Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Message-ID: <9c7cb68c-6516-6087-92ad-e707d8b122ed@linaro.org>
Date:   Fri, 23 Dec 2022 20:15:32 +0200
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:102.0) Gecko/20100101
 Thunderbird/102.5.1
Subject: Re: [PATCH v2 3/3] arm64: dts: qcom: sm8450: Use GIC-ITS for PCIe0
 and PCIe1
Content-Language: en-GB
To:     Manivannan Sadhasivam <manivannan.sadhasivam@linaro.org>
Cc:     andersson@kernel.org, robh+dt@kernel.org,
        krzysztof.kozlowski+dt@linaro.org, bhelgaas@google.com,
        konrad.dybcio@linaro.org, linux-arm-msm@vger.kernel.org,
        linux-pci@vger.kernel.org, devicetree@vger.kernel.org,
        linux-kernel@vger.kernel.org
References: <20221222133123.50676-1-manivannan.sadhasivam@linaro.org>
 <20221222133123.50676-4-manivannan.sadhasivam@linaro.org>
 <e756516a-a5e2-a6ac-fd7f-71726766fa81@linaro.org>
 <20221223174555.GE4587@thinkpad>
From:   Dmitry Baryshkov <dmitry.baryshkov@linaro.org>
In-Reply-To: <20221223174555.GE4587@thinkpad>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 7bit
Precedence: bulk

On 23/12/2022 19:45, Manivannan Sadhasivam wrote:
> On Fri, Dec 23, 2022 at 07:18:32PM +0200, Dmitry Baryshkov wrote:
>> On 22/12/2022 15:31, Manivannan Sadhasivam wrote:
>>> Both PCIe0 and PCIe1 controllers are capable of receiving MSIs from
>>> endpoint devices using GIC-ITS MSI controller. Add support for it.
>>>
>>> Currently, BDF (0:0.0) and BDF (1:0.0) are enabled and with the
>>> msi-map-mask of 0xff00, all the 32 devices under these two busses can
>>> share the same Device ID.
>>>
>>> The GIC-ITS MSI implementation provides an advantage over internal MSI
>>> implementation using Locality-specific Peripheral Interrupts (LPI) that
>>> would allow MSIs to be targeted for each CPU core.
>>>
>>> It should be noted that the MSIs for BDF (1:0.0) only works with Device
>>> ID of 0x5980 and 0x5a00. Hence, the IDs are swapped.
>>>
>>> Signed-off-by: Manivannan Sadhasivam <manivannan.sadhasivam@linaro.org>
>>> ---
>>>    arch/arm64/boot/dts/qcom/sm8450.dtsi | 20 ++++++++++++++------
>>>    1 file changed, 14 insertions(+), 6 deletions(-)
>>>
>>> diff --git a/arch/arm64/boot/dts/qcom/sm8450.dtsi b/arch/arm64/boot/dts/qcom/sm8450.dtsi
>>> index 570475040d95..c4dd5838fac6 100644
>>> --- a/arch/arm64/boot/dts/qcom/sm8450.dtsi
>>> +++ b/arch/arm64/boot/dts/qcom/sm8450.dtsi
>>> @@ -1733,9 +1733,13 @@ pcie0: pci@1c00000 {
>>>    			ranges = <0x01000000 0x0 0x60200000 0 0x60200000 0x0 0x100000>,
>>>    				 <0x02000000 0x0 0x60300000 0 0x60300000 0x0 0x3d00000>;
>>> -			interrupts = <GIC_SPI 141 IRQ_TYPE_LEVEL_HIGH>;
>>> -			interrupt-names = "msi";
>>> -			#interrupt-cells = <1>;
>>> +			/*
>>> +			 * MSIs for BDF (1:0.0) only works with Device ID 0x5980.
>>> +			 * Hence, the IDs are swapped.
>>> +			 */
>>> +			msi-map = <0x0 &gic_its 0x5981 0x1>,
>>> +				  <0x100 &gic_its 0x5980 0x1>;
>>
>> This definitely doesn't match what has been used in the downstream.
>>
> 
> Yes, I do not know why the downstream Device ID doesn't work. I tried finding
> the answer within Qcom but didn't get any answer so far :/ So I just went with
> the value that works on multiple boards.

Ugh :-(

> 
>> Also if I understand correctly this change would prevent us from using
>> multiple MSI interrupts for the connected device, as the last value of the
>> 0x100 mapping is 0x1, while the vendor kernel uses <0x100 &its 0x5981 0x20>.
>>
> 
> Not true. The controller can still support multiple MSIs for the endpoint
> devices but the only difference is, it would use the same Device ID for all.

I see, please excuse me then. But don't we have to define multiple MSI 
vectors here too?

> 
> The Qcom GIC-ITS implementation could only support 32 Device IDs. By specifying
> the size of 0x20, a separate Device ID would be used for each devices of bus 1.
> But if a PCIe switch is connected and the bus count becomes > 1, then the MSI
> allocation would fail because Device IDs are exhausted.
> 
> The downstream implementation just assumes that there will be only bus 1 and I
> do not want to follow that assumption.
> 
> That's why I used "msi-map-mask" property of value "0xff00" here, as that will
> allow all the devices under the bus 1 to share the same Device ID. For now I
> only mapped bus 1, but extending that in the future for other busses is simple.
> 
> Thanks,
> Mani
> 
>> Do you know by chance, why do we differ from the vendor dtsi?
>>
>>> +			msi-map-mask = <0xff00>;
>>>    			interrupt-map-mask = <0 0 0 0x7>;
>>>    			interrupt-map = <0 0 0 1 &intc 0 0 0 149 IRQ_TYPE_LEVEL_HIGH>, /* int_a */
>>>    					<0 0 0 2 &intc 0 0 0 150 IRQ_TYPE_LEVEL_HIGH>, /* int_b */
>>> @@ -1842,9 +1846,13 @@ pcie1: pci@1c08000 {
>>>    			ranges = <0x01000000 0x0 0x40200000 0 0x40200000 0x0 0x100000>,
>>>    				 <0x02000000 0x0 0x40300000 0 0x40300000 0x0 0x1fd00000>;
>>> -			interrupts = <GIC_SPI 307 IRQ_TYPE_LEVEL_HIGH>;
>>> -			interrupt-names = "msi";
>>> -			#interrupt-cells = <1>;
>>> +			/*
>>> +			 * MSIs for BDF (1:0.0) only works with Device ID 0x5a00.
>>> +			 * Hence, the IDs are swapped.
>>> +			 */
>>> +			msi-map = <0x0 &gic_its 0x5a01 0x1>,
>>> +				  <0x100 &gic_its 0x5a00 0x1>;
>>> +			msi-map-mask = <0xff00>;
>>>    			interrupt-map-mask = <0 0 0 0x7>;
>>>    			interrupt-map = <0 0 0 1 &intc 0 0 0 434 IRQ_TYPE_LEVEL_HIGH>, /* int_a */
>>>    					<0 0 0 2 &intc 0 0 0 435 IRQ_TYPE_LEVEL_HIGH>, /* int_b */
>>
>> -- 
>> With best wishes
>> Dmitry
>>
> 

-- 
With best wishes
Dmitry