diff options
Diffstat (limited to 'sys/contrib/octeon-sdk/cvmx-log-arc.S')
| -rw-r--r-- | sys/contrib/octeon-sdk/cvmx-log-arc.S | 178 |
1 files changed, 178 insertions, 0 deletions
diff --git a/sys/contrib/octeon-sdk/cvmx-log-arc.S b/sys/contrib/octeon-sdk/cvmx-log-arc.S new file mode 100644 index 0000000..bfb068e --- /dev/null +++ b/sys/contrib/octeon-sdk/cvmx-log-arc.S @@ -0,0 +1,178 @@ +/***********************license start*************** + * Copyright (c) 2003-2010 Cavium Inc. (support@cavium.com). All rights + * reserved. + * + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions are + * met: + * + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * + * * Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following + * disclaimer in the documentation and/or other materials provided + * with the distribution. + + * * Neither the name of Cavium Inc. nor the names of + * its contributors may be used to endorse or promote products + * derived from this software without specific prior written + * permission. + + * This Software, including technical data, may be subject to U.S. export control + * laws, including the U.S. Export Administration Act and its associated + * regulations, and may be subject to export or import regulations in other + * countries. + + * TO THE MAXIMUM EXTENT PERMITTED BY LAW, THE SOFTWARE IS PROVIDED "AS IS" + * AND WITH ALL FAULTS AND CAVIUM INC. MAKES NO PROMISES, REPRESENTATIONS OR + * WARRANTIES, EITHER EXPRESS, IMPLIED, STATUTORY, OR OTHERWISE, WITH RESPECT TO + * THE SOFTWARE, INCLUDING ITS CONDITION, ITS CONFORMITY TO ANY REPRESENTATION OR + * DESCRIPTION, OR THE EXISTENCE OF ANY LATENT OR PATENT DEFECTS, AND CAVIUM + * SPECIFICALLY DISCLAIMS ALL IMPLIED (IF ANY) WARRANTIES OF TITLE, + * MERCHANTABILITY, NONINFRINGEMENT, FITNESS FOR A PARTICULAR PURPOSE, LACK OF + * VIRUSES, ACCURACY OR COMPLETENESS, QUIET ENJOYMENT, QUIET POSSESSION OR + * CORRESPONDENCE TO DESCRIPTION. THE ENTIRE RISK ARISING OUT OF USE OR + * PERFORMANCE OF THE SOFTWARE LIES WITH YOU. + ***********************license end**************************************/ + + + + + + + +// +// The function defined here is called for every function as it is executed. +// These calls are automatically inserted by GCC when the switch "-pg" is +// used. This allows cvmx-log to add a PC entry as each function is executed. +// This information, along with the timestamps can give the user a good idea +// of the performance characteristics of their program. This function normally +// takes about 22 cycles to execute. +// + +#ifdef __linux__ +#include <asm/asm.h> +#include <asm/regdef.h> +#define LA dla +#else +#include <machine/asm.h> +#include <machine/regdef.h> +#define LA la +#endif + +.set noreorder +.set noat +LEAF(_mcount) + // + // All registers we use must be saved since calls are added by gcc + // after register allocation. The at register ($3) will contain the + // original ra register before the _mcount call. Also the compiler + // automatically performs a "dsubu sp, sp, 16" before we're called. + // At the end of this function all registers must have their original + // values and the stack pointr must be adjusted by 16. This code is + // pretty unreadable since it has been arranged to promote dual issue. + // +#ifdef __linux__ + dsubu sp, sp, 32 +#else + dsubu sp, sp, 16 +#endif + sd s3, 24(sp) // Save register + rdhwr s3, $31 // Read the cycle count + sd s0, 0(sp) // Save register + LA s0, cvmx_log_buffer_end_ptr // Load the address of the end of the log buffer + sd s1, 8(sp) // Save register + LA s1, cvmx_log_buffer_write_ptr // Load the address of the location in the log buffer + sd s2, 16(sp) // Save register + ld s0, 0(s0) // Get value of the current log buffer end location + ld s2, 0(s1) // Get value of the current log buffer location + dsubu s0, s0, s2 // Subtract the end pointer and the write pointer + sltiu s0, s0, 16 // Check if there are at least 16 bytes + bne s0, $0, call_c_pc // Call the slow C function if we don't have room in the log + li s0, 0x001 // 11 bit constant that matches the first 11 bits of a CVMX_LOG_TYPE_PC header + sd ra, 8(s2) // Write the pc to the log + dins s3, s0, 53, 11 // Overwrite the upper cycle count bits with the CVMX_LOG_TYPE_PC header + sd s3, 0(s2) // Write the log header + daddu s2, s2, 16 // Increment the write location ptr + sd s2, 0(s1) // Store the write location ptr +return_c_pc: + ld s0, 0(sp) // Restore register + ld s1, 8(sp) // Restore register + ld s2, 16(sp) // Restore register + ld s3, 24(sp) // Restore register + daddu sp, sp, 32 // Pop everything off the stack, even the 16 bytes done by gcc + jr ra // Return to the caller and + or ra, $1, $1 // make sure the ra is back to its original value + +call_c_pc: + // The registers used by the C code may change based on optimizations. To be + // safe, I'll save all registers. We're in the slow path case anyway. + dsubu sp, sp, 216 + sd $1, 0(sp) + sd $2, 8(sp) + sd $3, 16(sp) + sd $4, 24(sp) + sd $5, 32(sp) + sd $6, 40(sp) + sd $7, 48(sp) + sd $8, 56(sp) + sd $9, 64(sp) + sd $10, 72(sp) + sd $11, 80(sp) + sd $12, 88(sp) + sd $13, 96(sp) + sd $14, 104(sp) + sd $15, 112(sp) + // s0, s1, s2, s3 are already saved + sd $20, 120(sp) + sd $21, 128(sp) + sd $22, 136(sp) + sd $23, 144(sp) + sd $24, 152(sp) + sd $25, 160(sp) + sd $26, 168(sp) + sd $27, 176(sp) + sd $28, 184(sp) + sd $29, 192(sp) + sd $30, 200(sp) + sd $31, 208(sp) + + or a0, ra, ra + jal cvmx_log_pc + nop + + ld $1, 0(sp) + ld $2, 8(sp) + ld $3, 16(sp) + ld $4, 24(sp) + ld $5, 32(sp) + ld $6, 40(sp) + ld $7, 48(sp) + ld $8, 56(sp) + ld $9, 64(sp) + ld $10, 72(sp) + ld $11, 80(sp) + ld $12, 88(sp) + ld $13, 96(sp) + ld $14, 104(sp) + ld $15, 112(sp) + // s0, s1, s2, s3 will be restored later + ld $20, 120(sp) + ld $21, 128(sp) + ld $22, 136(sp) + ld $23, 144(sp) + ld $24, 152(sp) + ld $25, 160(sp) + ld $26, 168(sp) + ld $27, 176(sp) + ld $28, 184(sp) + ld $29, 192(sp) + ld $30, 200(sp) + ld $31, 208(sp) + b return_c_pc + daddu sp, sp, 216 + +END(_mcount) + |
