Rev 4418 | Details | Compare with Previous | Last modification | View Log | RSS feed
Rev | Author | Line No. | Line |
---|---|---|---|
3520 | clevermous | 1 | ; Implementation of periodic transaction scheduler for USB. |
2 | ; Bandwidth dedicated to periodic transactions is limited, so |
||
3 | ; different pipes should be scheduled as uniformly as possible. |
||
4 | |||
5 | ; USB2 scheduler. |
||
6 | ; There are two parts: high-speed pipes and split-transaction pipes. |
||
3826 | clevermous | 7 | ; |
3520 | clevermous | 8 | ; High-speed scheduler uses the same algorithm as USB1 scheduler: |
9 | ; when adding a pipe, optimize the following quantity: |
||
10 | ; * for every microframe, take all bandwidth scheduled to periodic transfers, |
||
3826 | clevermous | 11 | ; * calculate maximum over all microframes, |
3520 | clevermous | 12 | ; * select a variant which minimizes that maximum; |
3826 | clevermous | 13 | ; * if there are several such variants, |
14 | ; prefer those that are closer to end of frame |
||
15 | ; to minimize collisions with split transactions; |
||
3520 | clevermous | 16 | ; when removing a pipe, do nothing (except for bookkeeping). |
17 | ; in: esi -> usb_controller |
||
18 | ; out: edx -> usb_static_ep, eax = S-Mask |
||
19 | proc ehci_select_hs_interrupt_list |
||
20 | ; inherit some variables from usb_open_pipe |
||
21 | virtual at ebp-12 |
||
22 | .targetsmask dd ? |
||
23 | .bandwidth dd ? |
||
24 | .target dd ? |
||
25 | dd ? |
||
26 | dd ? |
||
27 | .config_pipe dd ? |
||
28 | .endpoint dd ? |
||
29 | .maxpacket dd ? |
||
30 | .type dd ? |
||
31 | .interval dd ? |
||
32 | end virtual |
||
33 | ; prolog, initialize local vars |
||
34 | or [.bandwidth], -1 |
||
35 | or [.target], -1 |
||
36 | or [.targetsmask], -1 |
||
37 | push ebx edi ; save used registers to be stdcall |
||
38 | ; 1. In EHCI, every list describes one millisecond = 8 microframes. |
||
39 | ; Thus, there are two significantly different branches: |
||
40 | ; for pipes with interval >= 8 microframes, advance to 2, |
||
41 | ; for pipes which should be planned in every frame (one or more microframes), |
||
42 | ; go to 9. |
||
43 | ; Note: the actual interval for high-speed devices is 2^([.interval]-1), |
||
44 | ; (the core specification forbids [.interval] == 0) |
||
45 | mov ecx, [.interval] |
||
46 | dec ecx |
||
47 | cmp ecx, 3 |
||
48 | jb .every_frame |
||
49 | ; 2. Determine the actual interval in milliseconds. |
||
50 | sub ecx, 3 |
||
51 | cmp ecx, 5 ; maximum 32ms |
||
52 | jbe @f |
||
3598 | clevermous | 53 | movi ecx, 5 |
3520 | clevermous | 54 | @@: |
55 | ; There are four nested loops, |
||
56 | ; * Loop #4 (the innermost one) calculates the total periodic bandwidth |
||
57 | ; scheduled in the given microframe of the given millisecond. |
||
58 | ; * Loop #3 calculates the maximum over all milliseconds |
||
59 | ; in the given variant, that is the quantity we're trying to minimize. |
||
60 | ; * Loops #1 and #2 check all variants; |
||
61 | ; loop #1 is responsible for the target millisecond, |
||
62 | ; loop #2 is responsible for the microframe within millisecond. |
||
63 | ; 3. Prepare for loops. |
||
64 | ; ebx = number of iterations of loop #1 |
||
65 | ; [esp] = delta of counter for loop #3, in bytes |
||
66 | ; [esp+4] = delta between the first group and the target group, in bytes |
||
3598 | clevermous | 67 | movi ebx, 1 |
68 | movi edx, sizeof.ehci_static_ep |
||
3520 | clevermous | 69 | shl ebx, cl |
70 | shl edx, cl |
||
71 | mov eax, 64*sizeof.ehci_static_ep |
||
72 | sub eax, edx |
||
73 | sub eax, edx |
||
74 | push eax |
||
75 | push edx |
||
76 | ; 4. Select the best variant. |
||
77 | ; 4a. Loop #1: initialize counter = pointer to ehci_static_ep for |
||
78 | ; the target millisecond in the first group. |
||
79 | lea edx, [esi+ehci_controller.IntEDs-sizeof.ehci_controller] |
||
80 | .varloop0: |
||
81 | ; 4b. Loop #2: initialize counter = microframe within the target millisecond. |
||
82 | xor ecx, ecx |
||
83 | .varloop: |
||
84 | ; 4c. Loop #3: save counter of loop #1, |
||
85 | ; initialize counter with the value of loop #1 counter, |
||
86 | ; initialize maximal bandwidth = zero. |
||
87 | xor edi, edi |
||
88 | push edx |
||
89 | virtual at esp |
||
90 | .saved_counter1 dd ? ; step 4c |
||
91 | .loop3_delta dd ? ; step 3 |
||
92 | .target_delta dd ? ; step 3 |
||
93 | end virtual |
||
94 | .calc_max_bandwidth: |
||
95 | ; 4d. Loop #4: initialize counter with the value of loop #3 counter, |
||
96 | ; initialize total bandwidth = zero. |
||
97 | xor eax, eax |
||
98 | push edx |
||
99 | .calc_bandwidth: |
||
100 | ; 4e. Loop #4: add the bandwidth from the current list |
||
101 | ; and advance to the next list, while there is one. |
||
102 | add ax, [edx+ehci_static_ep.Bandwidths+ecx*2] |
||
103 | mov edx, [edx+ehci_static_ep.NextList] |
||
104 | test edx, edx |
||
105 | jnz .calc_bandwidth |
||
106 | ; 4f. Loop #4 end: restore counter of loop #3. |
||
107 | pop edx |
||
108 | ; 4g. Loop #3: update maximal bandwidth. |
||
109 | cmp eax, edi |
||
110 | jb @f |
||
111 | mov edi, eax |
||
112 | @@: |
||
113 | ; 4h. Loop #3: advance the counter and repeat while within the first group. |
||
114 | lea eax, [esi+ehci_controller.IntEDs+32*sizeof.ehci_static_ep-sizeof.ehci_controller] |
||
115 | add edx, [.loop3_delta] |
||
116 | cmp edx, eax |
||
117 | jb .calc_max_bandwidth |
||
118 | ; 4i. Loop #3 end: restore counter of loop #1. |
||
119 | pop edx |
||
120 | ; 4j. Loop #2: if the current variant is better (maybe not strictly) |
||
121 | ; then the previous optimum, update the optimal bandwidth and the target. |
||
122 | cmp edi, [.bandwidth] |
||
123 | ja @f |
||
3826 | clevermous | 124 | jb .update |
125 | cmp ecx, [.targetsmask] |
||
126 | jb @f |
||
127 | .update: |
||
3520 | clevermous | 128 | mov [.bandwidth], edi |
129 | mov [.target], edx |
||
3826 | clevermous | 130 | mov [.targetsmask], ecx |
3520 | clevermous | 131 | @@: |
132 | ; 4k. Loop #2: continue 8 times for every microframe. |
||
133 | inc ecx |
||
134 | cmp ecx, 8 |
||
135 | jb .varloop |
||
136 | ; 4l. Loop #1: advance counter and repeat ebx times, |
||
137 | ; ebx was calculated in step 3. |
||
138 | add edx, sizeof.ehci_static_ep |
||
139 | dec ebx |
||
140 | jnz .varloop0 |
||
3816 | clevermous | 141 | ; 5. Calculate bandwidth for the new pipe. |
3520 | clevermous | 142 | mov eax, [.maxpacket] |
3816 | clevermous | 143 | call calc_hs_bandwidth |
144 | mov ecx, [.maxpacket] |
||
3520 | clevermous | 145 | shr ecx, 11 |
146 | inc ecx |
||
147 | and ecx, 3 |
||
148 | imul eax, ecx |
||
3816 | clevermous | 149 | ; 6. Get the pointer to the best list. |
150 | pop edx ; restore value from step 3 |
||
151 | pop edx ; get delta calculated in step 3 |
||
152 | add edx, [.target] |
||
153 | ; 7. Check that bandwidth for the new pipe plus old bandwidth |
||
3520 | clevermous | 154 | ; still fits to maximum allowed by the core specification |
155 | ; current [.bandwidth] + new bandwidth <= limit; |
||
156 | ; USB2 specification allows maximum 60000*80% bit times for periodic microframe |
||
3816 | clevermous | 157 | mov ecx, [.bandwidth] |
158 | add ecx, eax |
||
159 | cmp ecx, 48000 |
||
160 | ja .no_bandwidth |
||
3520 | clevermous | 161 | ; 8. Convert {o|u}hci_static_ep to usb_static_ep, update bandwidth and return. |
162 | mov ecx, [.targetsmask] |
||
163 | add [edx+ehci_static_ep.Bandwidths+ecx*2], ax |
||
164 | add edx, ehci_static_ep.SoftwarePart |
||
3598 | clevermous | 165 | movi eax, 1 |
3520 | clevermous | 166 | shl eax, cl |
167 | pop edi ebx ; restore used registers to be stdcall |
||
168 | ret |
||
3816 | clevermous | 169 | .no_bandwidth: |
170 | dbgstr 'Periodic bandwidth limit reached' |
||
171 | xor eax, eax |
||
172 | xor edx, edx |
||
173 | pop edi ebx |
||
174 | ret |
||
3520 | clevermous | 175 | .every_frame: |
176 | ; The pipe should be scheduled every frame in two or more microframes. |
||
177 | ; 9. Calculate maximal bandwidth for every microframe: three nested loops. |
||
178 | ; 9a. The outermost loop: ebx = microframe to calculate. |
||
179 | xor ebx, ebx |
||
180 | .calc_all_bandwidths: |
||
181 | ; 9b. The intermediate loop: |
||
182 | ; edx = pointer to ehci_static_ep in the first group, [esp] = counter, |
||
183 | ; edi = maximal bandwidth |
||
184 | lea edx, [esi+ehci_controller.IntEDs-sizeof.ehci_controller] |
||
185 | xor edi, edi |
||
186 | push 32 |
||
187 | .calc_max_bandwidth2: |
||
188 | ; 9c. The innermost loop: calculate bandwidth for the given microframe |
||
189 | ; in the given frame. |
||
190 | xor eax, eax |
||
191 | push edx |
||
192 | .calc_bandwidth2: |
||
193 | add ax, [edx+ehci_static_ep.Bandwidths+ebx*2] |
||
194 | mov edx, [edx+ehci_static_ep.NextList] |
||
195 | test edx, edx |
||
196 | jnz .calc_bandwidth2 |
||
197 | pop edx |
||
198 | ; 9d. The intermediate loop continued: update maximal bandwidth. |
||
199 | cmp eax, edi |
||
200 | jb @f |
||
201 | mov edi, eax |
||
202 | @@: |
||
203 | add edx, sizeof.ehci_static_ep |
||
204 | dec dword [esp] |
||
205 | jnz .calc_max_bandwidth2 |
||
206 | pop eax |
||
207 | ; 9e. Push the calculated maximal bandwidth and continue the outermost loop. |
||
208 | push edi |
||
209 | inc ebx |
||
210 | cmp ebx, 8 |
||
211 | jb .calc_all_bandwidths |
||
212 | virtual at esp |
||
213 | .bandwidth7 dd ? |
||
214 | .bandwidth6 dd ? |
||
215 | .bandwidth5 dd ? |
||
216 | .bandwidth4 dd ? |
||
217 | .bandwidth3 dd ? |
||
218 | .bandwidth2 dd ? |
||
219 | .bandwidth1 dd ? |
||
220 | .bandwidth0 dd ? |
||
221 | end virtual |
||
222 | ; 10. Select the best variant. |
||
223 | ; edx = S-Mask = bitmask of scheduled microframes |
||
3598 | clevermous | 224 | movi edx, 0x11 |
3520 | clevermous | 225 | cmp ecx, 1 |
226 | ja @f |
||
227 | mov dl, 0x55 |
||
228 | jz @f |
||
229 | mov dl, 0xFF |
||
230 | @@: |
||
231 | ; try all variants edx, edx shl 1, edx shl 2, ... |
||
3826 | clevermous | 232 | ; while they fit in the lower byte (8 microframes per frame) |
3520 | clevermous | 233 | .select_best_mframe: |
234 | xor edi, edi |
||
235 | mov ecx, edx |
||
236 | mov eax, esp |
||
237 | .calc_mframe: |
||
238 | add cl, cl |
||
239 | jnc @f |
||
240 | cmp edi, [eax] |
||
241 | jae @f |
||
242 | mov edi, [eax] |
||
243 | @@: |
||
244 | add eax, 4 |
||
245 | test cl, cl |
||
246 | jnz .calc_mframe |
||
247 | cmp [.bandwidth], edi |
||
248 | jb @f |
||
249 | mov [.bandwidth], edi |
||
250 | mov [.targetsmask], edx |
||
251 | @@: |
||
252 | add dl, dl |
||
253 | jnc .select_best_mframe |
||
254 | ; 11. Restore stack after step 9. |
||
255 | add esp, 8*4 |
||
256 | ; 12. Get the pointer to the target list (responsible for every microframe). |
||
257 | lea edx, [esi+ehci_controller.IntEDs.SoftwarePart+62*sizeof.ehci_static_ep-sizeof.ehci_controller] |
||
3816 | clevermous | 258 | ; 13. Calculate bandwidth on the bus. |
3520 | clevermous | 259 | mov eax, [.maxpacket] |
3816 | clevermous | 260 | call calc_hs_bandwidth |
261 | mov ecx, [.maxpacket] |
||
3520 | clevermous | 262 | shr ecx, 11 |
263 | inc ecx |
||
264 | and ecx, 3 |
||
265 | imul eax, ecx |
||
3816 | clevermous | 266 | ; 14. Check that current [.bandwidth] + new bandwidth <= limit; |
3520 | clevermous | 267 | ; USB2 specification allows maximum 60000*80% bit times for periodic microframe. |
3816 | clevermous | 268 | mov ecx, [.bandwidth] |
269 | add ecx, eax |
||
270 | cmp ecx, 48000 |
||
271 | ja .no_bandwidth |
||
272 | ; 15. Update bandwidths including the new pipe. |
||
3520 | clevermous | 273 | mov ecx, [.targetsmask] |
274 | lea edi, [edx+ehci_static_ep.Bandwidths-ehci_static_ep.SoftwarePart] |
||
275 | .update_bandwidths: |
||
276 | shr ecx, 1 |
||
277 | jnc @f |
||
278 | add [edi], ax |
||
279 | @@: |
||
280 | add edi, 2 |
||
281 | test ecx, ecx |
||
282 | jnz .update_bandwidths |
||
3816 | clevermous | 283 | ; 16. Return target list and target S-Mask. |
3520 | clevermous | 284 | mov eax, [.targetsmask] |
285 | pop edi ebx ; restore used registers to be stdcall |
||
286 | ret |
||
287 | endp |
||
288 | |||
289 | ; Pipe is removing, update the corresponding lists. |
||
290 | ; We do not reorder anything, so just update book-keeping variable |
||
291 | ; in the list header. |
||
292 | proc ehci_hs_interrupt_list_unlink |
||
3653 | clevermous | 293 | movzx eax, word [ebx+ehci_pipe.Token-sizeof.ehci_pipe+2] |
3816 | clevermous | 294 | ; calculate bandwidth |
295 | call calc_hs_bandwidth |
||
3653 | clevermous | 296 | mov ecx, [ebx+ehci_pipe.Flags-sizeof.ehci_pipe] |
3520 | clevermous | 297 | shr ecx, 30 |
298 | imul eax, ecx |
||
3653 | clevermous | 299 | movzx ecx, byte [ebx+ehci_pipe.Flags-sizeof.ehci_pipe] |
3826 | clevermous | 300 | ; get target list |
4547 | clevermous | 301 | mov edx, [ebx+usb_pipe.BaseList] |
3520 | clevermous | 302 | ; update bandwidth |
303 | .dec_bandwidth: |
||
304 | shr ecx, 1 |
||
305 | jnc @f |
||
3826 | clevermous | 306 | sub word [edx+ehci_static_ep.Bandwidths - ehci_static_ep.SoftwarePart], ax |
3520 | clevermous | 307 | @@: |
308 | add edx, 2 |
||
309 | test ecx, ecx |
||
310 | jnz .dec_bandwidth |
||
311 | ; return |
||
312 | ret |
||
313 | endp |
||
314 | |||
3816 | clevermous | 315 | ; Helper procedure for USB2 scheduler: calculate bandwidth on the bus. |
316 | ; in: low 11 bits of eax = payload size in bytes |
||
317 | ; out: eax = maximal bandwidth in HS-bits |
||
318 | proc calc_hs_bandwidth |
||
319 | and eax, (1 shl 11) - 1 ; get payload for one transaction |
||
320 | add eax, 3 ; add 3 bytes for other fields in data packet, PID+CRC16 |
||
321 | ; Multiply by 8 for bytes -> bits and then by 7/6 to accomodate bit stuffing; |
||
322 | ; total 28/3 = 9+1/3 |
||
323 | mov edx, 55555556h |
||
324 | lea ecx, [eax*9] |
||
325 | mul edx |
||
326 | ; Add 989 extra bits: 68 bits for Token packet (32 for SYNC, 24 for token+address, |
||
327 | ; 4 extra bits for possible bit stuffing in token+address, 8 for EOP), |
||
328 | ; 736 bits for bus turn-around, 40 bits for SYNC+EOP in Data packet, |
||
329 | ; 8 bits for inter-packet delay, 49 bits for Handshake packet, |
||
330 | ; 88 bits for another inter-packet delay. |
||
331 | lea eax, [ecx+edx+989] |
||
332 | ret |
||
333 | endp |
||
334 | |||
3826 | clevermous | 335 | ; Split-transaction scheduler (aka TT scheduler, TT stands for Transaction |
336 | ; Translator, section 11.14 of the core spec) needs to schedule three event |
||
337 | ; types on two buses: Start-Split and Complete-Split on HS bus and normal |
||
338 | ; transaction on FS/LS bus. |
||
339 | ; Assume that FS/LS bus is more restricted and more important to be scheduled |
||
340 | ; uniformly, so select the variant which minimizes maximal used bandwidth |
||
341 | ; on FS/LS bus and does not overflow HS bus. |
||
342 | ; If there are several such variants, prefer variants which is closest to |
||
343 | ; start of frame, and within the same microframe consider HS bandwidth |
||
344 | ; utilization as a last criteria. |
||
3520 | clevermous | 345 | |
3826 | clevermous | 346 | ; The procedure ehci_select_tt_interrupt_list has been splitted into several |
347 | ; macro, each representing a logical step of the procedure, |
||
348 | ; to simplify understanding what is going on. Consider all the following macro |
||
349 | ; as logical parts of one procedure, they are meaningless outside the context. |
||
350 | |||
351 | ; Given a frame, calculate bandwidth occupied by already opened pipes |
||
352 | ; in every microframe. |
||
353 | ; Look for both HS and FS/LS buses: there are 16 words of information, |
||
354 | ; 8 for HS bus, 8 for FS/LS bus, for every microframe. |
||
355 | ; Since we count already opened pipes, the total bandwidth in every microframe |
||
356 | ; is less than 60000 bits (and even 60000*80% bits), otherwise the scheduler |
||
357 | ; would not allow to open those pipes. |
||
358 | ; edi -> first list for the frame |
||
359 | macro tt_calc_bandwidth_in_frame |
||
360 | { |
||
361 | local .lists, .pipes, .pipes_done, .carry |
||
362 | ; 1. Zero everything. |
||
363 | xor eax, eax |
||
364 | mov edx, edi |
||
365 | repeat 4 |
||
366 | mov dword [.budget+(%-1)*4], eax |
||
367 | end repeat |
||
368 | repeat 4 |
||
369 | mov dword [.hs_bandwidth+(%-1)*4], eax |
||
370 | end repeat |
||
371 | mov [.total_budget], ax |
||
372 | ; Loop over all lists for the given frame. |
||
373 | .lists: |
||
374 | ; 2. Total HS bandwidth for all pipes in one list is kept inside list header, |
||
375 | ; add it. Note that overflow is impossible, so we may add entire dwords. |
||
376 | mov ebx, [edx+ehci_static_ep.SoftwarePart+usb_static_ep.NextVirt] |
||
377 | repeat 4 |
||
378 | mov eax, dword [edx+ehci_static_ep.Bandwidths+(%-1)*4] |
||
379 | add dword [.hs_bandwidth+(%-1)*4], eax |
||
380 | end repeat |
||
381 | ; Loop over all pipes in the given list. |
||
382 | add edx, ehci_static_ep.SoftwarePart |
||
383 | .pipes: |
||
384 | cmp ebx, edx |
||
385 | jz .pipes_done |
||
386 | ; 3. For every pipe in every list for the given frame: |
||
387 | ; 3a. Check whether the pipe resides on the same FS/LS bus as the new pipe. |
||
388 | ; If not, skip this pipe. |
||
389 | mov eax, [ebx+usb_pipe.DeviceData] |
||
390 | mov eax, [eax+usb_device_data.TTHub] |
||
391 | cmp eax, [.tthub] |
||
392 | jnz @f |
||
393 | ; 3b. Calculate FS/LS budget for the opened pipe. |
||
394 | ; Note that eax = TTHub after 3a. |
||
395 | call tt_calc_budget |
||
396 | ; 3c. Update total budget: add the value from 3b |
||
397 | ; to the budget of the first microframe scheduled for this pipe. |
||
398 | bsf ecx, [ebx+ehci_pipe.Flags-sizeof.ehci_pipe] |
||
399 | add [.budget+ecx*2], ax |
||
400 | @@: |
||
401 | mov ebx, [ebx+usb_pipe.NextVirt] |
||
402 | jmp .pipes |
||
403 | .pipes_done: |
||
404 | mov edx, [edx+ehci_static_ep.NextList-ehci_static_ep.SoftwarePart] |
||
405 | test edx, edx |
||
406 | jnz .lists |
||
407 | ; 4. If the budget for some microframe is exceeded, carry it to the following |
||
408 | ; microframe(s). The actual size of one microframe is 187.5 raw bytes; |
||
409 | ; the core spec says that 188 bytes should be scheduled in every microframe. |
||
410 | xor eax, eax |
||
411 | xor ecx, ecx |
||
412 | .carry: |
||
413 | xor edx, edx |
||
414 | add ax, [.budget+ecx*2] |
||
415 | cmp ax, 188 |
||
416 | jbe @f |
||
417 | mov dx, ax |
||
418 | mov ax, 188 |
||
419 | sub dx, ax |
||
420 | @@: |
||
421 | mov [.budget+ecx*2], ax |
||
422 | add [.total_budget], ax |
||
423 | mov ax, dx |
||
424 | inc ecx |
||
425 | cmp ecx, 8 |
||
426 | jb .carry |
||
427 | } |
||
428 | |||
429 | ; Checks whether the new pipe fits in the existing FS budget |
||
430 | ; starting from the given microframe. If not, mark the microframe |
||
431 | ; as impossible for scheduling. |
||
432 | ; in: ecx = microframe |
||
433 | macro tt_exclude_microframe_if_no_budget |
||
434 | { |
||
435 | local .loop, .good, .bad |
||
436 | ; 1. If the new budget plus the current budget does not exceed 188 bytes, |
||
437 | ; the variant is possible. |
||
438 | mov ax, [.budget+ecx*2] |
||
439 | mov edx, ecx |
||
440 | add ax, [.new_budget] |
||
441 | sub ax, 188 |
||
442 | jbe .good |
||
443 | ; 2. Otherwise, |
||
444 | ; a) nothing should be scheduled in some following microframes, |
||
445 | ; b) after adding the new budget everything should fit in first 6 microframes, |
||
446 | ; this guarantees that even in the worst case 90% limit is satisfied. |
||
447 | .loop: |
||
448 | cmp edx, 5 |
||
449 | jae .bad |
||
450 | cmp [.budget+(edx+1)*2], 0 |
||
451 | jnz .bad |
||
452 | inc edx |
||
453 | sub ax, 188 |
||
454 | ja .loop |
||
455 | .bad: |
||
456 | btr [.possible_microframes], ecx |
||
457 | .good: |
||
458 | } |
||
459 | |||
460 | ; Calculate data corresponding to the particular scheduling variant for the new pipe. |
||
461 | ; Data describe the current scheduling state collected over all frames touched |
||
462 | ; by the given variant: maximal HS bandwidth, maximal FS/LS budget, |
||
463 | ; which microframes fit in the current FS/LS budget for all frames. |
||
464 | macro tt_calc_statistics_for_one_variant |
||
465 | { |
||
466 | local .frames, .microframes |
||
467 | ; 1. Initialize: zero maximal bandwidth, |
||
468 | ; first 6 microframes are possible for scheduling. |
||
469 | xor eax, eax |
||
470 | repeat 4 |
||
471 | mov dword [.max_hs_bandwidth+(%-1)*4], eax |
||
472 | end repeat |
||
473 | mov [.max_fs_bandwidth], ax |
||
474 | mov [.possible_microframes], 0x3F |
||
475 | ; Loop over all frames starting with [.variant] advancing by [.variant_delta]. |
||
476 | mov edi, [.variant] |
||
477 | .frames: |
||
478 | ; 2. Calculate statistics for one frame. |
||
479 | tt_calc_bandwidth_in_frame |
||
480 | ; 3. Update maximal FS budget. |
||
481 | mov ax, [.total_budget] |
||
482 | cmp ax, [.max_fs_bandwidth] |
||
483 | jb @f |
||
484 | mov [.max_fs_bandwidth], ax |
||
485 | @@: |
||
486 | ; 4. For every microframe, update maximal HS bandwidth |
||
487 | ; and check whether the microframe is allowed for scheduling. |
||
488 | xor ecx, ecx |
||
489 | .microframes: |
||
490 | mov ax, [.hs_bandwidth+ecx*2] |
||
491 | cmp ax, [.max_hs_bandwidth+ecx*2] |
||
492 | jb @f |
||
493 | mov [.max_hs_bandwidth+ecx*2], ax |
||
494 | @@: |
||
495 | tt_exclude_microframe_if_no_budget |
||
496 | inc ecx |
||
497 | cmp ecx, 8 |
||
498 | jb .microframes |
||
499 | ; Stop loop when outside of first descriptor group. |
||
500 | lea eax, [esi+ehci_controller.IntEDs+32*sizeof.ehci_static_ep-sizeof.ehci_controller] |
||
501 | add edi, [.variant_delta] |
||
502 | cmp edi, eax |
||
503 | jb .frames |
||
504 | } |
||
505 | |||
506 | struct usb_split_info |
||
507 | microframe_mask dd ? ; lower byte is S-mask, second byte is C-mask |
||
508 | ssplit_bandwidth dd ? |
||
509 | csplit_bandwidth dd ? |
||
510 | ends |
||
511 | |||
512 | ; Check whether the current variant and the current microframe are allowed |
||
513 | ; for scheduling. If so, check whether they are better than the previously |
||
514 | ; selected variant+microframe, if any. If so, update the previously selected |
||
515 | ; variant+microframe to current ones. |
||
516 | ; ecx = microframe, [.variant] = variant |
||
517 | macro tt_check_variant_microframe |
||
518 | { |
||
519 | local .nothing, .update, .ssplit, .csplit, .csplit_done |
||
520 | ; 1. If the current microframe does not fit in existing FS budget, do nothing. |
||
521 | bt [.possible_microframes], ecx |
||
522 | jnc .nothing |
||
523 | ; 2. Calculate maximal HS bandwidth over all affected microframes. |
||
524 | ; 2a. Start-split phase: one or more microframes starting with ecx, |
||
525 | ; coded in lower byte of .info.microframe_mask. |
||
526 | xor ebx, ebx |
||
527 | xor edx, edx |
||
528 | .ssplit: |
||
529 | lea eax, [ecx+edx] |
||
530 | movzx eax, [.max_hs_bandwidth+eax*2] |
||
531 | add eax, [.info.ssplit_bandwidth] |
||
532 | cmp ebx, eax |
||
533 | ja @f |
||
534 | mov ebx, eax |
||
535 | @@: |
||
536 | inc edx |
||
537 | bt [.info.microframe_mask], edx |
||
538 | jc .ssplit |
||
539 | ; 2b. Complete-split phase: zero or more microframes starting with |
||
540 | ; ecx+(last start-split microframe)+2, |
||
541 | ; coded in second byte of .info.microframe_mask. |
||
542 | add edx, 8 |
||
543 | .csplit: |
||
544 | inc edx |
||
545 | bt [.info.microframe_mask], edx |
||
546 | jnc .csplit_done |
||
547 | lea eax, [ecx+edx] |
||
548 | cmp eax, 8 |
||
549 | jae .csplit_done |
||
550 | movzx eax, [.max_hs_bandwidth+(eax-8)*2] |
||
551 | add eax, [.info.csplit_bandwidth] |
||
552 | cmp ebx, eax |
||
553 | ja .csplit |
||
554 | mov ebx, eax |
||
555 | jmp .csplit |
||
556 | .csplit_done: |
||
557 | ; 3. Check that current HS bandwidth + new bandwidth <= limit; |
||
558 | ; USB2 specification allows maximum 60000*80% bit times for periodic microframe. |
||
559 | cmp ebx, 48000 |
||
560 | ja .nothing |
||
561 | ; 4. This variant is possible for scheduling. |
||
562 | ; Check whether it is better than the currently selected one. |
||
563 | ; 4a. The primary criteria: FS/LS bandwidth. |
||
564 | mov ax, [.max_fs_bandwidth] |
||
565 | cmp ax, [.best_fs_bandwidth] |
||
566 | ja .nothing |
||
567 | jb .update |
||
568 | ; 4b. The secondary criteria: prefer microframes which are closer to start of frame. |
||
569 | cmp ecx, [.targetsmask] |
||
570 | ja .nothing |
||
571 | jb .update |
||
572 | ; 4c. The last criteria: HS bandwidth. |
||
573 | cmp ebx, [.bandwidth] |
||
574 | ja .nothing |
||
575 | .update: |
||
576 | ; 5. This variant is better than the previously selected. |
||
577 | ; Update the best variant with current data. |
||
578 | mov [.best_fs_bandwidth], ax |
||
579 | mov [.bandwidth], ebx |
||
580 | mov [.targetsmask], ecx |
||
581 | mov eax, [.variant] |
||
582 | mov [.target], eax |
||
583 | .nothing: |
||
584 | } |
||
585 | |||
586 | ; TT scheduler: add new pipe. |
||
587 | ; in: esi -> usb_controller, edi -> usb_pipe |
||
588 | ; out: edx -> usb_static_ep, eax = S-Mask |
||
589 | proc ehci_select_tt_interrupt_list |
||
590 | virtual at ebp-12-.local_vars_size |
||
591 | .local_vars_start: |
||
592 | .info usb_split_info |
||
593 | .new_budget dw ? |
||
594 | .total_budget dw ? |
||
595 | .possible_microframes dd ? |
||
596 | .tthub dd ? |
||
597 | .budget rw 8 |
||
598 | .hs_bandwidth rw 8 |
||
599 | .max_hs_bandwidth rw 8 |
||
600 | .max_fs_bandwidth dw ? |
||
601 | .best_fs_bandwidth dw ? |
||
602 | .variant dd ? |
||
603 | .variant_delta dd ? |
||
604 | .target_delta dd ? |
||
605 | .local_vars_size = $ - .local_vars_start |
||
4418 | clevermous | 606 | if .local_vars_size > 24*4 |
607 | err Modify stack frame size in |
||
608 | end if |
||
3826 | clevermous | 609 | |
3520 | clevermous | 610 | .targetsmask dd ? |
611 | .bandwidth dd ? |
||
612 | .target dd ? |
||
613 | dd ? |
||
614 | dd ? |
||
615 | .config_pipe dd ? |
||
616 | .endpoint dd ? |
||
617 | .maxpacket dd ? |
||
618 | .type dd ? |
||
619 | .interval dd ? |
||
620 | end virtual |
||
3826 | clevermous | 621 | mov eax, [edi+ehci_pipe.Token-sizeof.ehci_pipe] |
622 | shr eax, 16 |
||
623 | and eax, (1 shl 11) - 1 |
||
624 | push ebx edi |
||
625 | ; 1. Compute the real interval. FS/LS devices encode the interval as |
||
626 | ; number of milliseconds. Use the maximal power of two that is not greater than |
||
627 | ; the given interval and EHCI scheduling area = 32 frames. |
||
3520 | clevermous | 628 | cmp [.interval], 1 |
629 | adc [.interval], 0 |
||
630 | mov ecx, 64 |
||
3826 | clevermous | 631 | mov eax, 64 * sizeof.ehci_static_ep |
3520 | clevermous | 632 | @@: |
633 | shr ecx, 1 |
||
634 | cmp [.interval], ecx |
||
635 | jb @b |
||
3826 | clevermous | 636 | mov [.interval], ecx |
637 | ; 2. Compute variables for further calculations. |
||
638 | ; 2a. [.variant_delta] is delta between two lists from the first group |
||
639 | ; that correspond to the same variant. |
||
640 | imul ecx, sizeof.ehci_static_ep |
||
641 | mov [.variant_delta], ecx |
||
642 | ; 2b. [.target_delta] is delta between the final answer from the group |
||
643 | ; corresponding to [.interval] and the item from the first group. |
||
3520 | clevermous | 644 | sub eax, ecx |
645 | sub eax, ecx |
||
3826 | clevermous | 646 | mov [.target_delta], eax |
647 | ; 2c. [.variant] is the first list from the first group that corresponds |
||
648 | ; to the current variant. |
||
649 | lea eax, [esi+ehci_controller.IntEDs-sizeof.ehci_controller] |
||
650 | mov [.variant], eax |
||
651 | ; 2d. [.tthub] identifies TT hub for new pipe, [.new_budget] is FS budget |
||
652 | ; for new pipe. |
||
653 | mov eax, [edi+usb_pipe.DeviceData] |
||
654 | mov eax, [eax+usb_device_data.TTHub] |
||
655 | mov ebx, edi |
||
656 | mov [.tthub], eax |
||
657 | call tt_calc_budget |
||
658 | mov [.new_budget], ax |
||
659 | ; 2e. [.usb_split_info] describes bandwidth used by new pipe on HS bus. |
||
660 | lea edi, [.info] |
||
661 | call tt_fill_split_info |
||
662 | test eax, eax |
||
663 | jz .no_bandwidth |
||
664 | ; 2f. There is no best variant yet, put maximal possible values, |
||
665 | ; so any variant would be better than the "current". |
||
666 | or [.best_fs_bandwidth], -1 |
||
667 | or [.target], -1 |
||
668 | or [.bandwidth], -1 |
||
669 | or [.targetsmask], -1 |
||
670 | ; 3. Loop over all variants, for every variant decide whether it is acceptable, |
||
671 | ; select the best variant from all acceptable variants. |
||
672 | .check_variants: |
||
673 | tt_calc_statistics_for_one_variant |
||
674 | xor ecx, ecx |
||
675 | .check_microframes: |
||
676 | tt_check_variant_microframe |
||
677 | inc ecx |
||
678 | cmp ecx, 6 |
||
679 | jb .check_microframes |
||
680 | add [.variant], sizeof.ehci_static_ep |
||
681 | dec [.interval] |
||
682 | jnz .check_variants |
||
683 | ; 4. If there is no acceptable variants, return error. |
||
684 | mov ecx, [.targetsmask] |
||
685 | mov edx, [.target] |
||
686 | cmp ecx, -1 |
||
687 | jz .no_bandwidth |
||
688 | ; 5. Calculate the answer: edx -> selected list, eax = S-Mask and C-Mask. |
||
689 | mov eax, [.info.microframe_mask] |
||
690 | add edx, [.target_delta] |
||
691 | shl eax, cl |
||
692 | and eax, 0xFFFF |
||
693 | ; 6. Update HS bandwidths in the selected list. |
||
694 | xor ecx, ecx |
||
695 | mov ebx, [.info.ssplit_bandwidth] |
||
696 | .update_ssplit: |
||
697 | bt eax, ecx |
||
698 | jnc @f |
||
699 | add [edx+ehci_static_ep.Bandwidths+ecx*2], bx |
||
700 | @@: |
||
701 | inc ecx |
||
702 | cmp ecx, 8 |
||
703 | jb .update_ssplit |
||
704 | mov ebx, [.info.csplit_bandwidth] |
||
705 | .update_csplit: |
||
706 | bt eax, ecx |
||
707 | jnc @f |
||
708 | add [edx+ehci_static_ep.Bandwidths+(ecx-8)*2], bx |
||
709 | @@: |
||
710 | inc ecx |
||
711 | cmp ecx, 16 |
||
712 | jb .update_csplit |
||
713 | ; 7. Return. |
||
714 | add edx, ehci_static_ep.SoftwarePart |
||
715 | pop edi ebx |
||
3520 | clevermous | 716 | ret |
3826 | clevermous | 717 | .no_bandwidth: |
718 | dbgstr 'Periodic bandwidth limit reached' |
||
719 | xor eax, eax |
||
720 | xor edx, edx |
||
721 | pop edi ebx |
||
722 | ret |
||
3520 | clevermous | 723 | endp |
724 | |||
3826 | clevermous | 725 | ; Pipe is removing, update the corresponding lists. |
726 | ; We do not reorder anything, so just update book-keeping variable |
||
727 | ; in the list header. |
||
3520 | clevermous | 728 | proc ehci_fs_interrupt_list_unlink |
3826 | clevermous | 729 | ; calculate bandwidth |
730 | push edi |
||
731 | sub esp, sizeof.usb_split_info |
||
732 | mov edi, esp |
||
733 | call tt_fill_split_info |
||
734 | ; get target list |
||
4547 | clevermous | 735 | mov edx, [ebx+usb_pipe.BaseList] |
3826 | clevermous | 736 | ; update bandwidth for Start-Split |
737 | mov eax, [edi+usb_split_info.ssplit_bandwidth] |
||
738 | xor ecx, ecx |
||
739 | .dec_bandwidth_1: |
||
740 | bt [ebx+ehci_pipe.Flags-sizeof.ehci_pipe], ecx |
||
741 | jnc @f |
||
742 | sub word [edx+ecx*2+ehci_static_ep.Bandwidths - ehci_static_ep.SoftwarePart], ax |
||
743 | @@: |
||
744 | inc ecx |
||
745 | cmp ecx, 8 |
||
746 | jb .dec_bandwidth_1 |
||
747 | ; update bandwidth for Complete-Split |
||
748 | mov eax, [edi+usb_split_info.csplit_bandwidth] |
||
749 | .dec_bandwidth_2: |
||
750 | bt [ebx+ehci_pipe.Flags-sizeof.ehci_pipe], ecx |
||
751 | jnc @f |
||
752 | sub word [edx+(ecx-8)*2+ehci_static_ep.Bandwidths - ehci_static_ep.SoftwarePart], ax |
||
753 | @@: |
||
754 | inc ecx |
||
755 | cmp ecx, 16 |
||
756 | jb .dec_bandwidth_2 |
||
757 | add esp, sizeof.usb_split_info |
||
758 | pop edi |
||
3520 | clevermous | 759 | ret |
760 | endp |
||
3826 | clevermous | 761 | |
762 | ; Helper procedure for ehci_select_tt_interrupt_list. |
||
763 | ; Calculates "best-case budget" according to the core spec, |
||
764 | ; that is, number of bytes (not bits) corresponding to "optimistic" transaction |
||
765 | ; time, including inter-packet delays/bus turn-around time, |
||
766 | ; but without bit stuffing and timers drift. |
||
767 | ; One extra TT-specific delay is added: TT think time from the hub descriptor. |
||
768 | ; Similar to calc_usb1_bandwidth with corresponding changes. |
||
769 | ; eax -> usb_hub with TT, ebx -> usb_pipe |
||
770 | proc tt_calc_budget |
||
4418 | clevermous | 771 | invoke usbhc_api.usb_get_tt_think_time ; ecx = TT think time in FS-bytes |
3826 | clevermous | 772 | mov eax, [ebx+ehci_pipe.Token-sizeof.ehci_pipe] |
773 | shr eax, 16 |
||
774 | and eax, (1 shl 11) - 1 ; get data length |
||
775 | bt [ebx+ehci_pipe.Token-sizeof.ehci_pipe], 12 |
||
776 | jc .low_speed |
||
777 | ; Full-speed interrupt IN/OUT: |
||
778 | ; 33 bits for Token packet (8 for SYNC, 24 for token+address, 3 for EOP), |
||
779 | ; 18 bits for bus turn-around, 11 bits for SYNC+EOP in Data packet, |
||
780 | ; 2 bits for inter-packet delay, 19 bits for Handshake packet, |
||
781 | ; 2 bits for another inter-packet delay. 85 bits total, pad to 11 bytes. |
||
4418 | clevermous | 782 | lea eax, [eax+11+ecx] |
3826 | clevermous | 783 | ; 1 byte is minimal TT think time in addition to ecx. |
784 | ret |
||
785 | .low_speed: |
||
786 | ; Low-speed interrupt IN/OUT: |
||
787 | ; multiply by 8 for LS -> FS, |
||
788 | ; add 85 bytes as in full-speed interrupt and extra 5 bytes for two PRE packets |
||
789 | ; and two hub delays. |
||
790 | ; 1 byte is minimal TT think time in addition to ecx. |
||
4418 | clevermous | 791 | lea eax, [eax*8+90+ecx] |
3826 | clevermous | 792 | ret |
793 | endp |
||
794 | |||
795 | ; Helper procedure for TT scheduler. |
||
796 | ; Calculates Start-Split/Complete-Split masks and HS bandwidths. |
||
797 | ; ebx -> usb_pipe, edi -> usb_split_info |
||
798 | proc tt_fill_split_info |
||
799 | ; Interrupt endpoints. |
||
800 | ; The core spec says in 5.7.3 "Interrupt Transfer Packet Size Constraints" that: |
||
801 | ; The maximum allowable interrupt data payload size is 64 bytes or less for full-speed. |
||
802 | ; Low-speed devices are limited to eight bytes or less maximum data payload size. |
||
803 | ; This is important for scheduling, it guarantees that in any case transaction fits |
||
804 | ; in two microframes (usually one, two if transaction has started too late in the first |
||
805 | ; microframe), so check it. |
||
806 | mov eax, [ebx+ehci_pipe.Token-sizeof.ehci_pipe] |
||
807 | mov ecx, 8 |
||
808 | bt eax, 12 |
||
809 | jc @f |
||
810 | mov ecx, 64 |
||
811 | @@: |
||
812 | shr eax, 16 |
||
813 | and eax, (1 shl 11) - 1 ; get data length |
||
814 | cmp eax, ecx |
||
815 | ja .error |
||
816 | add eax, 3 ; add 3 bytes for other fields in data packet, PID+CRC16 |
||
817 | ; Multiply by 8 for bytes -> bits and then by 7/6 to accomodate bit stuffing; |
||
818 | ; total 28/3 = 9+1/3 |
||
819 | mov edx, 55555556h |
||
820 | lea ecx, [eax*9] |
||
821 | mul edx |
||
822 | ; One start-split, three complete-splits (unless the last is too far, |
||
823 | ; but this is handled by the caller). |
||
824 | mov eax, [ebx+usb_pipe.LastTD] |
||
825 | mov [edi+usb_split_info.microframe_mask], 0x1C01 |
||
826 | ; Structure and HS bandwidth of packets depends on the direction. |
||
827 | bt [eax+ehci_gtd.Token-sizeof.ehci_gtd], 8 |
||
828 | jc .interrupt_in |
||
829 | .interrupt_out: |
||
830 | ; Start-Split phase: |
||
831 | ; 77 bits for SPLIT packet (32 for SYNC, 8 for EOP, 32 for data, 5 for bit stuffing), |
||
832 | ; 88 bits for inter-packet delay, 68 bits for Token packet, |
||
833 | ; 88 bits for inter-packet delay, 40 bits for SYNC+EOP in Data packet, |
||
834 | ; 88 bits for last inter-packet delay, total 449 bits. |
||
835 | lea eax, [edx+ecx+449] |
||
836 | mov [edi+usb_split_info.ssplit_bandwidth], eax |
||
837 | ; Complete-Split phase: |
||
838 | ; 77 bits for SPLIT packet, |
||
839 | ; 88 bits for inter-packet delay, 68 bits for Token packet, |
||
840 | ; 736 bits for bus turn-around, 49 bits for Handshake packet, |
||
841 | ; 8 bits for inter-packet delay, total 1026 bits. |
||
842 | mov [edi+usb_split_info.csplit_bandwidth], 1026 |
||
843 | ret |
||
844 | .interrupt_in: |
||
845 | ; Start-Split phase: |
||
846 | ; 77 bits for SPLIT packet, 88 bits for inter-packet delay, |
||
847 | ; 68 bits for Token packet, 88 bits for another inter-packet delay, |
||
848 | ; total 321 bits. |
||
849 | mov [edi+usb_split_info.ssplit_bandwidth], 321 |
||
850 | ; Complete-Split phase: |
||
851 | ; 77 bits for SPLIT packet, 88 bits for inter-packet delay, |
||
852 | ; 68 bits for Token packet, 736 bits for bus turn-around, |
||
853 | ; 40 bits for SYNC+EOP in Data packet, 8 bits for inter-packet delay, |
||
854 | ; total 1017 bits. |
||
855 | lea eax, [edx+ecx+1017] |
||
856 | mov [edi+usb_split_info.csplit_bandwidth], eax |
||
857 | ret |
||
858 | .error: |
||
859 | xor eax, eax |
||
860 | ret |
||
861 | endp=>=>=> |