# 2024-06-25,27 討論簡記
> [課程期末專題](https://hackmd.io/@sysprog/linux2024-projects)

root fs (rootfs)

u-boot 是 boot loader，初始化 SRAM, DRAM, flash, eMMC, 等等硬體

root of trust (RoT), initiatated in HW
Trusted Firmware for A profile (TF-A)

> https://hackmd.io/@sysprog/linux-kvm
> https://hackmd.io/@sysprog/linux-pid0

SMP: multi-processor
multi-core : 未必對稱 (如 Arm big.LITTLE)
multi-core 相對的硬體組態不是 single core，而是 uniprocessor
single => island

`[_]` `[_]` `[Active]`
`[_]` `[_]`

Microsoft Xbox, CE, Mobile, NT, ...

我的問題：
https://hackmd.io/@Henry0922/linux-final-question-list

> 先分析遇到的問題，例如 Mailbox 的中斷處理以及佇列配置，不要急著下手寫程式。

---

https://github.com/sysprog21/concurrent-programs

如何擴充為 MPMC？

-/* Returns an index to push or -1 on failure (Queue is full) */
-static inline int mp_try_prepare_push(uint queue_size,
-                                      atomic_uint *tail_committed,
-                                      atomic_uint *head_pending)
-{
-    /* Tail needs to be loaded first, otherwise we might overestimate the amount
-     * of free space */
-    uint tail = atomic_load(tail_committed);
-    uint head = atomic_load(head_pending);
-
-    while (1) {
-        if (queue_get_free_explicit(queue_size, head, tail) <= 0)
-            return -1;
-
-        /* As this is an MP queue another thread might have taken our index. */
-        /* This will update `head` on failure */
-        if (atomic_compare_exchange_strong(head_pending, &head, head + 1))
-            return head;
-    }
-}

-/* Returns an index to consume */
-static inline int mc_prepare_consume(atomic_uint *head_committed,
-                                     atomic_uint *tail_pending,
-                                     atomic_uint *head_waiters)
-{
-    /* Head needs to be loaded first, otherwise we might overestimate the amount
-     * of committed space */
-    uint head = atomic_load(head_committed);
-    uint tail = atomic_load(tail_pending);
-
-    while (1) {
-        while (queue_get_committed_explicit(head, tail) == 0) {
-            QUEUE_WAIT_FOR_HEAD(head_waiters, head_comitted, head);
-            tail = atomic_load(tail_pending);
-        }
-
-        /* As this is an MC queue another thread might have taken our index. */
-        /* This will update `tail` on failure */
-        if (atomic_compare_exchange_strong(tail_pending, &tail, tail + 1))
-            return tail;
-    }
-}

-static inline void mc_commit_consume(uint prepared_index,
-                                     atomic_uint *tail_committed,
-                                     atomic_uint *tail_waiters)
-{
-    uint tail = atomic_load(tail_committed);
-
-    /* Wait for sequential increment */
-    while (prepared_index != tail)
-        QUEUE_WAIT_FOR_TAIL(tail_waiters, tail_committed, tail);
-
-    atomic_fetch_add(tail_committed, 1);
-
-    if (atomic_load(tail_waiters))
-        QUEUE_WAKE_ALL_TAIL_WAITERS(tail_committed);
-}

https://github.com/facebook/folly/blob/main/folly/docs/Synchronized.md

MPMC
SPSC (specialized)

---

quick sort 的 stability
> https://github.com/sysprog21/linux-list

---

老師好，我的 github 帳號名稱是 jujuegg，我的期末專題的開發紀錄還沒被放至課程期末專題的頁面中，我在昨日 13:00 左右有發電子郵件給老師，但老師尚未回覆。
> 知道！我要建立新的 HackMD 頁面 -- jserv

https://wiki.csie.ncku.edu.tw/User/HenryChaing
* 沒有針對作業共筆/測驗進行檢討，投入的狀況 (不要說「我花了大量時間」) ，儘量列舉客觀事實

> 首先是作業的部份，雖然並未達成全部老師的要求，但是我認為不論是我<s>投入的時間</s>或是產出都有達到一定的要求，因此會在這點自評高分。再來是上課與老師的互動，主要是針對測驗題目以及期末專題進行討論，測驗檢討雖然上課當下一知半解，但是在經過與老師的互動下有了表現自己的機會（雖然還是不盡人意），最後課堂結束也有機會好好再研究題目，甚至將其改寫成不同版本但原理相同的程式, <s>也謝謝老師還有批改課後我所加上的紀錄。</s>
$\to$ 客觀事實在哪？

## Kuanch
https://hackmd.io/@sysprog/rkJd7TFX0
EEVDF since Linux v6.6 => 解決 CFS 的哪些問題？

用 stress-ng 創造 workload
Linux v5.1 引入 Energy-Aware Scheduling (EAS)，當時的 CPU scheduler 是 CFS
$\to$ load balancing, task migration

## weihsinyeh
https://github.com/sysprog21/concurrency-primer/pull/7
Fetch and…
Transform RMW to directly modify the shared variable (such as addition, subtraction, or bitwise AND, OR, XOR)
and return its previous value, all as part of a single atomic operation. Compare with Exchange §5.1, when programmers
only need to make simple modification to the shared variable, they can use Fetch and….

哪裡 weird ?

## yeh-sudo
https://hackmd.io/@sysprog/B1W2HKTER

根據[測試結果](https://github.com/yeh-sudo/mt-redis?tab=readme-ov-file#performance) ，吞吐量雖然上升，但是延遲也顯著增加，不知道延遲跟什麼因素有關係?

============================================================================================================================
Type         Ops/sec     Hits/sec   Misses/sec    Avg. Latency     p50 Latency     p99 Latency   p99.9 Latency       KB/sec 
----------------------------------------------------------------------------------------------------------------------------
Sets        38465.04          ---          ---         4.11819         2.28700        23.80700        30.71900      2962.48 
Gets       384227.73         0.00    384227.73         0.52990         0.27900        16.31900        24.31900     14967.33 
Waits           0.00          ---          ---             ---             ---             ---             ---          --- 
Totals     422692.77         0.00    384227.73         0.85643         0.28700        18.04700        26.36700     17929.81

如何對 Redis 作效能評比？

        --*-- cpu0  >>>>>>>                |
        --*-- cpu1  >>>                    | Redis
        --*-- cpu2  >>>                    | urcu 
        --*-- cpu2  >>>>>                  | concurrency level = 4
        (concurrency level = 4)
        4                                    4

taskset : https://man7.org/linux/man-pages/man1/taskset.1.html
HT (hyperthreading) 開啟的狀況，找出 logical core
https://man7.org/linux/man-pages/man1/chrt.1.html (選擇 RT class)

git-bisect
ln -s (symbolic) : inode