Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Message-ID: <8dfb28d5-b654-746c-03d8-aeee3d438240@ozlabs.ru>
Date:   Wed, 21 Jul 2021 13:32:36 +1000
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:91.0) Gecko/20100101
 Thunderbird/91.0
Subject: Re: [PATCH v5 10/11] powerpc/pseries/iommu: Make use of DDW for
 indirect mapping
Content-Language: en-US
To:     Frederic Barrat <fbarrat@linux.ibm.com>,
        Leonardo Bras <leobras.c@gmail.com>,
        Michael Ellerman <mpe@ellerman.id.au>,
        Benjamin Herrenschmidt <benh@kernel.crashing.org>,
        Paul Mackerras <paulus@samba.org>,
        David Gibson <david@gibson.dropbear.id.au>,
        kernel test robot <lkp@intel.com>,
        Nicolin Chen <nicoleotsuka@gmail.com>
Cc:     linuxppc-dev@lists.ozlabs.org, linux-kernel@vger.kernel.org
References: <20210716082755.428187-1-leobras.c@gmail.com>
 <20210716082755.428187-11-leobras.c@gmail.com>
 <b98f696a-ed64-4c9e-ccb6-549ae8bc7fd6@linux.ibm.com>
From:   Alexey Kardashevskiy <aik@ozlabs.ru>
In-Reply-To: <b98f696a-ed64-4c9e-ccb6-549ae8bc7fd6@linux.ibm.com>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 8bit
Precedence: bulk


On 21/07/2021 04:12, Frederic Barrat wrote:
> 
> 
> On 16/07/2021 10:27, Leonardo Bras wrote:
>> So far it's assumed possible to map the guest RAM 1:1 to the bus, which
>> works with a small number of devices. SRIOV changes it as the user can
>> configure hundreds VFs and since phyp preallocates TCEs and does not
>> allow IOMMU pages bigger than 64K, it has to limit the number of TCEs
>> per a PE to limit waste of physical pages.
>>
>> As of today, if the assumed direct mapping is not possible, DDW creation
>> is skipped and the default DMA window "ibm,dma-window" is used instead.
>>
>> By using DDW, indirect mapping  can get more TCEs than available for the
>> default DMA window, and also get access to using much larger pagesizes
>> (16MB as implemented in qemu vs 4k from default DMA window), causing a
>> significant increase on the maximum amount of memory that can be IOMMU
>> mapped at the same time.
>>
>> Indirect mapping will only be used if direct mapping is not a
>> possibility.
>>
>> For indirect mapping, it's necessary to re-create the iommu_table with
>> the new DMA window parameters, so iommu_alloc() can use it.
>>
>> Removing the default DMA window for using DDW with indirect mapping
>> is only allowed if there is no current IOMMU memory allocated in
>> the iommu_table. enable_ddw() is aborted otherwise.
>>
>> Even though there won't be both direct and indirect mappings at the
>> same time, we can't reuse the DIRECT64_PROPNAME property name, or else
>> an older kexec()ed kernel can assume direct mapping, and skip
>> iommu_alloc(), causing undesirable behavior.
>> So a new property name DMA64_PROPNAME "linux,dma64-ddr-window-info"
>> was created to represent a DDW that does not allow direct mapping.
>>
>> Signed-off-by: Leonardo Bras <leobras.c@gmail.com>
>> ---
>>   arch/powerpc/platforms/pseries/iommu.c | 87 +++++++++++++++++++++-----
>>   1 file changed, 72 insertions(+), 15 deletions(-)
>>
>> diff --git a/arch/powerpc/platforms/pseries/iommu.c 
>> b/arch/powerpc/platforms/pseries/iommu.c
>> index 22d251e15b61..a67e71c49aeb 100644
>> --- a/arch/powerpc/platforms/pseries/iommu.c
>> +++ b/arch/powerpc/platforms/pseries/iommu.c
>> @@ -375,6 +375,7 @@ static DEFINE_SPINLOCK(direct_window_list_lock);
>>   /* protects initializing window twice for same device */
>>   static DEFINE_MUTEX(direct_window_init_mutex);
>>   #define DIRECT64_PROPNAME "linux,direct64-ddr-window-info"
>> +#define DMA64_PROPNAME "linux,dma64-ddr-window-info"
>>   static int tce_clearrange_multi_pSeriesLP(unsigned long start_pfn,
>>                       unsigned long num_pfn, const void *arg)
>> @@ -925,6 +926,7 @@ static int find_existing_ddw_windows(void)
>>           return 0;
>>       find_existing_ddw_windows_named(DIRECT64_PROPNAME);
>> +    find_existing_ddw_windows_named(DMA64_PROPNAME);
>>       return 0;
>>   }
>> @@ -1211,14 +1213,17 @@ static bool enable_ddw(struct pci_dev *dev, 
>> struct device_node *pdn)
>>       struct ddw_create_response create;
>>       int page_shift;
>>       u64 win_addr;
>> +    const char *win_name;
>>       struct device_node *dn;
>>       u32 ddw_avail[DDW_APPLICABLE_SIZE];
>>       struct direct_window *window;
>>       struct property *win64;
>>       bool ddw_enabled = false;
>>       struct failed_ddw_pdn *fpdn;
>> -    bool default_win_removed = false;
>> +    bool default_win_removed = false, direct_mapping = false;
>>       bool pmem_present;
>> +    struct pci_dn *pci = PCI_DN(pdn);
>> +    struct iommu_table *tbl = pci->table_group->tables[0];
>>       dn = of_find_node_by_type(NULL, "ibm,pmemory");
>>       pmem_present = dn != NULL;
>> @@ -1227,6 +1232,7 @@ static bool enable_ddw(struct pci_dev *dev, 
>> struct device_node *pdn)
>>       mutex_lock(&direct_window_init_mutex);
>>       if (find_existing_ddw(pdn, &dev->dev.archdata.dma_offset, &len)) {
>> +        direct_mapping = (len >= max_ram_len);
>>           ddw_enabled = true;
>>           goto out_unlock;
>>       }
>> @@ -1307,8 +1313,7 @@ static bool enable_ddw(struct pci_dev *dev, 
>> struct device_node *pdn)
>>                 query.page_size);
>>           goto out_failed;
>>       }
>> -    /* verify the window * number of ptes will map the partition */
>> -    /* check largest block * page size > max memory hotplug addr */
>> +
>>       /*
>>        * The "ibm,pmemory" can appear anywhere in the address space.
>>        * Assuming it is still backed by page structs, try 
>> MAX_PHYSMEM_BITS
>> @@ -1324,13 +1329,25 @@ static bool enable_ddw(struct pci_dev *dev, 
>> struct device_node *pdn)
>>               dev_info(&dev->dev, "Skipping ibm,pmemory");
>>       }
>> +    /* check if the available block * number of ptes will map 
>> everything */
>>       if (query.largest_available_block < (1ULL << (len - page_shift))) {
>>           dev_dbg(&dev->dev,
>>               "can't map partition max 0x%llx with %llu %llu-sized 
>> pages\n",
>>               1ULL << len,
>>               query.largest_available_block,
>>               1ULL << page_shift);
>> -        goto out_failed;
>> +
>> +        /* DDW + IOMMU on single window may fail if there is any 
>> allocation */
>> +        if (default_win_removed && iommu_table_in_use(tbl)) {
>> +            dev_dbg(&dev->dev, "current IOMMU table in use, can't be 
>> replaced.\n");
>> +            goto out_failed;
>> +        }
>> +
>> +        len = order_base_2(query.largest_available_block << page_shift);
>> +        win_name = DMA64_PROPNAME;
>> +    } else {
>> +        direct_mapping = true;
>> +        win_name = DIRECT64_PROPNAME;
>>       }
>>       ret = create_ddw(dev, ddw_avail, &create, page_shift, len);
>> @@ -1341,8 +1358,7 @@ static bool enable_ddw(struct pci_dev *dev, 
>> struct device_node *pdn)
>>             create.liobn, dn);
>>       win_addr = ((u64)create.addr_hi << 32) | create.addr_lo;
>> -    win64 = ddw_property_create(DIRECT64_PROPNAME, create.liobn, 
>> win_addr,
>> -                    page_shift, len);
>> +    win64 = ddw_property_create(win_name, create.liobn, win_addr, 
>> page_shift, len);
>>       if (!win64) {
>>           dev_info(&dev->dev,
>>                "couldn't allocate property, property name, or value\n");
>> @@ -1360,12 +1376,51 @@ static bool enable_ddw(struct pci_dev *dev, 
>> struct device_node *pdn)
>>       if (!window)
>>           goto out_del_prop;
>> -    ret = walk_system_ram_range(0, memblock_end_of_DRAM() >> PAGE_SHIFT,
>> -            win64->value, tce_setrange_multi_pSeriesLP_walk);
>> -    if (ret) {
>> -        dev_info(&dev->dev, "failed to map direct window for %pOF: 
>> %d\n",
>> -             dn, ret);
>> -        goto out_del_list;
>> +    if (direct_mapping) {
>> +        /* DDW maps the whole partition, so enable direct DMA mapping */
>> +        ret = walk_system_ram_range(0, memblock_end_of_DRAM() >> 
>> PAGE_SHIFT,
>> +                        win64->value, 
>> tce_setrange_multi_pSeriesLP_walk);
>> +        if (ret) {
>> +            dev_info(&dev->dev, "failed to map direct window for 
>> %pOF: %d\n",
>> +                 dn, ret);
>> +            goto out_del_list;
>> +        }
>> +    } else {
>> +        struct iommu_table *newtbl;
>> +        int i;
>> +
>> +        for (i = 0; i < ARRAY_SIZE(pci->phb->mem_resources); i++) {
>> +            const unsigned long mask = IORESOURCE_MEM_64 | 
>> IORESOURCE_MEM;
>> +
>> +            /* Look for MMIO32 */
>> +            if ((pci->phb->mem_resources[i].flags & mask) == 
>> IORESOURCE_MEM)
>> +                break;
>> +        }
>> +
>> +        if (i == ARRAY_SIZE(pci->phb->mem_resources))
>> +            goto out_del_list;
> 
> 
> So we exit and do nothing if there's no MMIO32 bar?
> Isn't the intent just to figure out the MMIO32 area to reserve it when 
> init'ing the table? In which case we could default to 0,0
> 
> I'm actually not clear why we are reserving this area on pseries.


If we do not reserve it, then the iommu code will allocate DMA pages 
from there and these addresses are MMIO32 from the kernel pov at least. 
I saw crashes when (I think) a device tried DMAing to the top 2GB of the 
bus space which happened to be a some other device's BAR.


> 
> 
>> +
>> +        /* New table for using DDW instead of the default DMA window */
>> +        newtbl = iommu_pseries_alloc_table(pci->phb->node);
>> +        if (!newtbl) {
>> +            dev_dbg(&dev->dev, "couldn't create new IOMMU table\n");
>> +            goto out_del_list;
>> +        }
>> +
>> +        iommu_table_setparms_common(newtbl, pci->phb->bus->number, 
>> create.liobn, win_addr,
>> +                        1UL << len, page_shift, NULL, 
>> &iommu_table_lpar_multi_ops);
>> +        iommu_init_table(newtbl, pci->phb->node, 
>> pci->phb->mem_resources[i].start,
>> +                 pci->phb->mem_resources[i].end);
>> +
>> +        pci->table_group->tables[1] = newtbl;
>> +
>> +        /* Keep default DMA window stuct if removed */
>> +        if (default_win_removed) {
>> +            tbl->it_size = 0;
>> +            kfree(tbl->it_map);
>> +        }
>> +
>> +        set_iommu_table_base(&dev->dev, newtbl);
>>       }
>>       spin_lock(&direct_window_list_lock);
> 
> 
> 
> 
> Somewhere around here, we have:
> 
> out_remove_win:
>      remove_ddw(pdn, true, DIRECT64_PROPNAME);
> 
> We should replace with:
>      remove_ddw(pdn, true, win_name);


True. Good spotting. Or rework remove_dma_window() to take just a liobn. 
Thanks,

> 
> 
>    Fred
> 
> 
> 
>> @@ -1408,10 +1463,10 @@ static bool enable_ddw(struct pci_dev *dev, 
>> struct device_node *pdn)
>>        * as RAM, then we failed to create a window to cover persistent
>>        * memory and need to set the DMA limit.
>>        */
>> -    if (pmem_present && ddw_enabled && (len == max_ram_len))
>> +    if (pmem_present && ddw_enabled && direct_mapping && len == 
>> max_ram_len)
>>           dev->dev.bus_dma_limit = dev->dev.archdata.dma_offset + 
>> (1ULL << len);
>> -    return ddw_enabled;
>> +    return ddw_enabled && direct_mapping;
>>   }
>>   static void pci_dma_dev_setup_pSeriesLP(struct pci_dev *dev)
>> @@ -1553,7 +1608,9 @@ static int iommu_reconfig_notifier(struct 
>> notifier_block *nb, unsigned long acti
>>            * we have to remove the property when releasing
>>            * the device node.
>>            */
>> -        remove_ddw(np, false, DIRECT64_PROPNAME);
>> +        if (remove_ddw(np, false, DIRECT64_PROPNAME))
>> +            remove_ddw(np, false, DMA64_PROPNAME);
>> +
>>           if (pci && pci->table_group)
>>               iommu_pseries_free_group(pci->table_group,
>>                       np->full_name);
>>

-- 
Alexey