25, 16)
+/* AMD MDB (Xilinx) specific defines */
+#define DW_PCIE_XILINX_MDB_VSEC_DMA_ID 0x6
+#define DW_PCIE_XILINX_MDB_VSEC_ID 0x20
+#define PCI_DEVICE_ID_AMD_MDB_B054 0xb054
+#define DW_PCIE_AMD_MDB_INVALID_ADDR (~0ULL)
+#define DW_PCIE_XILINX_LL_OFF_GAP 0x200000
+#define DW_PCIE_XILINX_LL_SIZE 0x800
+#define DW_PCIE_XILINX_DT_OFF_GAP 0x100000
+#define DW_PCIE_XILINX_DT_SIZE 0x800
+#define DW_PCIE_XILINX_MDB_VSEC_HDR_ID 0x20
+#define DW_PCIE_XILINX_MDB_VSEC_REV 0x1
+#define DW_PCIE_XILINX_MDB_DEVMEM_OFF_REG_HIGH 0xc
+#define DW_PCIE_XILINX_MDB_DEVMEM_OFF_REG_LOW 0x8
+
#define DW_BLOCK(a, b, c) \
{ \
.bar = a, \
@@ -50,6 +65,7 @@ struct dw_edma_pcie_data {
u8 irqs;
u16 wr_ch_cnt;
u16 rd_ch_cnt;
+ u64 devmem_phys_off;
};
static const struct dw_edma_pcie_data snps_edda_data = {
@@ -90,6 +106,64 @@ struct dw_edma_pcie_data {
.rd_ch_cnt = 2,
};
+static const struct dw_edma_pcie_data amd_mdb_data = {
+ /* MDB registers location */
+ .rg.bar = BAR_0,
+ .rg.off = 0x00001000, /* 4 Kbytes */
+ .rg.sz = 0x00002000, /* 8 Kbytes */
+
+ /* Other */
+ .mf = EDMA_MF_HDMA_NATIVE,
+ .irqs = 1,
+ .wr_ch_cnt = 8,
+ .rd_ch_cnt = 8,
+};
+
+static void dw_edma_set_chan_region_offset(struct dw_edma_pcie_data *pdata,
+ enum pci_barno bar, off_t start_off,
+ off_t ll_off_gap, size_t ll_size,
+ off_t dt_off_gap, size_t dt_size)
+{
+ u16 wr_ch = pdata->wr_ch_cnt;
+ u16 rd_ch = pdata->rd_ch_cnt;
+ off_t off;
+ u16 i;
+
+ off = start_off;
+
+ /* Write channel LL region */
+ for (i = 0; i < wr_ch; i++) {
+ pdata->ll_wr[i].bar = bar;
+ pdata->ll_wr[i].off = off;
+ pdata->ll_wr[i].sz = ll_size;
+ off += ll_off_gap;
+ }
+
+ /* Read channel LL region */
+ for (i = 0; i < rd_ch; i++) {
+ pdata->ll_rd[i].bar = bar;
+ pdata->ll_rd[i].off = off;
+ pdata->ll_rd[i].sz = ll_size;
+ off += ll_off_gap;
+ }
+
+ /* Write channel data region */
+ for (i = 0; i < wr_ch; i++) {
+ pdata->dt_wr[i].bar = bar;
+ pdata->dt_wr[i].off = off;
+ pdata->dt_wr[i].sz = dt_size;
+ off += dt_off_gap;
+ }
+
+ /* Read channel data region */
+ for (i = 0; i < rd_ch; i++) {
+ pdata->dt_rd[i].bar = bar;
+ pdata->dt_rd[i].off = off;
+ pdata->dt_rd[i].sz = dt_size;
+ off += dt_off_gap;
+ }
+}
+
static int dw_edma_pcie_irq_vector(struct device *dev, unsigned int nr)
{
return pci_irq_vector(to_pci_dev(dev), nr);
@@ -120,9 +194,24 @@ static void dw_edma_pcie_get_vsec_dma_data(struct pci_dev *pdev,
u32 val, map;
u16 vsec;
u64 off;
+ int cap;
- vsec = pci_find_vsec_capability(pdev, PCI_VENDOR_ID_SYNOPSYS,
- DW_PCIE_VSEC_DMA_ID);
+ /*
+ * Synopsys and AMD (Xilinx) use the same VSEC ID for the purpose
+ * of map, channel counts, etc.
+ */
+ switch (pdev->vendor) {
+ case PCI_VENDOR_ID_SYNOPSYS:
+ cap = DW_PCIE_VSEC_DMA_ID;
+ break;
+ case PCI_VENDOR_ID_XILINX:
+ cap = DW_PCIE_XILINX_MDB_VSEC_DMA_ID;
+ break;
+ default:
+ return;
+ }
+
+ vsec = pci_find_vsec_capability(pdev, pdev->vendor, cap);
if (!vsec)
return;
@@ -155,6 +244,28 @@ static void dw_edma_pcie_get_vsec_dma_data(struct pci_dev *pdev,
off <<= 32;
off |= val;
pdata->rg.off = off;
+
+ /* Xilinx specific VSEC capability */
+ vsec = pci_find_vsec_capability(pdev, PCI_VENDOR_ID_XILINX,
+ DW_PCIE_XILINX_MDB_VSEC_ID);
+ if (!vsec)
+ return;
+
+ pci_read_config_dword(pdev, vsec + PCI_VNDR_HEADER, &val);
+ if (PCI_VNDR_HEADER_ID(val) != DW_PCIE_XILINX_MDB_VSEC_HDR_ID ||
+ PCI_VNDR_HEADER_REV(val) != DW_PCIE_XILINX_MDB_VSEC_REV)
+ return;
+
+ pci_read_config_dword(pdev,
+ vsec + DW_PCIE_XILINX_MDB_DEVMEM_OFF_REG_HIGH,
+ &val);
+ off = val;
+ pci_read_config_dword(pdev,
+ vsec + DW_PCIE_XILINX_MDB_DEVMEM_OFF_REG_LOW,
+ &val);
+ off <<= 32;
+ off |= val;
+ pdata->devmem_phys_off = off;
}
static int dw_edma_pcie_probe(struct pci_dev *pdev,
@@ -179,6 +290,7 @@ static int dw_edma_pcie_probe(struct pci_dev *pdev,
}
memcpy(vsec_data, pdata, sizeof(struct dw_edma_pcie_data));
+ vsec_data->devmem_phys_off = DW_PCIE_AMD_MDB_INVALID_ADDR;
/*
* Tries to find if exists a PCIe Vendor-Specific Extended Capability
@@ -186,6 +298,26 @@ static int dw_edma_pcie_probe(struct pci_dev *pdev,
*/
dw_edma_pcie_get_vsec_dma_data(pdev, vsec_data);
+ if (pdev->vendor == PCI_VENDOR_ID_XILINX) {
+ /*
+ * There is no valid address found for the LL memory
+ * space on the device side.
+ */
+ if (vsec_data->devmem_phys_off == DW_PCIE_AMD_MDB_INVALID_ADDR)
+ return -ENOMEM;
+
+ /*
+ * Configure the channel LL and data blocks if number of
+ * channels enabled in VSEC capability are more than the
+ * channels configured in amd_mdb_data.
+ */
+ dw_edma_set_chan_region_offset(vsec_data, BAR_2, 0,
+ DW_PCIE_XILINX_LL_OFF_GAP,
+ DW_PCIE_XILINX_LL_SIZE,
+ DW_PCIE_XILINX_DT_OFF_GAP,
+ DW_PCIE_XILINX_DT_SIZE);
+ }
+
/* Mapping PCI BAR regions */
mask = BIT(vsec_data->rg.bar);
for (i = 0; i < vsec_data->wr_ch_cnt; i++) {
@@ -367,6 +499,8 @@ static void dw_edma_pcie_remove(struct pci_dev *pdev)
static const struct pci_device_id dw_edma_pcie_id_table[] = {
{ PCI_DEVICE_DATA(SYNOPSYS, EDDA, &snps_edda_data) },
+ { PCI_VDEVICE(XILINX, PCI_DEVICE_ID_AMD_MDB_B054),
+ (kernel_ulong_t)&amd_mdb_data },
{ }
};
MODULE_DEVICE_TABLE(pci, dw_edma_pcie_id_table);
--
1.8.3.1
^ permalink raw reply related [flat|nested] 7+ messages in thread
* [PATCH v5 2/2] dmaengine: dw-edma: Add non-LL mode
2025-10-28 11:28 [PATCH v5 0/2] Add AMD MDB Endpoint and non-LL mode Support Devendra K Verma
2025-10-28 11:28 ` [PATCH v5 1/2] dmaengine: dw-edma: Add AMD MDB Endpoint Support Devendra K Verma
@ 2025-10-28 11:28 ` Devendra K Verma
2025-10-28 13:32 ` Ilpo Järvinen
1 sibling, 1 reply; 7+ messages in thread
From: Devendra K Verma @ 2025-10-28 11:28 UTC (permalink / raw)
To: bhelgaas, mani, vkoul
Cc: dmaengine, linux-pci, linux-kernel, michal.simek, Devendra.Verma
AMD MDB IP supports Linked List (LL) mode as well as non-LL mode.
The current code does not have the mechanisms to enable the
DMA transactions using the non-LL mode. The following two cases
are added with this patch:
- When a valid physical base address is not configured via the
Xilinx VSEC capability then the IP can still be used in non-LL
mode. The default mode for all the DMA transactions and for all
the DMA channels then is non-LL mode.
- When a valid physical base address is configured but the client
wants to use the non-LL mode for DMA transactions then also the
flexibility is provided via the peripheral_config struct member of
dma_slave_config. In this case the channels can be individually
configured in non-LL mode. This use case is desirable for single
DMA transfer of a chunk, this saves the effort of preparing the
Link List. This particular scenario is applicable to AMD as well
as Synopsys IP.
Signed-off-by: Devendra K Verma <devendra.verma@amd.com>
---
Changes in v5
Variable name 'nollp' changed to 'non_ll'.
In the dw_edma_device_config() WARN_ON replaced with dev_err().
Comments follow the 80-column guideline.
Changes in v4
No change
Changes in v3
No change
Changes in v2
Reverted the function return type to u64 for
dw_edma_get_phys_addr().
Changes in v1
Changed the function return type for dw_edma_get_phys_addr().
Corrected the typo raised in review.
---
drivers/dma/dw-edma/dw-edma-core.c | 41 ++++++++++++++++++++---
drivers/dma/dw-edma/dw-edma-core.h | 1 +
drivers/dma/dw-edma/dw-edma-pcie.c | 44 +++++++++++++++++--------
drivers/dma/dw-edma/dw-hdma-v0-core.c | 62 ++++++++++++++++++++++++++++++++++-
include/linux/dma/edma.h | 1 +
5 files changed, 130 insertions(+), 19 deletions(-)
diff --git a/drivers/dma/dw-edma/dw-edma-core.c b/drivers/dma/dw-edma/dw-edma-core.c
index b43255f..60a3279 100644
--- a/drivers/dma/dw-edma/dw-edma-core.c
+++ b/drivers/dma/dw-edma/dw-edma-core.c
@@ -223,8 +223,31 @@ static int dw_edma_device_config(struct dma_chan *dchan,
struct dma_slave_config *config)
{
struct dw_edma_chan *chan = dchan2dw_edma_chan(dchan);
+ int non_ll = 0;
+
+ if (config->peripheral_config &&
+ config->peripheral_size != sizeof(int)) {
+ dev_err(dchan->device->dev,
+ "config param peripheral size mismatch\n");
+ return -EINVAL;
+ }
memcpy(&chan->config, config, sizeof(*config));
+
+ /*
+ * When there is no valid LLP base address available then the default
+ * DMA ops will use the non-LL mode.
+ * Cases where LL mode is enabled and client wants to use the non-LL
+ * mode then also client can do so via providing the peripheral_config
+ * param.
+ */
+ if (config->peripheral_config)
+ non_ll = *(int *)config->peripheral_config;
+
+ chan->non_ll = false;
+ if (chan->dw->chip->non_ll || (!chan->dw->chip->non_ll && non_ll))
+ chan->non_ll = true;
+
chan->configured = true;
return 0;
@@ -353,7 +376,7 @@ static void dw_edma_device_issue_pending(struct dma_chan *dchan)
struct dw_edma_chan *chan = dchan2dw_edma_chan(xfer->dchan);
enum dma_transfer_direction dir = xfer->direction;
struct scatterlist *sg = NULL;
- struct dw_edma_chunk *chunk;
+ struct dw_edma_chunk *chunk = NULL;
struct dw_edma_burst *burst;
struct dw_edma_desc *desc;
u64 src_addr, dst_addr;
@@ -419,9 +442,11 @@ static void dw_edma_device_issue_pending(struct dma_chan *dchan)
if (unlikely(!desc))
goto err_alloc;
- chunk = dw_edma_alloc_chunk(desc);
- if (unlikely(!chunk))
- goto err_alloc;
+ if (!chan->non_ll) {
+ chunk = dw_edma_alloc_chunk(desc);
+ if (unlikely(!chunk))
+ goto err_alloc;
+ }
if (xfer->type == EDMA_XFER_INTERLEAVED) {
src_addr = xfer->xfer.il->src_start;
@@ -450,7 +475,13 @@ static void dw_edma_device_issue_pending(struct dma_chan *dchan)
if (xfer->type == EDMA_XFER_SCATTER_GATHER && !sg)
break;
- if (chunk->bursts_alloc == chan->ll_max) {
+ /*
+ * For non-LL mode, only a single burst can be handled
+ * in a single chunk unlike LL mode where multiple bursts
+ * can be configured in a single chunk.
+ */
+ if ((chunk && chunk->bursts_alloc == chan->ll_max) ||
+ chan->non_ll) {
chunk = dw_edma_alloc_chunk(desc);
if (unlikely(!chunk))
goto err_alloc;
diff --git a/drivers/dma/dw-edma/dw-edma-core.h b/drivers/dma/dw-edma/dw-edma-core.h
index 71894b9..c8e3d19 100644
--- a/drivers/dma/dw-edma/dw-edma-core.h
+++ b/drivers/dma/dw-edma/dw-edma-core.h
@@ -86,6 +86,7 @@ struct dw_edma_chan {
u8 configured;
struct dma_slave_config config;
+ bool non_ll;
};
struct dw_edma_irq {
diff --git a/drivers/dma/dw-edma/dw-edma-pcie.c b/drivers/dma/dw-edma/dw-edma-pcie.c
index 7b991a0..b02977c 100644
--- a/drivers/dma/dw-edma/dw-edma-pcie.c
+++ b/drivers/dma/dw-edma/dw-edma-pcie.c
@@ -268,6 +268,15 @@ static void dw_edma_pcie_get_vsec_dma_data(struct pci_dev *pdev,
pdata->devmem_phys_off = off;
}
+static u64 dw_edma_get_phys_addr(struct pci_dev *pdev,
+ struct dw_edma_pcie_data *pdata,
+ enum pci_barno bar)
+{
+ if (pdev->vendor == PCI_VENDOR_ID_XILINX)
+ return pdata->devmem_phys_off;
+ return pci_bus_address(pdev, bar);
+}
+
static int dw_edma_pcie_probe(struct pci_dev *pdev,
const struct pci_device_id *pid)
{
@@ -277,6 +286,7 @@ static int dw_edma_pcie_probe(struct pci_dev *pdev,
struct dw_edma_chip *chip;
int err, nr_irqs;
int i, mask;
+ bool non_ll = false;
vsec_data = kmalloc(sizeof(*vsec_data), GFP_KERNEL);
if (!vsec_data)
@@ -301,21 +311,24 @@ static int dw_edma_pcie_probe(struct pci_dev *pdev,
if (pdev->vendor == PCI_VENDOR_ID_XILINX) {
/*
* There is no valid address found for the LL memory
- * space on the device side.
+ * space on the device side. In the absence of LL base
+ * address use the non-LL mode or simple mode supported by
+ * the HDMA IP.
*/
if (vsec_data->devmem_phys_off == DW_PCIE_AMD_MDB_INVALID_ADDR)
- return -ENOMEM;
+ non_ll = true;
/*
* Configure the channel LL and data blocks if number of
* channels enabled in VSEC capability are more than the
* channels configured in amd_mdb_data.
*/
- dw_edma_set_chan_region_offset(vsec_data, BAR_2, 0,
- DW_PCIE_XILINX_LL_OFF_GAP,
- DW_PCIE_XILINX_LL_SIZE,
- DW_PCIE_XILINX_DT_OFF_GAP,
- DW_PCIE_XILINX_DT_SIZE);
+ if (!non_ll)
+ dw_edma_set_chan_region_offset(vsec_data, BAR_2, 0,
+ DW_PCIE_XILINX_LL_OFF_GAP,
+ DW_PCIE_XILINX_LL_SIZE,
+ DW_PCIE_XILINX_DT_OFF_GAP,
+ DW_PCIE_XILINX_DT_SIZE);
}
/* Mapping PCI BAR regions */
@@ -363,6 +376,7 @@ static int dw_edma_pcie_probe(struct pci_dev *pdev,
chip->mf = vsec_data->mf;
chip->nr_irqs = nr_irqs;
chip->ops = &dw_edma_pcie_plat_ops;
+ chip->non_ll = non_ll;
chip->ll_wr_cnt = vsec_data->wr_ch_cnt;
chip->ll_rd_cnt = vsec_data->rd_ch_cnt;
@@ -371,7 +385,7 @@ static int dw_edma_pcie_probe(struct pci_dev *pdev,
if (!chip->reg_base)
return -ENOMEM;
- for (i = 0; i < chip->ll_wr_cnt; i++) {
+ for (i = 0; i < chip->ll_wr_cnt && !non_ll; i++) {
struct dw_edma_region *ll_region = &chip->ll_region_wr[i];
struct dw_edma_region *dt_region = &chip->dt_region_wr[i];
struct dw_edma_block *ll_block = &vsec_data->ll_wr[i];
@@ -382,7 +396,8 @@ static int dw_edma_pcie_probe(struct pci_dev *pdev,
return -ENOMEM;
ll_region->vaddr.io += ll_block->off;
- ll_region->paddr = pci_bus_address(pdev, ll_block->bar);
+ ll_region->paddr = dw_edma_get_phys_addr(pdev, vsec_data,
+ ll_block->bar);
ll_region->paddr += ll_block->off;
ll_region->sz = ll_block->sz;
@@ -391,12 +406,13 @@ static int dw_edma_pcie_probe(struct pci_dev *pdev,
return -ENOMEM;
dt_region->vaddr.io += dt_block->off;
- dt_region->paddr = pci_bus_address(pdev, dt_block->bar);
+ dt_region->paddr = dw_edma_get_phys_addr(pdev, vsec_data,
+ dt_block->bar);
dt_region->paddr += dt_block->off;
dt_region->sz = dt_block->sz;
}
- for (i = 0; i < chip->ll_rd_cnt; i++) {
+ for (i = 0; i < chip->ll_rd_cnt && !non_ll; i++) {
struct dw_edma_region *ll_region = &chip->ll_region_rd[i];
struct dw_edma_region *dt_region = &chip->dt_region_rd[i];
struct dw_edma_block *ll_block = &vsec_data->ll_rd[i];
@@ -407,7 +423,8 @@ static int dw_edma_pcie_probe(struct pci_dev *pdev,
return -ENOMEM;
ll_region->vaddr.io += ll_block->off;
- ll_region->paddr = pci_bus_address(pdev, ll_block->bar);
+ ll_region->paddr = dw_edma_get_phys_addr(pdev, vsec_data,
+ ll_block->bar);
ll_region->paddr += ll_block->off;
ll_region->sz = ll_block->sz;
@@ -416,7 +433,8 @@ static int dw_edma_pcie_probe(struct pci_dev *pdev,
return -ENOMEM;
dt_region->vaddr.io += dt_block->off;
- dt_region->paddr = pci_bus_address(pdev, dt_block->bar);
+ dt_region->paddr = dw_edma_get_phys_addr(pdev, vsec_data,
+ dt_block->bar);
dt_region->paddr += dt_block->off;
dt_region->sz = dt_block->sz;
}
diff --git a/drivers/dma/dw-edma/dw-hdma-v0-core.c b/drivers/dma/dw-edma/dw-hdma-v0-core.c
index e3f8db4..47ecc84 100644
--- a/drivers/dma/dw-edma/dw-hdma-v0-core.c
+++ b/drivers/dma/dw-edma/dw-hdma-v0-core.c
@@ -225,7 +225,7 @@ static void dw_hdma_v0_sync_ll_data(struct dw_edma_chunk *chunk)
readl(chunk->ll_region.vaddr.io);
}
-static void dw_hdma_v0_core_start(struct dw_edma_chunk *chunk, bool first)
+static void dw_hdma_v0_core_ll_start(struct dw_edma_chunk *chunk, bool first)
{
struct dw_edma_chan *chan = chunk->chan;
struct dw_edma *dw = chan->dw;
@@ -263,6 +263,66 @@ static void dw_hdma_v0_core_start(struct dw_edma_chunk *chunk, bool first)
SET_CH_32(dw, chan->dir, chan->id, doorbell, HDMA_V0_DOORBELL_START);
}
+static void dw_hdma_v0_core_non_ll_start(struct dw_edma_chunk *chunk)
+{
+ struct dw_edma_chan *chan = chunk->chan;
+ struct dw_edma *dw = chan->dw;
+ struct dw_edma_burst *child;
+ u32 val;
+
+ list_for_each_entry(child, &chunk->burst->list, list) {
+ SET_CH_32(dw, chan->dir, chan->id, ch_en, BIT(0));
+
+ /* Source address */
+ SET_CH_32(dw, chan->dir, chan->id, sar.lsb,
+ lower_32_bits(child->sar));
+ SET_CH_32(dw, chan->dir, chan->id, sar.msb,
+ upper_32_bits(child->sar));
+
+ /* Destination address */
+ SET_CH_32(dw, chan->dir, chan->id, dar.lsb,
+ lower_32_bits(child->dar));
+ SET_CH_32(dw, chan->dir, chan->id, dar.msb,
+ upper_32_bits(child->dar));
+
+ /* Transfer size */
+ SET_CH_32(dw, chan->dir, chan->id, transfer_size, child->sz);
+
+ /* Interrupt setup */
+ val = GET_CH_32(dw, chan->dir, chan->id, int_setup) |
+ HDMA_V0_STOP_INT_MASK |
+ HDMA_V0_ABORT_INT_MASK |
+ HDMA_V0_LOCAL_STOP_INT_EN |
+ HDMA_V0_LOCAL_ABORT_INT_EN;
+
+ if (!(dw->chip->flags & DW_EDMA_CHIP_LOCAL)) {
+ val |= HDMA_V0_REMOTE_STOP_INT_EN |
+ HDMA_V0_REMOTE_ABORT_INT_EN;
+ }
+
+ SET_CH_32(dw, chan->dir, chan->id, int_setup, val);
+
+ /* Channel control setup */
+ val = GET_CH_32(dw, chan->dir, chan->id, control1);
+ val &= ~HDMA_V0_LINKLIST_EN;
+ SET_CH_32(dw, chan->dir, chan->id, control1, val);
+
+ /* Ring the doorbell */
+ SET_CH_32(dw, chan->dir, chan->id, doorbell,
+ HDMA_V0_DOORBELL_START);
+ }
+}
+
+static void dw_hdma_v0_core_start(struct dw_edma_chunk *chunk, bool first)
+{
+ struct dw_edma_chan *chan = chunk->chan;
+
+ if (!chan->non_ll)
+ dw_hdma_v0_core_ll_start(chunk, first);
+ else
+ dw_hdma_v0_core_non_ll_start(chunk);
+}
+
static void dw_hdma_v0_core_ch_config(struct dw_edma_chan *chan)
{
struct dw_edma *dw = chan->dw;
diff --git a/include/linux/dma/edma.h b/include/linux/dma/edma.h
index 3080747..78ce31b 100644
--- a/include/linux/dma/edma.h
+++ b/include/linux/dma/edma.h
@@ -99,6 +99,7 @@ struct dw_edma_chip {
enum dw_edma_map_format mf;
struct dw_edma *dw;
+ bool non_ll;
};
/* Export to the platform drivers */
--
1.8.3.1
^ permalink raw reply related [flat|nested] 7+ messages in thread
* Re: [PATCH v5 1/2] dmaengine: dw-edma: Add AMD MDB Endpoint Support
2025-10-28 11:28 ` [PATCH v5 1/2] dmaengine: dw-edma: Add AMD MDB Endpoint Support Devendra K Verma
@ 2025-10-28 13:28 ` Ilpo Järvinen
2025-10-29 8:37 ` Verma, Devendra
0 siblings, 1 reply; 7+ messages in thread
From: Ilpo Järvinen @ 2025-10-28 13:28 UTC (permalink / raw)
To: Devendra K Verma
Cc: bhelgaas, mani, vkoul, dmaengine, linux-pci, LKML, michal.simek
On Tue, 28 Oct 2025, Devendra K Verma wrote:
> AMD MDB PCIe endpoint support. For AMD specific support
> added the following
> - AMD supported PCIe Device IDs and Vendor ID (Xilinx).
> - AMD MDB specific driver data
> - AMD MDB specific VSEC capability to retrieve the device DDR
> base address.
>
> Signed-off-by: Devendra K Verma <devendra.verma@amd.com>
> ---
> Changes in v5:
> Added the definitions for Xilinx specific VSEC header id,
> revision, and register offsets.
> Corrected the error type when no physical offset found for
> device side memory.
> Corrected the order of variables.
>
> Changes in v4:
> Configured 8 read and 8 write channels for Xilinx vendor
> Added checks to validate vendor ID for vendor
> specific vsec id.
> Added Xilinx specific vendor id for vsec specific to Xilinx
> Added the LL and data region offsets, size as input params to
> function dw_edma_set_chan_region_offset().
> Moved the LL and data region offsets assignment to function
> for Xilinx specific case.
> Corrected comments.
>
> Changes in v3:
> Corrected a typo when assigning AMD (Xilinx) vsec id macro
> and condition check.
>
> Changes in v2:
> Reverted the devmem_phys_off type to u64.
> Renamed the function appropriately to suit the
> functionality for setting the LL & data region offsets.
>
> Changes in v1:
> Removed the pci device id from pci_ids.h file.
> Added the vendor id macro as per the suggested method.
> Changed the type of the newly added devmem_phys_off variable.
> Added to logic to assign offsets for LL and data region blocks
> in case more number of channels are enabled than given in
> amd_mdb_data struct.
> ---
> drivers/dma/dw-edma/dw-edma-pcie.c | 138 ++++++++++++++++++++++++++++++++++++-
> 1 file changed, 136 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/dma/dw-edma/dw-edma-pcie.c b/drivers/dma/dw-edma/dw-edma-pcie.c
> index 3371e0a7..7b991a0 100644
> --- a/drivers/dma/dw-edma/dw-edma-pcie.c
> +++ b/drivers/dma/dw-edma/dw-edma-pcie.c
> @@ -17,12 +17,27 @@
>
> #include "dw-edma-core.h"
>
> +/* Synopsys */
> #define DW_PCIE_VSEC_DMA_ID 0x6
> #define DW_PCIE_VSEC_DMA_BAR GENMASK(10, 8)
> #define DW_PCIE_VSEC_DMA_MAP GENMASK(2, 0)
> #define DW_PCIE_VSEC_DMA_WR_CH GENMASK(9, 0)
> #define DW_PCIE_VSEC_DMA_RD_CH GENMASK(25, 16)
>
> +/* AMD MDB (Xilinx) specific defines */
> +#define DW_PCIE_XILINX_MDB_VSEC_DMA_ID 0x6
> +#define DW_PCIE_XILINX_MDB_VSEC_ID 0x20
> +#define PCI_DEVICE_ID_AMD_MDB_B054 0xb054
> +#define DW_PCIE_AMD_MDB_INVALID_ADDR (~0ULL)
> +#define DW_PCIE_XILINX_LL_OFF_GAP 0x200000
> +#define DW_PCIE_XILINX_LL_SIZE 0x800
> +#define DW_PCIE_XILINX_DT_OFF_GAP 0x100000
> +#define DW_PCIE_XILINX_DT_SIZE 0x800
> +#define DW_PCIE_XILINX_MDB_VSEC_HDR_ID 0x20
> +#define DW_PCIE_XILINX_MDB_VSEC_REV 0x1
> +#define DW_PCIE_XILINX_MDB_DEVMEM_OFF_REG_HIGH 0xc
> +#define DW_PCIE_XILINX_MDB_DEVMEM_OFF_REG_LOW 0x8
> +
> #define DW_BLOCK(a, b, c) \
> { \
> .bar = a, \
> @@ -50,6 +65,7 @@ struct dw_edma_pcie_data {
> u8 irqs;
> u16 wr_ch_cnt;
> u16 rd_ch_cnt;
> + u64 devmem_phys_off;
> };
>
> static const struct dw_edma_pcie_data snps_edda_data = {
> @@ -90,6 +106,64 @@ struct dw_edma_pcie_data {
> .rd_ch_cnt = 2,
> };
>
> +static const struct dw_edma_pcie_data amd_mdb_data = {
> + /* MDB registers location */
> + .rg.bar = BAR_0,
> + .rg.off = 0x00001000, /* 4 Kbytes */
> + .rg.sz = 0x00002000, /* 8 Kbytes */
Please use SZ_* + check that this file #includes correct header for them.
You can then drop those comments.
--
i.
> +
> + /* Other */
> + .mf = EDMA_MF_HDMA_NATIVE,
> + .irqs = 1,
> + .wr_ch_cnt = 8,
> + .rd_ch_cnt = 8,
> +};
> +
> +static void dw_edma_set_chan_region_offset(struct dw_edma_pcie_data *pdata,
> + enum pci_barno bar, off_t start_off,
> + off_t ll_off_gap, size_t ll_size,
> + off_t dt_off_gap, size_t dt_size)
> +{
> + u16 wr_ch = pdata->wr_ch_cnt;
> + u16 rd_ch = pdata->rd_ch_cnt;
> + off_t off;
> + u16 i;
> +
> + off = start_off;
> +
> + /* Write channel LL region */
> + for (i = 0; i < wr_ch; i++) {
> + pdata->ll_wr[i].bar = bar;
> + pdata->ll_wr[i].off = off;
> + pdata->ll_wr[i].sz = ll_size;
> + off += ll_off_gap;
> + }
> +
> + /* Read channel LL region */
> + for (i = 0; i < rd_ch; i++) {
> + pdata->ll_rd[i].bar = bar;
> + pdata->ll_rd[i].off = off;
> + pdata->ll_rd[i].sz = ll_size;
> + off += ll_off_gap;
> + }
> +
> + /* Write channel data region */
> + for (i = 0; i < wr_ch; i++) {
> + pdata->dt_wr[i].bar = bar;
> + pdata->dt_wr[i].off = off;
> + pdata->dt_wr[i].sz = dt_size;
> + off += dt_off_gap;
> + }
> +
> + /* Read channel data region */
> + for (i = 0; i < rd_ch; i++) {
> + pdata->dt_rd[i].bar = bar;
> + pdata->dt_rd[i].off = off;
> + pdata->dt_rd[i].sz = dt_size;
> + off += dt_off_gap;
> + }
> +}
> +
> static int dw_edma_pcie_irq_vector(struct device *dev, unsigned int nr)
> {
> return pci_irq_vector(to_pci_dev(dev), nr);
> @@ -120,9 +194,24 @@ static void dw_edma_pcie_get_vsec_dma_data(struct pci_dev *pdev,
> u32 val, map;
> u16 vsec;
> u64 off;
> + int cap;
>
> - vsec = pci_find_vsec_capability(pdev, PCI_VENDOR_ID_SYNOPSYS,
> - DW_PCIE_VSEC_DMA_ID);
> + /*
> + * Synopsys and AMD (Xilinx) use the same VSEC ID for the purpose
> + * of map, channel counts, etc.
> + */
> + switch (pdev->vendor) {
> + case PCI_VENDOR_ID_SYNOPSYS:
> + cap = DW_PCIE_VSEC_DMA_ID;
> + break;
> + case PCI_VENDOR_ID_XILINX:
> + cap = DW_PCIE_XILINX_MDB_VSEC_DMA_ID;
> + break;
> + default:
> + return;
> + }
> +
> + vsec = pci_find_vsec_capability(pdev, pdev->vendor, cap);
> if (!vsec)
> return;
>
> @@ -155,6 +244,28 @@ static void dw_edma_pcie_get_vsec_dma_data(struct pci_dev *pdev,
> off <<= 32;
> off |= val;
> pdata->rg.off = off;
> +
> + /* Xilinx specific VSEC capability */
> + vsec = pci_find_vsec_capability(pdev, PCI_VENDOR_ID_XILINX,
> + DW_PCIE_XILINX_MDB_VSEC_ID);
> + if (!vsec)
> + return;
> +
> + pci_read_config_dword(pdev, vsec + PCI_VNDR_HEADER, &val);
> + if (PCI_VNDR_HEADER_ID(val) != DW_PCIE_XILINX_MDB_VSEC_HDR_ID ||
> + PCI_VNDR_HEADER_REV(val) != DW_PCIE_XILINX_MDB_VSEC_REV)
> + return;
> +
> + pci_read_config_dword(pdev,
> + vsec + DW_PCIE_XILINX_MDB_DEVMEM_OFF_REG_HIGH,
> + &val);
> + off = val;
> + pci_read_config_dword(pdev,
> + vsec + DW_PCIE_XILINX_MDB_DEVMEM_OFF_REG_LOW,
> + &val);
> + off <<= 32;
> + off |= val;
> + pdata->devmem_phys_off = off;
> }
>
> static int dw_edma_pcie_probe(struct pci_dev *pdev,
> @@ -179,6 +290,7 @@ static int dw_edma_pcie_probe(struct pci_dev *pdev,
> }
>
> memcpy(vsec_data, pdata, sizeof(struct dw_edma_pcie_data));
> + vsec_data->devmem_phys_off = DW_PCIE_AMD_MDB_INVALID_ADDR;
>
> /*
> * Tries to find if exists a PCIe Vendor-Specific Extended Capability
> @@ -186,6 +298,26 @@ static int dw_edma_pcie_probe(struct pci_dev *pdev,
> */
> dw_edma_pcie_get_vsec_dma_data(pdev, vsec_data);
>
> + if (pdev->vendor == PCI_VENDOR_ID_XILINX) {
> + /*
> + * There is no valid address found for the LL memory
> + * space on the device side.
> + */
> + if (vsec_data->devmem_phys_off == DW_PCIE_AMD_MDB_INVALID_ADDR)
> + return -ENOMEM;
> +
> + /*
> + * Configure the channel LL and data blocks if number of
> + * channels enabled in VSEC capability are more than the
> + * channels configured in amd_mdb_data.
> + */
> + dw_edma_set_chan_region_offset(vsec_data, BAR_2, 0,
> + DW_PCIE_XILINX_LL_OFF_GAP,
> + DW_PCIE_XILINX_LL_SIZE,
> + DW_PCIE_XILINX_DT_OFF_GAP,
> + DW_PCIE_XILINX_DT_SIZE);
> + }
> +
> /* Mapping PCI BAR regions */
> mask = BIT(vsec_data->rg.bar);
> for (i = 0; i < vsec_data->wr_ch_cnt; i++) {
> @@ -367,6 +499,8 @@ static void dw_edma_pcie_remove(struct pci_dev *pdev)
>
> static const struct pci_device_id dw_edma_pcie_id_table[] = {
> { PCI_DEVICE_DATA(SYNOPSYS, EDDA, &snps_edda_data) },
> + { PCI_VDEVICE(XILINX, PCI_DEVICE_ID_AMD_MDB_B054),
> + (kernel_ulong_t)&amd_mdb_data },
> { }
> };
> MODULE_DEVICE_TABLE(pci, dw_edma_pcie_id_table);
>
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [PATCH v5 2/2] dmaengine: dw-edma: Add non-LL mode
2025-10-28 11:28 ` [PATCH v5 2/2] dmaengine: dw-edma: Add non-LL mode Devendra K Verma
@ 2025-10-28 13:32 ` Ilpo Järvinen
2025-10-29 8:35 ` Verma, Devendra
0 siblings, 1 reply; 7+ messages in thread
From: Ilpo Järvinen @ 2025-10-28 13:32 UTC (permalink / raw)
To: Devendra K Verma
Cc: bhelgaas, mani, vkoul, dmaengine, linux-pci, LKML, michal.simek
On Tue, 28 Oct 2025, Devendra K Verma wrote:
> AMD MDB IP supports Linked List (LL) mode as well as non-LL mode.
> The current code does not have the mechanisms to enable the
> DMA transactions using the non-LL mode. The following two cases
> are added with this patch:
> - When a valid physical base address is not configured via the
> Xilinx VSEC capability then the IP can still be used in non-LL
> mode. The default mode for all the DMA transactions and for all
> the DMA channels then is non-LL mode.
> - When a valid physical base address is configured but the client
> wants to use the non-LL mode for DMA transactions then also the
> flexibility is provided via the peripheral_config struct member of
> dma_slave_config. In this case the channels can be individually
> configured in non-LL mode. This use case is desirable for single
> DMA transfer of a chunk, this saves the effort of preparing the
> Link List. This particular scenario is applicable to AMD as well
> as Synopsys IP.
>
> Signed-off-by: Devendra K Verma <devendra.verma@amd.com>
> ---
> Changes in v5
> Variable name 'nollp' changed to 'non_ll'.
> In the dw_edma_device_config() WARN_ON replaced with dev_err().
> Comments follow the 80-column guideline.
>
> Changes in v4
> No change
>
> Changes in v3
> No change
>
> Changes in v2
> Reverted the function return type to u64 for
> dw_edma_get_phys_addr().
>
> Changes in v1
> Changed the function return type for dw_edma_get_phys_addr().
> Corrected the typo raised in review.
> ---
> drivers/dma/dw-edma/dw-edma-core.c | 41 ++++++++++++++++++++---
> drivers/dma/dw-edma/dw-edma-core.h | 1 +
> drivers/dma/dw-edma/dw-edma-pcie.c | 44 +++++++++++++++++--------
> drivers/dma/dw-edma/dw-hdma-v0-core.c | 62 ++++++++++++++++++++++++++++++++++-
> include/linux/dma/edma.h | 1 +
> 5 files changed, 130 insertions(+), 19 deletions(-)
>
> diff --git a/drivers/dma/dw-edma/dw-edma-core.c b/drivers/dma/dw-edma/dw-edma-core.c
> index b43255f..60a3279 100644
> --- a/drivers/dma/dw-edma/dw-edma-core.c
> +++ b/drivers/dma/dw-edma/dw-edma-core.c
> @@ -223,8 +223,31 @@ static int dw_edma_device_config(struct dma_chan *dchan,
> struct dma_slave_config *config)
> {
> struct dw_edma_chan *chan = dchan2dw_edma_chan(dchan);
> + int non_ll = 0;
> +
> + if (config->peripheral_config &&
> + config->peripheral_size != sizeof(int)) {
> + dev_err(dchan->device->dev,
> + "config param peripheral size mismatch\n");
> + return -EINVAL;
> + }
>
> memcpy(&chan->config, config, sizeof(*config));
> +
> + /*
> + * When there is no valid LLP base address available then the default
> + * DMA ops will use the non-LL mode.
> + * Cases where LL mode is enabled and client wants to use the non-LL
> + * mode then also client can do so via providing the peripheral_config
> + * param.
> + */
> + if (config->peripheral_config)
> + non_ll = *(int *)config->peripheral_config;
> +
> + chan->non_ll = false;
> + if (chan->dw->chip->non_ll || (!chan->dw->chip->non_ll && non_ll))
> + chan->non_ll = true;
> +
> chan->configured = true;
>
> return 0;
> @@ -353,7 +376,7 @@ static void dw_edma_device_issue_pending(struct dma_chan *dchan)
> struct dw_edma_chan *chan = dchan2dw_edma_chan(xfer->dchan);
> enum dma_transfer_direction dir = xfer->direction;
> struct scatterlist *sg = NULL;
> - struct dw_edma_chunk *chunk;
> + struct dw_edma_chunk *chunk = NULL;
> struct dw_edma_burst *burst;
> struct dw_edma_desc *desc;
> u64 src_addr, dst_addr;
> @@ -419,9 +442,11 @@ static void dw_edma_device_issue_pending(struct dma_chan *dchan)
> if (unlikely(!desc))
> goto err_alloc;
>
> - chunk = dw_edma_alloc_chunk(desc);
> - if (unlikely(!chunk))
> - goto err_alloc;
> + if (!chan->non_ll) {
> + chunk = dw_edma_alloc_chunk(desc);
> + if (unlikely(!chunk))
> + goto err_alloc;
> + }
>
> if (xfer->type == EDMA_XFER_INTERLEAVED) {
> src_addr = xfer->xfer.il->src_start;
> @@ -450,7 +475,13 @@ static void dw_edma_device_issue_pending(struct dma_chan *dchan)
> if (xfer->type == EDMA_XFER_SCATTER_GATHER && !sg)
> break;
>
> - if (chunk->bursts_alloc == chan->ll_max) {
> + /*
> + * For non-LL mode, only a single burst can be handled
> + * in a single chunk unlike LL mode where multiple bursts
> + * can be configured in a single chunk.
> + */
> + if ((chunk && chunk->bursts_alloc == chan->ll_max) ||
> + chan->non_ll) {
> chunk = dw_edma_alloc_chunk(desc);
> if (unlikely(!chunk))
> goto err_alloc;
> diff --git a/drivers/dma/dw-edma/dw-edma-core.h b/drivers/dma/dw-edma/dw-edma-core.h
> index 71894b9..c8e3d19 100644
> --- a/drivers/dma/dw-edma/dw-edma-core.h
> +++ b/drivers/dma/dw-edma/dw-edma-core.h
> @@ -86,6 +86,7 @@ struct dw_edma_chan {
> u8 configured;
>
> struct dma_slave_config config;
> + bool non_ll;
> };
>
> struct dw_edma_irq {
> diff --git a/drivers/dma/dw-edma/dw-edma-pcie.c b/drivers/dma/dw-edma/dw-edma-pcie.c
> index 7b991a0..b02977c 100644
> --- a/drivers/dma/dw-edma/dw-edma-pcie.c
> +++ b/drivers/dma/dw-edma/dw-edma-pcie.c
> @@ -268,6 +268,15 @@ static void dw_edma_pcie_get_vsec_dma_data(struct pci_dev *pdev,
> pdata->devmem_phys_off = off;
> }
>
> +static u64 dw_edma_get_phys_addr(struct pci_dev *pdev,
> + struct dw_edma_pcie_data *pdata,
> + enum pci_barno bar)
> +{
> + if (pdev->vendor == PCI_VENDOR_ID_XILINX)
> + return pdata->devmem_phys_off;
> + return pci_bus_address(pdev, bar);
> +}
> +
> static int dw_edma_pcie_probe(struct pci_dev *pdev,
> const struct pci_device_id *pid)
> {
> @@ -277,6 +286,7 @@ static int dw_edma_pcie_probe(struct pci_dev *pdev,
> struct dw_edma_chip *chip;
> int err, nr_irqs;
> int i, mask;
> + bool non_ll = false;
>
> vsec_data = kmalloc(sizeof(*vsec_data), GFP_KERNEL);
> if (!vsec_data)
> @@ -301,21 +311,24 @@ static int dw_edma_pcie_probe(struct pci_dev *pdev,
> if (pdev->vendor == PCI_VENDOR_ID_XILINX) {
> /*
> * There is no valid address found for the LL memory
> - * space on the device side.
> + * space on the device side. In the absence of LL base
> + * address use the non-LL mode or simple mode supported by
> + * the HDMA IP.
> */
> if (vsec_data->devmem_phys_off == DW_PCIE_AMD_MDB_INVALID_ADDR)
> - return -ENOMEM;
> + non_ll = true;
>
> /*
> * Configure the channel LL and data blocks if number of
> * channels enabled in VSEC capability are more than the
> * channels configured in amd_mdb_data.
> */
> - dw_edma_set_chan_region_offset(vsec_data, BAR_2, 0,
> - DW_PCIE_XILINX_LL_OFF_GAP,
> - DW_PCIE_XILINX_LL_SIZE,
> - DW_PCIE_XILINX_DT_OFF_GAP,
> - DW_PCIE_XILINX_DT_SIZE);
> + if (!non_ll)
> + dw_edma_set_chan_region_offset(vsec_data, BAR_2, 0,
> + DW_PCIE_XILINX_LL_OFF_GAP,
> + DW_PCIE_XILINX_LL_SIZE,
> + DW_PCIE_XILINX_DT_OFF_GAP,
> + DW_PCIE_XILINX_DT_SIZE);
> }
>
> /* Mapping PCI BAR regions */
> @@ -363,6 +376,7 @@ static int dw_edma_pcie_probe(struct pci_dev *pdev,
> chip->mf = vsec_data->mf;
> chip->nr_irqs = nr_irqs;
> chip->ops = &dw_edma_pcie_plat_ops;
> + chip->non_ll = non_ll;
>
> chip->ll_wr_cnt = vsec_data->wr_ch_cnt;
> chip->ll_rd_cnt = vsec_data->rd_ch_cnt;
> @@ -371,7 +385,7 @@ static int dw_edma_pcie_probe(struct pci_dev *pdev,
> if (!chip->reg_base)
> return -ENOMEM;
>
> - for (i = 0; i < chip->ll_wr_cnt; i++) {
> + for (i = 0; i < chip->ll_wr_cnt && !non_ll; i++) {
> struct dw_edma_region *ll_region = &chip->ll_region_wr[i];
> struct dw_edma_region *dt_region = &chip->dt_region_wr[i];
> struct dw_edma_block *ll_block = &vsec_data->ll_wr[i];
> @@ -382,7 +396,8 @@ static int dw_edma_pcie_probe(struct pci_dev *pdev,
> return -ENOMEM;
>
> ll_region->vaddr.io += ll_block->off;
> - ll_region->paddr = pci_bus_address(pdev, ll_block->bar);
> + ll_region->paddr = dw_edma_get_phys_addr(pdev, vsec_data,
> + ll_block->bar);
> ll_region->paddr += ll_block->off;
> ll_region->sz = ll_block->sz;
>
> @@ -391,12 +406,13 @@ static int dw_edma_pcie_probe(struct pci_dev *pdev,
> return -ENOMEM;
>
> dt_region->vaddr.io += dt_block->off;
> - dt_region->paddr = pci_bus_address(pdev, dt_block->bar);
> + dt_region->paddr = dw_edma_get_phys_addr(pdev, vsec_data,
> + dt_block->bar);
> dt_region->paddr += dt_block->off;
> dt_region->sz = dt_block->sz;
> }
>
> - for (i = 0; i < chip->ll_rd_cnt; i++) {
> + for (i = 0; i < chip->ll_rd_cnt && !non_ll; i++) {
> struct dw_edma_region *ll_region = &chip->ll_region_rd[i];
> struct dw_edma_region *dt_region = &chip->dt_region_rd[i];
> struct dw_edma_block *ll_block = &vsec_data->ll_rd[i];
> @@ -407,7 +423,8 @@ static int dw_edma_pcie_probe(struct pci_dev *pdev,
> return -ENOMEM;
>
> ll_region->vaddr.io += ll_block->off;
> - ll_region->paddr = pci_bus_address(pdev, ll_block->bar);
> + ll_region->paddr = dw_edma_get_phys_addr(pdev, vsec_data,
> + ll_block->bar);
> ll_region->paddr += ll_block->off;
> ll_region->sz = ll_block->sz;
>
> @@ -416,7 +433,8 @@ static int dw_edma_pcie_probe(struct pci_dev *pdev,
> return -ENOMEM;
>
> dt_region->vaddr.io += dt_block->off;
> - dt_region->paddr = pci_bus_address(pdev, dt_block->bar);
> + dt_region->paddr = dw_edma_get_phys_addr(pdev, vsec_data,
> + dt_block->bar);
> dt_region->paddr += dt_block->off;
> dt_region->sz = dt_block->sz;
> }
> diff --git a/drivers/dma/dw-edma/dw-hdma-v0-core.c b/drivers/dma/dw-edma/dw-hdma-v0-core.c
> index e3f8db4..47ecc84 100644
> --- a/drivers/dma/dw-edma/dw-hdma-v0-core.c
> +++ b/drivers/dma/dw-edma/dw-hdma-v0-core.c
> @@ -225,7 +225,7 @@ static void dw_hdma_v0_sync_ll_data(struct dw_edma_chunk *chunk)
> readl(chunk->ll_region.vaddr.io);
> }
>
> -static void dw_hdma_v0_core_start(struct dw_edma_chunk *chunk, bool first)
> +static void dw_hdma_v0_core_ll_start(struct dw_edma_chunk *chunk, bool first)
> {
> struct dw_edma_chan *chan = chunk->chan;
> struct dw_edma *dw = chan->dw;
> @@ -263,6 +263,66 @@ static void dw_hdma_v0_core_start(struct dw_edma_chunk *chunk, bool first)
> SET_CH_32(dw, chan->dir, chan->id, doorbell, HDMA_V0_DOORBELL_START);
> }
>
> +static void dw_hdma_v0_core_non_ll_start(struct dw_edma_chunk *chunk)
> +{
> + struct dw_edma_chan *chan = chunk->chan;
> + struct dw_edma *dw = chan->dw;
> + struct dw_edma_burst *child;
> + u32 val;
> +
> + list_for_each_entry(child, &chunk->burst->list, list) {
> + SET_CH_32(dw, chan->dir, chan->id, ch_en, BIT(0));
Please name BIT(0) with a define.
> +
> + /* Source address */
> + SET_CH_32(dw, chan->dir, chan->id, sar.lsb,
> + lower_32_bits(child->sar));
> + SET_CH_32(dw, chan->dir, chan->id, sar.msb,
> + upper_32_bits(child->sar));
> +
> + /* Destination address */
> + SET_CH_32(dw, chan->dir, chan->id, dar.lsb,
> + lower_32_bits(child->dar));
> + SET_CH_32(dw, chan->dir, chan->id, dar.msb,
> + upper_32_bits(child->dar));
> +
> + /* Transfer size */
> + SET_CH_32(dw, chan->dir, chan->id, transfer_size, child->sz);
> +
> + /* Interrupt setup */
> + val = GET_CH_32(dw, chan->dir, chan->id, int_setup) |
> + HDMA_V0_STOP_INT_MASK |
> + HDMA_V0_ABORT_INT_MASK |
> + HDMA_V0_LOCAL_STOP_INT_EN |
> + HDMA_V0_LOCAL_ABORT_INT_EN;
> +
> + if (!(dw->chip->flags & DW_EDMA_CHIP_LOCAL)) {
> + val |= HDMA_V0_REMOTE_STOP_INT_EN |
> + HDMA_V0_REMOTE_ABORT_INT_EN;
> + }
> +
> + SET_CH_32(dw, chan->dir, chan->id, int_setup, val);
> +
> + /* Channel control setup */
> + val = GET_CH_32(dw, chan->dir, chan->id, control1);
> + val &= ~HDMA_V0_LINKLIST_EN;
> + SET_CH_32(dw, chan->dir, chan->id, control1, val);
> +
> + /* Ring the doorbell */
> + SET_CH_32(dw, chan->dir, chan->id, doorbell,
> + HDMA_V0_DOORBELL_START);
Doesn't the code explain itself already, why you need to have that comment
above it, it doesn't seem to add any real value?
> + }
> +}
> +
> +static void dw_hdma_v0_core_start(struct dw_edma_chunk *chunk, bool first)
> +{
> + struct dw_edma_chan *chan = chunk->chan;
> +
> + if (!chan->non_ll)
> + dw_hdma_v0_core_ll_start(chunk, first);
> + else
> + dw_hdma_v0_core_non_ll_start(chunk);
> +}
> +
> static void dw_hdma_v0_core_ch_config(struct dw_edma_chan *chan)
> {
> struct dw_edma *dw = chan->dw;
> diff --git a/include/linux/dma/edma.h b/include/linux/dma/edma.h
> index 3080747..78ce31b 100644
> --- a/include/linux/dma/edma.h
> +++ b/include/linux/dma/edma.h
> @@ -99,6 +99,7 @@ struct dw_edma_chip {
> enum dw_edma_map_format mf;
>
> struct dw_edma *dw;
> + bool non_ll;
> };
>
> /* Export to the platform drivers */
>
--
i.
^ permalink raw reply [flat|nested] 7+ messages in thread
* RE: [PATCH v5 2/2] dmaengine: dw-edma: Add non-LL mode
2025-10-28 13:32 ` Ilpo Järvinen
@ 2025-10-29 8:35 ` Verma, Devendra
0 siblings, 0 replies; 7+ messages in thread
From: Verma, Devendra @ 2025-10-29 8:35 UTC (permalink / raw)
To: Ilpo Järvinen
Cc: bhelgaas@google.com, mani@kernel.org, vkoul@kernel.org,
dmaengine@vger.kernel.org, linux-pci@vger.kernel.org, LKML,
Simek, Michal, Verma, Devendra
[AMD Official Use Only - AMD Internal Distribution Only]
Thank you @Ilpo Järvinen for the inputs.
Please check my response inline.
Regards,
Devendra
> -----Original Message-----
> From: Ilpo Järvinen <ilpo.jarvinen@linux.intel.com>
> Sent: Tuesday, October 28, 2025 7:03 PM
> To: Verma, Devendra <Devendra.Verma@amd.com>
> Cc: bhelgaas@google.com; mani@kernel.org; vkoul@kernel.org;
> dmaengine@vger.kernel.org; linux-pci@vger.kernel.org; LKML <linux-
> kernel@vger.kernel.org>; Simek, Michal <michal.simek@amd.com>
> Subject: Re: [PATCH v5 2/2] dmaengine: dw-edma: Add non-LL mode
>
> Caution: This message originated from an External Source. Use proper
> caution when opening attachments, clicking links, or responding.
>
>
> On Tue, 28 Oct 2025, Devendra K Verma wrote:
>
> > AMD MDB IP supports Linked List (LL) mode as well as non-LL mode.
> > The current code does not have the mechanisms to enable the DMA
> > transactions using the non-LL mode. The following two cases are added
> > with this patch:
> > - When a valid physical base address is not configured via the
> > Xilinx VSEC capability then the IP can still be used in non-LL
> > mode. The default mode for all the DMA transactions and for all
> > the DMA channels then is non-LL mode.
> > - When a valid physical base address is configured but the client
> > wants to use the non-LL mode for DMA transactions then also the
> > flexibility is provided via the peripheral_config struct member of
> > dma_slave_config. In this case the channels can be individually
> > configured in non-LL mode. This use case is desirable for single
> > DMA transfer of a chunk, this saves the effort of preparing the
> > Link List. This particular scenario is applicable to AMD as well
> > as Synopsys IP.
> >
> > Signed-off-by: Devendra K Verma <devendra.verma@amd.com>
> > ---
> > Changes in v5
> > Variable name 'nollp' changed to 'non_ll'.
> > In the dw_edma_device_config() WARN_ON replaced with dev_err().
> > Comments follow the 80-column guideline.
> >
> > Changes in v4
> > No change
> >
> > Changes in v3
> > No change
> >
> > Changes in v2
> > Reverted the function return type to u64 for
> > dw_edma_get_phys_addr().
> >
> > Changes in v1
> > Changed the function return type for dw_edma_get_phys_addr().
> > Corrected the typo raised in review.
> > ---
> > drivers/dma/dw-edma/dw-edma-core.c | 41 ++++++++++++++++++++---
> > drivers/dma/dw-edma/dw-edma-core.h | 1 +
> > drivers/dma/dw-edma/dw-edma-pcie.c | 44 +++++++++++++++++--------
> > drivers/dma/dw-edma/dw-hdma-v0-core.c | 62
> ++++++++++++++++++++++++++++++++++-
> > include/linux/dma/edma.h | 1 +
> > 5 files changed, 130 insertions(+), 19 deletions(-)
> >
> > diff --git a/drivers/dma/dw-edma/dw-edma-core.c
> > b/drivers/dma/dw-edma/dw-edma-core.c
> > index b43255f..60a3279 100644
> > --- a/drivers/dma/dw-edma/dw-edma-core.c
> > +++ b/drivers/dma/dw-edma/dw-edma-core.c
> > @@ -223,8 +223,31 @@ static int dw_edma_device_config(struct
> dma_chan *dchan,
> > struct dma_slave_config *config) {
> > struct dw_edma_chan *chan = dchan2dw_edma_chan(dchan);
> > + int non_ll = 0;
> > +
> > + if (config->peripheral_config &&
> > + config->peripheral_size != sizeof(int)) {
> > + dev_err(dchan->device->dev,
> > + "config param peripheral size mismatch\n");
> > + return -EINVAL;
> > + }
> >
> > memcpy(&chan->config, config, sizeof(*config));
> > +
> > + /*
> > + * When there is no valid LLP base address available then the default
> > + * DMA ops will use the non-LL mode.
> > + * Cases where LL mode is enabled and client wants to use the non-LL
> > + * mode then also client can do so via providing the peripheral_config
> > + * param.
> > + */
> > + if (config->peripheral_config)
> > + non_ll = *(int *)config->peripheral_config;
> > +
> > + chan->non_ll = false;
> > + if (chan->dw->chip->non_ll || (!chan->dw->chip->non_ll && non_ll))
> > + chan->non_ll = true;
> > +
> > chan->configured = true;
> >
> > return 0;
> > @@ -353,7 +376,7 @@ static void dw_edma_device_issue_pending(struct
> dma_chan *dchan)
> > struct dw_edma_chan *chan = dchan2dw_edma_chan(xfer->dchan);
> > enum dma_transfer_direction dir = xfer->direction;
> > struct scatterlist *sg = NULL;
> > - struct dw_edma_chunk *chunk;
> > + struct dw_edma_chunk *chunk = NULL;
> > struct dw_edma_burst *burst;
> > struct dw_edma_desc *desc;
> > u64 src_addr, dst_addr;
> > @@ -419,9 +442,11 @@ static void dw_edma_device_issue_pending(struct
> dma_chan *dchan)
> > if (unlikely(!desc))
> > goto err_alloc;
> >
> > - chunk = dw_edma_alloc_chunk(desc);
> > - if (unlikely(!chunk))
> > - goto err_alloc;
> > + if (!chan->non_ll) {
> > + chunk = dw_edma_alloc_chunk(desc);
> > + if (unlikely(!chunk))
> > + goto err_alloc;
> > + }
> >
> > if (xfer->type == EDMA_XFER_INTERLEAVED) {
> > src_addr = xfer->xfer.il->src_start; @@ -450,7 +475,13
> > @@ static void dw_edma_device_issue_pending(struct dma_chan *dchan)
> > if (xfer->type == EDMA_XFER_SCATTER_GATHER && !sg)
> > break;
> >
> > - if (chunk->bursts_alloc == chan->ll_max) {
> > + /*
> > + * For non-LL mode, only a single burst can be handled
> > + * in a single chunk unlike LL mode where multiple bursts
> > + * can be configured in a single chunk.
> > + */
> > + if ((chunk && chunk->bursts_alloc == chan->ll_max) ||
> > + chan->non_ll) {
> > chunk = dw_edma_alloc_chunk(desc);
> > if (unlikely(!chunk))
> > goto err_alloc; diff --git
> > a/drivers/dma/dw-edma/dw-edma-core.h
> > b/drivers/dma/dw-edma/dw-edma-core.h
> > index 71894b9..c8e3d19 100644
> > --- a/drivers/dma/dw-edma/dw-edma-core.h
> > +++ b/drivers/dma/dw-edma/dw-edma-core.h
> > @@ -86,6 +86,7 @@ struct dw_edma_chan {
> > u8 configured;
> >
> > struct dma_slave_config config;
> > + bool non_ll;
> > };
> >
> > struct dw_edma_irq {
> > diff --git a/drivers/dma/dw-edma/dw-edma-pcie.c
> > b/drivers/dma/dw-edma/dw-edma-pcie.c
> > index 7b991a0..b02977c 100644
> > --- a/drivers/dma/dw-edma/dw-edma-pcie.c
> > +++ b/drivers/dma/dw-edma/dw-edma-pcie.c
> > @@ -268,6 +268,15 @@ static void
> dw_edma_pcie_get_vsec_dma_data(struct pci_dev *pdev,
> > pdata->devmem_phys_off = off;
> > }
> >
> > +static u64 dw_edma_get_phys_addr(struct pci_dev *pdev,
> > + struct dw_edma_pcie_data *pdata,
> > + enum pci_barno bar) {
> > + if (pdev->vendor == PCI_VENDOR_ID_XILINX)
> > + return pdata->devmem_phys_off;
> > + return pci_bus_address(pdev, bar); }
> > +
> > static int dw_edma_pcie_probe(struct pci_dev *pdev,
> > const struct pci_device_id *pid) { @@
> > -277,6 +286,7 @@ static int dw_edma_pcie_probe(struct pci_dev *pdev,
> > struct dw_edma_chip *chip;
> > int err, nr_irqs;
> > int i, mask;
> > + bool non_ll = false;
> >
> > vsec_data = kmalloc(sizeof(*vsec_data), GFP_KERNEL);
> > if (!vsec_data)
> > @@ -301,21 +311,24 @@ static int dw_edma_pcie_probe(struct pci_dev
> *pdev,
> > if (pdev->vendor == PCI_VENDOR_ID_XILINX) {
> > /*
> > * There is no valid address found for the LL memory
> > - * space on the device side.
> > + * space on the device side. In the absence of LL base
> > + * address use the non-LL mode or simple mode supported by
> > + * the HDMA IP.
> > */
> > if (vsec_data->devmem_phys_off ==
> DW_PCIE_AMD_MDB_INVALID_ADDR)
> > - return -ENOMEM;
> > + non_ll = true;
> >
> > /*
> > * Configure the channel LL and data blocks if number of
> > * channels enabled in VSEC capability are more than the
> > * channels configured in amd_mdb_data.
> > */
> > - dw_edma_set_chan_region_offset(vsec_data, BAR_2, 0,
> > - DW_PCIE_XILINX_LL_OFF_GAP,
> > - DW_PCIE_XILINX_LL_SIZE,
> > - DW_PCIE_XILINX_DT_OFF_GAP,
> > - DW_PCIE_XILINX_DT_SIZE);
> > + if (!non_ll)
> > + dw_edma_set_chan_region_offset(vsec_data, BAR_2, 0,
> > + DW_PCIE_XILINX_LL_OFF_GAP,
> > + DW_PCIE_XILINX_LL_SIZE,
> > + DW_PCIE_XILINX_DT_OFF_GAP,
> > +
> > + DW_PCIE_XILINX_DT_SIZE);
> > }
> >
> > /* Mapping PCI BAR regions */
> > @@ -363,6 +376,7 @@ static int dw_edma_pcie_probe(struct pci_dev
> *pdev,
> > chip->mf = vsec_data->mf;
> > chip->nr_irqs = nr_irqs;
> > chip->ops = &dw_edma_pcie_plat_ops;
> > + chip->non_ll = non_ll;
> >
> > chip->ll_wr_cnt = vsec_data->wr_ch_cnt;
> > chip->ll_rd_cnt = vsec_data->rd_ch_cnt; @@ -371,7 +385,7 @@
> > static int dw_edma_pcie_probe(struct pci_dev *pdev,
> > if (!chip->reg_base)
> > return -ENOMEM;
> >
> > - for (i = 0; i < chip->ll_wr_cnt; i++) {
> > + for (i = 0; i < chip->ll_wr_cnt && !non_ll; i++) {
> > struct dw_edma_region *ll_region = &chip->ll_region_wr[i];
> > struct dw_edma_region *dt_region = &chip->dt_region_wr[i];
> > struct dw_edma_block *ll_block = &vsec_data->ll_wr[i];
> > @@ -382,7 +396,8 @@ static int dw_edma_pcie_probe(struct pci_dev
> *pdev,
> > return -ENOMEM;
> >
> > ll_region->vaddr.io += ll_block->off;
> > - ll_region->paddr = pci_bus_address(pdev, ll_block->bar);
> > + ll_region->paddr = dw_edma_get_phys_addr(pdev, vsec_data,
> > + ll_block->bar);
> > ll_region->paddr += ll_block->off;
> > ll_region->sz = ll_block->sz;
> >
> > @@ -391,12 +406,13 @@ static int dw_edma_pcie_probe(struct pci_dev
> *pdev,
> > return -ENOMEM;
> >
> > dt_region->vaddr.io += dt_block->off;
> > - dt_region->paddr = pci_bus_address(pdev, dt_block->bar);
> > + dt_region->paddr = dw_edma_get_phys_addr(pdev, vsec_data,
> > + dt_block->bar);
> > dt_region->paddr += dt_block->off;
> > dt_region->sz = dt_block->sz;
> > }
> >
> > - for (i = 0; i < chip->ll_rd_cnt; i++) {
> > + for (i = 0; i < chip->ll_rd_cnt && !non_ll; i++) {
> > struct dw_edma_region *ll_region = &chip->ll_region_rd[i];
> > struct dw_edma_region *dt_region = &chip->dt_region_rd[i];
> > struct dw_edma_block *ll_block = &vsec_data->ll_rd[i];
> > @@ -407,7 +423,8 @@ static int dw_edma_pcie_probe(struct pci_dev
> *pdev,
> > return -ENOMEM;
> >
> > ll_region->vaddr.io += ll_block->off;
> > - ll_region->paddr = pci_bus_address(pdev, ll_block->bar);
> > + ll_region->paddr = dw_edma_get_phys_addr(pdev, vsec_data,
> > + ll_block->bar);
> > ll_region->paddr += ll_block->off;
> > ll_region->sz = ll_block->sz;
> >
> > @@ -416,7 +433,8 @@ static int dw_edma_pcie_probe(struct pci_dev
> *pdev,
> > return -ENOMEM;
> >
> > dt_region->vaddr.io += dt_block->off;
> > - dt_region->paddr = pci_bus_address(pdev, dt_block->bar);
> > + dt_region->paddr = dw_edma_get_phys_addr(pdev, vsec_data,
> > + dt_block->bar);
> > dt_region->paddr += dt_block->off;
> > dt_region->sz = dt_block->sz;
> > }
> > diff --git a/drivers/dma/dw-edma/dw-hdma-v0-core.c
> > b/drivers/dma/dw-edma/dw-hdma-v0-core.c
> > index e3f8db4..47ecc84 100644
> > --- a/drivers/dma/dw-edma/dw-hdma-v0-core.c
> > +++ b/drivers/dma/dw-edma/dw-hdma-v0-core.c
> > @@ -225,7 +225,7 @@ static void dw_hdma_v0_sync_ll_data(struct
> dw_edma_chunk *chunk)
> > readl(chunk->ll_region.vaddr.io); }
> >
> > -static void dw_hdma_v0_core_start(struct dw_edma_chunk *chunk, bool
> > first)
> > +static void dw_hdma_v0_core_ll_start(struct dw_edma_chunk *chunk,
> > +bool first)
> > {
> > struct dw_edma_chan *chan = chunk->chan;
> > struct dw_edma *dw = chan->dw;
> > @@ -263,6 +263,66 @@ static void dw_hdma_v0_core_start(struct
> dw_edma_chunk *chunk, bool first)
> > SET_CH_32(dw, chan->dir, chan->id, doorbell,
> > HDMA_V0_DOORBELL_START); }
> >
> > +static void dw_hdma_v0_core_non_ll_start(struct dw_edma_chunk
> *chunk)
> > +{
> > + struct dw_edma_chan *chan = chunk->chan;
> > + struct dw_edma *dw = chan->dw;
> > + struct dw_edma_burst *child;
> > + u32 val;
> > +
> > + list_for_each_entry(child, &chunk->burst->list, list) {
> > + SET_CH_32(dw, chan->dir, chan->id, ch_en, BIT(0));
>
> Please name BIT(0) with a define.
>
Sure, defined a macro HDMA_V0_CH_EN for clarity.
> > +
> > + /* Source address */
> > + SET_CH_32(dw, chan->dir, chan->id, sar.lsb,
> > + lower_32_bits(child->sar));
> > + SET_CH_32(dw, chan->dir, chan->id, sar.msb,
> > + upper_32_bits(child->sar));
> > +
> > + /* Destination address */
> > + SET_CH_32(dw, chan->dir, chan->id, dar.lsb,
> > + lower_32_bits(child->dar));
> > + SET_CH_32(dw, chan->dir, chan->id, dar.msb,
> > + upper_32_bits(child->dar));
> > +
> > + /* Transfer size */
> > + SET_CH_32(dw, chan->dir, chan->id, transfer_size,
> > + child->sz);
> > +
> > + /* Interrupt setup */
> > + val = GET_CH_32(dw, chan->dir, chan->id, int_setup) |
> > + HDMA_V0_STOP_INT_MASK |
> > + HDMA_V0_ABORT_INT_MASK |
> > + HDMA_V0_LOCAL_STOP_INT_EN |
> > + HDMA_V0_LOCAL_ABORT_INT_EN;
> > +
> > + if (!(dw->chip->flags & DW_EDMA_CHIP_LOCAL)) {
> > + val |= HDMA_V0_REMOTE_STOP_INT_EN |
> > + HDMA_V0_REMOTE_ABORT_INT_EN;
> > + }
> > +
> > + SET_CH_32(dw, chan->dir, chan->id, int_setup, val);
> > +
> > + /* Channel control setup */
> > + val = GET_CH_32(dw, chan->dir, chan->id, control1);
> > + val &= ~HDMA_V0_LINKLIST_EN;
> > + SET_CH_32(dw, chan->dir, chan->id, control1, val);
> > +
> > + /* Ring the doorbell */
> > + SET_CH_32(dw, chan->dir, chan->id, doorbell,
> > + HDMA_V0_DOORBELL_START);
>
> Doesn't the code explain itself already, why you need to have that comment
> above it, it doesn't seem to add any real value?
>
I followed the existing comment pattern and applied the same here.
I will remove this comment as pointed out it is self-explanatory.
> > + }
> > +}
> > +
> > +static void dw_hdma_v0_core_start(struct dw_edma_chunk *chunk, bool
> > +first) {
> > + struct dw_edma_chan *chan = chunk->chan;
> > +
> > + if (!chan->non_ll)
> > + dw_hdma_v0_core_ll_start(chunk, first);
> > + else
> > + dw_hdma_v0_core_non_ll_start(chunk);
> > +}
> > +
> > static void dw_hdma_v0_core_ch_config(struct dw_edma_chan *chan) {
> > struct dw_edma *dw = chan->dw;
> > diff --git a/include/linux/dma/edma.h b/include/linux/dma/edma.h index
> > 3080747..78ce31b 100644
> > --- a/include/linux/dma/edma.h
> > +++ b/include/linux/dma/edma.h
> > @@ -99,6 +99,7 @@ struct dw_edma_chip {
> > enum dw_edma_map_format mf;
> >
> > struct dw_edma *dw;
> > + bool non_ll;
> > };
> >
> > /* Export to the platform drivers */
> >
>
> --
> i.
^ permalink raw reply [flat|nested] 7+ messages in thread
* RE: [PATCH v5 1/2] dmaengine: dw-edma: Add AMD MDB Endpoint Support
2025-10-28 13:28 ` Ilpo Järvinen
@ 2025-10-29 8:37 ` Verma, Devendra
0 siblings, 0 replies; 7+ messages in thread
From: Verma, Devendra @ 2025-10-29 8:37 UTC (permalink / raw)
To: Ilpo Järvinen
Cc: bhelgaas@google.com, mani@kernel.org, vkoul@kernel.org,
dmaengine@vger.kernel.org, linux-pci@vger.kernel.org, LKML,
Simek, Michal, Verma, Devendra
[AMD Official Use Only - AMD Internal Distribution Only]
Thank you for the inputs Ilpo.
Please check my response inline.
Regards,
Devendra
> -----Original Message-----
> From: Ilpo Järvinen <ilpo.jarvinen@linux.intel.com>
> Sent: Tuesday, October 28, 2025 6:58 PM
> To: Verma, Devendra <Devendra.Verma@amd.com>
> Cc: bhelgaas@google.com; mani@kernel.org; vkoul@kernel.org;
> dmaengine@vger.kernel.org; linux-pci@vger.kernel.org; LKML <linux-
> kernel@vger.kernel.org>; Simek, Michal <michal.simek@amd.com>
> Subject: Re: [PATCH v5 1/2] dmaengine: dw-edma: Add AMD MDB Endpoint
> Support
>
> Caution: This message originated from an External Source. Use proper caution
> when opening attachments, clicking links, or responding.
>
>
> On Tue, 28 Oct 2025, Devendra K Verma wrote:
>
> > AMD MDB PCIe endpoint support. For AMD specific support added the
> > following
> > - AMD supported PCIe Device IDs and Vendor ID (Xilinx).
> > - AMD MDB specific driver data
> > - AMD MDB specific VSEC capability to retrieve the device DDR
> > base address.
> >
> > Signed-off-by: Devendra K Verma <devendra.verma@amd.com>
> > ---
> > Changes in v5:
> > Added the definitions for Xilinx specific VSEC header id, revision,
> > and register offsets.
> > Corrected the error type when no physical offset found for device side
> > memory.
> > Corrected the order of variables.
> >
> > Changes in v4:
> > Configured 8 read and 8 write channels for Xilinx vendor Added checks
> > to validate vendor ID for vendor specific vsec id.
> > Added Xilinx specific vendor id for vsec specific to Xilinx Added the
> > LL and data region offsets, size as input params to function
> > dw_edma_set_chan_region_offset().
> > Moved the LL and data region offsets assignment to function for Xilinx
> > specific case.
> > Corrected comments.
> >
> > Changes in v3:
> > Corrected a typo when assigning AMD (Xilinx) vsec id macro and
> > condition check.
> >
> > Changes in v2:
> > Reverted the devmem_phys_off type to u64.
> > Renamed the function appropriately to suit the functionality for
> > setting the LL & data region offsets.
> >
> > Changes in v1:
> > Removed the pci device id from pci_ids.h file.
> > Added the vendor id macro as per the suggested method.
> > Changed the type of the newly added devmem_phys_off variable.
> > Added to logic to assign offsets for LL and data region blocks in case
> > more number of channels are enabled than given in amd_mdb_data struct.
> > ---
> > drivers/dma/dw-edma/dw-edma-pcie.c | 138
> > ++++++++++++++++++++++++++++++++++++-
> > 1 file changed, 136 insertions(+), 2 deletions(-)
> >
> > diff --git a/drivers/dma/dw-edma/dw-edma-pcie.c
> > b/drivers/dma/dw-edma/dw-edma-pcie.c
> > index 3371e0a7..7b991a0 100644
> > --- a/drivers/dma/dw-edma/dw-edma-pcie.c
> > +++ b/drivers/dma/dw-edma/dw-edma-pcie.c
> > @@ -17,12 +17,27 @@
> >
> > #include "dw-edma-core.h"
> >
> > +/* Synopsys */
> > #define DW_PCIE_VSEC_DMA_ID 0x6
> > #define DW_PCIE_VSEC_DMA_BAR GENMASK(10, 8)
> > #define DW_PCIE_VSEC_DMA_MAP GENMASK(2, 0)
> > #define DW_PCIE_VSEC_DMA_WR_CH GENMASK(9, 0)
> > #define DW_PCIE_VSEC_DMA_RD_CH GENMASK(25, 16)
> >
> > +/* AMD MDB (Xilinx) specific defines */
> > +#define DW_PCIE_XILINX_MDB_VSEC_DMA_ID 0x6
> > +#define DW_PCIE_XILINX_MDB_VSEC_ID 0x20
> > +#define PCI_DEVICE_ID_AMD_MDB_B054 0xb054
> > +#define DW_PCIE_AMD_MDB_INVALID_ADDR (~0ULL)
> > +#define DW_PCIE_XILINX_LL_OFF_GAP 0x200000
> > +#define DW_PCIE_XILINX_LL_SIZE 0x800
> > +#define DW_PCIE_XILINX_DT_OFF_GAP 0x100000
> > +#define DW_PCIE_XILINX_DT_SIZE 0x800
> > +#define DW_PCIE_XILINX_MDB_VSEC_HDR_ID 0x20
> > +#define DW_PCIE_XILINX_MDB_VSEC_REV 0x1
> > +#define DW_PCIE_XILINX_MDB_DEVMEM_OFF_REG_HIGH 0xc
> > +#define DW_PCIE_XILINX_MDB_DEVMEM_OFF_REG_LOW 0x8
> > +
> > #define DW_BLOCK(a, b, c) \
> > { \
> > .bar = a, \
> > @@ -50,6 +65,7 @@ struct dw_edma_pcie_data {
> > u8 irqs;
> > u16 wr_ch_cnt;
> > u16 rd_ch_cnt;
> > + u64 devmem_phys_off;
> > };
> >
> > static const struct dw_edma_pcie_data snps_edda_data = { @@ -90,6
> > +106,64 @@ struct dw_edma_pcie_data {
> > .rd_ch_cnt = 2,
> > };
> >
> > +static const struct dw_edma_pcie_data amd_mdb_data = {
> > + /* MDB registers location */
> > + .rg.bar = BAR_0,
> > + .rg.off = 0x00001000, /* 4 Kbytes */
> > + .rg.sz = 0x00002000, /* 8 Kbytes */
>
> Please use SZ_* + check that this file #includes correct header for them.
> You can then drop those comments.
>
> --
> i.
>
Included the header file and used the appropriate macros instead of constants.
>
> > +
> > + /* Other */
> > + .mf = EDMA_MF_HDMA_NATIVE,
> > + .irqs = 1,
> > + .wr_ch_cnt = 8,
> > + .rd_ch_cnt = 8,
> > +};
> > +
> > +static void dw_edma_set_chan_region_offset(struct dw_edma_pcie_data
> *pdata,
> > + enum pci_barno bar, off_t start_off,
> > + off_t ll_off_gap, size_t ll_size,
> > + off_t dt_off_gap, size_t
> > +dt_size) {
> > + u16 wr_ch = pdata->wr_ch_cnt;
> > + u16 rd_ch = pdata->rd_ch_cnt;
> > + off_t off;
> > + u16 i;
> > +
> > + off = start_off;
> > +
> > + /* Write channel LL region */
> > + for (i = 0; i < wr_ch; i++) {
> > + pdata->ll_wr[i].bar = bar;
> > + pdata->ll_wr[i].off = off;
> > + pdata->ll_wr[i].sz = ll_size;
> > + off += ll_off_gap;
> > + }
> > +
> > + /* Read channel LL region */
> > + for (i = 0; i < rd_ch; i++) {
> > + pdata->ll_rd[i].bar = bar;
> > + pdata->ll_rd[i].off = off;
> > + pdata->ll_rd[i].sz = ll_size;
> > + off += ll_off_gap;
> > + }
> > +
> > + /* Write channel data region */
> > + for (i = 0; i < wr_ch; i++) {
> > + pdata->dt_wr[i].bar = bar;
> > + pdata->dt_wr[i].off = off;
> > + pdata->dt_wr[i].sz = dt_size;
> > + off += dt_off_gap;
> > + }
> > +
> > + /* Read channel data region */
> > + for (i = 0; i < rd_ch; i++) {
> > + pdata->dt_rd[i].bar = bar;
> > + pdata->dt_rd[i].off = off;
> > + pdata->dt_rd[i].sz = dt_size;
> > + off += dt_off_gap;
> > + }
> > +}
> > +
> > static int dw_edma_pcie_irq_vector(struct device *dev, unsigned int
> > nr) {
> > return pci_irq_vector(to_pci_dev(dev), nr); @@ -120,9 +194,24 @@
> > static void dw_edma_pcie_get_vsec_dma_data(struct pci_dev *pdev,
> > u32 val, map;
> > u16 vsec;
> > u64 off;
> > + int cap;
> >
> > - vsec = pci_find_vsec_capability(pdev, PCI_VENDOR_ID_SYNOPSYS,
> > - DW_PCIE_VSEC_DMA_ID);
> > + /*
> > + * Synopsys and AMD (Xilinx) use the same VSEC ID for the purpose
> > + * of map, channel counts, etc.
> > + */
> > + switch (pdev->vendor) {
> > + case PCI_VENDOR_ID_SYNOPSYS:
> > + cap = DW_PCIE_VSEC_DMA_ID;
> > + break;
> > + case PCI_VENDOR_ID_XILINX:
> > + cap = DW_PCIE_XILINX_MDB_VSEC_DMA_ID;
> > + break;
> > + default:
> > + return;
> > + }
> > +
> > + vsec = pci_find_vsec_capability(pdev, pdev->vendor, cap);
> > if (!vsec)
> > return;
> >
> > @@ -155,6 +244,28 @@ static void
> dw_edma_pcie_get_vsec_dma_data(struct pci_dev *pdev,
> > off <<= 32;
> > off |= val;
> > pdata->rg.off = off;
> > +
> > + /* Xilinx specific VSEC capability */
> > + vsec = pci_find_vsec_capability(pdev, PCI_VENDOR_ID_XILINX,
> > + DW_PCIE_XILINX_MDB_VSEC_ID);
> > + if (!vsec)
> > + return;
> > +
> > + pci_read_config_dword(pdev, vsec + PCI_VNDR_HEADER, &val);
> > + if (PCI_VNDR_HEADER_ID(val) != DW_PCIE_XILINX_MDB_VSEC_HDR_ID
> ||
> > + PCI_VNDR_HEADER_REV(val) != DW_PCIE_XILINX_MDB_VSEC_REV)
> > + return;
> > +
> > + pci_read_config_dword(pdev,
> > + vsec + DW_PCIE_XILINX_MDB_DEVMEM_OFF_REG_HIGH,
> > + &val);
> > + off = val;
> > + pci_read_config_dword(pdev,
> > + vsec + DW_PCIE_XILINX_MDB_DEVMEM_OFF_REG_LOW,
> > + &val);
> > + off <<= 32;
> > + off |= val;
> > + pdata->devmem_phys_off = off;
> > }
> >
> > static int dw_edma_pcie_probe(struct pci_dev *pdev, @@ -179,6 +290,7
> > @@ static int dw_edma_pcie_probe(struct pci_dev *pdev,
> > }
> >
> > memcpy(vsec_data, pdata, sizeof(struct dw_edma_pcie_data));
> > + vsec_data->devmem_phys_off = DW_PCIE_AMD_MDB_INVALID_ADDR;
> >
> > /*
> > * Tries to find if exists a PCIe Vendor-Specific Extended
> > Capability @@ -186,6 +298,26 @@ static int dw_edma_pcie_probe(struct
> pci_dev *pdev,
> > */
> > dw_edma_pcie_get_vsec_dma_data(pdev, vsec_data);
> >
> > + if (pdev->vendor == PCI_VENDOR_ID_XILINX) {
> > + /*
> > + * There is no valid address found for the LL memory
> > + * space on the device side.
> > + */
> > + if (vsec_data->devmem_phys_off ==
> DW_PCIE_AMD_MDB_INVALID_ADDR)
> > + return -ENOMEM;
> > +
> > + /*
> > + * Configure the channel LL and data blocks if number of
> > + * channels enabled in VSEC capability are more than the
> > + * channels configured in amd_mdb_data.
> > + */
> > + dw_edma_set_chan_region_offset(vsec_data, BAR_2, 0,
> > + DW_PCIE_XILINX_LL_OFF_GAP,
> > + DW_PCIE_XILINX_LL_SIZE,
> > + DW_PCIE_XILINX_DT_OFF_GAP,
> > + DW_PCIE_XILINX_DT_SIZE);
> > + }
> > +
> > /* Mapping PCI BAR regions */
> > mask = BIT(vsec_data->rg.bar);
> > for (i = 0; i < vsec_data->wr_ch_cnt; i++) { @@ -367,6 +499,8 @@
> > static void dw_edma_pcie_remove(struct pci_dev *pdev)
> >
> > static const struct pci_device_id dw_edma_pcie_id_table[] = {
> > { PCI_DEVICE_DATA(SYNOPSYS, EDDA, &snps_edda_data) },
> > + { PCI_VDEVICE(XILINX, PCI_DEVICE_ID_AMD_MDB_B054),
> > + (kernel_ulong_t)&amd_mdb_data },
> > { }
> > };
> > MODULE_DEVICE_TABLE(pci, dw_edma_pcie_id_table);
> >
^ permalink raw reply [flat|nested] 7+ messages in thread
end of thread, other threads:[~2025-10-29 8:37 UTC | newest]
Thread overview: 7+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2025-10-28 11:28 [PATCH v5 0/2] Add AMD MDB Endpoint and non-LL mode Support Devendra K Verma
2025-10-28 11:28 ` [PATCH v5 1/2] dmaengine: dw-edma: Add AMD MDB Endpoint Support Devendra K Verma
2025-10-28 13:28 ` Ilpo Järvinen
2025-10-29 8:37 ` Verma, Devendra
2025-10-28 11:28 ` [PATCH v5 2/2] dmaengine: dw-edma: Add non-LL mode Devendra K Verma
2025-10-28 13:32 ` Ilpo Järvinen
2025-10-29 8:35 ` Verma, Devendra
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).