All of lore.kernel.org
 help / color / mirror / Atom feed
From: Andrew Morton <akpm@osdl.org>
To: James Nelson <james4765@verizon.net>
Cc: kernel-janitors@lists.osdl.org, linux-kernel@vger.kernel.org
Subject: [KJ] Re: [PATCH] esp: Make driver SMP-correct
Date: Fri, 31 Dec 2004 09:46:11 +0000	[thread overview]
Message-ID: <20041231014611.003281e5.akpm@osdl.org> (raw)
In-Reply-To: <20041231014403.3309.58245.96163@localhost.localdomain>

[-- Attachment #1: Type: text/plain, Size: 1875 bytes --]

James Nelson <james4765@verizon.net> wrote:
>
> This is an attempt to make the esp serial driver SMP-correct.  It also removes
>  some cruft left over from the serial_write() conversion.

From a quick scan:

- startup() does multiple sleeping allocations and request_irq() under
  spin_lock_irqsave().  Maybe fixed by this:

--- 25/drivers/char/esp.c~esp-make-driver-smp-correct-fixes	2004-12-31 01:40:57.987232152 -0800
+++ 25-akpm/drivers/char/esp.c	2004-12-31 01:42:21.444544712 -0800
@@ -851,16 +851,14 @@ static int startup(struct esp_struct * i
 	int	retval=0;
         unsigned int num_chars;
 
-	spin_lock_irqsave(&info->irq_lock, flags);
-
 	if (info->flags & ASYNC_INITIALIZED)
-		goto out;
+		goto out_unlocked;
 
 	if (!info->xmit_buf) {
 		info->xmit_buf = (unsigned char *)get_zeroed_page(GFP_KERNEL);
 		retval = -ENOMEM;
 		if (!info->xmit_buf)
-			goto out;
+			goto out_unlocked;
 	}
 
 #ifdef SERIAL_DEBUG_OPEN
@@ -907,7 +905,7 @@ static int startup(struct esp_struct * i
 					&info->tty->flags);
 			retval = 0;
 		}
-		goto out;
+		goto out_unocked;
 	}
 
 	if (!(info->stat_flags & ESP_STAT_USE_PIO) && !dma_buffer) {
@@ -926,6 +924,8 @@ static int startup(struct esp_struct * i
 			
 	}
 
+	spin_lock_irqsave(&info->irq_lock, flags);
+
 	info->MCR = UART_MCR_DTR | UART_MCR_RTS | UART_MCR_OUT2;
 	serial_out(info, UART_ESI_CMD1, ESI_WRITE_UART);
 	serial_out(info, UART_ESI_CMD2, UART_MCR);
@@ -965,7 +965,9 @@ static int startup(struct esp_struct * i
 
 	info->flags |= ASYNC_INITIALIZED;
 	retval = 0;
-out:	spin_unlock_irqrestore(&info->irq_lock, flags);
+out:
+	spin_unlock_irqrestore(&info->irq_lock, flags);
+out_unocked:
 	return retval;
 }
 
_


- startup() calls change_speed() under info->irq_lock, but change_speed()
  also takes info->irq_lock.  Instant deadlock on driver initialisation.

The driver needs more serious surgery than this.

[-- Attachment #2: Type: text/plain, Size: 167 bytes --]

_______________________________________________
Kernel-janitors mailing list
Kernel-janitors@lists.osdl.org
http://lists.osdl.org/mailman/listinfo/kernel-janitors

WARNING: multiple messages have this Message-ID (diff)
From: Andrew Morton <akpm@osdl.org>
To: James Nelson <james4765@verizon.net>
Cc: kernel-janitors@lists.osdl.org, linux-kernel@vger.kernel.org,
	james4765@verizon.net
Subject: Re: [PATCH] esp: Make driver SMP-correct
Date: Fri, 31 Dec 2004 01:46:11 -0800	[thread overview]
Message-ID: <20041231014611.003281e5.akpm@osdl.org> (raw)
In-Reply-To: <20041231014403.3309.58245.96163@localhost.localdomain>

James Nelson <james4765@verizon.net> wrote:
>
> This is an attempt to make the esp serial driver SMP-correct.  It also removes
>  some cruft left over from the serial_write() conversion.

>From a quick scan:

- startup() does multiple sleeping allocations and request_irq() under
  spin_lock_irqsave().  Maybe fixed by this:

--- 25/drivers/char/esp.c~esp-make-driver-smp-correct-fixes	2004-12-31 01:40:57.987232152 -0800
+++ 25-akpm/drivers/char/esp.c	2004-12-31 01:42:21.444544712 -0800
@@ -851,16 +851,14 @@ static int startup(struct esp_struct * i
 	int	retval=0;
         unsigned int num_chars;
 
-	spin_lock_irqsave(&info->irq_lock, flags);
-
 	if (info->flags & ASYNC_INITIALIZED)
-		goto out;
+		goto out_unlocked;
 
 	if (!info->xmit_buf) {
 		info->xmit_buf = (unsigned char *)get_zeroed_page(GFP_KERNEL);
 		retval = -ENOMEM;
 		if (!info->xmit_buf)
-			goto out;
+			goto out_unlocked;
 	}
 
 #ifdef SERIAL_DEBUG_OPEN
@@ -907,7 +905,7 @@ static int startup(struct esp_struct * i
 					&info->tty->flags);
 			retval = 0;
 		}
-		goto out;
+		goto out_unocked;
 	}
 
 	if (!(info->stat_flags & ESP_STAT_USE_PIO) && !dma_buffer) {
@@ -926,6 +924,8 @@ static int startup(struct esp_struct * i
 			
 	}
 
+	spin_lock_irqsave(&info->irq_lock, flags);
+
 	info->MCR = UART_MCR_DTR | UART_MCR_RTS | UART_MCR_OUT2;
 	serial_out(info, UART_ESI_CMD1, ESI_WRITE_UART);
 	serial_out(info, UART_ESI_CMD2, UART_MCR);
@@ -965,7 +965,9 @@ static int startup(struct esp_struct * i
 
 	info->flags |= ASYNC_INITIALIZED;
 	retval = 0;
-out:	spin_unlock_irqrestore(&info->irq_lock, flags);
+out:
+	spin_unlock_irqrestore(&info->irq_lock, flags);
+out_unocked:
 	return retval;
 }
 
_


- startup() calls change_speed() under info->irq_lock, but change_speed()
  also takes info->irq_lock.  Instant deadlock on driver initialisation.

The driver needs more serious surgery than this.

  parent reply	other threads:[~2004-12-31  9:46 UTC|newest]

Thread overview: 30+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2004-12-31  1:43 [KJ] [PATCH] esp: Make driver SMP-correct James Nelson
2004-12-31  1:43 ` James Nelson
2004-12-31  9:45 ` [KJ] " Jan Dittmer
2004-12-31  9:45   ` Jan Dittmer
2004-12-31  9:46 ` Andrew Morton [this message]
2004-12-31  9:46   ` Andrew Morton
2004-12-31 10:00   ` [KJ] " Russell King
2004-12-31 10:00     ` Russell King
2004-12-31 11:00     ` [KJ] " Jim Nelson
2004-12-31 11:00       ` Jim Nelson
2004-12-31 17:01     ` [KJ] " Russell King
2004-12-31 17:01       ` Russell King
2004-12-31 22:16       ` [KJ] " Jim Nelson
2004-12-31 22:16         ` Jim Nelson
2005-01-01  0:01         ` Gene Heskett
2005-01-01  0:01           ` Gene Heskett
2005-01-01  0:13           ` Russell King
2005-01-01  0:13             ` Russell King
2005-01-01  1:14             ` Gene Heskett
2005-01-01  1:14               ` Gene Heskett
2005-01-01  1:33               ` Russell King
2005-01-01  1:33                 ` Russell King
2005-01-01  1:49                 ` Jim Nelson
2005-01-01  1:49                   ` Jim Nelson
2005-01-01  1:35               ` Gene Heskett
2005-01-01  1:35                 ` Gene Heskett
2005-01-01  1:47                 ` Jim Nelson
2005-01-01  1:47                   ` Jim Nelson
2005-01-01  3:11                   ` Gene Heskett
2005-01-01  3:11                     ` Gene Heskett

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20041231014611.003281e5.akpm@osdl.org \
    --to=akpm@osdl.org \
    --cc=james4765@verizon.net \
    --cc=kernel-janitors@lists.osdl.org \
    --cc=linux-kernel@vger.kernel.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body. 
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.