lists.openwall.net   lists  /  announce  owl-users  owl-dev  john-users  john-dev  passwdqc-users  yescrypt  popa3d-users  /  oss-security  kernel-hardening  musl  sabotage  tlsify  passwords  /  crypt-dev  xvendor  /  Bugtraq  Full-Disclosure  linux-kernel  linux-netdev  linux-ext4  linux-hardening  linux-cve-announce  PHC 
Open Source and information security mailing list archives
 
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <fe8a7217-51af-1102-80eb-db4475bb47fc@quicinc.com>
Date:   Thu, 2 Nov 2023 17:11:24 +0530
From:   Mukesh Ojha <quic_mojha@...cinc.com>
To:     <neil.armstrong@...aro.org>, Andy Gross <agross@...nel.org>,
        Bjorn Andersson <andersson@...nel.org>,
        Konrad Dybcio <konrad.dybcio@...aro.org>,
        Mathieu Poirier <mathieu.poirier@...aro.org>,
        Rob Herring <robh+dt@...nel.org>,
        Krzysztof Kozlowski <krzysztof.kozlowski+dt@...aro.org>,
        Conor Dooley <conor+dt@...nel.org>,
        Manivannan Sadhasivam <mani@...nel.org>
CC:     <linux-arm-msm@...r.kernel.org>,
        <linux-remoteproc@...r.kernel.org>, <devicetree@...r.kernel.org>,
        <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH v2 2/3] remoteproc: qcom: pas: make region assign more
 generic



On 11/2/2023 3:56 PM, neil.armstrong@...aro.org wrote:
> On 01/11/2023 15:42, Mukesh Ojha wrote:
>>
>>
>> On 10/31/2023 10:36 PM, Neil Armstrong wrote:
>>> Hi,
>>>
>>> On 30/10/2023 14:10, Mukesh Ojha wrote:
>>>>
>>>>
>>>> On 10/30/2023 3:33 PM, Neil Armstrong wrote:
>>>>> The current memory region assign only supports a single
>>>>> memory region.
>>>>>
>>>>> But new platforms introduces more regions to make the
>>>>> memory requirements more flexible for various use cases.
>>>>> Those new platforms also shares the memory region between the
>>>>> DSP and HLOS.
>>>>>
>>>>> To handle this, make the region assign more generic in order
>>>>> to support more than a single memory region and also permit
>>>>> setting the regions permissions as shared.
>>>>>
>>>>> Signed-off-by: Neil Armstrong <neil.armstrong@...aro.org>
>>>>> ---
>>>>>   drivers/remoteproc/qcom_q6v5_pas.c | 102 
>>>>> ++++++++++++++++++++++++-------------
>>>>>   1 file changed, 66 insertions(+), 36 deletions(-)
>>>>>
>>>>> diff --git a/drivers/remoteproc/qcom_q6v5_pas.c 
>>>>> b/drivers/remoteproc/qcom_q6v5_pas.c
>>>>> index 913a5d2068e8..4829fd26e17d 100644
>>>>> --- a/drivers/remoteproc/qcom_q6v5_pas.c
>>>>> +++ b/drivers/remoteproc/qcom_q6v5_pas.c
>>>>> @@ -33,6 +33,8 @@
>>>>>   #define ADSP_DECRYPT_SHUTDOWN_DELAY_MS    100
>>>>> +#define MAX_ASSIGN_COUNT 2
>>>>> +
>>>>>   struct adsp_data {
>>>>>       int crash_reason_smem;
>>>>>       const char *firmware_name;
>>>>> @@ -51,6 +53,9 @@ struct adsp_data {
>>>>>       int ssctl_id;
>>>>>       int region_assign_idx;
>>>>> +    int region_assign_count;
>>>>> +    bool region_assign_shared;
>>>>> +    int region_assign_vmid;
>>>>>   };
>>>>>   struct qcom_adsp {
>>>>> @@ -87,15 +92,18 @@ struct qcom_adsp {
>>>>>       phys_addr_t dtb_mem_phys;
>>>>>       phys_addr_t mem_reloc;
>>>>>       phys_addr_t dtb_mem_reloc;
>>>>> -    phys_addr_t region_assign_phys;
>>>>> +    phys_addr_t region_assign_phys[MAX_ASSIGN_COUNT];
>>>>>       void *mem_region;
>>>>>       void *dtb_mem_region;
>>>>>       size_t mem_size;
>>>>>       size_t dtb_mem_size;
>>>>> -    size_t region_assign_size;
>>>>> +    size_t region_assign_size[MAX_ASSIGN_COUNT];
>>>>>       int region_assign_idx;
>>>>> -    u64 region_assign_perms;
>>>>> +    int region_assign_count;
>>>>> +    bool region_assign_shared;
>>>>> +    int region_assign_vmid;
>>>>> +    u64 region_assign_perms[MAX_ASSIGN_COUNT];
>>>>>       struct qcom_rproc_glink glink_subdev;
>>>>>       struct qcom_rproc_subdev smd_subdev;
>>>>> @@ -590,37 +598,52 @@ static int adsp_alloc_memory_region(struct 
>>>>> qcom_adsp *adsp)
>>>>>   static int adsp_assign_memory_region(struct qcom_adsp *adsp)
>>>>>   {
>>>>> -    struct reserved_mem *rmem = NULL;
>>>>> -    struct qcom_scm_vmperm perm;
>>>>> +    struct qcom_scm_vmperm perm[MAX_ASSIGN_COUNT];
>>>>> +    unsigned int perm_size = 1;
>>>>
>>>> AFAICS, not need of initialization.
>>>
>>> Indeed, removed
>>>
>>>>
>>>>>       struct device_node *node;
>>>>> -    int ret;
>>>>> +    int offset, ret;
>>>>
>>>> Nit: one variable per line.
>>>
>>> Done
>>>
>>>>
>>>>>       if (!adsp->region_assign_idx)
>>>>
>>>> Not related to this patch..
>>>> Should not this be valid only for > 1 ?
>>>
>>> I don't understand, only region_assign_idx > 1 triggers a memory_assign,
>>> and this check discards configurations with region_assign_idx == 0 as
>>> expected.
>>
>> Ah, you can ignore the comments, I got the intention after commenting
>> here ..
>>
>>>
>>>>
>>>>
>>>>>           return 0;
>>>>> -    node = of_parse_phandle(adsp->dev->of_node, "memory-region", 
>>>>> adsp->region_assign_idx);
>>>>> -    if (node)
>>>>> -        rmem = of_reserved_mem_lookup(node);
>>>>> -    of_node_put(node);
>>>>> -    if (!rmem) {
>>>>> -        dev_err(adsp->dev, "unable to resolve shareable 
>>>>> memory-region\n");
>>>>> -        return -EINVAL;
>>>>> -    }
>>>>> +    for (offset = 0; offset < adsp->region_assign_count; ++offset) {
>>>>> +        struct reserved_mem *rmem = NULL;
>>>>> +
>>>>> +        node = of_parse_phandle(adsp->dev->of_node, "memory-region",
>>>>> +                    adsp->region_assign_idx + offset);
>>>>> +        if (node)
>>>>> +            rmem = of_reserved_mem_lookup(node);
>>>>> +        of_node_put(node);
>>>>> +        if (!rmem) {
>>>>> +            dev_err(adsp->dev, "unable to resolve shareable 
>>>>> memory-region index %d\n",
>>>>> +                offset);
>>>>> +            return -EINVAL; > +        }
>>>>
>>>>
>>>>> -    perm.vmid = QCOM_SCM_VMID_MSS_MSA;
>>>>> -    perm.perm = QCOM_SCM_PERM_RW;
>>>>> +        if (adsp->region_assign_shared)  {
>>>>> +            perm[0].vmid = QCOM_SCM_VMID_HLOS;
>>>>> +            perm[0].perm = QCOM_SCM_PERM_RW;
>>>>> +            perm[1].vmid = adsp->region_assign_vmid;
>>>>> +            perm[1].perm = QCOM_SCM_PERM_RW;
>>>>> +            perm_size = 2;
>>>>> +        } else {
>>>>> +            perm[0].vmid = adsp->region_assign_vmid;
>>>>> +            perm[0].perm = QCOM_SCM_PERM_RW;
>>>>> +            perm_size = 1;
>>>>> +        }
>>>>> -    adsp->region_assign_phys = rmem->base;
>>>>> -    adsp->region_assign_size = rmem->size;
>>>>> -    adsp->region_assign_perms = BIT(QCOM_SCM_VMID_HLOS);
>>>>> +        adsp->region_assign_phys[offset] = rmem->base;
>>>>> +        adsp->region_assign_size[offset] = rmem->size;
>>>>> +        adsp->region_assign_perms[offset] = BIT(QCOM_SCM_VMID_HLOS);
>>>>
>>>> Do we need array for this, is this changing ?
>>>
>>> We need to keep region_assign_perms for unassign, but for the other 2 
>>> we would
>>> need to duplicate the code from adsp_assign_memory_region into
>>> adsp_unassign_memory_region.
>>
>> Thanks got it.
>>
>>>
>>>>
>>>>> -    ret = qcom_scm_assign_mem(adsp->region_assign_phys,
>>>>> -                  adsp->region_assign_size,
>>>>> -                  &adsp->region_assign_perms,
>>>>> -                  &perm, 1);
>>>>> -    if (ret < 0) {
>>>>> -        dev_err(adsp->dev, "assign memory failed\n");
>>>>> -        return ret;
>>>>> +        ret = qcom_scm_assign_mem(adsp->region_assign_phys[offset],
>>>>> +                      adsp->region_assign_size[offset],
>>>>> +                      &adsp->region_assign_perms[offset],
>>>>> +                      perm, perm_size);
>>>>> +        if (ret < 0) {
>>>>> +            dev_err(adsp->dev, "assign memory %d failed\n", offset);
>>>>> +            return ret;
>>>>> +        }
>>>>>       }
>>>>>       return 0;
>>>>> @@ -629,20 +652,22 @@ static int adsp_assign_memory_region(struct 
>>>>> qcom_adsp *adsp)
>>>>>   static void adsp_unassign_memory_region(struct qcom_adsp *adsp)
>>>>>   {
>>>>>       struct qcom_scm_vmperm perm;
>>>>> -    int ret;
>>>>> +    int offset, ret;
>>>>> -    if (!adsp->region_assign_idx)
>>>>> +    if (!adsp->region_assign_idx || adsp->region_assign_shared)
>>>>>           return;
>>>>> -    perm.vmid = QCOM_SCM_VMID_HLOS;
>>>>> -    perm.perm = QCOM_SCM_PERM_RW;
>>>>> +    for (offset = 0; offset < adsp->region_assign_count; ++offset) {
>>>>> +        perm.vmid = QCOM_SCM_VMID_HLOS;
>>>>> +        perm.perm = QCOM_SCM_PERM_RW;
>>>>
>>>>> -    ret = qcom_scm_assign_mem(adsp->region_assign_phys,
>>>>> -                  adsp->region_assign_size,
>>>>> -                  &adsp->region_assign_perms,
>>>>> -                  &perm, 1);
>>>>> -    if (ret < 0)
>>>>> -        dev_err(adsp->dev, "unassign memory failed\n");
>>>>> +        ret = qcom_scm_assign_mem(adsp->region_assign_phys[offset],
>>>>> +                      adsp->region_assign_size[offset],
>>>>> +                      &adsp->region_assign_perms[offset],
>>>>> +                      &perm, 1);
>>>>> +        if (ret < 0)
>>>>> +            dev_err(adsp->dev, "unassign memory failed\n");
>>>>> +    }
>>>>>   }
>>>>>   static int adsp_probe(struct platform_device *pdev)
>>>>> @@ -696,6 +721,9 @@ static int adsp_probe(struct platform_device 
>>>>> *pdev)
>>>>>       adsp->info_name = desc->sysmon_name;
>>>>>       adsp->decrypt_shutdown = desc->decrypt_shutdown;
>>>>>       adsp->region_assign_idx = desc->region_assign_idx;
>>
>> Should this also need
>> min_t(int, MAX_ASSIGN_COUNT - 1, desc->region_assign_idx);
>> as no where boundary check is being done.

I was wrong here.. MAX_ASSIGN_COUNT was relative to assign index.

> 
> region_idx is the offset in the memory-region DT property where assigned 
> memory starts,
> so for example there's 2 memory regions on SM8650 CDSP, but only a 
> single shared memory region
> so we have the following:
>   - region_assign_idx = 2
>   - region_assign_count = 1
> and in DT:
>   memory-region = <&cdsp_mem>, <&q6_cdsp_dtb_mem>, <&global_sync_mem>;
> -------------------------------------------------/\
>                                         region_assign_idx
> ------------------------------------------------[                    ]
>                                         region_assign_count
> 
> and for MPSS, there's 2 of both:
>   - region_assign_idx = 2
>   - region_assign_count = 2
> and in DT:
> memory-region = <&mpss_mem>, <&q6_mpss_dtb_mem>, <&mpss_dsm_mem>, 
> <&mpss_dsm_mem_2>;
> -------------------------------------------------/\
>                                         region_assign_idx
> 
> ------------------------------------------------[                                   ]
>                                         region_assign_count
> 
> so we cannot add a bounday check.
> 
> In any case of_parse_phandle() will do the boundary check if DT has less 
> phandles than expected.

Thanks for explaining.

-Mukesh
> 
> Neil
> 
>>
>> -Mukesh
>>>>> +    adsp->region_assign_count = min_t(int, MAX_ASSIGN_COUNT, 
>>>>> desc->region_assign_count);
>>>>> +    adsp->region_assign_vmid = desc->region_assign_vmid;
>>>>> +    adsp->region_assign_shared = desc->region_assign_shared;
>>>>>       if (dtb_fw_name) {
>>>>>           adsp->dtb_firmware_name = dtb_fw_name;
>>>>>           adsp->dtb_pas_id = desc->dtb_pas_id;
>>>>> @@ -1163,6 +1191,8 @@ static const struct adsp_data 
>>>>> sm8550_mpss_resource = {
>>>>>       .sysmon_name = "modem",
>>>>>       .ssctl_id = 0x12,
>>>>>       .region_assign_idx = 2,
>>>>> +    .region_assign_count = 1,
>>>>> +    .region_assign_vmid = QCOM_SCM_VMID_MSS_MSA,
>>>>>   };
>>>>>   static const struct of_device_id adsp_of_match[] = {
>>>>>
>>>>
>>>> -Mukesh
>>>
>>> Thanks,
>>> Neil
>>>
> 

Powered by blists - more mailing lists

Powered by Openwall GNU/*/Linux Powered by OpenVZ