Re: [PATCH] iommu/io-pgtable-arm: Fix table descriptor paddr formatting

[Robin Murphy]

On 2021-11-17 17:12, Hector Martin wrote:
> Table descriptors were being installed without properly formatting the
> address using paddr_to_iopte, which does not match up with the
> iopte_deref in __arm_lpae_map. This is incorrect for the LPAE pte
> format, as it does not handle the high bits properly.

Oh, I guess whatever system it was tested on can't have exercised it all
that thoroughly. IIRC I couldn't test it myself since at the time none
of the Fast Model builds with SMMUs actually implemented any memory
above 48 bits.

> This was found on Apple T6000 DARTs, which require a new pte format
> (different shift); adding support for that to
> paddr_to_iopte/iopte_to_paddr caused it to break badly, as even <48-bit
> addresses would end up incorrect in that case.

...I look forward to not wanting to look at that patch :)

> Signed-off-by: Hector Martin <marcan@xxxxxxxxx>
> ---
>   drivers/iommu/io-pgtable-arm.c | 14 +++++++-------
>   1 file changed, 7 insertions(+), 7 deletions(-)
>
> diff --git a/drivers/iommu/io-pgtable-arm.c b/drivers/iommu/io-pgtable-arm.c
> index dd9e47189d0d..b636e2737607 100644
> --- a/drivers/iommu/io-pgtable-arm.c
> +++ b/drivers/iommu/io-pgtable-arm.c
> @@ -315,12 +315,12 @@ static int arm_lpae_init_pte(struct arm_lpae_io_pgtable *data,
>   static arm_lpae_iopte arm_lpae_install_table(arm_lpae_iopte *table,
>   					     arm_lpae_iopte *ptep,
>   					     arm_lpae_iopte curr,
> -					     struct io_pgtable_cfg *cfg)
> +					     struct arm_lpae_io_pgtable *data)
>   {

Please just define a local "cfg" variable here like in most other
places, to avoid the rest of the churn in this function. Other than
that,

Acked-by: Robin Murphy <robin.murphy@xxxxxxx>

Also,

Fixes: 6c89928ff7a0 ("iommu/io-pgtable-arm: Support 52-bit physical address")

Thanks,
Robin.
(currently elbow-deep in other parts of io-pgtable-arm...)

>   	arm_lpae_iopte old, new;
>   
> -	new = __pa(table) | ARM_LPAE_PTE_TYPE_TABLE;
> -	if (cfg->quirks & IO_PGTABLE_QUIRK_ARM_NS)
> +	new = paddr_to_iopte(__pa(table), data) | ARM_LPAE_PTE_TYPE_TABLE;
> +	if (data->iop.cfg.quirks & IO_PGTABLE_QUIRK_ARM_NS)
>   		new |= ARM_LPAE_PTE_NSTABLE;
>   
>   	/*
> @@ -332,11 +332,11 @@ static arm_lpae_iopte arm_lpae_install_table(arm_lpae_iopte *table,
>   
>   	old = cmpxchg64_relaxed(ptep, curr, new);
>   
> -	if (cfg->coherent_walk || (old & ARM_LPAE_PTE_SW_SYNC))
> +	if (data->iop.cfg.coherent_walk || (old & ARM_LPAE_PTE_SW_SYNC))
>   		return old;
>   
>   	/* Even if it's not ours, there's no point waiting; just kick it */
> -	__arm_lpae_sync_pte(ptep, 1, cfg);
> +	__arm_lpae_sync_pte(ptep, 1, &data->iop.cfg);
>   	if (old == curr)
>   		WRITE_ONCE(*ptep, new | ARM_LPAE_PTE_SW_SYNC);
>   
> @@ -380,7 +380,7 @@ static int __arm_lpae_map(struct arm_lpae_io_pgtable *data, unsigned long iova,
>   		if (!cptep)
>   			return -ENOMEM;
>   
> -		pte = arm_lpae_install_table(cptep, ptep, 0, cfg);
> +		pte = arm_lpae_install_table(cptep, ptep, 0, data);
>   		if (pte)
>   			__arm_lpae_free_pages(cptep, tblsz, cfg);
>   	} else if (!cfg->coherent_walk && !(pte & ARM_LPAE_PTE_SW_SYNC)) {
> @@ -592,7 +592,7 @@ static size_t arm_lpae_split_blk_unmap(struct arm_lpae_io_pgtable *data,
>   		__arm_lpae_init_pte(data, blk_paddr, pte, lvl, 1, &tablep[i]);
>   	}
>   
> -	pte = arm_lpae_install_table(tablep, ptep, blk_pte, cfg);
> +	pte = arm_lpae_install_table(tablep, ptep, blk_pte, data);
>   	if (pte != blk_pte) {
>   		__arm_lpae_free_pages(tablep, tablesz, cfg);
>   		/*