Re: [PATCH v4] irqchip/gicv3-its: Avoid memory over allocation for ITEs
From: Marc Zyngier
Date: Wed Jun 21 2017 - 03:30:51 EST
On 21/06/17 02:22, Shanker Donthineni wrote:
> Hi Marc,
>
> On 05/06/2017 06:25 AM, Marc Zyngier wrote:
>> On Fri, May 05 2017 at 11:04:22 pm BST, Shanker Donthineni <shankerd@xxxxxxxxxxxxxx> wrote:
>>> Hi Marc,
>>>
>>>
>>> On 05/02/2017 11:16 AM, Marc Zyngier wrote:
>>>> On Sun, Apr 30 2017 at 3:36:15 pm BST, Shanker Donthineni <shankerd@xxxxxxxxxxxxxx> wrote:
>>>>> We are always allocating extra 255Bytes of memory to handle ITE
>>>>> physical address alignment requirement. The kmalloc() satisfies
>>>>> the ITE alignment since the ITS driver is requesting a minimum
>>>>> size of ITS_ITT_ALIGN bytes.
>>>>>
>>>>> Let's try to allocate the exact amount of memory that is required
>>>>> for ITEs to avoid wastage.
>>>>>
>>>>> Signed-off-by: Shanker Donthineni <shankerd@xxxxxxxxxxxxxx>
>>>>> ---
>>>>> Changes:
>>>>> v2: removed 'Change-Id: Ia8084189833f2081ff13c392deb5070c46a64038' from commit.
>>>>> v3: changed from IITE to ITE.
>>>>> v3: removed fallback since kmalloc() guarantees the right alignment.
>>>>>
>>>>> drivers/irqchip/irq-gic-v3-its.c | 6 +++---
>>>>> 1 file changed, 3 insertions(+), 3 deletions(-)
>>>>>
>>>>> diff --git a/drivers/irqchip/irq-gic-v3-its.c b/drivers/irqchip/irq-gic-v3-its.c
>>>>> index 45ea1933..72e56f03 100644
>>>>> --- a/drivers/irqchip/irq-gic-v3-its.c
>>>>> +++ b/drivers/irqchip/irq-gic-v3-its.c
>>>>> @@ -261,7 +261,6 @@ static struct its_collection *its_build_mapd_cmd(struct its_cmd_block *cmd,
>>>>> u8 size = ilog2(desc->its_mapd_cmd.dev->nr_ites);
>>>>>
>>>>> itt_addr = virt_to_phys(desc->its_mapd_cmd.dev->itt);
>>>>> - itt_addr = ALIGN(itt_addr, ITS_ITT_ALIGN);
>>>>>
>>>>> its_encode_cmd(cmd, GITS_CMD_MAPD);
>>>>> its_encode_devid(cmd, desc->its_mapd_cmd.dev->device_id);
>>>>> @@ -1329,13 +1328,14 @@ static struct its_device *its_create_device(struct its_node *its, u32 dev_id,
>>>>> */
>>>>> nr_ites = max(2UL, roundup_pow_of_two(nvecs));
>>>>> sz = nr_ites * its->ite_size;
>>>>> - sz = max(sz, ITS_ITT_ALIGN) + ITS_ITT_ALIGN - 1;
>>>>> + sz = max(sz, ITS_ITT_ALIGN);
>>>>> itt = kzalloc(sz, GFP_KERNEL);
>>>>> lpi_map = its_lpi_alloc_chunks(nvecs, &lpi_base, &nr_lpis);
>>>>> if (lpi_map)
>>>>> col_map = kzalloc(sizeof(*col_map) * nr_lpis, GFP_KERNEL);
>>>>>
>>>>> - if (!dev || !itt || !lpi_map || !col_map) {
>>>>> + if (!dev || !itt || !lpi_map || !col_map ||
>>>>> + !IS_ALIGNED(virt_to_phys(itt), ITS_ITT_ALIGN)) {
>>>>> kfree(dev);
>>>>> kfree(itt);
>>>>> kfree(lpi_map);
>>>> I'm confused. Either the alignment is guaranteed (and you should
>>>> document why it is so), or it is not, and we need to handle the
>>>> non-alignment (instead of failing).
>>>
>>> Sorry for confusion, alignment is guaranteed by kmalloc(), added a
>>> check for readability purpose only can be removed.
>>
>> My question still remains. Where exactly is that alignment guarantee
>> documented and enforced? I can't see anything giving that certainty.
>>
>
> The internal implementation of kmalloc() uses the slab/slub feature
> to allocate memory from 2^N size pool. Linux kernel maintains the
> fixed size of kmem_cache pools to serve the kmalloc(), It allocates
> minimum size of 128Bytes and maximum size depends on the system
> configuration and memory availability. In fact SMMUv3 driver has a
> similar requirement and absolutely there no problem using kmalloc()
> to meet the address alignment requirement.
>
> Call trace:
> kmalloc()
> kmalloc_slab() --> convert size to kmem_cache
> slab_alloc() ---> allocate 2^N size kmem_cache object
>
> root@null-8cfdf006971f:~# cat /proc/slabinfo | grep kmall
> dma-kmalloc-131072 0 0 131072 4 8 : tunables 0 0 0 : slabdata 0 0 0
> dma-kmalloc-65536 0 0 65536 8 8 : tunables 0 0 0 : slabdata 0 0 0
> dma-kmalloc-32768 0 0 32768 16 8 : tunables 0 0 0 : slabdata 0 0 0
> dma-kmalloc-16384 0 0 16384 32 8 : tunables 0 0 0 : slabdata 0 0 0
> dma-kmalloc-8192 0 0 8192 32 4 : tunables 0 0 0 : slabdata 0 0 0
> dma-kmalloc-4096 0 0 4096 32 2 : tunables 0 0 0 : slabdata 0 0 0
> dma-kmalloc-2048 0 0 2048 32 1 : tunables 0 0 0 : slabdata 0 0 0
> dma-kmalloc-1024 0 0 1024 64 1 : tunables 0 0 0 : slabdata 0 0 0
> dma-kmalloc-512 128 128 512 128 1 : tunables 0 0 0 : slabdata 1 1 0
> dma-kmalloc-256 0 0 256 256 1 : tunables 0 0 0 : slabdata 0 0 0
> dma-kmalloc-128 512 512 128 512 1 : tunables 0 0 0 : slabdata 1 1 0
> kmalloc-131072 4 4 131072 4 8 : tunables 0 0 0 : slabdata 1 1 0
> kmalloc-65536 376 376 65536 8 8 : tunables 0 0 0 : slabdata 47 47 0
> kmalloc-32768 320 320 32768 16 8 : tunables 0 0 0 : slabdata 20 20 0
> kmalloc-16384 5248 5248 16384 32 8 : tunables 0 0 0 : slabdata 164 164 0
> kmalloc-8192 2176 2176 8192 32 4 : tunables 0 0 0 : slabdata 68 68 0
> kmalloc-4096 4452 4576 4096 32 2 : tunables 0 0 0 : slabdata 143 143 0
> kmalloc-2048 4416 4416 2048 32 1 : tunables 0 0 0 : slabdata 138 138 0
> kmalloc-1024 10048 10176 1024 64 1 : tunables 0 0 0 : slabdata 159 159 0
> kmalloc-512 19071 19584 512 128 1 : tunables 0 0 0 : slabdata 153 153 0
> kmalloc-256 75873 77312 256 256 1 : tunables 0 0 0 : slabdata 302 302 0
> kmalloc-128 82078 85504 128 512 1 : tunables 0 0 0 : slabdata 167 167 0
>
>
>> I would expect kmalloc to give you something that is cache-line aligned,
>> but probably nothing more than that. Now, I'd happily be proven wrong,
>> but so far, all I can see is that:
>>
>> - ARCH_KMALLOC_MINALIGN is defined as ARCH_DMA_MINALIGN
>> - ARCH_DMA_MINALIGN is defined as L1_CACHE_BYTES
>> - L1_CACHE_BYTES is 128 on arm64, and either 32, 64, or 128 on arm.
>>
>
> Kmalloc always allocates memory with size=roundup_pow_of_two(size)
> and address alignment roundup_pow_of_two(size).
Again, where is that enforced? The slob allocator explicitly uses
ARCH_KMALLOC_MINALIGN to compute its alignment. How does that match your
description above? Where is this roundup_pow_of_two(size) you're
quoting? Does it actually apply to all 3 allocators we have in the kernel?
Please don't give me any of this "it works for me". Show me the code ;-)
Thanks,
M.
--
Jazz is not dead. It just smells funny...