]> ###2026.09.19### **Changed** - HBAviewer moved off its own top-level **Tools** entry into **Tools → Disk Utilities**, alongside the other disk tools. The old entry was a body-less stub that only created a separate group; the Monitor and Firmware pages now join the shared `DiskUtilities` menu instead. (PR #23, joshmans) ###2026.08.23.3### **Fixed** - **Every tab was slow, and now is not.** Reading a controller enumerated the storcli controller list twice, back to back, for the same number -- once to find out the tool worked and again to ask how many cards there are. Every tab goes through that path. On a 9300-16i a Health read measured 56 seconds before and 2-4 seconds after. Boxes with a slow enumeration will notice this most; nothing about what is displayed has changed. - **The plugin on a phone.** Tab labels ran outside the buttons the Unraid theme draws around them, and the Overview and HBA Health cards were wider than the screen, so the page scrolled sideways and the right-hand half of every card sat off the edge. Three separate causes, all fixed: the tab strip now scrolls instead of squeezing, the card grids collapse below their column minimum, and the dual-controller card is sized by its border box rather than overflowing by its own padding. Long values wrap instead of running off, and a phone gets a narrower page gutter. Desktop is unchanged. (issue #21, reported by jac2424) **Added** - **Track link-error history** on the Settings page, off by default. Samples PHY counters every ten minutes so HBA Health can judge trends rather than comparing whatever two moments you happened to open the tab. Previously this sampling only ran if you had Notifications switched on, which is not something anyone would guess: the two are separate features and now have separate switches. With both off, the cron still does one small file read and touches no hardware. **Internal** - The health ring is written atomically, so a cron sample landing mid-render can no longer silently discard the accumulated history. - `docs/review-policy.md` records which agent review tool governs which layer and which findings are rejected on sight, and `tests/review_policy_test.php` fails if the code it describes is renamed away. Three claims it made about the code were verified wrong and corrected. - The install verifier proves WHICH build it checked by diffing the package against the installed tree; it had reported PASS twice against a stale tarball. Codacy removed. ###2026.08.23.2### **Added** - Tabs with nothing to show now say why. The Event Log in particular: it used to print "No log entries" even when entries WERE archived on /boot but had been written by a different backend, so a box that switched backend was told its history was gone. It now says how many are held and why they cannot be displayed. - The Overview says when a controller read did not finish, instead of showing "Reading controller information..." for as long as the tab stays open. A card or tool that stops responding now looks like a failure rather than like a slow plugin. **Fixed** - A background read that wrote anything to stderr -- a shell notice, a storcli message -- could corrupt its own cached result and surface as "Backend unavailable" even though the read had succeeded. stderr is kept separately now. - **HBA 9500 owners:** the bundled firmware index listed 28.00.00.00 as newest for the 9500-8i and recorded the board as having no legacy option ROM. A real card reported firmware 31.00.00.00 and a BIOS of 09.27.00.00_14.00.00.00, so both were wrong. A 9500-8i on current firmware now reads as current instead of "newer than index". The 9500-16i is unchanged -- no 16i has been read here, and this index only records what has actually been seen. **Internal** - Every plan in `docs/superpowers/` now states whether it shipped, ARCHITECTURE.md covers the files added over the last week, and the release procedure documents the trap that `main` moves on its own when a release patches the plugin file. ###2026.08.23.1### **Fixed** - The dashboard tile's cold-start message printed `Reading controller information\u2026` instead of an ellipsis. Cosmetic, but it showed on every reboot until the first controller read completed. Thanks to the user who reported it. ###2026.08.23### **The dashboard tile could freeze the whole webGui.** It read the controller synchronously while Unraid rendered its own Dashboard page, so once a minute -- when the reader's 60-second cache expired -- a PHP worker was held for as long as the card took to answer and everything else queued behind it. Reported as ~10 seconds of an unresponsive server, once a minute, on every install that had the tile. It reads through the plugin's existing non-blocking cache now and can never wait on hardware again; a test fails if that call is ever put back. **Added** - **Temperature unit** (Settings): °C or °F, default °C. Contributed by [Josh Mans](https://github.com/joshmans) as PR #22. Every sensor, stored value and threshold comparison stays in Celsius -- only the printed number changes. - **Sortable tables.** Every column header on Drives, SMART, PHY and the Event Log sorts, ascending then descending, by click or keyboard. Capacities sort by real size, so 9.095 TB comes before 12.733 TB. - **Unassigned drives are named.** A drive mounted by the Unassigned Devices plugin shows its mount label (`media1`, `media9`) in the Unraid column and on the Array Map, instead of an em dash. On a box with no array that column was previously empty on every row. - **One HBA is one dashboard tile.** A dual-IOC board (SAS9300-16i and similar) reported as two tiles; it now shows as one, with a section per controller carrying its own temperature, address, drive count and health -- matching what the Overview tab has always shown. **Fixed** - Drives whose serial contains a space -- several HGST SAS models do -- showed no `/dev` name on the Drives tab and no name in the Array Map tray, while SMART named them correctly. Four of nine drives on the reporting box. - Long session URLs on the Settings page ran out through the right edge of the Export/API card. - The dashboard tile showed one controller's PCI address as though it belonged to the board. **Accessibility** The plugin had no visible focus ring anywhere, and two inputs actively removed the browser's. Bay cells and tray chips were plain `div`s with click handlers, so the Array Map could not be operated by keyboard at all -- and emptying a bay was double-click only, with no keyboard equivalent. The tab strip announced itself to a screen reader as ten unrelated buttons. - Focus rings on everything the keyboard can reach. - The Array Map is fully keyboard-operable: Enter picks up and places a drive, Delete empties a bay, and focus stays where you were after the grid redraws. - The tab strip is a proper tab list with arrow-key movement. - Small controls meet the 24px pointer-target minimum. - The two warning/gear glyphs are now SVG. As text they took *emoji* presentation on Windows and Android, which meant the warning sign beside the firmware flasher was rendered in the font's own colour and could not be made to look like a warning. **Internal** Design rules the CSS already followed are written down in `design-system/MASTER.md`, and the colour tokens, the button, and the icon sprite each have one home instead of two or three copies that had already drifted apart. `docs/foreground-reads.md` records which code may block a request, which is the rule whose absence caused the freeze above. ###2026.08.17### SAS4 / 9600-series (`mpi3mr`) cards are now read instead of being named and refused. Ported from techanonymous's `Unraid-HBAviewer-sas4` fork (MIT, commit 882f88c) — the backend, the parsers, the Settings-page install helper and the StorCLI2 installer script are their work, adapted into this repo's existing structure and tests. **Important:** no 9600-series card has run this port on this repo's own hardware. techanonymous verified their original code on their own 9600-24i; that is a real and useful data point, but it is not the same claim as this repo having tested it. Treat this release as beta for SAS4 owners and report back on [issue #19](https://github.com/FugginOld/Unraid-HBAviewer/issues/19). **Added** - Every tab except the firmware Event Log works against the Lite StorCLI2 that the dkaser/unraid-storcli plugin ships. - The Event Log needs Broadcom's FULL StorCLI2, which is proprietary and not downloadable automatically (its page is Cloudflare-gated with no stable URL). `scripts/install_storcli2.sh` does everything after you fetch that archive yourself: unpack, install to `/opt` (RAM on Unraid), and restore it at boot since the flash is FAT32 and cannot keep the execute bit. Backed up, idempotent, and driven from Settings → HBA Connection → Firmware Event Log. ###2026.08.16### Every HBA in the machine is now monitored, not just one. Thanks to @brianara3 and @masterwishx, who reported the problem and then ran every test build on their own hardware until it was right. Firmware/BIOS flashing remains off. **Fixed** - **A machine with more than one SAS2 card monitored only one of them.** The Overview, HBA Health, PHY Health, Event Log and Drives tabs all read the single card named by the "lsiutil Port" setting, while Detected Hardware listed them all — so the plugin looked complete and was watching one card. Every card is now read, each with its own temperature, firmware, PCIe link, drive count and PHY counters. Verified on a three-card box and a two-card box before release. - **The Event Log was empty on every SAS2 card.** The log was being read correctly and then discarded, because the plugin expected a layout lsiutil does not print. One reporter's card held 85 entries behind that empty tab. - **A card whose name contains a space showed only the first word.** A 9400 read as "HBA" instead of "HBA 9400-16i", which also cost it its firmware check, since no name like that can be matched. - **The two backends printed PCI addresses differently** — the same card read as "00:c1:00:00" or "c1:00" depending on which tool could see it. Both now print the short form that lspci starts with. - **The diagnostic bundle collected only one card's telemetry**, which is why the bug above took three rounds of hand-written commands to pin down. It now captures every card. **Changed** - **The PHY tab's error counts now read zero after you reset the baseline.** The big number is what has happened since you last pressed the button, and a link you have fixed stops being flagged, instead of staying flagged until the next reboot. The driver's own running total keeps a line underneath, labelled "since driver load" — because that is what it is: it returns to zero when the driver reloads, with no cable having changed. - **A dual-controller board renders as one card**, with a section per controller, rather than as two unrelated cards that happen to share a slot. - The "lsiutil Port" setting no longer chooses which card is watched, since they all are. It stays as the fallback for a card table the plugin cannot read, and says so. ###2026.08.09### Firmware/BIOS flashing is switched off in this release while it gets more testing. Nothing else changes. **Changed** - **Firmware/BIOS flashing is disabled for now.** It is the one part of this plugin that writes to your controller, and a wrong or mismatched image bricks it permanently — so while it is being tested on more hardware it stays off rather than staying available and unverified. The Settings card is greyed out and the firmware page explains itself instead of offering the wizard. If you had flashing enabled, that setting is remembered and comes back with it; you do not need to do anything. - Every request to the flashing endpoint is refused outright, not merely hidden — a page left open in another tab cannot reach it either. Monitor, Array Map, Locate, SMART, PHY Health, notifications and the diagnostic bundle are all unaffected. ###2026.08.08.1### A card that cannot measure its own temperature now says so instead of reporting zero. Thanks to @NotAnotherForumName, whose diagnostics bundle showed a SAS9211-8i being asked for a reading it has no way to take. **Fixed** - SAS9211-8i and other SAS2008 cards showed a temperature of 0 °C in a "normal" band. Those cards have no onboard temperature sensor — but they answer the query anyway and return zero, which was read as a real measurement. That is worse than showing nothing, because a number in a green band looks like a live reading. They now show "N/A — no sensor". The card still counts as read, no temperature alert can fire from it, and cards that do have a sensor are unaffected. (#17) **Changed** - The Monitor and firmware pages load their behaviour from separate files now, rather than carrying it inline. Nothing changes on screen. The point is that this is the largest body of code in the plugin and, until now, not one of the tools that check this project for mistakes could open any of it. ###2026.08.08### If you have drives sharing a controller port, your Array Map placements were being overwritten by each other. Thanks to @PaliKinG3, who diagnosed the cause correctly in the report and suggested the fix that was used, and to @jac2424, whose SAS9207-8i turned up a second bug on the same page. **Fixed** - Assigning one drive to a bay placed every other drive on the same controller port into that same bay. The map keyed a placement on the CONTROLLER's Connected Port Number, which is not the drive's position and is shared — a SAS9305-16i reports 0(path0) for most of its drives, so one assignment captured all of them and emptied the Unassigned list. It now keys on the drive's own enclosure/slot, which is the backplane position and matches the label on the chassis. Nothing was ever lost: the map held a single entry the whole time. Existing maps are carried over automatically the first time you open the Array Map — a port that belonged to exactly one drive keeps its bay silently, and drives whose port collided return to the tray to be placed once. A drive that simply isn't present is never dropped. (#15) - Bay cards on SAS2 cards driven by lsiutil (9207-8i and similar) showed no model, serial, temperature, health or capacity. That backend reports no serial at all, and the SMART cache was joined by serial only, so the join missed every drive on the card. It falls back to /dev now, and the collector caches each drive's capacity. Values the controller reports itself still win. (#15) - Host Link reported no PCIe maximum on SAS2 cards, so a downtrained link could not be detected there. The card's maximum was in sysfs the whole time — the backend just never resolved a PCI address, because that was derived from storcli output. It now walks there from the scsi_host, so both backends read the link, the card's ceiling and the slot's ceiling the same way. (#14) ###2026.08.06### Everything below the Array Map is the same map you already built — placement got easier, and it can now leave the machine. Thanks to @Waz and @jac2424, whose sysfs output made the Host Link fix possible on hardware nobody here owns. **Fixed** - Host Link no longer warns forever about a card in a slot narrower than itself. It compared the negotiated link against the CARD's maximum, so an x8 card the motherboard runs at x4 — normal on OEM servers and not fixable — reported a permanent fault, while a chipset-limited x4 was told it was "running at its full x4 width" when the card could do x8. The slot's own ceiling is now read from its PCIe bridge and the link judged against whichever of the two is lower, so both read correctly with no configuration. A link below what it could actually reach still warns. (#13, #14) - Host Link claimed a SAS2 card was at "its full x4 width" when nothing had told it what that card's maximum was. The lsiutil backend reports no maximum at all, so there was nothing to be full of. It now says the link and stops there, rather than dressing an unmeasured link as a checked one. (#14) - Double-clicking a bay to empty it did nothing, from 2026.08.05. Picking a drive up redraws the whole grid, so the two clicks landed on different elements and the browser never delivered the double-click to the bay. - The Locate STOP button needed two presses. Any click on the Array Map redrew the Locate buttons from the state at the last refresh, so a button could show "Locate" over a drive that was still blinking — and pressing it then started a locate that was already running instead of stopping it. - The bay map, Locate and Reset Baseline could all silently fail to save when firmware flashing was turned off. They share Unraid's CSRF token, which was only being read when flashing was enabled. **Added** - Drag a drive into its bay on the Array Map, or drag it back to the unassigned list to empty it. Click-then-click still works and is still the way to do it on a touch screen. - **Copy map** and **Restore map**. The layout is built by hand and nothing on the machine can regenerate it, so it can now be copied out as text and pasted back — into a note, another server, or the file itself. - **Clear map**, which empties the grid in one go after telling you how many drives it is about to unplace, and **Undo**, which puts the map back after a Clear or a grid resize. One level deep, and consumed when used. - Expected PCIe Width and Generation in Settings, for boards whose PCIe bridge reports nothing and where Host Link therefore cannot work the slot out for itself. Leave both at 0 unless Host Link is wrong about your link; they are corrections, not mutes, and a link below what you set still warns. **Changed** - Firmware/BIOS Update is its own page now, under Settings → User Utilities, instead of a tab on the Monitor. Once flashing is enabled, a red button at the bottom of the Settings page opens it — and that is the only way in; the Monitor no longer links to it at all. It is the one part of this plugin that writes to hardware, so getting to it means passing the warning that explains what it costs to get wrong. With flashing off, neither the button nor the menu entry exists. Nothing about how flashing works has changed. - HBA Health lays its controller cards out the same way, for the same reason. A narrow window still gets a single column. On a frame too tight to fit two cards at full width, the gauge and the temperature meter stack inside each card rather than the cards dropping back to one column — the card gets a little taller and the page loses a whole card's height. - The unassigned drive list on the Array Map is ordered the way Unraid's Main page lists disks — Parity, Disk 1, Disk 2, then pools, then anything the array does not use — instead of the order the controller happened to report them in. - Every tab is one width now. The panel used to size itself to whichever tab was open, so it changed shape as you moved along the strip. - Tables and health rows lost the dead space between them, and the bay cards are shorter: the Unraid slot and the port share a line, and the model and serial keep their own so neither gets truncated. - Placing a drive no longer reloads the whole Array Map. The bay updates immediately and the save happens behind it. ###2026.08.05### Two new tabs, plus the topology fix that 2026.08.04.1 did not actually deliver on SAS2 cards. Thanks to @jac2424, who confirmed the SAS2 fixes on his own 9207-8i, and to @TheIlluminate92, whose diagnosis and diagnostic bundle located the expander problem below where it was reported. **Added** - An Array Map tab. Enclosure/slot and PHY numbers are the controller's own addressing, and knowing a drive sits on PHY 14 does not tell you which of 24 physical bays to walk over to. The Array Map is a grid you shape to match your chassis and then assign each drive to the bay it actually occupies. Bays are coloured by health and show temperature, so a warning becomes a place rather than a row. Once built it can be locked, since the map is the one thing here that cannot be regenerated by re-reading the hardware: it is you, walking to the rack. - Locate, on any Drives row and on any assigned bay. The drive's activity light starts blinking in a steady rhythm so the row becomes a tray you can find. It needs nothing from the backplane and drives no LED — the plugin reads the drive twice a second and the tray light flickers because the drive is busy, which is why it works where a proper locate LED does not. Press again to stop; it also stops itself after five minutes so a forgotten blink cannot hold a disk awake. It does keep that drive awake while it runs, which is inherent to the technique. **Fixed** - Topology still read "0 drives" on SAS2 cards. The fix in 2026.08.04.1 was written against a sysfs layout these cards do not have — the kernel places two further directory levels between a SAS host and its drives, and the plugin looked only as deep as the first. The test meant to catch this built its fixture in the same wrong shape, so it passed, and that is why the earlier fix shipped believing it worked. The fixture was rebuilt from a real card's sysfs before a line of code changed. (#14) - Drives showed the wrong PHY number on those same cards. Where the PHY could not be read, the plugin quietly fell back to the SCSI target id and displayed it as a PHY — a wrong number rather than a missing one, on six of eight drives on the reporting hardware. (#14) - Link Integrity never left "Not enough samples yet" on a box with an expander. The card's own PHYs and the PHYs on an expander behind it were read into one list under colliding numbers, so every poll compared a real counter against an unrelated one, read the difference as a decrease, and discarded the history as unreliable. Hours of samples were thrown away each poll and it looked like patience was the answer. Only the card's own PHYs are read now. (#12) - Array Map bay assignments now record which expander a drive sits behind. An expander numbers its own PHYs from zero, exactly as the card does, so on a chassis with one two drives could claim the same bay — and whichever was read last took it, leaving the other missing from the grid and from the unassigned tray both. On the reporting hardware — two expanders on one card — that was seven of nineteen drives missing. Cards without an expander are unaffected and every existing assignment keeps its bay, confirmed unchanged on a direct-attached card. (#12) **Changed** - HBA Health and PHY Health could report different error rates for the same controller at the same moment, with neither saying why. They answer different questions: one measures the recent samples it holds, the other averages across everything since the counters were first seen, so a burst that ended days ago keeps showing in one and not the other. Both now state the window they cover, and the PHY tab shows a recent rate on its own line under the long-run one. Neither number changed. - Firmware/BIOS Update puts the controller cards side by side instead of stacking them. Nothing in a card is as wide as the frame, so on a two-HBA box half the page was empty and the second card sat below the fold. A narrow window still gets a single column. ###2026.08.04.1### Everything reported on issue #11, plus two things found while fixing it. Thanks to @jac2424, whose screenshots of a SAS9207-8i showed a healthy card reporting nonsense. **Fixed** - Topology read "0 drives" on SAS2 cards. Cards read through lsiutil never counted their drives at all, so the row judged a hard-coded zero and still called it "all drives present" on a card with eight disks attached. The count now comes from the drives the controller's own SAS host actually presents. A controller that cannot report a count says so, instead of claiming zero. (#11) - HBA Health rows gave numbers with nothing to read them against. "Link Integrity 0/hr" never said 0 of what, and "Read Health ok" said nothing at all. Every row now prints the sentence behind its value — what was measured, over what window, and why it is judged the way it is. Values that are words are capitalised to match the rest of the plugin. (#11) - Drive names never appeared on the PHY Health tab. The tab read its drive list through a 60-second cache that it was the only user of, so the cache had always expired by the next visit and every visit got an empty answer. This also affected the drive names on the top-offenders list. **Added** - A Device column on the Drives and PHY Health tabs. Enclosure/slot, bus/target and PHY index are the controller's own addressing and line up with nothing on Unraid's Main page, so a row could not be matched to a disk without comparing SAS addresses by eye. Both tabs now lead with the same /dev name Main and the SMART tab use. Rows that cannot be resolved with certainty show a dash rather than a guess. (#11) **Changed** - The Settings page no longer leaves holes between its cards. Sections sat in a grid whose rows were as tall as their tallest card, so every short section left empty space beside it. They now pack. ###2026.08.04### Everything reported on issue #10, plus a Health tab fix found while investigating it. Thanks to @PaliKinG3 and @jac2424, who ran diagnostics on their own hardware and found several of these. **Fixed** - Mode was missing entirely on 9200-series cards. The Mode row only ever appeared on cards read through storcli, so a SAS9207-8i showed the next field where Mode should be and it looked like a 93xx-only feature. It is now read from the flashed firmware image name, which states the personality directly. Verified on a SAS9207-8i. (#10) - Mode could say IR on a card flashed to IT. On cards read through storcli the personality was inferred from drive states, and an unconfigured drive reports "UGood" whether it sits on an IT-only HBA or on an IR card with no arrays — so an IT-flashed SAS9305-16i was labelled IR. Mode is now reported only when the drive states actually prove one, and the row is hidden otherwise. A hidden row is better than a confident wrong answer. (#10) - Chip was blank on some controllers. The chip name came from a built-in list of five device IDs, and any card outside it — a SAS9305-16i among them — showed nothing at all. It now comes from storcli's own report, which covers every chipset. (#10) - SATA drives were labelled SAS, and every drive showed "Grown Defects" — a metric only SAS drives have. Drive type was taken from the bus, and every drive behind a SAS HBA reports itself as SAS whatever it actually is. The SMART tab now asks the drive rather than the bus, shows a Type column, and the column is renamed Reallocated, which is true on both. (#10) - Link Integrity could read "errors rising (0/hr)". A genuine sub-1/hr error rate was rounded down to zero, so the HBA Health tab warned about a rising counter while showing it as nothing. Rates below 1/hr are now shown to one decimal, and a non-zero rate never displays as zero. **Changed** - The diagnostic bundle now collects the firmware personality, the drive transport and the SAS device class. Diagnosing issue #10 needed all three and none were in the bundle, so they had to be requested by hand. A test now fails the build if the bundle stops collecting something the plugin reads. **Known** - Opening the SMART tab can spin up a sleeping SATA drive if it sits behind a SAS HBA. Whether to use smartctl's "do not wake a sleeping disk" guard is decided from the bus, and every drive behind a SAS HBA reports as SAS — a bus whose health reads genuinely do not touch the platters. A SATA drive on that path is therefore read without the guard. This is not new in this release and nothing here changed it. Fixing it means changing when a sleeping array disk gets woken, which needs measuring on real hardware first. If you see drives spin up when you open the tab, please say so on the tracker — that report is the measurement. ###2026.08.03### - Fixed: 2026.08.02 could not be installed. Unraid rejected the plugin file with "XML file doesn't exist or xml parse error" before downloading anything. A shell comment inside one of the plg's INLINE blocks contained angle brackets, which an XML parser reads as an unclosed tag; INLINE content is XML text, so a bare "<" or "&" anywhere in it — even in a comment — breaks installation for everyone. The comment is reworded; nothing else changed. - Added: the test suite now parses hbaviewer.plg and fails on a bare "<" or "&" inside any INLINE block, on a missing version/pkgURL/md5 entity, on a pkgURL whose path does not match the version, and on a missing changelog block. CI linted every .php and .sh but never the one file that decides whether the plugin installs at all. ###2026.08.02### The largest release so far: a new HBA Health tab, PHY error baselines, an export endpoint, diagnostic bundles, notifications, and fixes for every open issue on the tracker. **New** - HBA Health tab: five independent indicators (thermal, link integrity, topology, host link, controller read) with a worst-of rollup and a reason string naming the offending PHY. Anything that cannot be measured reads grey rather than green — a collector that times out must never look healthy. - PHY error baselines: press Set Baseline per controller and every counter is shown as a delta and an errors/hour rate instead of a number that has been accumulating since the driver loaded. Stored on /boot so it survives reboots; a reboot or driver reload invalidates it rather than showing a negative delta. - Top offenders: above the PHY table, the PHYs with the highest error rate since the baseline, each named by the drive it serves. A PHY that cannot be matched to exactly one drive says so rather than guessing at a bay. - Notifications (opt-in, off by default): one Unraid notification each time a controller's health status changes, never a repeat while it persists. - Diagnostic bundle: one button collects the raw storcli/lsiutil output, the sysfs state and the plugin's parsed JSON into one archive for a bug report, anonymised by default with one length-preserving map for the whole bundle. - Export / API: a read-only JSON snapshot of every controller, plus the same data in Prometheus text format, linked from Settings. Session-gated, so an unauthenticated scraper cannot poll it. - HOWTO.md and ARCHITECTURE.md. **Fixed** - Drives tab empty on controllers that report no enclosure. storcli answers the enclosure form with "No drive found!" on those cards and addresses drives as /cN/sN instead; that form is now used as a fallback. (#5, #6) - PCIe width and speed missing on storcli cards, now read from sysfs. (#5) - PCIe generation reported one notch low on SAS2/lsiutil cards — the IOUnit Page 7 speed field is an enum, not a bitmask like the width field beside it, so a card negotiating 8 GT/s read as Gen2. Gen1 rendered as nothing at all. (#9) - False "warning" temperature notifications: five fixed bands now separate how hot the card is from whether the badge should complain, and the PHY error floor no longer trips on a handful of counts. (#8) - Plugin colours now follow the Unraid theme instead of hardcoded values that only worked on one of them. (#7) - The enclosure summary no longer reports "8 slots, 0 drives" above a table of drives; those counts describe a synthesised enclosure the drives are not attached to. - Event archive no longer mixes entry shapes when a system changes backend. - Firmware/BIOS Step 3 is greyed out while the array is running. The server-side guard was always there; the UI now says so before you fill the form in. **Changed** - Every per-controller tab renders one card per HBA. - Two-column Settings layout. - Dashboard tile: per-controller model and temperature, status colour, and a "no sensor" reading instead of 0 °C on cards without one. - Test fixtures are now real controller output from reporters wherever possible, with identifiers masked length-preservingly. ###2026.07.27### - Fixed: the Performance tab showed no temperature on SAS2 (9200-series) cards. Temperatures were matched with a pattern that never fit the lsiutil backend's output, so those controllers graphed nothing while the Overview and dashboard tile showed the reading correctly. On multi-controller systems a card that failed to report could also shift the next card's temperature onto it. (#2) - Fixed: firmware version misread on SAS2 cards. The version is packed as four hex bytes, so a card reporting 14000700 is running 20.00.07.00 — it was shown as 14.00.07.00 and then flagged with a false "pre-P20" warning. (#3) - Fixed: Settings no longer tells SAS2-only systems to install storcli. The check keyed off the mpt3sas driver being present, but mpt3sas also drives SAS2 hardware, so cards that need nothing extra were reported as SAS3. (#3) - Dashboard tile: new HBA card icon that takes the controller's status colour, and the header now shows model and temperature per controller — so the tile still answers at a glance when it is minimised. Cards with no onboard sensor say "no sensor" instead of 0 °C. ###2026.07.23### - UI: one consistent text color across the Monitor tabs, Settings and the dashboard tile, replacing the several shades of gray that were harder to read on some themes. Appearance only — no change to what's collected or shown. ###2026.07.10.2### - NEW: Performance tab — real-time per-controller graphs (throughput, IOPS, %util, latency, PHY error-rate, temperature). Sampled ~2s in the browser from /proc/diskstats + sysfs (zero-dependency, no sampler daemon, no flash writes); history is in-browser only and resets on reload. Toggle in Settings (default on). - NEW (advanced, opt-in, OFF by default): Firmware/BIOS Update tab. Enable it in Settings to detect the card, run a read-only per-controller verify, upload a model-correct firmware image (auto-detects storcli / sas2flash / sas3flash, or upload the tool), and flash ONE controller behind hard guardrails — array must be stopped, typed-FLASH confirmation, single-flight lock — with a live log and reboot prompt. Flash tools are proprietary and never bundled. Read the README safety section first: a wrong image can permanently brick the controller. - UI refresh: refreshed Monitor / Settings / dashboard-tile styling (rounded cards, cleaner tabs, temperature ring) and a branded plugin icon. No change to what's shown, just how it looks. ###2026.07.10.1### - Internal architecture refactor — no visible change, same output on every tab. The storcli/lsiutil backend choice, controller enumeration, and driver detection are now one shared module (scripts/lib.sh hba_each); the PHY/Drives/Events tabs pick their columns from an explicit "backend" field instead of inferring it from the data shape, so a new column can't misfire - Event-log archive and the cached background read are now standalone, unit-tested modules (event_archive.php, cached_read.php) - Removed the duplicated backend_storcli.sh and fixed a driver-string detection order that could differ between code paths ###2026.07.09.2### - Fixed "Backend produced no output" on the Overview when a slow storcli scan exceeded the web timeout — the hardware read now runs as a detached background job (the only hardware reader) and the page polls a result file until ready, so a slow first read can never fail or be thrown away ###2026.07.09.1### - Monitor page now opens instantly with a "Loading HBA information" banner; the hardware read runs in the background (no more page-load hang) - Settings page shows the Access Method (storcli / lsiutil) with zero lag - Open HBAviewer Monitor button shows a load-time confirmation popup ###2026.07.09### - Settings moved to User Utilities menu (was System Settings) - lsiutil Port field now hidden for non-SAS2 controllers - Redesigned settings buttons: 'Save Settings First' + 'Open HBAviewer Monitor' (conditional) - Added note about monitor load time for slow systems - Fixed XML compatibility for Unraid parser - Dashboard tile display with multi-controller support - Hard drive icon throughout interface &pkgURL; &md5; rm -rf /usr/local/emhttp/plugins/hbaviewer tar -xJf /boot/config/plugins/hbaviewer/hbaviewer.txz -C / chmod +x /usr/local/emhttp/plugins/hbaviewer/hbaviewer.x86_64 chmod +x /usr/local/emhttp/plugins/hbaviewer/scripts/*.sh # Unraid's plugin manager resolves the Plugins-page icon against # plugins/NAME/, where NAME is this plg's name attribute — "HBAviewer", # capital H — not the lowercase install directory. So the icon has to exist # there too. Deliberately created here rather than shipped in the repo: # HBAviewer/ and hbaviewer/ collide on case-insensitive filesystems, so a # maintainer on Windows or macOS would see a phantom deletion on every checkout. mkdir -p /usr/local/emhttp/plugins/HBAviewer cp -f /usr/local/emhttp/plugins/hbaviewer/icon.png /usr/local/emhttp/plugins/HBAviewer/hbaviewer.png if [ ! -f /boot/config/plugins/hbaviewer/hbaviewer.cfg ]; then mkdir -p /boot/config/plugins/hbaviewer printf "HBA_PORT=1\nALERT_THRESHOLD=76\n" > /boot/config/plugins/hbaviewer/hbaviewer.cfg fi # The one directory the user drops flash files into: the sas2flash/sas3flash # tool, the firmware image, the optional BIOS. Survives updates and reboots, and # stays mounted when the array is stopped — which flashing requires, and which # is why this cannot live under appdata. There is no upload form: a multipart # POST to any .php behind Unraid's nginx never completes. # tools/ is the pre-2026.08.09 name, still created and still read so nobody who # already put a flasher there is stranded. mkdir -p /boot/config/plugins/hbaviewer/flash mkdir -p /boot/config/plugins/hbaviewer/tools # Health-status notifications (Settings > HBAviewer > Notifications, OFF by # default). The entry is installed unconditionally and the script exits # immediately when the toggle is off, so enabling it needs no reinstall — and a # disabled feature costs one config read every 10 minutes, not a hardware scan. # /etc/cron.d is tmpfs and this block re-runs on every boot, which is how every # Unraid plugin schedules work; update_cron folds it into root's crontab. # printf, not a heredoc, and no ampersand-redirect: this INLINE block is XML # character data, where a raw opening angle bracket or ampersand is ill-formed. # (The comment above line 93 already contains one and Unraid's parser tolerates # it, but don't add more.) Same reason the cfg above uses printf. Stderr goes to # a second explicit /dev/null rather than the 2-ampersand-1 form. printf '%s\n' \ '# Generated by the HBAviewer plugin - removed on plugin uninstall.' \ '*/10 * * * * /usr/bin/php /usr/local/emhttp/plugins/hbaviewer/scripts/notify_check.php >/dev/null 2>/dev/null' \ > /etc/cron.d/hbaviewer /usr/local/sbin/update_cron removepkg hbaviewer rm -rf /usr/local/emhttp/plugins/hbaviewer rm -rf /usr/local/emhttp/plugins/HBAviewer # Pair of the install block's cron entry — an orphaned job pointing at a deleted # script would run (and fail) every 10 minutes forever. update_cron rebuilds # root's crontab without it. rm -f /etc/cron.d/hbaviewer /usr/local/sbin/update_cron