DOS ain't dead

Hi,
I found that my favourite USBASPI.SYS from Panasonic doesn't longer work on new MB with P67 chipset. The difference seems to be that P67 is pure USB 2.0 system with EHCI only compared to ICH7 that exposed also UHCI. USBASPI.SYS already supports EHCI but it failed to detect USB targets. It finds just hubs, nothing else. Seems that hub makes a firewall that cannot be passed. I tested only backpanel USB ports but probably it's screwed on all ports.
It reports:

Loading: C:\DOS\PLUS\USBASPI.SYS

ASPI Manager for USB mass-storage  Version 2.27
 (C)Copyright Panasonic Communications Co., Ltd. 2000-2008

    Controller  : 00-26-0 VID=8086h PID=1C2Dh (1458h-5006h) EHCI
                :         MEM=FBFFE000h-FBFFE3FFh(1024Bytes)
    Controller  : 00-29-0 VID=8086h PID=1C26h (1458h-5006h) EHCI
                :         MEM=FBFFD000h-FBFFD3FFh(1024Bytes)
    ... Initializing Host Controller 1/2 ...

    ... Initializing Host Controller 2/2 ...

    ... Scanning USB Devices ...

    USB Device  : VID=8087h PID=0024h HS(HUB)
    ... Scanning USB Devices ...

                : VID=8087h PID=0024h HS(HUB)
    ... Scanning USB Devices ...

    ERROR : Target USB device not found.

1 char device installed.
Driver loaded.



USBHOSTS 0.06, (C) 2007-2009, Bret E. Johnson.

       PCI BUS
      ÍÍÍÍÍÍÍÍÍÍ                           BASE           USB DRIVER
      I   B  D F                      BASE   PHYSICAL   ÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍ
HOST  d   u  v n              USB IRQ I/O    MEMORY     HST             BW
TYPE  x   s  c c  VENDR PROD  VER NUM ADDR   ADDRESS    IDX   STATUS   USED
ÍÍÍÍ  Í ÍÍÍ ÍÍ Í  ÍÍÍÍÍ ÍÍÍÍÍ ÍÍÍ ÍÍÍ ÍÍÍÍÍ ÍÍÍÍÍÍÍÍÍÍ        ÍÍÍ ÍÍÍÍÍÍÍÍÍÍ ÍÍÍÍ
EHCI  0   0 26 0  8086h 1C2Dh 2.0  10 ±±±±± FBFF_E000h     ±±± ±±±±±±±±±± ±±±±
                  Intel Corp

EHCI  1   0 29 0  8086h 1C26h 2.0   3 ±±±±± FBFF_D000h     ±±± ±±±±±±±±±± ±±±±
                  Intel Corp


I also tried USBMASS/386 and Iomega ASPIEHCI.SYS but without any success. Is there any working solution for newer chipsets? Bret, didn't you planned EHCI driver? :) BTW what about AMD, did they dropped UHCI too? This suxx. But BIOS USB disk support seems to work well and I never used other USB device under DOS so it's not big problem for me. But for future we would need some EHCI/XHCI solution...

---
DOS gives me freedom to unlimited HW access.

What about Georg Potthaust USB drivers?
They work with EHCI controler. Or maybe older driver DUSE49.EXE
If google will not help I can provide it.

---
DOS-u-akbar!

EHCI itself is not a problem but the hubs behind it, probably... I didn't take DOSUSB in mind as it's commercial SW but good to remember, I'll check if there is some new and free-trial version.

> What about Georg Potthaust USB drivers?
> They work with EHCI controler. Or maybe older driver DUSE49.EXE
> If google will not help I can provide it.

---
DOS gives me freedom to unlimited HW access.

Some Intel motherboards do not have OHCI/UHCI any more and just provide EHCI hubs. There is just one sentence in the ehci specs that allows to do that and it was also a surprise to me.

I implemented support for that in DOSUSB.

Georg

Yes, there is a demo version which will run for about 15 Minutes. Usually enough to transfer your files.

> EHCI itself is not a problem but the hubs behind it, probably... I didn't
> take DOSUSB in mind as it's commercial SW but good to remember, I'll check
> if there is some new and free-trial version.
>
> > What about Georg Potthaust USB drivers?
> > They work with EHCI controler. Or maybe older driver DUSE49.EXE
> > If google will not help I can provide it.

> Hi,
> I found that my favourite USBASPI.SYS from Panasonic doesn't longer work on
> new MB with P67 chipset. The difference seems to be that P67 is pure USB
> 2.0 system with EHCI only compared to ICH7 that exposed also UHCI.
> USBASPI.SYS already supports EHCI but it failed to detect USB targets. It
> finds just hubs, nothing else. Seems that hub makes a firewall that
> cannot be passed. I tested only backpanel USB ports but probably it's
> screwed on all ports.


There are several others here that you might have success with.

http://glennmcc.org/download/usb-dos/

---
--
http://glennmcc.org/

> Some Intel motherboards do not have OHCI/UHCI any more and just provide
> EHCI hubs. There is just one sentence in the ehci specs that allows to do
> that and it was also a surprise to me.
> I implemented support for that in DOSUSB.

Yes I just readed it in your PDF docs, you mention intel P55 but seems that all new intel-based MB's do that way. I downloaded and tried both yours DOSUSB 2.0 and 3.0 but none works.
DOSUSB 2.0 looked promissing, it detected EHCI and also the mass storage device but when I accessed drive letter (just typing D:) it hanged. I tried with and without himem.sys.

DOSUSB driver by G. Potthast - Rel. 2.0beta - Copyright 2010 - Not Registered

-> Time restricted Demo Version <-
Scanning usb ports ...

EHCI controller installed:
Bus Dev Func I/O      Devices
00/001A/0000 FBFFE000 Port00: device at high speed Port01: no device
00/001D/0000 FBFFD000 Port00: device at high speed Port01: no device

FBFFE000 01 H Unidentified device
     01 is a Hub with 06 ports
     02 H Generic - Mass Storage
FBFFD000 03 H Unidentified device

     03 is a Hub with 08 ports

DOSUSB driver Rel. 2.0 installed at int  0x65 - Demo Version

here's log http://rayer.g6.cz/1tmp/DOSUSB20.LOG
I can read valid USB descriptor of the flashdisk via USBVIEW!
I loaded USBDISK.SYS before via devload.

DOSUSB 3.0 wrongly detected non-existing UHCI even with /e switch

DOSUSB driver by Georg Potthast - Rel. 3.0 - Copyright 2011     
                                                               
-> Time restricted Demo Version <-                               
Scanning USB ports ...                                         
                                                               
UHCI controller installed:                                     
Bus Dev Func I/O      Devices                                   
00/0000/0000 00000000 PortF0: no device                         
                                                               
DOSUSB driver Rel. 3.0 installed at int  0x65 - Demo Version

and the log is quite empty, just only "DOSUSB resident now"

So I doubt that any existing DOS driver will work as this intel approach is new and incompatible. The Panasonic USBASPI.SYS was one of the latest developed drivers...

---
DOS gives me freedom to unlimited HW access.

> EHCI itself is not a problem but the hubs behind it, probably...

That seems to be the case. I just got a new computer a couple of weeks ago, and am having the same problem. It's EHCI/XCHI only, and I can see the EHCI hosts and the integrated hubs, but don't see anything attached to the hubs. I haven't gotten around to doing any troubleshooting yet, but apparently Intel has changed something about how the hubs work.

If I understand well, on older chipset there was no hubs between UHCI and ports? Or does the EHCI hub work differently than on UHCI? Would it be hard to implement the support of EHCI hub?
BTW what new machine do you have? I have currently i7 2600K that arrived from ebay/Germany yesterday. I just tested my FFMPEG port (single thread) and at the same freq. the i7 is just sligtly (few %) faster than my current C2D. It has better O'C potential but seems that turbo works only in ACPI systems, under DOS it always run at stock 3,4GHz even if turbo multiplier MSR is set higher. Of course I don't need ultra power under DOS, just curious how turbo works...

---
DOS gives me freedom to unlimited HW access.

> Of course I don't need ultra power under DOS, just
> curious how turbo works...

http://en.wikipedia.org/wiki/Turbocharger

---
--
http://glennmcc.org/

> I have currently i7 2600K that arrived from ebay/Germany yesterday.

Skylake?? What are the family/model/stepping?

> I just tested my FFMPEG port (single thread) and at the same freq.
> the i7 is just sligtly (few %) faster than my current C2D.

Yes, they've slightly increased performance. Even moreso with recompile.
What DJGPP options did you use?

> It has better O'C potential but seems that turbo works only in ACPI
> systems, under DOS it always run at stock 3,4GHz even if turbo multiplier
> MSR is set higher. Of course I don't need ultra power under DOS, just
> curious how turbo works...

Not sure. I thought I'd read it needs P0 power state, but maybe your motherboard must also be conforming. Dunno.

https://en.wikipedia.org/wiki/Intel_Turbo_Boost

> > I have currently i7 2600K that arrived from ebay/Germany yesterday.
>
> Skylake?? What are the family/model/stepping?

(i7 2600k is the topmodel of consumer Sandy Bridge)

Maybe the only solution will be the additional USB-UHCI card into PCI slot...

---
DOS-u-akbar!

> > > I have currently i7 2600K that arrived from ebay/Germany yesterday.
> >
> > Skylake?? What are the family/model/stepping?
>
> (i7 2600k is the topmodel of consumer Sandy Bridge)

(half-joking) Marco, why didn't you warn him that he was buying obsolete tech? It's from 2011! 32nm is bloated! Sure, it's probably cheaper, but you're paying for it in the long run. It doesn't even fully support AVX-512!

Anyways, don't blame me for not memorizing all the model numbers. Wikipedia does say "i7 6700k" is Skylake (Turbo Boost-able from 4.0 Ghz to 4.2, single core, while his ancient Sandy Bridge only goes from 3.4 to 3.8).

> If I understand well, on older chipset there was no hubs between UHCI and
> ports?

There could be hubs, but the weren't required. The way it works is that EHCI (USB 2.x) and OHCI/UHCI (USB 1.x) are set up as what they call companion controllers. EHCI can ONLY directly handle high-speed (USB 2.x) devices.

In older systems, when a device is first plugged in, the EHCI controller is in charge of the port and tests to see if the device is even capable of high-speed, and if it is the EHCI controller maintains control. If the device can only do full- or low-speed, the EHCI controller disconnects the port and passes control to the companion (UHCI/OHCI) controller which ONLY handles full- and low-speed devices.

A USB 2.x Hub, on the other hand, can do all three speeds on its downstream ports. Inside a USB 2.0 hub is something called a Transaction Translator, which is sort of like a mini Host Controller (sort of like EHCI and UHCI/OHCI in one box). In USB 2.0, the EHCI host controller talks the the USB 2.0 Hub Transaction Translator at high-speed, and then the hub does all of the speed conversions to send data to each of the downstream ports at the appropriate speed (high - 480 Mbps, full - 12 Mbps, or low - 1.5 Mbps).

Newer systems don't have companion (UHCI/OHCI) controllers at all, but still need to be able to talk to low- and full-speed devices (which EHCI can't do, at least not directly). So, instead of companion controllers they have an integrated hub (sometimes it's called a rate-matching hub, but all USB 2.x hubs are rate-matching so that name doesn't really make sense). The integrated hub on the new computers really don't work any different than regular USB 2.x hubs, except that the integrated hubs aren't removable. In addition, you can't attach anything directly to the host contoller ports like you could before because they aren't physically exposed -- the only thing attached to the host controller is the integrated hub, so you can only attach external devices to the integrated hub.

In the USB specifications, there are pretty tight restrictions on how hubs are supposed to work, so I suspect the reason the new Intel hubs don't seem to work right is probably something simple/subtle -- they couldn't have changed things very much and have it still be compliant with the USB standards. Like I said, I haven't had time to do any troubleshooting, but I don't thing it's going to be anything complicated to fix (assuming I can figure out exactly what the problem is).

> Would it be hard to implement the support of EHCI hub?

I have EHCI and USB 2.0 hubs and other things basically working, but am still doing lots of tweaks and updates to lots of different items as I find spare time (spare time is laughable concept these days). This Intel hub thing is yet another tweak on the list of things to do. Still not ready for an official release yet.

> BTW what new machine do you have? I have currently i7 2600K ...

My new computer has a Gigabyte motherboard with an I5 CPU. From a USB perspective, though, the CPU doesn't matter. It's all of the supporting chipsets on the motherboard that make the difference. The new computer also has an SSD, which makes a HUGE difference in speed -- just about makes disk caching in DOS a pointless endeavor.

> > > > I have currently i7 2600K that arrived from ebay/Germany yesterday.
> > >
> > > Skylake?? What are the family/model/stepping?
> >
> > (i7 2600k is the topmodel of consumer Sandy Bridge)
>
> (half-joking) Marco, why didn't you warn him that he was buying obsolete
> tech? It's from 2011! 32nm is bloated! Sure, it's probably cheaper, but
> you're paying for it in the long run. It doesn't even fully support
> AVX-512!

(neither does consumer Skylake to my best knowledge. Only the -E to be released in an year or so is rumoured to have it)

He said he bought it from ebay, so I assumed an economic metric (performance/$) rather than a technical or environmental one.

> Anyways, don't blame me for not memorizing all the model numbers. Wikipedia
> does say "i7 6700k" is Skylake (Turbo Boost-able from 4.0 Ghz to 4.2,
> single core, while his ancient Sandy Bridge only goes from 3.4 to 3.8).

I benchmarked skylake against its cousin i7 2500, and don't expect miracles from the CPU. Most changes nowadays are in the GPU, the rest is just a few percent per generation at best.

Thanks for the detailed description of USB topology.
I'm also busy and have some unfinished updates to my tools waiting an years for final release :) As I use USB under DOS only for mass storage I can use BIOS service and don't see this issue as urgent problem. But for future it would be nice to make it working as there will be most of MBs with this new topology. I don't want to bother you more with it. Mayebe if Georg would like to do some testing of his DOSUSB I'm ready to assist him.

> My new computer has a Gigabyte motherboard with an I5 CPU. From a USB
> perspective, though, the CPU doesn't matter. It's all of the supporting
> chipsets on the motherboard that make the difference.

Yes, CPU doesn't matter on USB stuff, I just asked by the way. And what chipset? I have P67 and also H55 at work available for testing.

> The new computer also has an SSD, which makes a HUGE difference in speed -- > just about makes disk caching in DOS a pointless endeavor.

Yeah, I use SSD Samsung 840 Pro for ~3 years and it was the best upgrade experience since stepping from Celeron Tualatin to C2D and it didn't cost a lot (comparing to complete replacement of CPU+MB+RAM+VGA...). Maybe Rux. wouldn't agree and will adore latest Skylake as the only way to upgrade but from my point of view intel CPU evolution slowed down last years a lot and if your programs are not heavily optimized to use newest instruction extension you wouldn't recognize significant speed difference (just some % at same freq.) so I see far much reasonable to spend money on SSD than on latest CPU model. In my case I got my CPU for ~40% of price that similar Haswell model is selling now, so I'm happy with it. BTW can Skylake easily overclock at 4,8-5GHz? :)

---
DOS gives me freedom to unlimited HW access.

> He said he bought it from ebay, so I assumed an economic metric
> (performance/$) rather than a technical or environmental one.

Yes, I replied this here
http://www.bttr-software.de/forum/forum_entry.php?id=14549

> I benchmarked skylake against its cousin i7 2500, and don't expect miracles
> from the CPU. Most changes nowadays are in the GPU, the rest is just a few
> percent per generation at best.

Yes, I would expect that. You can get maybe +20% in special optimised benchmark but maybe only 2% in common apps. I think intel had reached the dead end with current silicon technology. I would expect they will try to increase number of cores (at least in i7 line) but we still stucked at 4. Also the promising 32->22->14->10nm shrinking would not make such energy saving as it was expected, even AFAIK at higher clock rates the tiny litography perform worse than older (faster rise W/MHz). As you told Intel aimed on its GPU performance which is getting better and better (also it increase the % of used silicon beside CPU core).

---
DOS gives me freedom to unlimited HW access.

Btw, my motivation to buy and test was the memory bandwidth (I've been asked to bid on a memory intensive project ). The increase was about in the expected magnitude.

Hi RayerR,

USB 3.0 does XHCI only, no EHCI, OHCI and UHCI.

Did you try to load usbdisk.sys with the /D2 command line parameter to specify the address of the USB disk? I see from your log that the flash disk got this address.

Georg

> Did you try to load usbdisk.sys with the /D2 command line parameter

Yes I did try with and without /d2 - no difference.
Flashdisk has one FAT16 partition 2GB, under MS-DOS 6.22 and 7.1

---
DOS gives me freedom to unlimited HW access.

> Btw, my motivation to buy and test was the memory bandwidth...

Yes, it's the biggest architecture leap - integrating MCH (the idea that was pioneered by AMD on x86 years before intel) and later integrating PCIE. Even 1st core generation (Nehalem) offers 2x or more mem.BW than C2D/DDR2. With a single standard DDR3 module I get ~9000MB/s in MemTest while only ~4500MB/s on C2D/P31 dualchannel. So I think it would help me a lot in photoediting SW...

---
DOS gives me freedom to unlimited HW access.

Well, it has to be /D2 and not /d2. Please use uppercase.

Georg

> > Did you try to load usbdisk.sys with the /D2 command line parameter
>
> Yes I did try with and without /d2 - no difference.
> Flashdisk has one FAT16 partition 2GB, under MS-DOS 6.22 and 7.1

> I get ~9000MB/s in MemTest while only
> ~4500MB/s on C2D/P31 dualchannel.

(read, write or read/write ? Read/write is half the bandwidth obviously)

I think I got a 28MB image in slightly less than 3ms on skylake, so that would be 9333MB/s for a move.

Note that this was not really a fat one though. Just an average joe, but two matched dimms. (yes this is a case where this is measurable)


Yesterday in a kind of pre-blackfriday sale I bought a Kaveri (A10-7850K) to upgrade my "spare" machine, and tested it today, it did about 5ms (with DDR3-1600) which is about 5600MB/s in the same test.

Of note that one had to manually enable the usage of 1600 XMP memory in the UEFI bios.

> (read, write or read/write ? Read/write is half the bandwidth obviously)

I know that R/W matters, but I don't know what method Memtest use. It was just relative comparison of 2 systems. I will run mbench that makes detailed test and measure also the latency. Here's my last tuned result of my C2D/DDR2-2CH system:

Intel Pentium III (CPUID = 67a) @ 3553.0MHz (FSB:MEM = 2,4x)
Instruction set support : MMX SSE SSE2

Access latency                   65.9 ns (234 clocks)

Read datarate (INT)             6455 Mb/s
Write datarate (INT)            2818 Mb/s

Read datarate (MMX)             7182 Mb/s
Write datarate (MMX)            2818 Mb/s

Read datarate (SSE)             7242 Mb/s
Write datarate (SSE)            7627 Mb/s


But first I will need to make mounting for this little baby :) I don't have the right backplate and I really don't like intel box style plastic mounts. So I get a piece of 5mm Al and do some handwork... :)

> Yesterday in a kind of pre-blackfriday sale I bought a Kaveri (A10-7850K)
> to upgrade my "spare" machine, and tested it today, it did about 5ms (with
> DDR3-1600) which is about 5600MB/s in the same test.

Hm, I don't have any experiences with newer AMD machines but I would expect similar BW to intel from DDR3 - seems not...

> Of note that one had to manually enable the usage of 1600 XMP memory in the
> UEFI bios.

I didn't mess with XMS yet as I temporarly have some basic module at 1333, waiting to buy something better. If I understand well, XMS is another timing profile beyond JEDEC basic that is recorded in SPD EEPROM of the module and you can activate it easily instead of setting all the timings manually, right? If there is a datasheet from manufacturer it would't be problem to set timings manually according it.

---
DOS gives me freedom to unlimited HW access.

Well, I tried again with /D2 and it seemed that it hanged again but I wait a while and after about a half minute it finally acessed the drive. I wouldn't expect such delay. But then copying files from flash went fast according to USB 2.0 speed so it works but only with himem.sys - it doesn't like JEMM or EMM386 (in this case it hang forever).

---
DOS gives me freedom to unlimited HW access.

> ... from my point of view intel CPU evolution slowed down last years a lot and
> if your programs are not heavily optimized to use newest instruction
> extension you wouldn't recognize significant speed difference (just some %
> at same freq.).

Yes, the "rules of thumb" on how to upgrade have changed a lot over the years, especially in the last decade or so. CPU technology has basically maxed out with clock speeds in the 3~4 GHz range, and things only get significantly faster with multiple cores and/or multiple CPU's. Of course, you need to write special new software to take advantage of the multiple-CPU technology, which doesn't help with current DOS software (or older software of any sort). Adding more memory doesn't help with DOS either, since DOS can't even take full advantage of the already standard GB's of memory that come with new computers.

They're working on newer CPU technologies that may have some promise, but they're still a long ways off and may not really work all that well in the end anyway. If speed is your primary concern, an SSD might be a pretty good way to upgrade right now (maybe better than any other option for the next several years, anyway).

So there is a USB driver that works with your board. Yes, DOSUSB does not work in protected mode even if EMM386 puts the PC into protected mode.

> Well, I tried again with /D2 and it seemed that it hanged again but I wait
> a while and after about a half minute it finally acessed the drive. I
> wouldn't expect such delay. But then copying files from flash went fast
> according to USB 2.0 speed so it works but only with himem.sys - it doesn't
> like JEMM or EMM386 (in this case it hang forever).

> If speed is your primary concern, an SSD might be a pretty
> good way to upgrade right now (maybe better than any other option for the
> next several years, anyway).

Yes sure, as I wrote I'm using Samsung 840 Pro for near 3 years and it was the best upgrade I did since upgrade to C2D :)

---
DOS gives me freedom to unlimited HW access.

I just bought a pair of Kingston HyperX 1866 CL11 and they give me ~2x more performance than common single-channel 1333 module. Also it's big leap compared to DDR2.

Intel Xeon III (CPUID = 6a7) @ 3410.0MHz (i7-2600K, 2-ch, 1866MHz, CL11)
Instruction set support : MMX SSE SSE2

Access latency                   44.0 ns (150 clocks)

Read datarate (INT)             16030 Mb/s
Write datarate (INT)            11332 Mb/s

Read datarate (MMX)             17936 Mb/s
Write datarate (MMX)            11034 Mb/s

Read datarate (SSE)             19972 Mb/s
Write datarate (SSE)            22901 Mb/s


---
DOS gives me freedom to unlimited HW access.