| 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
| ||
|
Date: Wed, 15 Nov 2023 18:02:52 +0530
From: Bibek Kumar Patro <quic_bibekkum@...cinc.com>
To: Robin Murphy <robin.murphy@....com>, <will@...nel.org>,
<joro@...tes.org>, <dmitry.baryshkov@...aro.org>,
<a39.skl@...il.com>, <konrad.dybcio@...aro.org>,
<quic_pkondeti@...cinc.com>, <quic_molvera@...cinc.com>
CC: <linux-arm-msm@...r.kernel.org>,
<linux-arm-kernel@...ts.infradead.org>, <iommu@...ts.linux.dev>,
<linux-kernel@...r.kernel.org>, <qipl.kernel.upstream@...cinc.com>
Subject: Re: [PATCH v2 1/3] iommu/arm-smmu: introduction of ACTLR for custom
prefetcher settings
On 11/14/2023 10:25 PM, Robin Murphy wrote:
> On 14/11/2023 1:56 pm, Bibek Kumar Patro wrote:
>> Currently in Qualcomm SoCs the default prefetch is set to 1 which allows
>> the TLB to fetch just the next page table. MMU-500 features ACTLR
>> register which is implementation defined and is used for Qualcomm SoCs
>> to have a prefetch setting of 1/3/7/15 enabling TLB to prefetch
>> the next set of page tables accordingly allowing for faster translations.
>>
>> ACTLR value is unique for each SMR (Stream matching register) and stored
>> in a pre-populated table. This value is set to the register during
>> context bank initialisation.
>>
>> Signed-off-by: Bibek Kumar Patro <quic_bibekkum@...cinc.com>
>> ---
>> drivers/iommu/arm/arm-smmu/arm-smmu-qcom.c | 41 ++++++++++++++++++++++
>> drivers/iommu/arm/arm-smmu/arm-smmu-qcom.h | 2 ++
>> drivers/iommu/arm/arm-smmu/arm-smmu.c | 5 +--
>> drivers/iommu/arm/arm-smmu/arm-smmu.h | 5 +++
>> 4 files changed, 51 insertions(+), 2 deletions(-)
>>
>> diff --git a/drivers/iommu/arm/arm-smmu/arm-smmu-qcom.c
>> b/drivers/iommu/arm/arm-smmu/arm-smmu-qcom.c
>> index 549ae4dba3a6..578c662c7c30 100644
>> --- a/drivers/iommu/arm/arm-smmu/arm-smmu-qcom.c
>> +++ b/drivers/iommu/arm/arm-smmu/arm-smmu-qcom.c
>> @@ -14,6 +14,17 @@
>>
>> #define QCOM_DUMMY_VAL -1
>>
>> +struct actlr_config {
>> + const struct actlr_data *adata;
>> + size_t size;
>> +};
>> +
>> +struct actlr_data {
>> + u16 sid;
>> + u16 mask;
>
> Do we need to worry about masks? If you're already assuming that any SMR
> will be programmed to match a superset of the data here, surely a single
> unique ID per device would suffice?
>
If you refer to the arm_smmu_set_actlr below, mask would be needed as we
would check for mask along with sid as well while assigning actlr
configuration. Also with mask no of actlr entries can be reduces as
with out mask we have to have entry for each sid.
>> + u32 actlr;
>> +};
>> +
>> static struct qcom_smmu *to_qcom_smmu(struct arm_smmu_device *smmu)
>> {
>> return container_of(smmu, struct qcom_smmu, smmu);
>> @@ -261,9 +272,36 @@ static const struct of_device_id
>> qcom_smmu_client_of_match[] __maybe_unused = {
>> { }
>> };
>>
>> +static void arm_smmu_set_actlr(struct arm_smmu_device *smmu, int idx,
>> + const struct actlr_config *actlrcfg)
>> +{
>> + struct arm_smmu_smr *smr = smmu->smrs;
>> + int i;
>> + u16 id;
>> + u16 mask;
>> +
>> + for (i = 0; i < actlrcfg->size; ++i) {
>> + id = actlrcfg->adata[i].sid;
>> + mask = actlrcfg->adata[i].mask;
>> + if (!smr_is_subset(*smr, id, mask))
>
> How well have you tested this? ;)
>
Well, this logic has worked pretty good for us till now in our
downstream implementation. :) (During testing as well this logic helped
to better match the SMRs instead of manually mathcing the mask and SID
which missed some SIDs)
Also this is already being used to arm_smmu_find_sme hence packaged this
logic in a wrapper to be used in other places as well(including ACTLR
register setting case here.)
>> + arm_smmu_cb_write(smmu, idx, ARM_SMMU_CB_ACTLR,
>> + actlrcfg->adata[i].actlr);
>> + }
>> +}
>> +
>> static int qcom_smmu_init_context(struct arm_smmu_domain *smmu_domain,
>> struct io_pgtable_cfg *pgtbl_cfg, struct device *dev)
>> {
>> + struct arm_smmu_device *smmu = smmu_domain->smmu;
>> + struct qcom_smmu *qsmmu = to_qcom_smmu(smmu);
>> + const struct actlr_config *actlrcfg;
>> + int idx = smmu_domain->cfg.cbndx;
>> +
>> + if (qsmmu->actlrcfg) {
>> + actlrcfg = qsmmu->actlrcfg;
>> + arm_smmu_set_actlr(smmu, idx, actlrcfg);
>> + }
>> +
>> smmu_domain->cfg.flush_walk_prefer_tlbiasid = true;
>>
>> return 0;
>> @@ -467,6 +505,9 @@ static struct arm_smmu_device
>> *qcom_smmu_create(struct arm_smmu_device *smmu,
>> qsmmu->smmu.impl = impl;
>> qsmmu->cfg = data->cfg;
>>
>> + if (data->actlrcfg && (data->actlrcfg->size))
>> + qsmmu->actlrcfg = data->actlrcfg;
>
> Do we really need to replicate multiple parts of the data, or would it
> be sensible to just replace qsmmu->cfg with qsmmu->data and handle the
> further dereferences in the places that want them?
>
>> +
>> return &qsmmu->smmu;
>> }
>>
>> diff --git a/drivers/iommu/arm/arm-smmu/arm-smmu-qcom.h
>> b/drivers/iommu/arm/arm-smmu/arm-smmu-qcom.h
>> index 593910567b88..4b6862715070 100644
>> --- a/drivers/iommu/arm/arm-smmu/arm-smmu-qcom.h
>> +++ b/drivers/iommu/arm/arm-smmu/arm-smmu-qcom.h
>> @@ -9,6 +9,7 @@
>> struct qcom_smmu {
>> struct arm_smmu_device smmu;
>> const struct qcom_smmu_config *cfg;
>> + const struct actlr_config *actlrcfg;
>> bool bypass_quirk;
>> u8 bypass_cbndx;
>> u32 stall_enabled;
>> @@ -25,6 +26,7 @@ struct qcom_smmu_config {
>> };
>>
>> struct qcom_smmu_match_data {
>> + const struct actlr_config *actlrcfg;
>> const struct qcom_smmu_config *cfg;
>> const struct arm_smmu_impl *impl;
>> const struct arm_smmu_impl *adreno_impl;
>> diff --git a/drivers/iommu/arm/arm-smmu/arm-smmu.c
>> b/drivers/iommu/arm/arm-smmu/arm-smmu.c
>> index d6d1a2a55cc0..8e4faf015286 100644
>> --- a/drivers/iommu/arm/arm-smmu/arm-smmu.c
>> +++ b/drivers/iommu/arm/arm-smmu/arm-smmu.c
>> @@ -990,9 +990,10 @@ static int arm_smmu_find_sme(struct
>> arm_smmu_device *smmu, u16 id, u16 mask)
>> * expect simply identical entries for this case, but there's
>> * no harm in accommodating the generalisation.
>> */
>> - if ((mask & smrs[i].mask) == mask &&
>> - !((id ^ smrs[i].id) & ~smrs[i].mask))
>> +
>> + if (smr_is_subset(smrs[i], id, mask))
>> return i;
>> +
>> /*
>> * If the new entry has any other overlap with an existing one,
>> * though, then there always exists at least one stream ID
>> diff --git a/drivers/iommu/arm/arm-smmu/arm-smmu.h
>> b/drivers/iommu/arm/arm-smmu/arm-smmu.h
>> index 703fd5817ec1..b1638bbc41d4 100644
>> --- a/drivers/iommu/arm/arm-smmu/arm-smmu.h
>> +++ b/drivers/iommu/arm/arm-smmu/arm-smmu.h
>> @@ -501,6 +501,11 @@ static inline void arm_smmu_writeq(struct
>> arm_smmu_device *smmu, int page,
>> writeq_relaxed(val, arm_smmu_page(smmu, page) + offset);
>> }
>>
>> +static inline bool smr_is_subset(struct arm_smmu_smr smrs, u16 id,
>> u16 mask)
>
> Hmm, that name reads as implying the opposite of what it actually tests,
> not to mention that passing structs by value is a bit horrid as well :(
>
It might be okay to name it as subset_of_smr() though. You have any
other naming suggestion in mind which could correctly describe the
logic?
Thanks & regards,
Bibek
> Thanks,
> Robin.
>
>> +{
>> + return (mask & smrs.mask) == mask && !((id ^ smrs.id) & ~smrs.mask);
>> +}
>> +
>> #define ARM_SMMU_GR0 0
>> #define ARM_SMMU_GR1 1
>> #define ARM_SMMU_CB(s, n) ((s)->numpage + (n))
>> --
>> 2.17.1
>>
Powered by blists - more mailing lists