Date: Tue, 8 Oct 2013 18:59:41 +0000 (UTC) From: Gleb Smirnoff <glebius@FreeBSD.org> To: src-committers@freebsd.org, svn-src-user@freebsd.org Subject: svn commit: r256161 - user/glebius/course/02.entering_kernel Message-ID: <201310081859.r98Ixfd9098625@svn.freebsd.org>
next in thread | raw e-mail | index | archive | help
Author: glebius Date: Tue Oct 8 18:59:41 2013 New Revision: 256161 URL: http://svnweb.freebsd.org/changeset/base/256161 Log: Format notes. Modified: user/glebius/course/02.entering_kernel/lection.tex Modified: user/glebius/course/02.entering_kernel/lection.tex ============================================================================== --- user/glebius/course/02.entering_kernel/lection.tex Tue Oct 8 18:57:46 2013 (r256160) +++ user/glebius/course/02.entering_kernel/lection.tex Tue Oct 8 18:59:41 2013 (r256161) @@ -83,23 +83,23 @@ interrupts, traps, system calls} \draw [loosely dashed] (code2.south west) -- +(0,-1); \onslide<1> { - \note { - - CPU is like a Turing machine. - - Explain what IP or program counter means. + \note<1> { + - CPU is like a Turing machine.\\ + - Explain what IP or program counter means.\\ - Why pure TM doesn't represent what we want from a computer? We don't want computer to endlessly run a calculation, we want it to be interactive. User input should interrupt it. And not - only user input. + only user input.\\ } \draw [->] (sp.east) to [out=15, in=165] (stack1.west); \draw [->] (ip.east) to [out=-15, in=165] (code0.west); } \onslide<2> { - \note { - - So, interruption of computing is important. + \note<2> { + - So, interruption of computing is important.\\ - Interrupt should be transparent for current execution. How - can this be achieved? + can this be achieved?\\ } \draw [->, dashed] (sp.east) to [out=15, in=165] (stack1.west); \draw [->, dashed] (ip.east) to [out=-15, in=165] (code0.west); @@ -117,10 +117,10 @@ interrupts, traps, system calls} } \onslide<3> { - \note { - - Saving entire context is important in some cases. - - And what if on return interrupt handler provides another context? - - Handling multiple contexts == multitasking. + \note<3> { + - Saving entire context is important in some cases.\\ + - And what if on return interrupt handler provides another context?\\ + - Handling multiple contexts == multitasking.\\ } \node [name=ctx, draw, circle, color=red, minimum width=4cm] at (cpu.center) {}; @@ -138,12 +138,12 @@ interrupts, traps, system calls} \begin{frame} \frametitle{Interrupts} \note { - - Let's see what kind of interrupts we got. + - Let's see what kind of interrupts we got.\\ } \begin{itemize} \onslide<1->{ - \note { - - Okay, now we can do multitasking and interactivity. + \note<1> { + - Okay, now we can do multitasking and interactivity.\\ } \item{Hardware interrupt (asynchoronous, involuntary) \begin{itemize} @@ -154,8 +154,8 @@ interrupts, traps, system calls} } } \onslide<2->{ - \note { - - We can handle (and "fix") traps, and even implement paging. + \note<2> { + - We can handle (and "fix") traps, and even implement paging.\\ } \item{Trap (synchoronous, involuntary) \begin{itemize} @@ -165,10 +165,10 @@ interrupts, traps, system calls} } } \onslide<3->{ - \note { + \note<3> { - Via interrupts we can provide a library of functions, what actually MS DOS did via INT 0x21. Actually many software of - the DOS era were setting a library on a particular interrupt. + the DOS era were setting a library on a particular interrupt.\\ } \item{Software interrupt (synchoronous, voluntary)\\ On x86 the ``int'' instruction @@ -180,16 +180,16 @@ interrupts, traps, system calls} \begin{frame} \frametitle{Protection} -\note { +\note<1> { - What isn't enough for a true OS is protection of tasks against - each other. + each other.\\ - We need individual address space for each, and only a supervisor - (the kernel) that can set 'em up. + (the kernel) that can set 'em up.\\ - We need to forbid certain priveleged instructions to anyone - except supervisor. - - This is what 80386 processor gave us. + except supervisor.\\ + - This is what 80386 processor gave us.\\ - Now the library of functions lives in ring0, and is called - kernel. + kernel.\\ } \begin{tikzpicture}[very thick] \node[name=ring3, draw, rounded corners=3mm, @@ -223,14 +223,14 @@ interrupts, traps, system calls} \begin{frame} \frametitle{System call} -\note { +\note<1> { - Let's make an agreement that certain software interrupt has an - ABI that would provide a gate into kernel. + ABI that would provide a gate into kernel.\\ - Using variety of syscall numbers we can implement all needed - services that kernel offers to userland. + services that kernel offers to userland.\\ - The int 0x80 is already a history. Modern hardware provide special much more efficient instruction designed for syscall. - There also were "call gates". + There also were "call gates".\\ } \onslide<1->{ \begin{beamercolorbox}[rounded=true,shadow=true,sep=0pt,ht=2.9cm]{trace} @@ -264,18 +264,18 @@ lib/libc/amd64/sys/SYS.h, sys/amd64/amd6 \begin{frame}[fragile] \frametitle{System call example} -\note { +\note<1> { - Let's trace an application ping(8). It sends data to network, so we expect it to use send(2). Our outbound interface is Ethernet, - so we expect ether_output() to be in action. - - Arm gdb & dtrace appropriately! + so we expect ether\_output() to be in action.\\ + - Arm gdb \& dtrace appropriately!\\ - Note that data is taken from dtrace and gdb, but order is - reversed! + reversed!\\ - Note that we expected ICMP but got UDP on first send() due to resolving, but that still works for us as an example! Also, it - demonstrates that library calls can use syscalls implicitly. - - Explain user stack and kernel stack. - - Let's trace recvfrom(), too. + demonstrates that library calls can use syscalls implicitly.\\ + - Explain user stack and kernel stack.\\ + - Let's trace recvfrom(), too.\\ } \shellcmd{\# ping www.ru} \begin{scriptsize} @@ -338,17 +338,17 @@ lib/libc/amd64/sys/SYS.h, sys/amd64/amd6 \begin{frame} \frametitle{Hardware interrupts and system calls} -\note{ +\note<1> { - We can enter kernel from a hardware interrupt, or from a software interrupt. The former is usually a driver entry, - and the latter is a syscall. + and the latter is a syscall.\\ - Drivers can have pure interrupt handlers, but more often - have ISRs. + have ISRs.\\ - Kernel is somewhat divided into two parts... Well, it is really not. :) But some threads are always calling "down", - and some are calling "up". And they are tightly interconnected. + and some are calling "up". And they are tightly interconnected.\\ - They usually synchronise at a some data structure, a generic - buffer. + buffer.\\ } \begin{tikzpicture}[very thick] \node[name=user, draw, rounded corners=3mm, @@ -380,11 +380,11 @@ lib/libc/amd64/sys/SYS.h, sys/amd64/amd6 (buf.north east); \onslide<2>{ - \note{ + \note<2> { - A thread coming from userland can eventually reach the very - bottom of a driver. + bottom of a driver.\\ - A thread coming from a driver can't pass kernel to userland - boundary. + boundary.\\ } \draw [->] (ping.south) to [out=230, in=120] node [midway, sloped, below] {\footnotesize send()} @@ -392,8 +392,8 @@ lib/libc/amd64/sys/SYS.h, sys/amd64/amd6 } \onslide<3>{ - \note{ - - An interrupt isn't necessarily related to some user task. + \note<3> { + - An interrupt isn't necessarily related to some user task.\\ } \node [name=timer, right=5mm of ether, draw, rounded corners=1mm] { timer }; @@ -402,9 +402,9 @@ lib/libc/amd64/sys/SYS.h, sys/amd64/amd6 } \onslide<4>{ - \note{ + \note<4> { - Let students think theirselves on possibility of such - upcall. Provide hints if needed. + upcall. Provide hints if needed.\\ } \draw [color=red, ->] (bottom.north) to [out=140, in=270] node[midway, sloped, above] {is that possible?} @@ -416,9 +416,9 @@ lib/libc/amd64/sys/SYS.h, sys/amd64/amd6 \begin{frame}[fragile] \frametitle{Adding a system call in FreeBSD} -\note { - - It is simple. Let's go through the steps. - - Explain all fields. +\note<1> { + - It is simple. Let's go through the steps.\\ + - Explain all fields.\\ } \shellcmd{% \# cd /usr/src/sys/kern\\ @@ -452,16 +452,14 @@ lib/libc/amd64/sys/SYS.h, sys/amd64/amd6 \begin{frame} \frametitle{syscall ABI} -\note { - - What actually 'make sysent' does: - - /usr/include/sys/syscall.h - - /usr/include/sys/sysproto.h +\note<1> { + - What actually 'make sysent' does.\\ - ABI is a convention on how arguments are passed between userland - and kernel. - - A table of syscalls. - - What if provide multiple ABIs? Yep, and that's the case. + and kernel.\\ + - A table of syscalls.\\ + - What if provide multiple ABIs? Yep, and that's the case.\\ - Emphasize that Linux ABI isn't an emulator, but a compatibility - layer, very much like freebsd32 layer. The open(2) is special. + layer, very much like freebsd32 layer. The open(2) is special.\\ } \emph{make sysent} generates: \begin{itemize} @@ -487,7 +485,7 @@ lib/libc/amd64/sys/SYS.h, sys/amd64/amd6 \begin{frame}[fragile] \frametitle{Adding a system call in FreeBSD (dynamic way)} -\note{ +\note { } \begin{itemize} \begin{item}Kernel side
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?201310081859.r98Ixfd9098625>