qemu-devel
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Qemu-devel] [PATCH 3/3] megasas: LSI Megaraid SAS emulation


From: Alexander Graf
Subject: Re: [Qemu-devel] [PATCH 3/3] megasas: LSI Megaraid SAS emulation
Date: Sat, 2 Jul 2011 14:28:04 +0200


Am 02.07.2011 um 15:50 schrieb Hannes Reinecke <address@hidden>:

> On 07/01/2011 06:42 PM, Alexander Graf wrote:
>> 
>> On 01.07.2011, at 17:35, Hannes Reinecke wrote:
>> 
>>> This patch adds an emulation for the LSI Megaraid SAS 8708EM2 HBA.
>> 
>> Have you tried to execute the current version of megasas and actually
> > do something with it? I just booted up openSUSE 11.4 rescue from DVD
> > with a megasas adapter that contained a raw file backed by tmpfs.
> > Creating a partition worked fine, but when running mkfs.ext3 and
> > mounting afterwards, the mount fails saying there is no ext3 on the disk.
>> 
>> Sounds like data corruption to me :). I know that this used to work
> > a while back, so it might be a regression recently?
>> 
> It worked here, in the sense that I've booted up an existing openSUSE 11.4 
> installation. But I wouldn't be surprised if the degradation to use 
> bounce-buffers has some flaws.
> 
> My guess here is that we have problems when the transfersizes larger as the 
> internal bounce buffer.

Ah, might be a good idea to have some check for that. Just assert() out when 
the bounce buffer is too small - or at least tell the user about it. Can the 
iov functions tell you if you exceed the write/read side?

> 
> (I probably should be putting in some more references to 'bounce buffers' 
> here to alert people that using bounce buffers in SCSI is the best way of 
> killing performance)
> (And no, I will not getting into another dog-fight with Paul B. here.
> Virtio can do without bounce buffers. AHCI can. So I fail to see why SCSI has 
> to rely on bounce buffers.)

My hope is that by approaching this incrementally, we can move to a 0-copy 
approach later on. But for now, let's focus on getting the device emulation 
part well working an in.

> 
> But enough of this, Yeah, bugfixing is needed here. I see what I can do.

Thanks :)

Alex

> 



reply via email to

[Prev in Thread] Current Thread [Next in Thread]