ADCA.CTRLA = ADC_FLUSH_bm; // cancel any pending conversions, disable ADC

// set up exactly how Atmel did when they measured the calibration value
 ADCA.CTRLB = ADC_RESOLUTION_12BIT_gc; // unsigned conversion, produces result in range 0-2048
 ADCA.REFCTRL = ADC_REFSEL_INT1V_gc | ADC_TEMPREF_bm;
 ADCA.PRESCALER = ADC_PRESCALER_DIV512_gc;

ADCA.CH0.CTRL = ADC_CH_INPUTMODE_INTERNAL_gc | ADC_CH_GAIN_1X_gc;
 ADCA.CH0.MUXCTRL = ADC_CH_MUXINT_TEMP_gc;

ADCA.CTRLA |= ADC_ENABLE_bm;
 adc_wait_8mhz();

// get 358 K factory calibrated value
 ref = HW_read_signature_byte(PROD_SIGNATURES_START + TEMPSENSE0);
 ref += HW_read_signature_byte(PROD_SIGNATURES_START + TEMPSENSE1) << 8;
 kelvin_per_adc_x10 = ((273 + 85)*10) / (float)ref; // reference is ADC reading at 85C, scaled by 10 to get units of 0.1C

degrees_x10 = ADC_sample(&ADCA.CH0);
 ADCA.CTRLA = 0; // turn ADC off
 ADCA.REFCTRL = 0; // turn temperature sensor off

degrees_x10 *= kelvin_per_adc_x10;

degrees_x10 -= 2730;

return (degrees_x10);

The fans are loud. HP claim 22dB, but my basic meter rated the Microserver at 48dB from 1m. I swapped the 120mm fan for a quieter one which reduced it down to an acceptable level, but even so the PSU fan is far too loud to use as a media centre. Definitely the sort of thing you need to keep out of the living room and bedroom.

The PSU is also far less efficient than the review models. SPCR measured an idle power consumption of 25W with two hard drives. As shipping the Microserver comes with one 250GB HDD and idles at 52W. More than twice what the review samples used. This is a common trick – send review sites units with special high efficiency components but ship much cheaper parts to consumers. I confirmed that the PSU is at fault by trying a more efficient one which did manage to reach 27W at idle.

Other than these two serious issues the server is a very nice bit of kit. The CPU seems reasonably powerful, a bit quicker than my older Atom D510 despite that CPU having hyperthreading. Plenty of room for HDDs (you can actually have 6 if you use the 3.5″ bay and the external SATA) and fairly easy to work on. Seems solidly built. The unusual torx screw heads are a bit annoying and the multi-SATA cable is difficult to remove (necessary to slide the mobo out for RAM upgrades and the like). The GPU is powerful enough to do transcoding on the fly for DLNA. Some people have commented that the lack of a HDMI port makes it less useful as a media centre, but it is far too loud for that anyway.

The on-board network chip is a Broadcom NetXtream. It isn’t particularly good, strange for a machine basically built to serve files over a network. I can pull 60MB/sec over it, although CPU load is a bit high. There is no jumbo frame support, which isn’t as bad as it sounds because interrupt moderation makes it less important these days, but still… This is a budget chip. PCI-Express NICs are pretty cheap, but will of course add to the power consumption.

The BIOS is really basic. There is no fan speed control. Very little in the way of configurability for SATA operation, boot order and the like. The minimum shared memory for graphics is 32MB. The BIOS won’t boot with slow fans. SPCR says the cut-off is around 500 RPM, but I couldn’t go any lower than about 900 RPM.

Overall I’d say it’s a good product, if you are willing to accept the noise and either pay the higher electricity bills or shell out for a PicoPSU. As ever HP managed to take a good idea and almost ruin it.

Today I switched from using the Radeon 3200 based on-board graphics to a discrete Radeon 5300 card and the problem seems to have gone. It must be something about the HDMI sound output or otherwise related to the card. Very odd.

Ah! But we are closing all the loop holes too, they say. Okay, close them, and keep the rate at 50p as well. Just… fucking tax the bastards.

Sorry, that was uncalled for. They are not all bastards. I’m really upset with the betrayal.  You expect the Torys to be dicks, but the Lib Dems made me vote for them. To be fair I don’t think Charles Kennedy would have ever shacked up with Cameron, and I never voted for Clegg (admittedly due to tactical voting, but still my hands are relatively unsoiled).

1. Turn on receiver
2. Wait for IRQ to indicate data available
3. While RSSI bit in status word is set keep reading data as IRQ indicates it is coming in
4. Reset FIFO

The example code in the datasheet looks for an end-of-packet marker in step 3, which is also a workable option but has the disadvantage that the packet marker can never appear anywhere else in the packet. If your packets contain arbitrary data it’s possible an end-of-packet sequence like 0xAA55AA55 might appear in them, making those particular ones impossible to transmit. Instead I look for RSSI dropping and then decode what I have in the buffer independently of exactly how many bytes were received.

Step 4 is also rather critical. Without it you will keep getting IRQs signalling new data in the FIFO. The RFM12B picks up the preamble bytes which start the FIFO filling but does not seem to have any mechanism for stopping it again. You have to send the reset commands given in the datasheet or it will just keep receiving crap, even once RSSI disappears and it should be obvious that no new valid data is going to arrive.

unsigned char a,b,c,d; 
a = (unsigned char)((crc & 0x00FF) ^ data); 
b = (unsigned char)((a << 4) ^ a); 
c = (unsigned char)((b << 3) ^ (crc >> 8) ^ ((b & 0xFF) >> 4)); 
d = (unsigned char)(b ^ ((b & 0xFF) >> 5)); 
crc = (unsigned int)(((unsigned int)d << 8)|(c & 0xFF));

-Wl,--section-start=.text=$(BOOT_START),-e,boot,--defsym,zero_addr=0 -nostdlib

From http://old.nabble.com/Relocate-Data-into-bootloader-Section-td28511592.html