[PATCH v2 2/2] ath11k: Use reserved host DDR addresses from DT for PCI devices

[Anilkumar Kolli]

On 2021-11-17 13:35, Kalle Valo wrote:
> Anilkumar Kolli <akolli at codeaurora.org> writes:
> 
>> Host DDR memory (contiguous 45 MB in mode-0 or 15 MB in mode-2)
>> is reserved through DT entries for firmware usage. Send the base
>> address and size from DT entries.
>> 
>> If DT entry is available, PCI devices work with
>> fixed_mem_region else host allocates multiple segments.
>> 
>> IPQ8074 on HK10 board supports multiple PCI devices.
>> IPQ8074 + QCN9074 is tested with this patch.
>> 
>> Tested-on: QCN9074 hw1.0 PCI 
>> WLAN.HK.2.4.0.1-01838-QCAHKSWPL_SILICONZ-1
>> 
>> Signed-off-by: Anilkumar Kolli <akolli at codeaurora.org>
>> ---
>> V2:
>>   - Use of_ API to read from dt node (Rob)
>> 
>>  drivers/net/wireless/ath/ath11k/mhi.c | 42 ++++++++++++++++++--
>>  drivers/net/wireless/ath/ath11k/pci.c | 10 ++++-
>>  drivers/net/wireless/ath/ath11k/qmi.c | 75 
>> +++++++++++++++++++++++++++++++----
>>  drivers/net/wireless/ath/ath11k/qmi.h |  1 +
>>  4 files changed, 116 insertions(+), 12 deletions(-)
>> 
>> diff --git a/drivers/net/wireless/ath/ath11k/mhi.c 
>> b/drivers/net/wireless/ath/ath11k/mhi.c
>> index 26c7ae242db6..0ea6ab7edeb9 100644
>> --- a/drivers/net/wireless/ath/ath11k/mhi.c
>> +++ b/drivers/net/wireless/ath/ath11k/mhi.c
>> @@ -3,6 +3,7 @@
>> 
>>  #include <linux/msi.h>
>>  #include <linux/pci.h>
>> +#include <linux/of.h>
>> 
>>  #include "core.h"
>>  #include "debug.h"
>> @@ -316,7 +317,11 @@ int ath11k_mhi_register(struct ath11k_pci 
>> *ab_pci)
>>  	struct ath11k_base *ab = ab_pci->ab;
>>  	struct mhi_controller *mhi_ctrl;
>>  	struct mhi_controller_config *ath11k_mhi_config;
>> -	int ret;
>> +	struct device_node *np;
>> +	int ret, len, sw, aw;
>> +	u32 *reg, *reg_end;
>> +	unsigned long start, size;
>> +	bool no_dt_entry = 0;
> 
> Use only true or false with bools.
> 
Sure.

>>  	mhi_ctrl = mhi_alloc_controller();
>>  	if (!mhi_ctrl)
>> @@ -339,8 +344,39 @@ int ath11k_mhi_register(struct ath11k_pci 
>> *ab_pci)
>>  		return ret;
>>  	}
>> 
>> -	mhi_ctrl->iova_start = 0;
>> -	mhi_ctrl->iova_stop = 0xffffffff;
>> +	np = of_find_node_by_type(NULL, "memory");
>> +	if (!np) {
>> +		no_dt_entry = 1;
> 
> true
> 
>> +		goto no_dt_entry;
>> +	}
>> +
>> +	aw = of_n_addr_cells(np);
>> +	sw = of_n_size_cells(np);
>> +
>> +	reg = (unsigned int *)of_get_property(np, "reg", &len);
> 
> Always try to avoid casts. of_get_property() returns 'const void *' so
> you are casting out const here, which is a bug. You don't need to do
> casting with void pointers so remove it here.
> 
Sure.

> But is reg property really unsigned int? And not u32 or dma_addr_t, for
> example?
> 
 From DT specification, each value in reg is a u32.

In our example
     memory {
         device_type = "memory";
         reg = <0x0 0x40000000 0x0 0x20000000>;
     };

     0x0 is higer 32-bit addr of RAM
     0x40000000 is lower 32-bit addr of RAM
     0x0 is size
     0x20000000 is [1GB RAM] size

>> +	if (!reg) {
>> +		no_dt_entry = 1;
> 
> true
> 
>> +		goto no_dt_entry;
>> +	}
>> +
>> +	reg_end = reg + len / (aw * sw);
>> +
>> +	do {
>> +		start = of_read_number(reg, aw);
>> +		reg += aw;
>> +		size = of_read_number(reg, sw);
> 
> of_read_number() takes 'const __be32 *cell' but reg is 'u32 *'?
> 
Yes.

>> +		reg += sw;
>> +	} while (reg < reg_end);
>> +
>> +no_dt_entry:
>> +	if (no_dt_entry) {
>> +		mhi_ctrl->iova_start = 0;
>> +		mhi_ctrl->iova_stop = 0xFFFFFFFF;
>> +	} else {
>> +		mhi_ctrl->iova_start = (dma_addr_t)(start + 0x1000000);
>> +		mhi_ctrl->iova_stop = (dma_addr_t)(start + size);
> 
> I don't like casts, they hide bugs like the const issue above. Is there
> any way to do this without casts?
> 

u64 start, size;

     if (of_property_read_u32_array(np, "reg", reg, 4)) {
          ath11k_dbg(ab, ATH11K_DBG_QMI,
                     "qmi fail to get reg from hremote\n");
          return 0;
     }
     start = reg[0] + reg[1];
     size = reg[2] + reg[3];

I will this code in next patch. still needs typecast form u64 to 
dma_addr_t. no ?

>> @@ -1245,6 +1246,13 @@ static int ath11k_pci_probe(struct pci_dev 
>> *pdev,
>>  	pci_set_drvdata(pdev, ab);
>>  	spin_lock_init(&ab_pci->window_lock);
>> 
>> +	/* Set fixed_mem_region to true for platforms support reserved 
>> memory
>> +	 * from DT. If memory is reserved from DT for FW, ath11k driver need 
>> not
>> +	 * allocate memory.
>> +	 */
>> +	if (!of_property_read_u32(ab->dev->of_node, "memory-region", &addr))
>> +		ab->bus_params.fixed_mem_region = true;
> 
> ab->bus_params is supposed to be const to catch this, but clearly it's
> not. The idea of bus_params is to have static bus level configuration,
> please instead use dev_flags or something else.
> 
Sure.

>>  static int ath11k_qmi_assign_target_mem_chunk(struct ath11k_base *ab)
>>  {
>> -	int i, idx;
>> +	struct device *dev = ab->dev;
>> +	struct device_node *hremote_node = NULL;
>> +	phandle hremote_phandle;
>> +	int i, idx, len, sw, aw, host_ddr_sz;
>> +	u32 *reg, *reg_end;
>> +	u64 start, size;
>> 
>>  	for (i = 0, idx = 0; i < ab->qmi.mem_seg_count; i++) {
>>  		switch (ab->qmi.target_mem[i].type) {
>> +		case HOST_DDR_REGION_TYPE:
>> +			if (of_property_read_u32(dev->of_node, "memory-region",
>> +						 &hremote_phandle)) {
>> +				ath11k_dbg(ab, ATH11K_DBG_QMI,
>> +					   "qmi fail to get hremote phandle\n");
>> +				return 0;
>> +			}
>> +
>> +			hremote_node = of_find_node_by_phandle(hremote_phandle);
>> +			if (!hremote_node) {
>> +				ath11k_dbg(ab, ATH11K_DBG_QMI,
>> +					   "qmi fail to get hremote_node\n");
>> +				return 0;
>> +			}
>> +
>> +			aw = of_n_addr_cells(hremote_node);
>> +			sw = of_n_size_cells(hremote_node);
>> +
>> +			reg = (unsigned int *)of_get_property(hremote_node, "reg", &len);
> 
> No need to cast. And I guess you are again removing the const?


- Anil.