Differences

This shows you the differences between two versions of the page.

--- soc:2010:cooldavid:journal:week12 [2010/08/12 18:05]
cooldavid
+++ soc:2010:cooldavid:journal:week12 [2010/08/13 01:31] (current)
cooldavid
@@ Line 1: / Line 1: @@
 ==== Port wscale branch against current gPXE master ====
 Recent TCP updates is a lot different from previous base of wscale branch.
+=== Hit a GCC optimization trap ===
+== For the following code ==
+<code>
+#include <stdio.h>
+struct foo {
+        char huge_gap[0x4005d0];
+        int v;
+};
+void store_and_print(int *val, struct foo *f)
+{
+        *val = f->v;
+        if ( !f ) {
+                printf("Value is wrong.\n");
+        } else {
+                printf("Value is correct.\n");
+        }
+}
+int main()
+{
+        int unidentified_value;
+        printf("%p\n", main);
+        store_and_print(&unidentified_value, NULL);
+        return 0;
+}
+</code>
+The value ''0x4005d0'' is from ''printf("%p\n", main);'' in order to get an
+mapped memory address to prevent ''Segment fault'' error while running.
+== The output if compiled with ''-O2'' ==
+<code>
+$ ./null_deref_optimized_away
+x4005d0
+Value is correct.
+$
+</code>
+And its assembly code:
+<code>
+void store_and_print(int *val, struct foo *f)
+{
+        *val = f->v;
+b0:       8b 86 d0 05 40 00       mov    0x4005d0(%rsi),%eax
+b6:       89 07                   mov    %eax,(%rdi)
+        if ( !f ) {
+                printf("Value is wrong.\n");
+        } else {
+                printf("Value is correct.\n");
+b8:       bf ec 06 40 00          mov    $0x4006ec,%edi
+bd:       e9 c6 fe ff ff          jmpq   400488 <puts@plt>
+c2:       66 66 66 66 66 2e 0f    nopw   %cs:0x0(%rax,%rax,1)
+c9:       1f 84 00 00 00 00 00
+00000000004005d0 <main>:
+        }
+}
+</code>
+== The output if compiled with ''-O2 -fno-delete-null-pointer-checks'' ==
+<code>
+$ ./null_deref_optimized_away
+x4005d0
+Value is wrong.
+$
+</code>
+And its assembly code:
+<code>
+00000000004005b0 <store_and_print>:
+        int v;
+};
+void store_and_print(int *val, struct foo *f)
+{
+        *val = f->v;
+b0:       8b 86 d0 05 40 00       mov    0x4005d0(%rsi),%eax
+        if ( !f ) {
+b6:       48 85 f6                test   %rsi,%rsi
+        int v;
+};
+void store_and_print(int *val, struct foo *f)
+{
+        *val = f->v;
+b9:       89 07                   mov    %eax,(%rdi)
+        if ( !f ) {
+bb:       74 13                   je     4005d0 <store_and_print+0x20>
+                printf("Value is wrong.\n");
+        } else {
+                printf("Value is correct.\n");
+bd:       bf 0c 07 40 00          mov    $0x40070c,%edi
+c2:       e9 c1 fe ff ff          jmpq   400488 <puts@plt>
+c7:       66 0f 1f 84 00 00 00    nopw   0x0(%rax,%rax,1)
+ce:       00 00
+void store_and_print(int *val, struct foo *f)
+{
+        *val = f->v;
+        if ( !f ) {
+                printf("Value is wrong.\n");
+d0:       bf fc 06 40 00          mov    $0x4006fc,%edi
+d5:       e9 ae fe ff ff          jmpq   400488 <puts@plt>
+da:       66 0f 1f 44 00 00       nopw   0x0(%rax,%rax,1)
+00000000004005e0 <main>:
+                printf("Value is correct.\n");
+        }
+}
+</code>
 ==== Adding some facilities for testing convenience ====
@@ Line 8: / Line 122: @@
 [[http://git.etherboot.org/?p=people/cooldavid/gpxe.git;a=shortlog;h=refs/heads/wscale2_test|wscale2 testing branch]]
 [[http://git.etherboot.org/?p=people/cooldavid/gpxe.git;a=shortlog;h=refs/heads/tcpfix2_test|tcpfix2 testing branch]]
@@ Line 16: / Line 131: @@
   * Tested it with emulated rtl8139 of KVM, and Native on rtl8139 hardware.
-^ Window Size ^   tcpfix2 branch    ^^    wscale2 branch    ^^
+^ Window Size ^   tcpfix2 branch     ^^    wscale2 branch    ^^    sackws2 branch    ^^
-^             ^ KVM TAP/NAT ^  Native ^ KVM TAP/NAT ^  Native ^
+^             ^ KVM TAP/NAT ^  Native ^ KVM TAP/NAT ^   Native   ^ KVM TAP/NAT ^  Native ^
-|           8K|       17KB/s|   16KB/s|             |   16KB/s|
+|           8K|       16KB/s|   16KB/s|       16KB/s|      16KB/s|         KB/s|     KB/s|
-|          16K|       43KB/s|   43KB/s|             |   41KB/s|
+|          16K|       41KB/s|   41KB/s|       41KB/s|      41KB/s|         KB/s|     KB/s|
-|          32K|       94KB/s|   93KB/s|             |   91KB/s|
+|          32K|       90KB/s|   90KB/s|       90KB/s|      91KB/s|         KB/s|     KB/s|
-|          64K|      198KB/s|  192KB/s|             |  187KB/s|
+|          64K|      186KB/s|  186KB/s|      187KB/s|     187KB/s|         KB/s|     KB/s|
-|         128K|             |         |             |  374KB/s|
+|         128K|             |         |      374KB/s|     374KB/s|         KB/s|     KB/s|
-|         256K|             |         |             |  742KB/s|
+|         256K|             |         |      470KB/s|     742KB/s|         KB/s|     KB/s|
-|         512K|             |         |             |     KB/s|
+|         512K|             |         |      450KB/s|  [1]344KB/s|         KB/s|     KB/s|
-|        1024K|             |         |             |     KB/s|
+|        1024K|             |         |      498KB/s|  [1]403KB/s|         KB/s|     KB/s|
+[1]: While TCP rx window is large enough to hit the bandwidth limit(1MB/s),
+packets start to drop. According to captured data, the retransmittion is very slow.
+The sending host wait a long time for transmit next missing packet required by client.
+This issue should be able to fixed with Selective ACK support.

Trace:

Differences

Navigation

Search

Toolbox

QR Code