linux-kernel - Re: [PATCH] scsi: pm8001: Fix phys_to_virt() usage on dma_addr

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: <eb82b0ab-912a-4879-f1b2-d5cbef5bfc41@huawei.com>
Date:   Fri, 10 Dec 2021 10:44:15 +0000
From:   John Garry <john.garry@...wei.com>
To:     <Ajish.Koshy@...rochip.com>, <jinpu.wang@...ud.ionos.com>,
        <jejb@...ux.ibm.com>, <martin.petersen@...cle.com>
CC:     <Viswas.G@...rochip.com>, <linux-scsi@...r.kernel.org>,
        <linux-kernel@...r.kernel.org>,
        <Vasanthalakshmi.Tharmarajan@...rochip.com>
Subject: Re: [PATCH] scsi: pm8001: Fix phys_to_virt() usage on dma_addr_t

On 10/12/2021 10:23, Ajish.Koshy@...rochip.com wrote:
> Hi John,

Please config your editor to wrap at ~72 characters and don't top-post.

> 
> In my humble opinion what we observed very earlier was with respect smp_request()/response() crash and this patch resolved it. Given that the issue was not only specific to ARM, issue was observed on x86 too with IOMMU enabled. Device discovery went fine post application of this patch on x86.
> 
> What we are observing right now on error handling/timeouts for commands on drives will be altogether different issue that needs separate debugging on ARM platform with separate patch since this is a very initial execution of pm80xx driver on ARM platform.
> 
> This patch is acceptable. Let me know your further views.
> 

As I mentioned earlier about this patch, a v2 is needed as we need 
kmap_atomic() in both cases. But I would rather resolve all issues 
before getting that merged - it has been broken this way for a long time.

So I mentioned a new issue in my response to Damien - maxcpus=1 on the 
command line crashes on boot. I can imagine that x86 also has that issue.

As for the timeouts, why would the FW not respond in the way I 
described? I would always expect a completion, even for erroneous 
commands. Maybe it is an interrupt issue in the driver. Is there some 
diagnostics I can check - there seems to be a lot of "stuff" in the 
sysfs scsi_host folder.

Thanks,
John

> Thanks,
> Ajish
> 
> On 26/11/2021 15:35, John Garry wrote:
>>        /*
>> @@ -4280,8 +4283,9 @@ static int pm80xx_chip_smp_req(struct pm8001_hba_info *pm8001_ha,
>>                pm8001_ha->smp_exp_mode = SMP_INDIRECT;
>>
>>
>> -     tmp_addr = cpu_to_le64((u64)sg_dma_address(&task->smp_task.smp_req));
>> -     preq_dma_addr = (char *)phys_to_virt(tmp_addr);
>> +     smp_req = &task->smp_task.smp_req;
>> +     to = kmap(sg_page(smp_req));
> This should be a kmap_atomic() as well, as I see the following for when
> CONFIG_DEBUG_ATOMIC_SLEEP is enabled: