summary refs log tree commit diff stats
path: root/results/classifier/gemma3:12b/device/1247478
diff options
context:
space:
mode:
Diffstat (limited to 'results/classifier/gemma3:12b/device/1247478')
-rw-r--r--results/classifier/gemma3:12b/device/124747825
1 files changed, 25 insertions, 0 deletions
diff --git a/results/classifier/gemma3:12b/device/1247478 b/results/classifier/gemma3:12b/device/1247478
new file mode 100644
index 000000000..9ef07281d
--- /dev/null
+++ b/results/classifier/gemma3:12b/device/1247478
@@ -0,0 +1,25 @@
+
+usb passthrough mass storage write data corruption
+
+the windows 7 professional guest writes to usb high speed mass storage devices connected via host-libusb
+in bulk packages of either size 20480 or 4096 (as far as the actual file data is concerned and
+except for the last packet for odd-sized files). The pattern is:
+3 times bulk out 20480
+1 time bulk out 4096
+
+and that repeats for files longer than 65536 bytes.
+
+the file on the usb disk is corrupted and it is always corrupt in the last 4096 bytes of each
+20480 byte sized transfer. that means a file is corrupt at 16384-20480 and 36864-40960 and
+57344-61440.
+and so on. and because the 4096 sized  bulk out is always error free, the next corrupt span is from
+81920-86016.
+
+the last 4096 bytes of the 20480 sized transfer is always identical to the first 4096 bytes of the same
+transfer.
+
+to reproduce: run windows7 guest on and pass through usb2.0 disk with host-libusb. write a large file.
+(possibly check the bulk transfer sizes with usbmon).
+note that attaching usb disks with hw/usb/dev-storage does work just fine.
+cannot reproduce with linux as it always writes just 4096 bytes and writes with a linux guest are
+always ok even with usb passthrough.
\ No newline at end of file