[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <18c6b95a-961d-4544-856d-7dc08f7d5d74@oss.qualcomm.com>
Date: Mon, 8 Dec 2025 12:56:18 +0530
From: Kumari Pallavi <kumari.pallavi@....qualcomm.com>
To: Dmitry Baryshkov <dmitry.baryshkov@....qualcomm.com>
Cc: kpallavi@....qualcomm.com, srini@...nel.org, amahesh@....qualcomm.com,
arnd@...db.de, gregkh@...uxfoundation.org, robh@...nel.org,
krzk+dt@...nel.org, conor+dt@...nel.org, quic_bkumar@...cinc.com,
ekansh.gupta@....qualcomm.com, linux-kernel@...r.kernel.org,
quic_chennak@...cinc.com, dri-devel@...ts.freedesktop.org,
linux-arm-msm@...r.kernel.org, devicetree@...r.kernel.org,
jingyi.wang@....qualcomm.com, aiqun.yu@....qualcomm.com,
ktadakam@....qualcomm.com
Subject: Re: [PATCH v5 3/4] misc: fastrpc: Add support for new DSP IOVA
formatting
On 12/6/2025 8:00 AM, Dmitry Baryshkov wrote:
> On Tue, Dec 02, 2025 at 11:36:27AM +0530, Kumari Pallavi wrote:
>> Implement the new IOVA formatting required by the DSP architecture change
>> on Kaanapali SoC. Place the SID for DSP DMA transactions at bit 56 in the
>> physical address. This placement is necessary for the DSPs to correctly
>> identify streams and operate as intended.
>> To address this, set SID position to bit 56 via OF matching on the fastrpc
>> node; otherwise, default to legacy 32-bit placement.
>> This change ensures consistent SID placement across DSPs.
>>
>> Signed-off-by: Kumari Pallavi <kumari.pallavi@....qualcomm.com>
>> ---
>> drivers/misc/fastrpc.c | 48 ++++++++++++++++++++++++++++++++++++------
>> 1 file changed, 41 insertions(+), 7 deletions(-)
>>
>> diff --git a/drivers/misc/fastrpc.c b/drivers/misc/fastrpc.c
>> index eb9501fe79bc..6a67daafeaa1 100644
>> --- a/drivers/misc/fastrpc.c
>> +++ b/drivers/misc/fastrpc.c
>> @@ -33,7 +33,6 @@
>> #define FASTRPC_ALIGN 128
>> #define FASTRPC_MAX_FDLIST 16
>> #define FASTRPC_MAX_CRCLIST 64
>> -#define FASTRPC_PHYS(p) ((p) & 0xffffffff)
>> #define FASTRPC_CTX_MAX (256)
>> #define FASTRPC_INIT_HANDLE 1
>> #define FASTRPC_DSP_UTILITIES_HANDLE 2
>> @@ -105,6 +104,17 @@
>>
>> #define miscdev_to_fdevice(d) container_of(d, struct fastrpc_device, miscdev)
>>
>> +/* Extract smmu pa from consolidated iova */
>
> Nit: SMMU, PA, IOVA
>
ACK
>> +#define IPA_TO_DMA_ADDR(iova, sid_pos) (iova & ((1ULL << sid_pos) - 1ULL))
>
> GENMASK?
>
Thanks for the suggestion. I’ve updated the macro to use GENMASK_ULL
instead of the hand‑rolled (1ULL << sid_pos) - 1ULL.
The new version looks like this:
#define IPA_TO_DMA_ADDR(iova, sid_pos) \
((sid_pos) ? ((iova) & GENMASK_ULL((sid_pos) - 1, 0)) : 0)
>> +/*
>> + * Prepare the consolidated iova to send to dsp by prepending the sid
>
> Nit: DSP, SID, PA
>
ACK
>> + * to smmu pa at the appropriate position
>> + */
>> +static inline u64 fastrpc_compute_sid_offset(u64 sid, u32 sid_pos)
>> +{
>> + return sid << sid_pos;
>> +}
>> +
>> struct fastrpc_phy_page {
>> dma_addr_t addr; /* dma address */
>> u64 size; /* size of contiguous region */
>> @@ -257,6 +267,10 @@ struct fastrpc_session_ctx {
>> bool valid;
>> };
>>
>> +struct fastrpc_soc_data {
>> + u32 sid_pos;
>> +};
>> +
>> struct fastrpc_channel_ctx {
>> int domain_id;
>> int sesscount;
>> @@ -278,6 +292,7 @@ struct fastrpc_channel_ctx {
>> bool secure;
>> bool unsigned_support;
>> u64 dma_mask;
>> + const struct fastrpc_soc_data *soc_data;
>> };
>>
>> struct fastrpc_device {
>> @@ -390,7 +405,7 @@ static int fastrpc_map_lookup(struct fastrpc_user *fl, int fd,
>> static void fastrpc_buf_free(struct fastrpc_buf *buf)
>> {
>> dma_free_coherent(buf->dev, buf->size, buf->virt,
>> - FASTRPC_PHYS(buf->dma_addr));
>> + IPA_TO_DMA_ADDR(buf->dma_addr, buf->fl->cctx->soc_data->sid_pos));
>> kfree(buf);
>> }
>>
>> @@ -440,7 +455,8 @@ static int fastrpc_buf_alloc(struct fastrpc_user *fl, struct device *dev,
>> buf = *obuf;
>>
>> if (fl->sctx && fl->sctx->sid)
>> - buf->dma_addr += ((u64)fl->sctx->sid << 32);
>> + buf->dma_addr += fastrpc_compute_sid_offset((u64)fl->sctx->sid,
>
> Drop type conversion
>
ACK
>> + fl->cctx->soc_data->sid_pos);
>>
>> return 0;
>> }
>> @@ -685,7 +701,8 @@ static int fastrpc_dma_buf_attach(struct dma_buf *dmabuf,
>> return -ENOMEM;
>>
>> ret = dma_get_sgtable(buffer->dev, &a->sgt, buffer->virt,
>> - FASTRPC_PHYS(buffer->dma_addr), buffer->size);
>> + IPA_TO_DMA_ADDR(buffer->dma_addr,
>> + buffer->fl->cctx->soc_data->sid_pos), buffer->size);
>> if (ret < 0) {
>> dev_err(buffer->dev, "failed to get scatterlist from DMA API\n");
>> kfree(a);
>> @@ -734,7 +751,8 @@ static int fastrpc_mmap(struct dma_buf *dmabuf,
>> dma_resv_assert_held(dmabuf->resv);
>>
>> return dma_mmap_coherent(buf->dev, vma, buf->virt,
>> - FASTRPC_PHYS(buf->dma_addr), size);
>> + IPA_TO_DMA_ADDR(buf->dma_addr,
>> + buf->fl->cctx->soc_data->sid_pos), size);
>> }
>>
>> static const struct dma_buf_ops fastrpc_dma_buf_ops = {
>> @@ -789,7 +807,8 @@ static int fastrpc_map_attach(struct fastrpc_user *fl, int fd,
>> map->dma_addr = sg_phys(map->table->sgl);
>> } else {
>> map->dma_addr = sg_dma_address(map->table->sgl);
>> - map->dma_addr += ((u64)fl->sctx->sid << 32);
>> + map->dma_addr += fastrpc_compute_sid_offset((u64)fl->sctx->sid,
>
> Drop type conversion. Hmm. Looking at it, would it be better:
>
> map->dma_addr = fastrpc_compute_dma_addr(fl->sctx, sg_dma_address(map->table->sgl))
>
Thanks for the suggestion. To confirm: you’re proposing that the call
site use a single helper returning the final DMA address, i.e.
map->dma_addr = fastrpc_compute_dma_addr(fl,
sg_dma_address(map->table->sgl));
I can implement fastrpc_compute_dma_addr() as a thin wrapper that
internally calls our existing fastrpc_compute_sid_offset() and composes
the consolidated IOVA (SID in upper bits, base DMA/PA in lower bits)
?
>> + fl->cctx->soc_data->sid_pos);
>> }
>> for_each_sg(map->table->sgl, sgl, map->table->nents,
>> sgl_index)
>> @@ -2290,6 +2309,14 @@ static int fastrpc_get_domain_id(const char *domain)
>> return -EINVAL;
>> }
>>
>> +static const struct fastrpc_soc_data kaanapali_soc_data = {
>> + .sid_pos = 56,
>> +};
>> +
>> +static const struct fastrpc_soc_data default_soc_data = {
>> + .sid_pos = 32,
>> +};
>> +
>> static int fastrpc_rpmsg_probe(struct rpmsg_device *rpdev)
>> {
>> struct device *rdev = &rpdev->dev;
>> @@ -2298,6 +2325,11 @@ static int fastrpc_rpmsg_probe(struct rpmsg_device *rpdev)
>> const char *domain;
>> bool secure_dsp;
>> unsigned int vmids[FASTRPC_MAX_VMIDS];
>> + const struct fastrpc_soc_data *soc_data;
>> +
>> + soc_data = device_get_match_data(rdev);
>> + if (!soc_data)
>> + soc_data = &default_soc_data;
>
> Define it for all the entries, drop if-check.
>
ACK
>>
>> err = of_property_read_string(rdev->of_node, "label", &domain);
>> if (err) {
>> @@ -2350,6 +2382,7 @@ static int fastrpc_rpmsg_probe(struct rpmsg_device *rpdev)
>>
>> secure_dsp = !(of_property_read_bool(rdev->of_node, "qcom,non-secure-domain"));
>> data->secure = secure_dsp;
>> + data->soc_data = soc_data;
>>
>> switch (domain_id) {
>> case ADSP_DOMAIN_ID:
>> @@ -2487,7 +2520,8 @@ static int fastrpc_rpmsg_callback(struct rpmsg_device *rpdev, void *data,
>> }
>>
>> static const struct of_device_id fastrpc_rpmsg_of_match[] = {
>> - { .compatible = "qcom,fastrpc" },
>> + { .compatible = "qcom,kaanapali-fastrpc", .data = &kaanapali_soc_data },
>> + { .compatible = "qcom,fastrpc", .data = &default_soc_data },
>> { },
>> };
>> MODULE_DEVICE_TABLE(of, fastrpc_rpmsg_of_match);
>> --
>> 2.34.1
>>
>
Thanks,
Pallavi
Powered by blists - more mailing lists