linux-kernel - Re: [PATCH] scsi: libsas: Add SMP request allocation handler callback

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] [thread-next>] [day] [month] [year] [list]

Message-ID: <3ea54627-339f-4d19-896d-5843a8b9e55e@oracle.com>
Date: Mon, 25 Mar 2024 15:10:07 +0000
From: John Garry <john.g.garry@...cle.com>
To: Jason Yan <yanaijie@...wei.com>, Damien Le Moal <dlemoal@...nel.org>,
        Yihang Li <liyihang9@...wei.com>, jejb@...ux.ibm.com,
        martin.petersen@...cle.com, chenxiang66@...ilicon.com
Cc: linux-scsi@...r.kernel.org, linux-kernel@...r.kernel.org,
        linuxarm@...wei.com, prime.zeng@...wei.com, yangxingui@...wei.com
Subject: Re: [PATCH] scsi: libsas: Add SMP request allocation handler callback

On 25/03/2024 14:21, Jason Yan wrote:
> On 2024/3/25 22:03, Damien Le Moal wrote:
>> On 3/25/24 22:17, Yihang Li wrote:
>>> This series [1] reducing the kmalloc() minimum alignment on arm64 to 8
>>> (from 128).
>>
>> And ? What is the point you are trying to convey here ?
>>
>>> The hisi_sas has special requirements on the memory address alignment
>>> (must be 16-byte-aligned) of the command request frame, so add a SMP
>>> request allocation callback and fill it in for the hisi_sas driver.
>>
>> 128 is aligned to 16. So what is the problem you are trying to solve 
>> here ?
>> Can you clarify ? I suspect this is all about memory allocation 
>> optimization ?
> 
> After series [1] been merged, kmalloc is 8-byte-aligned, however 
> hisi_sas hardware needs 16-byte-aligned. That's the problem.
> 
>>
>>>
>>> Link: 
>>> https://urldefense.com/v3/__https://lkml.kernel.org/r/20230612153201.554742-1-catalin.marinas@arm.com__;!!ACWV5N9M2RV99hQ!L85qzmByNbkBbZByXKuNMAhvt8wxKPCsogKt3Pgn93DTkzfc54jA3of5XL8oEDDDDTMU1OtghdKiLZdKe5ub$  [1]
>>> Signed-off-by: Yihang Li <liyihang9@...wei.com>
>>> ---
>>>   drivers/scsi/hisi_sas/hisi_sas_main.c | 14 ++++++++++++
>>>   drivers/scsi/libsas/sas_expander.c    | 31 ++++++++++++++++++---------
>>>   include/scsi/libsas.h                 |  3 +++
>>>   3 files changed, 38 insertions(+), 10 deletions(-)
>>>
>>> diff --git a/drivers/scsi/hisi_sas/hisi_sas_main.c 
>>> b/drivers/scsi/hisi_sas/hisi_sas_main.c
>>> index 097dfe4b620d..40329558d435 100644
>>> --- a/drivers/scsi/hisi_sas/hisi_sas_main.c
>>> +++ b/drivers/scsi/hisi_sas/hisi_sas_main.c
>>> @@ -2031,6 +2031,19 @@ static int hisi_sas_write_gpio(struct 
>>> sas_ha_struct *sha, u8 reg_type,
>>>                   reg_index, reg_count, write_data);
>>>   }
>>> +static void *hisi_sas_alloc_smp_req(int size)
>>> +{
>>> +    u8 *p;
>>> +
>>> +    /* The address must be 16-byte-aligned. */
>>
>> ARCH_DMA_MINALIGN is not always 16, right ?
>>
>>> +    size = ALIGN(size, ARCH_DMA_MINALIGN);
>>> +    p = kzalloc(size, GFP_KERNEL);

Please make it clear that kmalloc() will return a naturally-aligned 
memory for power-of-2 sizes, and so ensuring that size is roundup to 16B 
will give a data which is aligned to 16B

>>> +    if (p)
>>> +        p[0] = SMP_REQUEST;
>>> +
>>> +    return p;
>>> +}
>>> +
>>>   static void hisi_sas_phy_disconnected(struct hisi_sas_phy *phy)
>>>   {
>>>       struct asd_sas_phy *sas_phy = &phy->sas_phy;
>>> @@ -2130,6 +2143,7 @@ static struct sas_domain_function_template 
>>> hisi_sas_transport_ops = {
>>>       .lldd_write_gpio    = hisi_sas_write_gpio,
>>>       .lldd_tmf_aborted    = hisi_sas_tmf_aborted,
>>>       .lldd_abort_timeout    = hisi_sas_internal_abort_timeout,
>>> +    .lldd_alloc_smp_req    = hisi_sas_alloc_smp_req,
>>
>> Why this complexity ? Why not simply modify alloc_smp_req() to have 
>> the required
>> alignment ? This will avoid a costly indirect function call.
> 
> Yeah, I think it's simpler to modify alloc_smp_req() directly too. 
> Yihang, Can you please cook a new one?
> 
> Thansk,
> Jason