performance improvements for the sglist API (Re: [PATCH] scsi: virtio-scsi: Fix address translation failure of HighMem pages used by sg list)

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

 



Il 25/07/2012 16:36, Boaz Harrosh ha scritto:
>> > 
>> > I did test the patch with value-assignment.
>> > 
> 
> Still you should use the sg_set_page()!!
> 1. It is not allowed to directly manipulate sg entries. One should always
>    use the proper accessor. Even if open coding does work and is not a bug
>    it should not be used anyway!
> 2. Future code that will support chaining will need to do as I say so why
>    change it then, again?

Future code that will support chaining will not copy anything at all.

Also, and more important, note that I am _not_ calling sg_init_table
before the loop, only once in the driver initialization.  That's because
memset in sg_init_table is an absolute performance killer, especially if
you have to do it in a critical section; and I'm not making this up, see
blk_rq_map_sg:

                          * If the driver previously mapped a shorter
                          * list, we could see a termination bit
                          * prematurely unless it fully inits the sg
                          * table on each mapping. We KNOW that there
                          * must be more entries here or the driver
                          * would be buggy, so force clear the
                          * termination bit to avoid doing a full
                          * sg_init_table() in drivers for each command.
                          */
                          sg->page_link &= ~0x02;
                          sg = sg_next(sg);

So let's instead fix the API so that I (and blk-merge.c) can touch
memory just once.  For example you could add __sg_set_page and
__sg_set_buf, basically the equivalent of

    memset(sg, 0, sizeof(*sg));
    sg_set_{page,buf}(sg, page, len, offset);

Calling these functions would be fine if you later add a manual call to
sg_mark_end, again the same as blk-merge.c does.  See the attached
untested/uncompiled patch.

And value assignment would be the same as a

    __sg_set_page(sg, sg_page(page), sg->length, sg->offset);

> Please don't change two things in one patch. The fix is for high-pages
> please fix only that here. You can blasphemy open-code the sg manipulation
> in a separate patch.

The blasphemy is already there (the scatterlist that is handed to virtio
won't have the right end-of-chain marker).  If anything,
value-assignment is trading a subtle blasphemy for a blatant one.
That's already an improvement, but let's just fix the API instead.

Paolo
--
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[Index of Archives]     [SCSI Target Devel]     [Linux SCSI Target Infrastructure]     [Kernel Newbies]     [IDE]     [Security]     [Git]     [Netfilter]     [Bugtraq]     [Yosemite News]     [MIPS Linux]     [ARM Linux]     [Linux Security]     [Linux RAID]     [Linux ATA RAID]     [Linux IIO]     [Samba]     [Device Mapper]
  Powered by Linux